30th Nov - 1 st Dec, Escuela Superior de Ingenieros de la Universidad de Sevilla (Spain)

Proceedings

Editors
A. Ollero K. Valavanis
A. Viguria G. Heredia
I. Maza M. Béjar
A. Rodrı́guez J. Gonzalo
.
 Foreword

The First Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-
UAS 2011) has attracted significant attention from those interested in the UAS technologies and their appli-
cations. More than 50 papers were submitted and the Program Committee selected 28 for oral presentation
and 14 for posters. The topics are New designs and prototypes, Navigation and Control, Methods for trajec-
tories/paths generation and planning, Multi-rotor systems and applications, Regulations, Applications, and
Education.

Furthermore the RED-UAS Program includes two plenary sessions on “Perception and Decision Mak-
ing for Autonomous Soaring - Towards Persistent Flight”, from Salah Sukkarieh, and “Challenges in Un-
manned Aircraft Systems: From Controller Design to Integration into the National Airspace”, from Kimon
Valavanis, a round table on “Capabilities needed for UAS development and applications (Facilities, Tech-
nologies, Regulations, R&D Funding Programs)” with 6 presentations, a special session on ”UAS and Aerial
Robotics” with 4 presentations, and 6 flight demonstrations.

Thus, RED-UAS has an attractive Program with 53 presentations from 12 countries combining UAS
methods, technologies, applications and education, as well as the 6 flight demonstrations.

Furthermore, RED-UAS is part of the European Robotics Week (28th Nov – 4th Dec, 2011). One week
with 340 events of robotics in 18 countries throughout Europe.

We would like to thank all the contributors to the RED-UAS for making this event possible. Thanks are
due to our colleagues at FADA-CATEC and the GRVC Group of the Universities of Seville and Pablo de
Olavide, that have worked very hard in the preparation of the event, and to the Sponsors and Organizations
that have supported RED-UAS.

We wish also to extend our gratitude to all the members of the Program Committee, Industrial and Ex-
hibition Committee and Organizing Committee.

Anibal Ollero
Program Committee Chair

Juan Pedro Vela

Industrial and Exhibition Committee Chair
Program Comittee Industrial and Exhibition Comittee
Chair: Anibal Ollero, FADA-CATEC and Univ. Sevilla Chair: Juan Pedro Vela, FADA-CATEC (Spain)
Antonio Barrientos, Univ. Politécnica Madrid (Spain) Pedro Becerra, Aertec (Spain)
Samir Bouabdallah, ETHZ (Switzerland) Francisco Cano, Cassidian an EADS Company (Spain)
Pascual Campoy, Univ. Politécnica Madrid (Spain) Ladislas Freire, Aries Ingenierı́a de Sistemas (Spain)
Cristina Cuerno, Univ. Politécnica Madrid (Spain) Pablo Gonzalez Sanchez-Cantalejo, INDRA (Spain)
Patrick Doherty, Linköpings Universitet (Sweden) Juan Alberto Herrerı́a, ISDEFE (Spain)
Patrick Fabiani, ONERA (France) Juan Luque, UAV Navigation (Spain)
Jose Patricio Gómez, Univ Politécnica Madrid (Spain) Juan Osuna, Elimco (Spain)
Jesús Gonzalo, Univ. León (Spain) Isabella Panella, SELEX Galileo (Spain)
Guillermo Heredia, Univ. Sevilla (Spain) Alfonso Picardo, FADA-CATEC (Spain)
Mo Jamshidi, University Texas at San Antonio (USA) Emmanuel Previnaire, Flying Cam (Belgium)
Konstantin Kondak, DLR (Germany) José Prieto, GMV (Spain)
Iván Maza, Univ. Sevilla (Spain) José E. Román, Boeing Research & Technology Europe (Spain)
Rogelio Lozano, Univ. Technologie Compiègne (France) Juan Sendagorta, SENER Ingenierı́a y Sistemas (Spain)
Francisco Muñoz, INTA (Spain) Francisco Serrano, Deimos Space (Spain)
Manuel Mulero, INTA (Spain) Peter Van Blyenburgh, UVS International
Paul Oh, Drexel University (USA)
Enric Pastor, Univ. Politécnica Catalunya (Spain)
Inna Sharf, McGill University (Canada)
João Sousa, Universidade do Porto (Portugal)
Rune Storvold, NORUT (Norway)
Salah Sukkarieh, University of Sydney (Australia)
Antonios Tsourdos, Cranfield University (UK)
Kimon Valavanis, University of Denver (USA)
Ilkay Yavrucuk, Middle East Technical Univ. (Turkey)
Antidio Viguria, FADA-CATEC (Spain)

Organizing Comittee
Dissemination: Silvia de los Santos, FADA-CATEC
Exhibition: Anastasio Sánchez, FADA-ATLAS
Publications: Manuel Béjar, Univ. Pablo de Olavide
Local Arrangement: Ramiro Martı́nez, Univ. Sevilla
Registration: Angel Rodrı́guez, Univ. Sevilla
Web: Pablo Rubiano, FADA-CATEC
Finance: Irene de Miguel, FADA-CATEC
Gold Sponsors Silver Sponsors

Other Support Entities European Robotics Week

Workshop Contents

Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Program Comittee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Industrial and Exhibition Comittee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Organizing Comittee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Sponsors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

SESSION Methods for trajectories/paths generation and planning . . . . . . . . . . . . . . . . . . . . 1

A new method for UAV path planning under uncertainties

J. A. Cobano, D. Alejo, R. Conde, and A. Ollero
GRVC, University of Seville and Center for Advanced Aerospace Technology FADA-CATEC, Spain . 1

Practical issues and improvements in farmland coverage with aerial vehicles

Joao Valente, Antonio Barrientos, Jaime del Cerro, David Sanz, Mario Garzón, and Claudio Rossi
Robotics & Cybernetics Research Group (CAR UPM-CSIC), Spain . . . . . . . . . . . . . . . . . . 10

3D Interpolation-Based Path Planning for a Fixed-Wing Unmanned Aircraft in Civil Airspace

Moritz Niendorf, Marius Bednary and Reinhard Reichel
Institute of Flight Systems, DLR, Germany . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Optimal Trajectory Generation for UAVs Using Dynamic Trajectory Modeling

Alfonso Valenzuela, Damián Rivas, Francisco Gavilán and Rafael Vázquez
Department of Aerospace Engineering, Escuela Superior de Ingenieros, Univ. Seville, Spain . . . . . 29

SESSION Multi-rotor systems and applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

Design, modeling and stabilization of a multi-rotor helicopter

M. Samano, R.Castro, R.Lozano and S. Salazar
CINVESTAV Mechatronics Section, Mexico, and UMI-LAFMIA, France . . . . . . . . . . . . . . . 41

I
 Towards an Indoor Testbed for Mobile Networked Control Systems
Martin Larsson, Jonas Lindberg, Jens Lycke, Karl Hansson, Aziz Khakulov, Emil Ringh, Fredrik Svens-
son, Isak Tjernberg, Assad Alam, Jose Araujo, Farhad Farokhi, Euhanna Gadhimi, Andre Teixeira,
Dimos V. Dimarogonas, Karl H. Johansson
KTH Royal Institute of Technology, Sweden . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

Aerial Object Following Using Visual Fuzzy Servoing

Miguel A. Olivares-Méndez, Iván F. Mondragón, Pascual Campoy, Carol Martinez
Centro de Automática y Robótica (CAR) UPM-CSIC, Spain . . . . . . . . . . . . . . . . . . . . . . 61

On the Use of UAVs for Vessel Inspection Assistance

Alberto Ortiz, Francisco Bonnin-Pascual and Emilio Garcia-Fidalgo
University of Balearic Islands, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

SESSION Regulations and Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

Review Paper: Present Perspective of the Airworthiness Regulatory Initiatives for Unmanned
Aerial Systems
Cuerno-Rejado, C. and Domingo-Leal, P.J
E.T.S.I. Aeronáuticos, Universidad Politécnica de Madrid, Spain . . . . . . . . . . . . . . . . . . . . 81

UAS Integration into Non-segregated airspace in the short term: SINUE Project
Javier Caina, Pablo Gonzalez, Sara Luis and Lourdes Garcia
INDRA and INECO, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

Potential UAS Applications vs. Air Power Security Trend

Alberto P. Martinez Borja
CASSIDIAN, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

Open architecture for border surveillance UAS

Jorge Bueno Gómez, Juan Alberto Herrerı́a Garcı́a, Olivier Reichert
Ingenierı́a de Sistemas para la Defensa de España ( ISDEFE ), Spain . . . . . . . . . . . . . . . . . . 124

Rapid Development of Integrated 3D Terrain Maps using Light Unmanned Aerial Vehicles
G. Heredia, J. A. Cobano, D. Pérez, G. Núñez, L. Durán, F. Caballero, V. Vega, F. Real and A. Ollero
University of Seville and Center for Advanced Aerospace Technology FADA-CATEC, Spain . . . . . 134

SESSION Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

Scalable UAV for Flight Mechanics Education

Jesús Gonzalo
University of Leon, Engineering School, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

Fuel Cells: Alternative Propulsion for Unmanned Aerial Vehicles

Oscar González Espasandin, Teresa J. Leo and Emilio Navarro
E.T.S.I. Aeronáuticos, Universidad Politécnica de Madrid, Spain . . . . . . . . . . . . . . . . . . . . 152

II
 Tutorial for the application of Kane’s Method to model a small-size helicopter
L.A. Sandino, M. Béjar and A. Ollero
Univ. of Seville and Center for Advanced Aerospace Technology FADA-CATEC, Spain . . . . . . . . 162

UAV piston engine test bench

Juan Carlos Celada Fernández and Pablo Fernández Rodrı́guez
Área de Ingenierı́a Aeroespacial. Escuela de Ingenierı́as Industrial, Informática y Aeronáutica. Cam-
pus de Vegazana. Universidad de León, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

Alternatives for programming a low-cost UAV in robotics undergraduate course

Mario Castro de Lera and Vicente Matellán
Vega Space GmbH. GCC, Germany and Universidad de León, Spain . . . . . . . . . . . . . . . . . . 182

SESSION New designs and prototypes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

Design and Experimental Validation of a Prototype of Ducted-Fan Aerial Service Robot

Roberto Naldi, Fabio Brué, Francesco Forte and Lorenzo Marconi
CASY-DEIS, Università di Bologna, Italy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

On the Use of Calculus of Variations to Determine Optimum Rotor Shapes for Micro Air Vehicles
Gregor Thurlow
Department of Aerospace Engineering, University of Glasgow, United Kingdom . . . . . . . . . . . . 199

Mini coaxial rocket-helicopter aeroydnamic modeling, hover control, and implementation

E. S. Espinoza, O. Garcia, R. Lozano, and A. Malo
Laboratoire Franco-Mexicain d’Informatique et Automatique, LAFMIA UMI, France and Mexico . . 211

Tools for UAV Operations

P. Dias, E. Marques, R. Martins, R. Bencatel and J.B. Sousa
Department of Electrical and Computer Engineering, University of Porto, Portugal. . . . . . . . . . . 224

Experience in Designing Autonomous Airplane

Róbert Schochmann, Jakub Pilka, Michal Kusenda, Ivan Pestún, Ing. Peter Volšı́k, Ján Malina
Slovak University of Technology in Bratislava, Faculty of Electrical Engineering and Information
Technology, Slovakia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232

SESSION Navigation and Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

Position-based velocity control system for rotary-wing UAVs

D. Santamarı́a, M. Garcı́a, A. Viguria and A. Ollero
Center for Advanced Aerospace Technologies (FADA-CATEC), Spain . . . . . . . . . . . . . . . . . 255

III
 Coordinated Standoff Target Tracking of Unmanned Aerial Vehicles Using Adaptive Sliding
Mode Control
Hyondong Oh, Seungkeun Kim, Antonios Tsourdos, and Brian A. White
Autonomous Systems Group, Department of Informatics and Systems Engineering, Cranfield Univer-
sity, United Kingdom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267

Nonlinear Optimal Control of Holomonic Indoor Airship

Yin Yang, Inna Sharf
Department of Mechanical Engineering, McGill University, Canada . . . . . . . . . . . . . . . . . . 277

Attitude Control of a Quad-rotor based on Speed Sensing in Brushless DC Motors

I. Gonzalez, S. Salazar H. Romero R. Lozano and J. Torres
UMI-LAFMIA 3175 CNRS, France, and CINVESTAV, Mexico . . . . . . . . . . . . . . . . . . . . 287

An INS / GPS Integrated Navigation Filter for a UAV Based on a Mobile Phone
B. Fons-Albert, J. Vila-Carbó, Enrique Hernández-Orallo, J. Tur-Massanet
Universidad Politécnica de Valencia, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299

POSTERS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305

A test bed for rapid flight testing of UAV control algorithms

Tiago Oliveira, Goncalo Cruz and Eduardo R.B. Marques
Academia da Forca Aérea Portuguesa, Portugal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305

Adaptive Control of Airplane Longitudinal Flight Dynamics

Francisco Gavilán, Rafael Vázquez, and Jose Angel Acosta
Departamento de Ingenierı́a Aeroespacial y de Ingenierı́a de Sistemas y Automática (Universidad de
Sevilla), Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316

Aircraft Structural Health Monitoring Use Case

Joaquı́n Pérez, Asier Rabanal and Ander Gorostiza
SISTEPLANT S.L. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326

Autonomic communications for joint maritime surveillance missions enabled by UAVs

José Manuel Sanchez and Vicente Baena
IntegraSys and University of Seville, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336

Building a UAS based on PBL concept

Rodrı́guez-Sevillano, A.A., Gandı́a-Agüera, F., Barcala-Montejano, M.A.
Universidad Politécnica de Madrid, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346

Comparison of different state estimation algorithms for quadrotor control

L.R. Garcı́a Carrillo, A. Dzul, R. Lozano
U. de T. de Compiègne, France, and Instituto Tecnológico de la Laguna, Mexico . . . . . . . . . . . . 356

IV
 Coordination of Multiple UAS for Tracking under Uncertainty
Jesús Capitán, Luis Merino and Anı́bal Ollero
GRVC University of Seville, University Pablo de Olavide and Center for Advanced Aerospace Tech-
nology FADA-CATEC, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366

Electromechanical Actuation Systems for Unmanned Systems with Zero Environmental Impact
Aitor Isturiz and Javier Viñals
Sener Ingenierı́a y Sistemas SA, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376

OpenFOAM as Free CFD Tool for Education, Research and Design in UAS
Diego Domı́nguez Fernández and Deibi López Rodrı́guez
Área de Ingenierı́a Aeroespacial. Escuela de Ingenierı́as Industrial, Informática y Aeronáutica (Uni-
versity of Leon), Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381

Real-time localization using a Wireless Sensor Network: Application to an Unmanned Aerial

Vehicle
J.-L. Rullán-Lara, G. Sanahuja, S. Salazar, and R. Lozano
Laboratoire HEUDIASYC UMR CNRS 6599, Université de Technologie de Compiègne, France . . . 389

Real Time Stabilization of a four-rotor UAV using rotor speed feedback

I. González, S. Salazar, H.Romero, R. Lozano and J. Torres
UMI-LAFMIA, France . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401

Small UAS for Montagu’s Harrier’s (Circus pygargus) nests monitoring

Mara Mulero and Juan José Negro
Doñana Biological Station. CSIC, Spain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411

Tail Motion Model Identification for Control Design of an Unmanned Helicopter

Bernardo Morcego, Eleazar Saldı́var and Fatiha Nejjari
Automatic Control Department, Universitat Politécnica de Catalunya, Spain . . . . . . . . . . . . . . 418

Towards ATM Automation: A Flight Management System for Unmanned Aircraft Systems (AT-
LANTIDA Project)
Juan José Calvente Plaza and Julio Antequera Galiano
AERTEC Ingenierı́a y Desarrollos, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429

DEMONSTRATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440

Pelicano demonstration
Gonzalo Pacios
INDRA, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440

Mantis Demonstration
César Mendez
INDRA, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442

V
K2B6 Demonstration
Jesús Garcı́a Muñoz
Unmanned Solutions, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445

Demonstration of multi-UAS cooperation system

C. Albo, A. Anderson Bell, A. Viguria and A. Ollero
Center for Advanced Aerospace Technology, FADA-CATEC, Spain . . . . . . . . . . . . . . . . . . 447

ALO Demonstration
Francisco Muñoz
INTA, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449

E100 and E300 Demonstration

José Marı́a Calero
ELIMCO, Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452

VI
VII
VIII
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

A new method for UAV trajectory planning under uncertainties

J. A. Cobano1, D. Alejo1, R. Conde1, and A. Ollero1,2

1
University of Seville, Camino de los Descubrimientos s/n,
41092 Seville, Spain
{jacobano, dalejo, rconde, aollero}@cartuja.us.es
2
Center for Advanced Aerospace Technologies (CATEC)
Aeropolis, Aerospace Technological Park of Andalusia
41309, La Rinconada (Seville), Spain
[email protected]

Abstract
In this paper a new collision-free 4D trajectory planning algorithm for an unmanned aerial
vehicle in a common airspace with other aerial vehicles and static obstacles is presented.
The collisions are detected by using an algorithm based on a grid model. The planning of a
safe trajectory is based on a genetic algorithm. A Monte-Carlo method is used to evaluate
the best simulated trajectory considering different sources of uncertainty as the wind, the
aerial vehicle model inaccuracies and the limitations of the sensor and control system on
board the aerial vehicle. Simulations have been carried out to study the characteristics of
the new method. The paper also presents the results of experiments carried out in the
airfield of Utrera in Seville (Spain) in February 2011.

Keywords: UAV, trajectory planning, conflict detection and resolution, genetic algorithms, uncertainty

1. INTRODUCTION
The research development and applications of Unmanned Aerial Vehicles (UAVs) have been
increasing in the last years. Applications with multiple UAVs present advantages [1,2,3]. In these
systems collision detection and avoidance by means of the computation of collision-free trajectories is
needed.
Several methods have been proposed for the planning of collision-free trajectories including
potential fields [4], evolutionary algorithmic techniques [5], multi-objective evolutionary algorithms
[6], ant colony algorithms [7], graph search methods, like A* and D* [8], or Rapidly-exploring
Random Tree (RRT) [9]. In [10] a review of planning algorithms with a comprehensive mathematical
discussion is presented. In [11] a motion planning problem for multiple UAVs sharing the space with
non-cooperative UAVs is solved. The velocity profile for all the UAVs involved in a potential
collision is calculated to solve it with a minimum change of the initial trajectory. Other works are

-1-

ISBN-xxx-y-zz-yyyyyyy-x 1
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

focused on game theory and interacting coordination between multi-agent hybrid systems [12] (see
[13] for a detailed survey).
Solving the planning of trajectories with multiple aerial vehicles is NP-hard [14]. Moreover, some
differential constraints given by the model of the aerial vehicle should be considered. These
constraints make the problem hardly tractable. Sampling-based techniques, as opposed to
combinatorial techniques, are usually preferred in these NP-hard problems [15]. Therefore, the
application of evolutionary techniques is an efficient and effective alternative for this problem [16].
The uncertainty related to the trajectory prediction should be taken into account. For that, a model
to predict aerial vehicles trajectories is required [17,18]. In this paper an algorithm based on the
Monte-Carlo method is used to simulate the trajectories of each UAV. The sources of uncertainty are
the atmospheric conditions, the UAV model used for prediction, and the limitations of the sensors and
on board control system. The most important source of uncertainty during the flight is the change in
the atmospheric conditions, mainly the wind.
The paper is focused on a decentralized approach, i.e., when an UAV detects a collision it can
change its 4D trajectory to solve the collision without collaborating with the rest of vehicles. The
detection algorithm used in this paper is based on a discretized space divided into squared cells, also
called grid model [11]. The resolution algorithm is based on a genetic algorithm. These algorithms are
appropriated for this problem because of their features to compute near-optimal trajectories in
airspaces with obstacles and dynamically changing environments. The information obtained from the
Monte-Carlo method is used to evaluate which is the best collision-free trajectory.
This paper is organized in six sections. Firstly, the problem formulation is stated in section 2.
Section 3 presents the used algorithms. Section 4 describes the proposed method. Section 5 shows the
simulations and experiments carried out and section 6 details the conclusions.
2. PROBLEM FORMULATION
Consider an UAV entering an airspace with other UAVs and static obstacles, such as mountains or
buildings in low altitude flights. These objects can be a threat to the UAV so a technique that ensures a
minimum separation between the UAV and them is necessary. It is also assumed that the solution only
considers the addition of intermediate waypoints.
The UAV needs the following information about the environment to detect and solve the conflicts:
1) sequence of waypoints that each UAV will follow, 2) parameters of the model of each UAV sharing
airspace, 3) initial and goal location of each UAV, 4) time of arrival to the goal location of the UAV
entering the airspace, 5) static obstacle centers and radii in the airspace.
The objective of this paper is to compute a collision-free trajectory, when collisions are detected
minimizing the changes with respect to the initial trajectory.
3. ALGORITHMS USED
The proposed method is split into three steps.

-2-

ISBN-xxx-y-zz-yyyyyyy-x 2
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3.1. Detection algorithm based on grid model

The detection algorithm is based on a discretized space divided into squared cells. The UAV trajectory
is described as a sequence of cells. Each UAV knows the list of cells that other UAVs cross. The
distance between two UAVs should be greater than the minimum required separation, dreq, to ensure a
collision-free trajectory. This separation dreq will be determined from the Monte-Carlo method as:

d req = d safe + σ max,i + σ max, j (1)

where dsafe is a fixed safety distance in cells, and σmax,i, σmax,j are the maximum spatial deviation of
the predicted trajectories from their means [11]. A conflict is detected whenever an UAV crosses a cell
C, while another UAV crosses the same or a neighbour cell of C at the same time. The neighbour cells
of C are the cells whose distance is less than dreq.

3.2. Monte-Carlo method

This sampling-based representation requires a model of both the UAV and the atmospheric
perturbations. The UAV model is needed to simulate and evaluate the suitability of the trajectories and
different models can be used [19,20]. The UAV model used is the one proposed in [21].
The atmospheric model includes the wind vector speed modulus, ωρ, and direction, ωφ. ωφ is
modeled using a Normal distribution of mean ωφm and standard deviation, ωφσ, without loss of
generality. On the other hand, at low altitudes, ωρ is modeled using a Weibull distribution [22]:
k x k
f(x; k , b) = ( )k −1 e−(x / b) ∀x ≥ 0 (2)
b b
which is determined by the shape factor, k, and the scale factor, b. Due to the stochastic nature of
the model, each specific simulation is affected by different wind disturbances.

3.3. Genetic algorithm

Firstly, an initial population is randomly generated and then its fitness is evaluated. An individual
represents a possible UAV trajectory.
Our approach constructs the cost function using the following weighted sum:

J i = ω 1Li + ω 2 D i
j
+ ω 3 Ai + ω 4∆ ti + ω 5 S i (3)

where Li is the length of the ith trajectory, Di is the deviation of the ith trajectory with respect to the
initial trajectory, Ai represents the turning angle between two consecutive segments, ∆ti is the
difference between the initial time of arrival and the time of arrival of the ith trajectory, Si is the
separation between the ith trajectory and the jth static obstacle, and Ci indicates if the UAV trajectory
is collision-free. ω1, ω2, ω3, ω4, ω5 are the weights. A collision with a static obstacle occurs when
dj<dreq. dj is the separation between an UAV and the jth static obstacle.
These objectives are in most cases confronted, so the GA has to reach a trade-off solution between

-3-

ISBN-xxx-y-zz-yyyyyyy-x 3
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

them. This characterization will be very helpful in the tuning process of the weight vectors. The four
three terms in (3) are related to the safety group, while the last two ω4, ω5, depends on the data
obtained of the Monte-Carlo method and are updated in each iteration.
4. PROPOSED METHOD
The proposed method starts checking for possible collisions from the trajectories obtained by
Monte-Carlo method. If a collision is detected, an iterative planning loop starts and will end when a
collision-free trajectory is computed. Then, a trajectory evaluation from the Monte-Carlo data is
performed to fulfill the conditions presented below (see Figure 1).

Figure 1: Flow chart of the proposed method.

Note that prediction by the Monte-Carlo method is not included in the GA due to the computation
time requirements. However, as the detection algorithm is fast computed, it is included in the GA,
enhancing the chances that the plan will meet the criteria after evaluation. The used information to
evaluate the trajectory and the conditions to choose the safe trajectory are:
1. Deviation: Mean separation, T, and standard deviation, σT, of the UAV trajectory with respect
to the initial UAV trajectory is less than Tmax and σTmax.
2. Obstacle: Number of instances of the model that do not fulfill the minimum required
separation from some moving or static obstacles should be zero, i.e., @obst=0. If @obst≠0 a
collision is detected so the trajectory is not valid.
3. Final location: Distance between the goal location and the mean final location, P, is less than
Pmax. This data is important because of the goal location will be the initial location of the
following trajectory to fly.
4. Time of arrival: Difference of the time of arrival with respect to the initial time of arrival.
If any condition is not satisfied, a new iteration is run but now considering the maximum deviation,
σmax, of the computed trajectory by the last evaluation from Monte-Carlo method (see Figure 1). Thus,
the maximum deviation is considered in the next iteration. Note that σmax,1=0 in the first iteration
providing the solution closest to the optimum .

-4-

ISBN-xxx-y-zz-yyyyyyy-x 4
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

5. SIMULATIONS AND EXPERIMENTS

Several simulations and experiments have been carried out to demonstrate the proposed method
taking into account the trajectory prediction under uncertainties. The time of arrival is not considered
because changes of speed are not performed in this paper.

5.1. Simulations

Consider UAV1 has to pass through airspace where there are three UAVs and three obstacles (see
Figure 2). UAV2, UAV3 and UAV4 can fly their collision-free trajectories. UAV1 should fly its initial
trajectory avoiding the other UAVs and the obstacles. The initial UAV1 trajectory crosses both
Obstacle1 and Obstacle2 if uncertainties are not considered.
The average and deviation of the wind speed and wind direction are given by (ωρm=10m/s
ωρσ=3m/s) and (ωφm=1.57rad, ωφσ=1rad) respectively. Initially, the parameters of the GA are set as:
ω1=1, ω2=1, ω3=1, ω4=0.1, ω5=1, and dsafe=40m. The parameters considered in the evaluation are:
Tmax=2000m, σTmax=1500m, Pmax=200m.
The mean UAV trajectories that are not modified and their maximum deviations, measured in
cells, σmax,2, σmax,3, σmax,4 , are considered as inputs for the detection algorithm.

Figure 2: Initial trajectories of each UAV.

A collision between UAV1 and UAV3 is detected at t=240s (see Figure 3(a)). Note that the
instances of each predicted trajectory are depicted every 20s. Then, the GA is executed. This new
UAV1 trajectory is evaluated using the Monte-Carlo method (see Figure 3(b)). The trajectory is not
safe because passes through Obstacle2. Note that this is because of the initialization of σmax,1=0, that
makes the GA to generate trajectories very close to the obstacles that can become invalid when the
uncertainty sources is considered, that is, σmax,1>0.
A new iteration is necessary considering σmax,1 obtained in the first iteration. Figure 4(a) shows
the different computed trajectories from GA (grey lines) and the near-optimal solution chosen (black
line). Finally, the chosen solution is evaluated by means of the Monte-Carlo method (see Figure 4(b)).
It is important to study how does the number of iterations of the GA influence on the Execution

-5-

ISBN-xxx-y-zz-yyyyyyy-x 5
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Time (ET) and the quality of the solution computed by GA (Cost). Table 1 shows the results obtained
increasing the number of iterations of the GA. Mi is the number of times that the proposed method is
executed.

(a) (b)

Figure 3: Collisions detected considering the predicted trajectories from Monte-Carlo method (a); and
collision-free UAV1 trajectory computes in the first iteration.
When the ET of the GA increases the cost of the solution decreases enhancing the solution (see
Table 1) while more iterations are considered. Therefore, when a collision is detected, the proposed
method computes different solutions for different number of iterations of the GA. The quality of the
solution obtained for UAV1 depends on the computation time, but it can be assured that a solution is
always obtained in less than a minute.

(a) (b)

Figure 4: Near-optimal trajectory for different iterations by GA (a) and collision-free UAV1 trajectory
computes by the proposed method (b).

5.2. Experiments

A flight experiment has been carried out using the Megastar UAV at the airfield of Utrera (Sevilla,
Spain). The experiment corresponds to the presented simulation (see Figure 2) scaled down 10 times.
Figure 5(a) depicts the solution for the UAV1 trajectory, the trajectories of the rest of UAVs as
well as the position of the obstacles. On the other hand, Figure 5(b) shows the distances between
UAV1 and the other UAVs flying in the same airspace. Also, the safety distance obtained by the

-6-

ISBN-xxx-y-zz-yyyyyyy-x 6
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Monte-Carlo Method is represented. As expected, the relative distance between UAV does not go
below the safety distance. Therefore, UAV1 avoids the collisions correctly.
Table 1: Scenario presented in Figure 2.
Iteration ET (s) Mi Ji (m)
20 50.360 2 21248.176
30 70.154 2 16126.371
40 85.966 2 16090.009
50 120.3481 3 17746.378
60 120.856 2 15477.069
70 135.979 2 15839.742
80 300.388 3 15925.737
90 340.967 3 16384.709
100 374.338 3 15522.491
110 408.524 3 16041.248
120 215.886 2 15239.710
130 479.141 3 15890.102
140 252.203 2 15287.979
6. CONCLUSIONS
The collision-free trajectory planning problem for an UAV sharing airspace with other UAVs has been
considered. The UAV can compute a collision-free trajectory without collaborating with other UAVs.
This approach presents advantages in applications where the rest of UAVs should not change its
trajectory to perform the mission satisfactorily.

(a) (b)

Figure 5: (a) Real trajectories of the UAVs and static obstacles; and (b) separation between UAV1 and
other UAVs over the time: UAV2 (blue line), UAV3 (red line) and UAV4 (black line).

-7-

ISBN-xxx-y-zz-yyyyyyy-x 7
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The uncertainty of the predicted trajectory is considered from the Monte-Carlo method. This
uncertainty modifies the minimum separation depending on the maximum deviation obtained from the
Monte-Carlo simulations in the detection algorithm used. The solution is computed from the genetic
algorithm. The algorithm also evaluates the fitness of the trajectory from Monte-Carlo method.
The most important characteristic is that its computation time does not depend significantly on
the scenario and with 20 iterations, performed in less than a minute, a solution is computed. Therefore,
this algorithm can be adapted to different situations by depending if the collision is detected in short
or long term. While more time to compute the solution will generate better trajectories, the generation
of trajectories avoiding obstacles can be assured in shorter time.
Future efforts will be focused on reducing the computation time and make a more exhaustive
study about the scalability.
7. ACKNOWLEDGMENT
Their work is partially supported by the European Commission ICT Programme under the Network of
Excellence CONET NoE (FP7-INFSO-ICT-224053) and the EC-SAFEMOBIL FP7 Integrated Project
288082, as well as the project of excellence of the Junta de Andalucía “Sistema para el Despliegue y
Operación Autónoma de Redes de Sensores empleando Vehículos aéreos no Tripulados”
(P09-TEP-5120).

REFERENCES
[1] I. Maza, F. Caballero, J. Capitán, J.R. de Dios and A. Ollero, “Experimental Results in Multi-UAV
Coordination for Disaster Management and Civil Security Applications”, Journal of Intelligent &
Robotic Systems, Vol. 61, Issue: 1-4, pp. 563-585, March 2011.
[2] X.C. Ding, A. Rahmani and M. Egerstedt, “Multi-UAV Convoy Protection: An Optimal Approach
to Path Planning and Coordination”, IEEE Transactions on Robotics, Vol. 26, I. 2, pp. 256-268, April,
2010.
[3] J.A. Cobano, J.R. Martínez-de Dios, R. Conde, J.M. Sánchez-Matamoros and A. Ollero, “Data
retrieving from heterogeneous wireless sensor network nodes using UAVs”, Journal of Intelligent &
Robotic Systems, vol. 60, no. 1, pp. 133-151, 2010.
[4] O. Khatib,“Real-Time Obstacle Avoidance for Manipulator and Mobile Robots,” International
Journal of Robotics Research., vol. 5, no. 1, pp. 90–98, 1986.
[5] C. Zheng, et al., “Evolutionary Route Planner for Unmanned Air Vehicles,” IEEE Transactions on
Robotics, vol. 21, no. 4, pp. 609–620, August 2005.
[6] G. B. Lamont, J.N. Slear and K. Melendez, “UAV Swarm Mission Planning and Routing using
Multi-Objective Evolutionary Algorithms,” in Proc. of the IEEE Symposium on Computational
Intelligence in Multicriteria Decision Making (MCDM 2007), pp. 10–20, 2007.
[7] A.L. Jennings, R. Ordoñez and N. Ceccarelli, “An Ant Colony Optimization Using Training Data

-8-

ISBN-xxx-y-zz-yyyyyyy-x 8
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Applied to UAV Way Point Path Planning in Wind”, IEEE Swarm Intelligent Symposium, pp.
179-186, St Louis (USA), Sep 21-23, 2008.
[8] A. Stentz, “Optimal and Efficient Path Planning for Partially-Known Environments,” in Proc. of
the International Conference on Robotics and Automation, vol. 4, Los Alamitos, CA., 2002, pp.
172-179.
[9] S. M. LaValle and J. Kuffner, “Rapidly-exploring random trees: Progress and prospects”, in
Proceedings of the International Workshop on the Algorithm Foundations of Robotics (WAFR), 2000.
[10] S. M. LaValle, “Planning Algorithms”, 2004. Available at http://msl.cs.uiuc.edu/planning/.
[11] D. Alejo, R. Conde, J.A. Cobano and A. Ollero, “Multi-UAV Collision Avoidance with
Separation Assurance under Uncertainties,” in Proc. of the IEEE International Conference on
Mechatronics, April 2009.
[12] G. J. Pappas, C. Tomlin, and S. S. Sastry, "Conflict resolution for multi-agent hybrid systems", in
Proceedings of the 35th IEEE Decision and Control, vol. 2, pp. 1184-1189, Kobe, Japan, 1996.
[13] J. Kuchar and L. C. Yang, “Survey of conflict detection and resolution modeling methods,” in
AIAA Guidance, Navigation & Control Conference, p.1388–1397, Aug. 1997.
[14] J. Reif and M. Sharir, “Motion planning in the presence of moving obstacles”, Journal of the
ACM, vol. 41, no. 4, pp. 764-790, 1994.
[15] R. J. Szczerba, “Threat neeting for real-time, intelligent route planners” in Proc. of the IEEE
Symposium Information, Decision and Control, pp. 377–382, Adelaide, Australia, 1999.
[16] R. Vivona, D. Karr, and D. Roscoe, “Pattern Based genetic algorithm for airborne conflict
resolution”, AIAA Guidance, Navigation and Control Conference and Exhibit, Keystone, Colorado,
August 2006.
[17] R. Paielli and H. Erzberger, “Conflict probability estimation for free flight,” Journal of Guidance,
Control and Dynamics, no. 3, pp. 558–596, 1997.
[18] M. Prandini, J. Hu, J. Lygeros and S. Sastry, “A probabilistic approach to aircraft conflict
detection,” IEEE Transactions on Intelligent Transportation Systems, vol. 1, no. 4, pp. 199–220, 2000.
[19] I. Lymperopoulos, A. Lecchini, W. Glover, J. Maciejowski, and J. Lygeros, “A Stochastic Hybrid
Model for Air Traffic Management Processes,” University of Cambridge, England, U.K., Technical
Report CUED/F-INFENG/TR.572, Feb. 2007.
[20] BADA website: http://www.eurocontrol.int/eec/public/standard_page/proj_BADA.html.
[21] T. W. McLain and R.W. Beard, “Coordination variables, coordination functions and cooperative
timming missions,” Journal of Guidance, Control and Dynamics, vol. 28, no. 1, 2005, pp. 150–161.
[22] I.Y.F. Lun and J.C. Lam, “A study of Weibull parameters using long-term wind observations”,
Renewable Energy, 20 (2), pp. 145-153, 2000.

-9-

ISBN-xxx-y-zz-yyyyyyy-x 9
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Practical issues and improvements in farmland coverage with

aerial vehicles

João Valente, Antonio Barrientos, Jaime del Cerro, David Sanz, Mario Garzón, and Claudio Rossi

Robotics & Cybernetics Research Group (CAR UPM-CSIC)

Madrid, Spain,

[email protected]

Abstract

This work presents a solution for the aerial coverage of a field by using a fleet of aerial vehicles. The use
of Unmanned Aerial Vehicles allows to obtain high resolution mosaics to be used in Precision Agriculture tech-
niques. This report is focus on providing a solution for the full simultaneous coverage problem taking into
account restrictions as the required spatial resolution and overlap while maintaining similar light conditions and
safety operation of the drones. Results obtained from real field tests are finally reported.

keywords: Quad-rotors, Coverage Path Planning, Precision Agriculture, Remote Sensing

1 INTRODUCTION
In recent years, precision agriculture (PA) researchers have found that the use of unmanned aerial vehicles (UAV)
based on quad-rotors could significantly help to improve the agricultural sciences research. Their motivation
was conceived by its availability, simple assemblage and maintenance, as well as their low cost compared with
traditional tools (e.g. Satellites or conventional aircrafts). For this reason, significant efforts are being made to use
a new generation of mini unmanned aerial vehicles (mUAV) to perform remote sensing (RS).
The main aim of an aerial survey is to obtain aerial images of the field, that can be used to generate a map of
the surface though mosaicking procedures, those maps can also be post-processed to extract interesting informa-
tion(e.g. biophysical parameters, shape and features detection).
Therefore, the aerial vehicles have to cover the full area to be surveyed by following a continuous and smooth
trajectory and avoiding obstacles or prohibited areas. In order ensure a minimum completion time for the survey,
it is desirable to minimize the number of changes in direction and to avoid revisiting points. Furthermore, not all
areas are suitable for taking off or landing with aerial vehicles, so the trajectory has to ensure starting and ending
locations that fulfill all the requirements (e.g. safety margins, space enough for operation, pick up and drop ability,
accessibility). The problem of covering an entire area, subjected to constraints established by the platform itself
and by the workspace, is known as the coverage path planning (CPP) issue.

1
ISBN-xxx-y-zz-yyyyyyy-x 10
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Aerial robots are mainly employed in agriculture for crop observation and map generation through imaging
surveys [2, 1, 8]. The maps are usually built by stitching a set of geo-referenced images (i.e. orthophotos) through
mosaicking procedures. Typically, they rely on information about the biophysical parameters of the crop field.
Moreover, the agricultural experiments reported with aerial vehicles fall mainly in waypoints based navigation
[2, 9, 4, 7], where the drones navigate autonomously through a predefined trajectory, composed by a set of points
in the environment.
This paper gives an overview about some practical aspects of coverage missions for agricultural sites using
single or multi aerial robot systems. A set of metrics have been proposed in order to evaluate those robotic
systems and make a comparative analysis between them. The metrics are mostly focused on the mission time, the
percentage of area covered, and the human effort in each mission, moreover, a Risk Analysis (RA) is presented to
study the possibility of improving the total area coverage by a fleet of aerial vehicles.
The organization of the paper is as follows: After this brief introduction, Section 2 introduces the conceptual
aerial framework with all their components and workload. Section 3 present the field trials with single and multi
aerial robots, while Section 4 provides an improvement approach for coverage mission with multi aerial robots,
and finally in Section 5 the issues reviewed are summed up .

2 AERIAL FRAMEWORK
The conceptual aerial framework is denoted as a set of actions and components (i.e. software and hardware) that
provide the achievement of a task inside a particular context in a feasible fashion. Hence, the framework intends
to support area coverage missions with aerial robotic systems in PA practices. This is a preliminary phase in the
design of a robotic system that can be seen as a top-down approach for solving the problem (i.e. through a step-
wise design). Abstractly speaking: Breaking up of overall goal, individualizing the requirements, coming up with
a concept, and finally starting the design phase.
As it was introduced in Section 1, there is an agricultural task demand and a fleet of aerial and ground robots
to carry out the proposed task is assumed. To make the concept description easier to follow, lets assume that there
is a general weed control task. This task falls mainly in two sub-tasks: the perception system (i.e. identify the
weed species and patches over a wide agricultural area) and the actuation system (i.e. kill the weeds). The second
sub-task will not be discussed since it is not considered in the framework design.
In order to identify the weed species and patches, the agricultural field has to be previously surveyed with
an image sensor. In this way, the aerial framework will be responsible for performing this operation. The main
components of the framework are: Aerial vehicles, Operation or Control station, and Mission planner.
The framework must be able to provide the operator in charge with the required tools to carry out the mission.
This is made up of an aerial vehicle with way-points navigation features, a visual sensor, a control station with
tools to define and monitor the mission, and finally a mission planner to generate the aerial trajectory. The general
inputs for the mission are: Number of aerial units available, field map characteristics (e.g. coordinates, undesired
areas), and images resolution and overlapping. The step-by-step mission can be summed up follows:

2
ISBN-xxx-y-zz-yyyyyyy-x 11
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

1. The aerial units and the control station are setup near the target agricultural site.

2. The operator introduces the before mentioned inputs to the mission planner.

3. The operator launches the mission planner.

4. The mission planner computes an aerial trajectory and sends it to the operator.

5. If the operator does not agree with the generated path, the procedure will jump back to 2.

6. Otherwise, the plan is sent to the aerial vehicles, the execution starts, and the operator supervises the mission
until it is finished.

Figure 1 illustrate an abstract schematic of the mission procedures.

Workstation setup
Mission execution

Mission assignment and

validation

Mission Real-time
feedback

Figure 1: The mission procedures.

3 AERIAL COVERAGE IN PRACTICE

3.1 Case studies on agricultural scenarios

The test scenario was selected to agree with the current goals of the research group. In this way the effort was
made to select a crop field with an irregular shape in order to test the CPP algorithms in a general scenario. The
goal of this study is to develop a feasible mission planner dedicated to aerial vehicles to be part of the current
European project RHEA in which the group is participating. On the other hand, the opportunity to contribute
to a crop with economic relevance in the Comunidad Autónoma de Madrid has also been considered. Thus, the
agricultural scenario to carry out the experiments was a vineyard located in Belmonte de Tajo, in the Northwest of
Madrid, Spain 1 . The agricultural field is enhanced with a red line in an orthophoto, see Figure 2.
1 The experiments on the vineyards have been possible due to the courtesy of Andrés Morate, http://www.andresmorate.com/

3
ISBN-xxx-y-zz-yyyyyyy-x 12
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Latitude Longitude Area

40◦ 060 47.4300 N 3◦ 170 02.3300 W 63765m2

Figure 2: Selected test field.

3.2 Performance Evaluation

In order to measure and evaluate the strategies employed four metrics were established: Number of Turns (Tu),
Quality of Service (QoS), Coverage Time (T), Human Effort (He), and trajectory length (Le).
The metric Tu measures the number and type of rotations made by the aerial vehicle around the z-axis (yaw
turns). The function that measures this magnitude can be given as follow,

m
X {i}
Γ= γk , k ∈ {135◦ , 90◦ , 45◦ , 0◦ } (1)
i=1

where,

γ±135◦ > γ±90◦ > γ±45◦ > γ0◦ (2)

The raster borders which divide the robots workspace in sub-areas will not be sampled. In addition to this, they
are used to divide the target area, and also employed as security strips for avoiding collisions during flight. An
upper bound (UB ) of unvisited cells can be estimated for the security strips as follows: let the area to be discretized
in a N × M grid, N < M . The maximum number of bounding lines is equal to the number of robots R, and the
maximum length of a line is equal to the diagonal. In a discrete world, the length in cells of the diagonal is equal
to M . Then,

R·M R
UB = = . (3)
N ·M N

The value UB /(N · M )(%) is referred as a QoS index, since it actually indicates the percentage of the remote
sensed field.
Moreover, the metric T is the mission completion time in seconds, which indeed can be split in coverage time
Tc and setup time Ts, as follows,

T = Tc + Ts (4)

Finally, the metric He denotes the number of qualified persons needed to carry out an aerial mission of this
kind and Le the length of the trajectory traveled by the quad-rotor in meters.

4
ISBN-xxx-y-zz-yyyyyyy-x 13
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3.3 Field trials

Figure 3 illustrates the coverage trajectory performed by one quad-rotor by a dark blue line and the desired tra-
jectory (i.e. path sent to the vehicle) by a black one. For further details about the approach employed in this
experiment the reader is referred to [5].

Metrics
Tu QoS Tc Ts Le He
21 100 540 120 1546 2

Figure 4: Metrics from the experiment with a

Figure 3: Farmland coverage with a single aerial vehicle. single aerial vehicle.

The experiment addresses an overall coverage of 45 cells, where each cell correspond to an image sample from
the ground. In each way-point the quad-rotor hovered for 1s with the purpose to simulate the camera shot. The
table in Figure 4 summarize the measured metrics.
As far as the same experiments with multi robots is concerned, a coverage path was computed with the purpose
to cover the vineyards with three aerial vehicles. In this experiment an area sampled into 66 cells was broke down
into three in an optimal fashion and then covered by three quad-rotors. The number of cells per area range from 14
to 25. The reader is referred to [6] for a general understanding of the methods applied. The experimental results
are shown in Figure 5. The table in Figure 6 summarize the parameters setup and the measured metrics. Should be
noted that total denotes the fleet Tc and Ts average value. And also, the sum of all Le.

Metrics
Tu QoS Tc Ts Le He
A 9 - 183 283 371 -
B 8 - 196 82 338 -
C 20 - 288 313 654 -
Total - 222 226 1363 5

Figure 6: Metrics from the experiment with

Figure 5: Farmland coverage with three aerial vehicles three aerial vehicle.

5
ISBN-xxx-y-zz-yyyyyyy-x 14
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3.4 Discussion

Taking the aforementioned field experiments into account, can be concluded that multi robot systems improve
significantly the mission completion time. Nevertheless, there are a couple of trade-offs that should be carefully
analyzed.
The first one is between the QoS and the the mission completion time. According with the results presented
in the previous section, is obvious that the mission completion time decreases with the number of aerial vehicles
(i.e. the coverage time with one robot is greater than when employing three robots). Nevertheless, the QoS is
from 100% for an area coverage with a single aerial vehicle, which mean that all area is full covered (i.e. major
sampling). On the other hand, when employing more than one aerial vehicle, the workspace is setup with security
strips which are defined as cells where the vehicles do not pass by. Therefore, the results from Figure 5 have a lack
of 22% of area to be covered. A workaround for this issue is to send the first robot that finishes its trajectory to
cover those areas. However, this is not the most elegant solution.
Another relevant point is the He in both missions. From the previous experiments is obvious that He increases
with the number of aerial units per mission. Therefore, there is a trade-off between T and He, since the employment
of multiple robots means higher costs in terms of personal. It should be highlighted that each aerial robot has to
have an experienced pilot and additionally there should be a ground base operator which supervises the mission.
In this sense, we are adopting a configuration proposed in [3], where a pilot is in charge of supervising (and, if
necessary, teleoperating) the UAVs, and a mission specialist is in charge of supervising the mission. Indeed, the
typical and safer modality is to have one person in charge for each drone and one more person as mission operator.
The pilot is responsible for supervising the mission of each UAV from the ground and switch from automatic
to manual mode in case there is some system failure or emergency. The base-station operator is in charge of
supervising the mission at a highest level (i.e. monitoring all the navigation data in real time). Although there is
not a specific safety regulation applied to mini aerial vehicles, all the tends are in the line that this measure will be
mandatory.

4 IMPROVING AREA COVERAGE

The rasterized borders represented in a multi robot workspace are meaningful, since they work as security strips,
where the vehicles are not allowed to enter it. However, as seen before there is still a considerable percentage of
area to cover when using them. The metric that address the percentage of area covered is the QoS index. Therefore,
borders have to be considered so as to to improve the QoS but at the same time safety for the robotic system has to
be ensured. A Risk Analysis (RA) has been carried out with the purpose of improving this index.
A risky situation is defined as a situation where two or more aerial robots are located in adjacent cells and
therefore, the distance between them is beyond a certain value. Lets denote Pn (Xn , Yn , Zn ) the 3-dimensional
position of a n aerial robot in an aerial fleet with N elements. A risk condition can be written as,

∀n ∈ N, ∃∆{i}
n ≤ δ, i = 1, 2, ...j

6
ISBN-xxx-y-zz-yyyyyyy-x 15
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

where i is the ith neighbor robot and j the length of a certain neighborhood (e.g. the Von-Neumann neighbor-
hood has a maximum of 4 neighbor robots per robot). Finally, ∆ is the distance between the robot n and a neighbor
robot i, and δ a stipulated minimum safety distance.
The first point to be analyzed is the number of occurrences found during a coverage trajectory (two or more
robot are in neighboring). From the analyses of 8 different coverage trajectories in 2 different crop fields (regular
shape maize fields (e.g. [6]) and non regular shape vineyards) it can be concluded that 37.5% of the times the
robot are in neighboring with other robot. Moreover, up to 12.5% of the times three robot are simultaneously in
neighboring. This non exhaustive analysis allowed to give an idea of the number of times that this situation occurs.
Secondly, lets suppose that a robot is adjacent to another (i.e. neighbor cells). For example, the vineyard has an
area of approximately 63765 m2 which corresponds to 195m length and 327m width. Lets assume that the aerial
robot carries a commercial digital camera that provides image resolutions up to 10.4 mega pixels. Moreover, if
choose the best image resolution to sample the field is chosen, it corresponds to an image size from 3368x3088
pixels. If the mission requires a certain percentage of overlapping, the effective size of the image will be reduce
in same percentage. If a spatial resolution of 1 pixel/cm is required, we will obtain a grid resolution of 6x10 cells
2
(each cell will have a dimension of 32.7x32.5m). With these magnitudes it will be difficult to have a collision
between two aerial robots. Even considering positions uncertainties of 3m and wind speeds of 10m/s, there is no
possibility that two aerial robots collide.
In conclusion, whenever a coverage mission where the cells dimension are above a margin of 5m is considered,
the security strips from can be removed from the planner and a QoS of 100% achieved. Figure 7 shows the same
coverage path computed in Figure 5 but without borders, where the borders have been distributed to each sub-area
in a load balancing fashion.

Figure 7: Multi robot coverage without borders

5 CONCLUSIONS
This paper highlights the importance of relying on a coverage path planning system for multiple aerial vehicles
with the ability of taking high-resolution images so as to create mosaics to be used in Precision Agriculture.
2 Assuming that the image dimension is 35mm and the focal length of the camera is from 50mm, and flying at 50 m height above the ground

7
ISBN-xxx-y-zz-yyyyyyy-x 16
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

A CPP approach has been presented and tested considering safety limitations with three quad-rotors. Some
metrics for evaluating the coverage mission has been presented and discussed.
Tests have been performed in real fields, considering no regular shapes with a fully autonomous planner for
several drones allowing 100% of coverage with no risk for the drones.

ACKNOWLEDGMENTS
This work have been supported by the Robotics and Cybernetics Research Group at Technique University of
Madrid (Spain), and funded under the projects “ROTOS: Multi-Robot system for outdoor infrastructures protec-
tion“, sponsored by Spain Ministry of Education and Science (DPI2010-17998), and ’Robot Fleets for Highly
Effective Agriculture and Forestry Management’, (RHEA) sponsored by the European Commission’s Seventh
Framework Programme (NMP-CP-IP 245986-2 RHEA). The authors want to thank all the project partners: Agen-
cia Estatal Consejo Superior de Investigaciones Cientı́ficas - CSIC (Centro de Automática y Robótica, Instituto
de Ciencias Agrarias, Instituto de Agricultura Sostenible), CogVis GmbH, Forschungszentrum Telekommunika-
tion Wien Ltd., Cyberbotics Ltd, Università di Pisa, Universidad Complutense de Madrid, Tropical, Soluciones
Agrı́colas de Precisión S.L., Universidad Politécnica de Madrid - UPM (ETS Ingenieros Agrónomos, ETS Inge-
nieros Industriales), AirRobot GmbH & Co. KG, Università degli Studi di Firenze, Centre National du Machinisme
Agricole, du Génie Rural, des Eaux et des Forèts -CEMAGREF, CNH Belgium NV, CNH France SA, Bluebotics
S.A. y CM Srl.

References
[1] S. R. Herwitz, L. F. Johnson, S. E. Dunagan, R. G. Higgins, D. V. Sullivan, J. Zheng, B. M. Lobitz, J. G.
Leung, B. A. Gallmeyer, M. Aoyagi, R. E. Slye, and J. A. Brass. Imaging from an unmanned aerial vehicle:
agricultural surveillance and decision support. Computers and Electronics in Agriculture, 44(1):49–61, 2004.

[2] L. F. Johnson, S. Herwitz, S. Dunagan, B. Lobitz, D. Sullivan, and R. Slye. Collection of ultra high spatial and
spectral resolution image data over california vineyards with a small uav. In Proceedings of the International
Symposium on Remote Sensing of Environment, pages 221–230, nov. 2003.

[3] R. Murphy and J. Burke. From remote tool to shared roles. IEEE Robotics & Automation Magazine, Jan 2008.

[4] S. Nebiker, A. Annen, M. Scherrer, and D. Oesch. A light-weight multispectral sensor for micro uav - opportu-
nities for very high resolution airborne remote sensing. In The International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences, volume XXXVII, pages 1193–1200, 2008.

[5] João Valente, Antonio Barrientos, Jaime Del Cerro, and David Sanz. A waypoint-based mission planner for
farmland coverage with an aerial robot - a precision farming tool. In 8th European Conference on Precision
Agriculture (ECPA’11), Prague, Czech Republic, 2011.

8
ISBN-xxx-y-zz-yyyyyyy-x 17
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[6] João Valente, Antonio Barrientos, Jaime del Cerro, Claudio Rossi, Julian Colorado, David Sanz, and Mario
Garzón. Multi-robot visual coverage path planning: Geometrical metamorphosis of the workspace through
raster graphics based approaches. In Beniamino Murgante, Osvaldo Gervasi, Andrés Iglesias, David Taniar,
and Bernady Apduhan, editors, Computational Science and Its Applications - ICCSA 2011, volume 6784 of
Lecture Notes in Computer Science, pages 58–73. Springer Berlin / Heidelberg, 2011.

[7] João Valente, David Sanz, Antonio Barrientos, Jaime del Cerro, Ángela Ribeiro, and Claudio Rossi. An
air-ground wireless sensor network for crop monitoring. Sensors, 11(6):6088–6108, 2011.

[8] H. Xiang and L. Tian. Development of autonomous unmanned helicopter based agricultural remote sensing
system. ASAE Annual Meeting, (063097), 2006.

[9] P. J. Zarco-Tejada, J. A. J. Berni, L. Suárez, and E. Fereres. A new era in remote sensing of crops with
unmanned robots. SPIE Newsroom, 2008.

9
ISBN-xxx-y-zz-yyyyyyy-x 18
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3D Interpolation-Based Path Planning for a Fixed-Wing

Unmanned Aircraft

∗1 †2 ‡2
Moritz Niendorf , Marius Bednar and Reinhard Reichel

1
Institute of Flight Systems, DLR, Braunschweig, Germany

2
Institute for Aircraft Systems, Universitaet Stuttgart, Stuttgart, Germany

Abstract
Path planning for unmanned aircraft has attracted a remarkable amount of interest from the
research community. However, planning in large environments such as the civil airspace has not
been addressed extensively. In this paper we apply a heuristic incremental interpolation-based
search algorithm with ecient replanning capabilities to the path planning problem for a xed-wing
aircraft operating in a natural environment to plan and re-plan long ight paths. We modied the
algorithm to account for the minimum turning radius and the limited ight path angles of a xed-
wing aircraft. Additionally, we present a method to consider a desired minimum cruising altitude.
These properties specic to aircraft operation could not be addressed with the original algorithm.
Simulation results show that the planner produces intuitive, short paths and is capable of exploiting
previous planning eorts, when unknown obstacles are encountered.

keywords: UAV, xed-wing, motion planning, 3D Field D*, kinematic constraints

1 INTRODUCTION
Planning intuitive ight paths for unmanned aircraft (UA) in large real-world scenarios and in the
presence of obstacles is hard due to the mostly open structure of the airspace. Casas et al. introduced
a procedure for extracting nodes and links from a digital elevation map forming a search graph to route
unmanned aerial vehicles [1]. A multi-resolution lattice representing discrete steady-state ight paths
of the aircraft is used in combination with a search algorithm based upon A* [2], a heuristic search
algorithm, to solve the global planning problem with respect to several objective functions in Ref. [3].
Algorithms of the D* lite family [4], an incremental heuristic search algorithm, have been successfully
applied to the global path planning and re-planning problem for ground-based vehicles. However, the
resulting paths are typcially limited to straight line segments with discrete azimuth angles. Field D* [5]
builds upon the same idea as D* lite, except that it allows for straight line segments with arbitrary
∗ [email protected]
† [email protected]
‡ [email protected]

1
ISBN-xxx-y-zz-yyyyyyy-x 19
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

path azimuth angles through the use of interpolation, while preserving the favorable properties of the
D* lite family such as ecient planning and re-planning. It has been successfully applied to ground
robots [5], [6]. A 3D version is described in Ref. [7].
The objective of this paper is to present a planning approach based on 3D Field D* for the path
planning problem for a medium-altitude long-endurance (MALE) xed-wing aircraft moving in a large
3D world with restricted airspace and a minimum en route ight altitude from a start to a goal position.
Therefore, we enhanced the algorithm to tailor it to the UA path planning tasks while preserving its
any-angle and re-planning capabilities. We introduce a method to extract paths from a eld created
with interpolation which accounts for the minimum turning radius and limited maximum ight path
angle of the aircraft. We carry out simulations to show the applicability of our path planninng approach.

2 Approach
We present our approach below by rst reviewing 3D Field D*. We then depict the modications we
made to adapt the algorithm to a xed-wing aircraft and describe our implementation.

2.1 3D Field D*
We will briey cover the details of the cost-eld creation process focussing on details which are relevant
for the overall approach. A complete algorithmic description of Field D*, the basis of 3D Field D*, can
be found in Ref. [5]. Like for all members of the D* lite family, a heuristic guided search is used to
direct exploration from the goal node sg towards the initial node si , i.e. the start node which represents
the aircraft's current location. D* lite rst computes a cost-eld which stores the cost-to-goal of the
shortest path originating at discrete nodes to the goal. A path is then found by iteratively moving from
the current position to the cheapest adjacent node.
For Field D* the nodes are placed at the corners of a rectangular grid. A traversal cost c is assigned
to each cell which when multiplied with the distance travelled within that cell yields the path cost for
this path segment. Linear interpolation along the edges between two adjacent nodes is used to compute
the cost-to-goal g(s0 ) for positions s0 along the grid's edges. In 2D an analytical closed-form solution to
determine the minimal cost-to-goal from a node s at a corner of the grid to the goal sg passing through
s0 which is located on one of the edges adjacent to s can be derived.
3D Field D* [7] builds upon this approach. However, a closed-form solution cannot be obtained
for the minimization problem. Instead, an approximation method is used. Numerical optimization
techniques are not applicable as the minimization has to be performed tens to hundreds of millions of
times when solving a moderately sized planning problem [7].
Considering the setup depicted in Fig. 1, the path of minimum cost from node s through one of
the adjacent faces − s− −−−−−−→
0 , s1 , s2 , s3 can be found by carrying out the following steps. Given the cost-to-goal

g(s0 ), g(s1 ), g(s2 ), g(s3 ) at the corners of the face, the cost g(sD ) at position sD on the face can be
computed assuming that the cost between two adjacent nodes can be interpolated linearly with t and u

2
ISBN-xxx-y-zz-yyyyyyy-x 20
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(0 ≤ t ≤ 1, 0 ≤ u ≤ 1) being the coordinates of sD in the local coordinate system on face −

s−−−−−−−→
0 , s1 , s2 , s3 :

g(sD ) = [(g(s1 ) + (g(s0 ) − g(s1 ))) · t] · (1 − u) + [g(s2 ) + (g(s3 ) − g(s2 )) · t] · u (1)

By adding the cost of the path from node s to sD cutting through the cell with traversal cost c, a
cost-to-goal estimate rhs(s) for any path originating at s cutting through the cell with cost c through
sD towards the goal sg can be computed:

p
rhs(s) = c · 1 + t2 + u2 + g(sD ) (2)

By minimizing Eq. 2 with respect to t and u, an approximation for the cost-to-goal rhs(s) of an
optimal path from s through an adjacent face is obtained. Recall that g(sD ) = f (t, u). As no closed-form
solution for this problem exists, the following approximation is used [7]:

1. Compute the minima along the edges of the face in question.

2. Compute the path cost for sD assuming that sD is located at the intersection of the edges, which
connect the minima on two opposing boundaries of the face, respectively.

3. The minimum of the 5 computed candidates (4 located on the edges and 1 located on the face) is
assumed to be the approximated minimum.

This approximation can then be used as rhs-value, an estimate of the cost-to-goal in the original D*
lite algorithm. These rhs-values in combination with the cost-to-goal g are used to compute a cost-eld
with estimates of the cost-to-goal at nodes s of the graph in the same fashion as with D* lite [4]. A
scaling factor k can be used to address directional cost dependencies [7]. We use such a scaling factor
to implicitly consider that height deviations are less preferrable than turning for typical MALE UA by
multiplying the lenght of the vertical path component of a cell traversal with the scaling factor k.

2.2 Application to the Fixed-Wing UA Path Planning Task

We demonstrate the application of 3D Field D* to the UA path planning problem to compute long
paths in real-world terrain in the following sections. The workspace, i.e. the airspace, is discretized by
decomposing it into cuboid cells. By scaling the cuboids and therefore the objects which they discretize,
a cubic grid with unit edge length as assumed by the interpolation equations is achieved. We chose to

s sgoal

t s1 sa sb
s0 sbasic
sD u
s3 s2 sstart

Figure 1: Basic Setup. Figure 2: Path Extraction Strategies.

3
ISBN-xxx-y-zz-yyyyyyy-x 21
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

represent the terrain by a rasterized digital elevation model (DEM) as it is directly applicable to our
rasterized workspace.

2.2.1 Path Extraction

Extracting a path from elds built with interpolation is not trivial [6]. In traditional grid-based planners
a path is found by sequential movement from the current node to the best adjacent node. However,
when using Field D* or 3D Field D* it is necessary to nd the best location on adjacent edges or faces
to obtain a path.
In order to extract a path, it needs to be ensured that all nodes adjacent to the current position
have been expanded, i.e. processed, and hold a cost-to-goal value. As the aircraft can exist anywhere
inside the workspace and not only at nodes, the interpolation equations used for eld creation need
to be modied accordingly. A path is then obtained by iteratively moving from the current position
to the cheapest adjacent position sequentially. However, unfavorable conditions can lead to suboptimal
solutions. Consider the example given in Fig. 2, where the black cell represents an obstacle. sbasic yields
the lowest path cost when solving the minimization problem starting at sstart . However, it is obvious
that a path going directly to sa is shorter by inspection. Therefore, before moving to sbasic , a virtual
path extraction step at position sbasic , called a one-step lookahead, is performed and used to conrm
the expected cost-to-goal from sbasic . This additional step, however, yields a higher cost-to-goal than
expected. The expected cost-to-goal for sa is lower than the updated cost-to-goal for sbasic . Therefore
sbasic is discarded. Such a one-step lookahead step eciently reduces the occurrence of suboptimal paths
during the path extraction [5].

2.3 Modications to 3D Field D*

2.3.1 Cruising Altitude

Typically, even if the actual takeo and landing are excluded from the planning problem, start and goal
of the planning problem remain close to the ground. Thus, the shortest path between start and goal is
usually located close to the ground, especially since with the introduction of the scaling factor in section
2.1 vertical movement is penalized. However, aircraft typically travel at a predened cruising altitude
during en route ight. We therefore need to model this behavior into the planner.
We bias the planner towards a desired minimum ight altitude as follows: By increasing the traversal
cost of cells lower than the cruising altitude using Eq. 3 with z being the z -coordinate of the cell and
altcruise being the desired cruising altitude, a virtual obstacle is created, which is not untraversable but
penalizes ying close to the ground.


1 + |altcruise − z| z < altcruise
c= (3)

1 otherwise

However, following the approach depicted above will typically create a cost-eld that yiels a solution

4
ISBN-xxx-y-zz-yyyyyyy-x 22
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

1 1 1 1 1 1 1 1 1 1 s2 s1
1 1 1 1 1 1 1 1 1 1 ŝ2 ŝ1
altcruise
1 1 1 2 2 2 1 1 1 1
z 1 1 3 3 3 3 3 ϑ
1 4 4 4 4 4 4 γcone 4 ŝ3 s ŝ0
x sstart s3 s0
sgoal

Figure 3: Exemplary adaption of the cell traversal cost Figure 4: Reachable Rectangle.
by the use of Eq. 3 and ascend/descend cones.

path including vertical ascends and descends, which is not desirable if planning paths for a MALE type
xed-wing aircraft. We found that setting the traversal cost of any cell, whose center is located inside
one of two circular cones originating at start and goal, to the traversal cost at cruising altitude leads
to smaller ight path angles during ascend and descend. The opening angle ϑ of the cone is dened
to be ϑ = π − 2γcone with γcone being a ight path angle which can be easily tracked by the aircraft,
thus γ <γmax . An example of this method is depicted in Fig. 3, where the traversal cost c of all cells
lower than the cruise altitude altcruise is modied as given in Eq. 3 except those located inside the
two cones. Preferred ight corridors can be included in the same fashion, i.e. biasing paths towards
predened corridors.

2.3.2 Turning Radius and Flight Path Angle

3D Field D* allows for any-angle planning. However, a xed-wing aircraft usually cannot track paths
with arbitrary ight path angles and and path azimuth changes under en route conditions, due to a
limited achievable ight path angle and minimum turning radius. Taking these two constraints into
consideration, we approximate the locally reachable positions by a rectangular cone, which originates
at the nose of the aircraft and widens in the direction of ight. The shape of this cone is dened as
follows: In the horizontal plane two circles with minimum turning radius tangent to a cell face's normal,
which is located at the center of this face, intersect the face opposite the rst face at two positions. The
opening angle of the triangle formed by the center of the rst face and the intersection points on the
opposite face is dened to be the opening angle of the reachable cone in the horizontal plane. We use
this method to roughly approximate an admissible azimuth angle change ∆χ in the horizontal plane.
The admissible ight path angles γmax and γmin dene the opening angle of the cone in the vertical
plane.
Following the setup depicted in Fig. 4 and given a maximum azimuth change ∆χ and a maximum
absolute ight path angle γmax , the intersection of the cone originating at s, which represents the locally
−−−−−−−−→
reachable positions, with the face −s− −−−−−−→
0 , s1 , s2 , s3 of the cell results in a reachable rectangle ŝ0 , ŝ1 , ŝ2 , ŝ3 . The

minimum height deviation ∆h = min {|ŝ0 − ŝ1 |, |ŝ2 − ŝ3 |} denes the height of the reachable rectangle.
The cost-to-goal, i.e. the g -values for each one of these intermediate vertices denoted ŝ can be computed
using the interpolation equation given in Eq. 1. The path cost minimization as described above can
then be executed for the reachable rectangle instead of the original cell face and thus a kinematically
feasible solution with respect to ∆χ, γmax and γmin can be computed. The examplary application of

5
ISBN-xxx-y-zz-yyyyyyy-x 23
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

this approach is depicted in Fig. 4 for the case that the aircraft is at position s and the ight path
at position s is given by χ = 90◦ and γ = 0◦ , i.e. level ight directly into the cell towards the face
−
s−−−−−−−→
0 , s1 , s2 , s3 .

However, we did not explicitly consider the aircraft's kinematic constraints during the cost-eld
creation. Thus, we cannot guarantee that the cost-eld yields a feasible solution for the kinematically
restricted path-extraction process. During eld creation the scaling factor k can be used to mitigate
this eect. Furthermore, the one-step lookahead path extraction penalizes locations in the cost-eld,
which yield low-cost paths after the rst extraction step, but lead to high-cost or infeasible paths after
a second step. If the path extraction process guides the solution away from the previously explored part
of the workspace, the D* lite-like behavior of the eld creation process which continues running until
the aircraft reaches its goal ensures that unexplored parts of the workspace are updated. Changes to
the workspace such as appearing obstacles are also considered by the continuously running eld creation
process.

3 SETUP
We implemented the suggested online planner in Matlab and performed simulations. The workspace is
discretized by 1N M × 1N M × 500f t cells. The vertical resolution allows for a choice of the cruising
altitude in 500f t increments.
Since the interpolation equation Eq. 2 assumes cells with unit edge length, scaling the cells to unit
cubes penalizes vertical movement heavily. We use the scaling factor k to balance the cost between
vertical and horizontal movement by choosing k = 0.9.
To determine the parameters for the planner, we consider an aircraft approximately of the size and
capabilities of a typical piston-engine MALE UA. We assume the cruise speed to be Vcruise = 120kts,
and the maximum rate of climb to be R/Cmax = 600f t/min at a speed of Vclimb = 60kts, leading
approximately to a maximum ight path angle γmax = 5◦ . We choose the minimum ight path angle
to be γmin = −5◦ in order to simplify the problem. Given Vcruise and assuming a maximum roll angle
φmax = 60◦ the minimum turning radius can be computed to be Rmin = Vcruise 2
/(g tan φmax ) ≈ 225m.
Thus, the aircraft can perform a complete turn within one cell at the given discretization, which is not
desirable. Therefore, the maximum heading change between two consecutive path segments is set to
∆χmax = 89◦ . The outer opening angle of the cones originating at the start and goal positions which is
depicted in Fig. 3 are set to γcone = 3◦ .

4 RESULTS
4.1 Articial Fractal Terrain
Table 1 lists the start position and initial path azimuth, the goal position, the cruising altitude altcruise ,
the number of node expansions until the solution was found nexp and the number of node expansions if
re-planning occured nexpreplan for all test cases. The initial ight path angle is set to γinitial = 0◦ for

6
ISBN-xxx-y-zz-yyyyyyy-x 24
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

test case start[x, y, z, χ] goal[x, y, z] altcruise nexp nexpreplan

1 10N M, 10N M, 2000f t, π2 50N M, 60N M, 2500f t 5500f t 67359 /
2 10N M, 10N M, 2000f t, π2 50N M, 60N M, 2500f t 5500f t 67359 16116
3 84N M, 4N M, 2000f t, π2 117N M, 119N M, 2000f t 5500f t 167061 /
4 82N M, 61N M, 1500f t, π2 84N M, 4N M, 2000f t 5500f t 132277 /
Table 1: Test Results: For each test case start position and azimuth, goal position, cruising altitude, number of
node expansions of the initial planning phase and number of node expansions for re-planning are given.

all cases. Test case 1 depicted in Fig. 5 shows the solution obtained in a 64N M × 64N M × 20000f t
environment with terrain data generated by the diamond square algorithm [9]. Four obstacles are
introduced. An intuitive and short path between start and goal is found which avoids the obstacles.
The desired cruising altitude is maintained en route and the ascending and descending ight phases
exhibit the desired ight path angles.
Test case 2 builds upon the previous test case and shows the re-planning capabilities in a realistic
environment. En route, when the aircraft reaches the position depicted by the black diamond in Fig. 6,
an additional obstacle depicted by the white cylinder is introduced, which renders the previous solution
infeasible. As only one third of the path is already traversed when the obstacle appears planning from
scratch would roughly need 40000 node expansions. However, building upon the previous search eort,
only nexpreplan = 16116 nodes are processed to repair the solution as given in table 1. The resulting path
consists mostly of long straight segments and several turns to avoid obstacles, thus closely resembling
the intuitive solution. The path remains at the desired cruising altitude during en route ight.

4.2 Terrain Based on Real Elevation Data

A 128N M × 128N M large part of SRTM tile 38_03 [8], representing Southern Germany, was extracted
and the resolution was downsampled to 1N M × 1N M using bicubic interpolation. The north-eastern
corner of the resulting tile is located at 50◦ N , 10◦ E . The airspace structure was approximately modeled
using cuboids and cylinders. Airspace classes C, D, restricted and controlled airspace are considered as
obstacles.
In test case 3 depicted in Fig. 7 we search for a route from Pfullendorf aireld (EDTP) to Hettstadt
aireld (EDGH) at a cruising altitude of 5500f t. The resulting path resembles the line-of-sight con-
nection between EDTP and the south-eastern corner of the class D airspace above Stuttgart airport
and the line-of-sight connection from there to EDGH, respectively. En route ight is conducted at the
desired altitude.
An additional test case is presented in Fig. 8. Starting at Pattonville aireld (EDTQ) a ight at
5500f t to Pfullendorf aireld (EDTP) is planned. It is interesting to note that the planner prefers to
rst climb to the cruising altitude rather than ying underneath the lower boundary of the class D
airspace. Thus, the cost of the shorter route underneath the airspace is higher than the cost of the path
at the desired cruising altitude for this conguration. Modifying the cell traversal cost computation for
cells lower than the desired cruising altitude as given in Eq. 3 would alter the behavior of the planner
to return a path underneath the airspace in case the cost for lower cells was lowered.

7
ISBN-xxx-y-zz-yyyyyyy-x 25
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

60 Flight Path
Update Position
50
40

y[NM]
30
Planned Flight Path 20
10000
5000 10
z[ft]

0
60 0 20 40 60
40 x[NM]
20 40 60
y[NM] 0 20 x[NM] Figure 6: Test Case 2.
Figure 5: Test Case 1.

5 DISCUSSION
The algorithm produces intuitive and short paths in all test cases. However, as the scaling factor k
alone did not yield satisfactory results especially in the vicinity of obstacles, we restricted the possible
movements in the path extraction process with constraints on ∆χ, γmax and γmin to tailor the algorithm
to a xed-wing aircraft. We did not implement a decoupled local planner as suggested in [5], which uses
the previously built cost-eld as cost-to-goal estimates, because we wished to preserve the any-angle
property of 3D Field D*. With the introduction of the modied path extraction, dependencies between
the scaling factor, the grid resolution and the allowable ight path angle and azimuth change occur.
We found that since most of the workspace W is free space Wf ree the algorithm expands a signicant
number of nodes before it terminates. Such behavior in case of mostly free workspace was also observed
in [5]. In free space multiple paths with approximately the same cost exist between start and goal.
Therefore, the shortest path is found only after exploring a large quantity of the workspace nodes.
This leads to long initial planning phases. But it eases the computational eort of re-planning, if the
precomputed ight path is left, since a high percentage of the nodes especially close to the path have
already been expanded. However, this behavior can lead to long re-planning periods, if the change to
the workspace is signicant. In order to fully exploit the re-planning capabilities of D* lite, changes to
the workspace must occur close to the position of the aircraft. However, changes in the airspace such as

Flight Path

10000 EDTP 0
5000
z[ft]

0
0 80
40 EDGH North
80 120
120
East y[NM] x[NM]
Figure 7: Test Case 3.

8
ISBN-xxx-y-zz-yyyyyyy-x 26
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

storm cells or a temporarily restricted airspace can occur anywhere. The application of forward-searching
algorithms such as LPA* [4] should be considered.
Test cases 1 and 2 show the deterministic behavior of the algorithm. As the setup is identical in
both test cases for the initial planning, it is expected that the algorithm will show the same behavior in
both cases. From table 1 it can be seen that the algorithm expands the same number of nodes in both
cases.
Considering the dimensions of the workspace in realistic planning cases, we observed that once the
cruising altitude is reached the planning problem reduces to a mostly planar problem. In case of limited
computational resources and time this property should be exploited. However, only a 3D representation
in contrast to a 2.5D consisting of only an elevation map allows for including obstacle congurations
which do not reach down to the ground. Typically such obstacle congurations can be found in this
specic planning problem - consider for example the stack of controlled and restricted airspace typically
found close to airports.
A side-eect of the airspace discretization becomes obvious in test case 4. As the obstacles are, other
than plotted in Fig. 8, represented by cells in the planner, the boundaries of obstacles are represented
by the axis-aligned faces of the cells. This leads to the unnecessary turning in the vicinity of the obstacle
close to the start location.
Due to the continuous nature of the cost-eld it is possible to extract a solution path from anywhere
inside the workspace rather than just from the vertices which hold the cost-to-goal information. This
makes 3D Field D* well suited to deal with tracking inaccuracies. Furthermore, tackling unknown obsta-
cles and re-planning at a global planning level has an advantage over reactive or local obstacle avoidance
approaches, as it will produce globally short and feasible solution paths. Local planners and reactive
methods are typically prone to failure due to local minima if unfavorable obstacle constellations appear
or produce larger detours to avoid the obstacles. Additionally the cell-based airspace representation
allows for a straightforward method of representing arbitrary obstacle shapes at the given resolution by
declaring cells within the obstacles' boundaries untraversable.

120 Flight Path

80
y[NM]

EDTQ
40

EDTP
0 40 80 120
x[NM]
Figure 8: Test Case 4.

9
ISBN-xxx-y-zz-yyyyyyy-x 27
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

6 CONCLUSION
In this paper we showed the application of 3D Field D* to the path planning problem for a xed-wing
unmanned aircraft operating in civilian airspace in a realistic environment. We found that the algorithm
produces natural looking and short paths after modications were made to account for the kinematic
constraints of a xed-wing aircraft. We provided test cases which show that the planner yields intuitive
paths and eciently uses previous planning eorts to ease re-planning.

ACKNOWLEDGMENTS
This work is based on a thesis conducted at the Institute of Aircraft Systems at Universitaet Stuttgart.

REFERENCES
[1] I. Casas, A. Malik, E. Delmelle, M. Karwan, R. Batta, An Automated Network Generation Proce-
dure for Routing of Unmanned Aerial Vehicles (UAVs) in a GIS Environment. Networks and Spatial
Economics, 7(2), 153-176 (2007).

[2] P. E. Hart, N. J. Nilsson, B. Raphael, A Formal Basis for the Heuristic Determination of Minimum
Cost Paths. IEEE Transactions on Systems Science and Cybernetics, 4(2), pp. 100-107 (1968).

[3] P. Wu, D. Campbell, T. Merz, On-board multi-objective mission planning for unmanned aerial
vehicles, in Proc. of the IEEE Aerospace Conference, Big Sky, Montana, pp.1-10 (2009).

[4] S. Koenig, M. Likhachev, D* lite, in Proc. of the AAAI Conference on Articial Intelligence,
Edmonton, Canada, pp.476-483 (2002).

[5] D. Ferguson, T. Stentz, Using Interpolation to Improve Path Planning: The Field D* Algorithm.
Journal of Field Robotics, 23(2), 79-101 (2006).

[6] M. Otte, G. Grudic, Extracting paths from elds built with linear interpolation, in Proc. of 2009
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS '09), St. Louis,
Missouri, pp.4406-4413 (2009).

[7] J. Carsten, D. Ferguson, T. Stentz, 3D Field D*: Improved Path Planning and Replanning in Three
Dimensions, in Proc. of the 2006 IEEE/RSJ International Conference on Intelligent Robots and
Systems (IROS '06), Beijing, pp.3381-3386 (2006).

[8] NASA Jet Propulsion Laboratory, Shuttle Radar Topography Mission [Online], Available:
http://www2.jpl.nasa.gov/srtm (2011).

[9] J. Shankel, Fractal Terrain Generation - Midpoint Displacement, in Game Programming Gems, M.
DeLoura (Ed.), pp.503-507, Charles River Media, Inc., Hingham, Massachusetts (2000).

10
ISBN-xxx-y-zz-yyyyyyy-x 28
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Optimal Trajectory Generation for UAVs Using Dynamic

Trajectory Modeling

Alfonso Valenzuela, Damián Rivas, Francisco Gavilán

and Rafael Vázquez

Department of Aerospace Engineering, Escuela Superior de Ingenieros, Universidad de Sevilla, 41092 Sevilla,
Spain, [email protected]

Abstract

The generation of optimal trajectories for UAVs in the presence of an average horizontal wind
is studied. The generation is performed off line (strategic planning). The approach is based on the
discretization of the aircraft intent and in the formulation of a parametric optimization problem.
For the computation of the trajectory, a dynamic, nonlinear, point-mass model with three degrees
of freedom is considered.

Keywords: UAV, optimal trajectory generation, dynamic modeling, wind effects.

1 INTRODUCTION
The operation of autonomous UAVs require the generation of flyable trajectories that serve as reference
for the guidance and control system. These trajectories are, in general, subject to constraints, which
complicate the generation process; these constraints arise, among other sources, from the environment
(obstacle clearance), from the UAV itself (operational envelope), and from the mission to be carried
out. For an overview of this problem see for example Tsourdos et al. [1], where they analyze the more
general problem of cooperative path planning for a group of UAVs. Path planning can be seen as one of
the layers of the UAV control system, along with the guidance and control (autopilot) layers. Voelker
et al. [2] analyze this 3-layer scheme in the case of horizontal trajectories, using a kinematic model.
In this work the problem of generation of optimal trajectories is studied, for a given optimality
criterium, in the presence of an average wind. The generation is performed off line (strategic planning).
The approach considered is based on the discretization of the aircraft intent (which guarantees that the
trajectory generated is flyable) and in the formulation of a parametric optimization problem. Once the
aircraft intent is discretized, it is defined in terms of a set of parameters, some of which may be fixed (by
the mission) whereas some others are free; the optimation is performed on this set of free parameters.
This approach has been used by Gill and Maddock [3] and Valenzuela et al. [4] in the optimization of 4D

ISBN-xxx-y-zz-yyyyyyy-x 29
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

aircraft trajectories, and by Valenzuela and Rivas [5] in the problem of conflict resolution in air traffic
management, in which one seeks optimal trajectories free of conflicts.
For the computation of the trajectory, the UAV motion is described by a dynamic, nonlinear, point-
mass model with three degrees of freedom, which is commonly used for trajectory prediction (see Slattery
and Zhao [6]). This model provides more accuracy than other simpler kinematic models also used for
path planning. Results are presented for a small fixed-wing UAV (14 kg of mass) powered by fuel cells.

2 PROBLEM DEFINITION
The general arquitecture of the problem of generating optimal trajectories is represented in Fig. 1. The
input to the optimal trajectory generator (OTG) is the data that define the mission to be performed,
which are supplied by the user, and the output is the optimal trajectory y∗ (t), where y is the state
vector of the UAV.

MISSION
OTG

TCI

Figure 1: OTG interfaces.

To obtain the trajectory the OTG requires the use of a trajectory computation infrastructure (TCI).
This module computes flight segments, which are the elements that form the trajectory. The input to
the TCI is in fact a set of parameters which define the trajectory; some of these are free (x), that is,
x). The output is the trajectory y
they are the optimization parameters, and some others are fixed (e
that corresponds to these parameters.
In the following sections the mission, the OTG and the TCI modules are described.

3 MISSION
To define the mission the user must specify the following sets of data: a) flight plan, b) UAV configu-
ration, c) optimality criterium, and d) wind.

a) The flight plan is defined, first, by the horizontal path (the two-dimensional waypoints to be flown), and,
second, by the altitudes and speeds with which the UAV is to fly between waypoints. The waypoints
must be fixed by the user, who can also fix both altitudes and speeds, although this is optional (those
which are not fixed, are free values to be determined by the optimization process).
b) For a given UAV platform and a given mission the user must define the payload and the fuel load (or,
the weight of the batteries or fuel cells). These two weights plus the known operating empty weight of

ISBN-xxx-y-zz-yyyyyyy-x 30
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

the UAV determine the take-off weight, which is an important parameter for the computation of the
trajectory. In this work the UAV is driven by an electric motor powered by fuel cells.
c) The user must define an optimality criterium, which in turn defines the objective function for the
optimization. In this work the optimality criterium is minimum energy consumption.
d) Wind is an important factor that greatly affects the trajectory of the UAV. The user must define an
average wind to be taken into account in the trajectory computation process, depending on the expected
wind during the mission. The difference with the actual wind during the flight must be taken care of by
the guidance and control system. In this work a nominal horizontal wind is considered, constant both
in speed and direction.

4 OPTIMAL TRAJECTORY GENERATOR

The optimal trajectory generator (OTG) is formed by two modules: the trajectory discretization module
(TDM) and the optimizer (OPT), as shown in Fig. 2. They are described next.

USER OTG
MISSION
OPT

TDM

TCI

Figure 2: OTG structure.

4.1 Trajectory discretization module

In this work, the discretization of the trajectory is based on the flight segments that form the trajectory.
These flight segments are defined by three flight constraints (in the case of symmetric flight), for example,
constant heading, altitude and speed; when a flight constraint changes a new flight segment is defined.
The flight constraints are defined by a number of parameters, some of which are fixed and some others
may be free, subject to the optimization process.
The TDM defines the two sets of free and fixed parameters, as well as the constraints to which the
free parameters are subject. Thus, once the flight segments are defined, the input to the TDM is the
x1 ) and the output is the set of fixed parameters of the
set of fixed parameters defined by the mission (e

ISBN-xxx-y-zz-yyyyyyy-x 31
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

x), the set of free parameters (x) and the constraints to which the free parameters are subject
problem (e
(g).
The flight segments are obtained after defining the horizontal and vertical profiles. These profiles
can be defined with different levels of detail, but if the level of detail increases then the computational
cost also increases (for example, the computation of turns usually requires an iterative procedure). To
simplify the computation process, turns, pull-ups and push-downs are not considered. Then, heading
and path angle changes are performed instantaneously. As a consequence, the heading angle and the
path angle will be discontinuous variables. These discontinuities will be smoothed out by the guidance
and control system.
The horizontal profile is defined by the waypoints (see Fig. 3). Waypoints WP1 , . . ., WPn are those
defined by the user. Waypoints WP0 and WPn+1 are needed to define the climb and descent phases, as
shown below. WPn+1 and the top-of-descent (TOD) point are calculated iteratively.

Figure 3: Horizontal profile.

The vertical profile is defined by the following phases (see Fig. 4): a) initial phase, b) mission phases,
and c) final phase. Next, these phases and its segments are described.

a) The initial phase consists of the following flight segments (the numeration is that shown in Fig. 4a):
1. Initial climb: rectilinear segment up to the altitude AGL ∆h0 , performed at runway heading, constant
path angle γ0 and constant airspeed V0 (γ0 and V0 are defined by default). 2. Turn: instantaneous
heading change towards WP1 . 3. Acceleration: rectilinear segment up to the airspeed V1 , performed at
constant heading, constant altitude and constant acceleration. 4. Climb: performed at constant heading,
constant airspeed (V1 ) and constant electric power, up to the altitude h1 (TOC point - Top Of Climb).
b) The mission phases can be executed at different altitudes and speeds. Each mission phase, for each
waypoint, consists of the following flight segments (the numeration is that shown in Fig. 4b): 1. Cruise:
horizontal rectilinear segment, performed at constant heading, constant altitude hi and constant airspeed
Vi , up to the waypoint WPi . 2. Turn: instantaneous heading change towards next waypoint WPi+1 .
3. Acceleration/deceleration: rectilinear segment up to the airspeed Vi+1 , performed at constant heading,
constant altitude hi and constant acceleration. 4. Climb/descent: performed at constant heading,
constant airspeed (Vi+1 ) and constant electric power, up to the altitude hi+1 .
c) In the final phase, the aircraft goes from the final waypoint to the runway. In this phase it is of great
importance to determine the position of the TOD (Top Of Descent), that is, the point at which the

ISBN-xxx-y-zz-yyyyyyy-x 32
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

UAV starts the descent to the final altitude hf . This phase consists of the following flight segments (the
numeration is that shown in Fig. 4c): 1. Cruise: horizontal rectilinear segment, performed at constant
heading, constant altitude hn+1 and constant airspeed Vn+1 , up to the TOD. 2. Descent: performed at
constant heading, constant airspeed (Vn+1 ) and constant electric power, up to the altitude AGL ∆hf .
3. Turn: instantaneous heading change towards runway. 4. Deceleration: rectilinear segment up to
the approach airspeed Vap , performed at constant heading, constant altitude and constant acceleration.
5. Final descent: rectilinear segment up to the touch-down point, performed at runway heading, constant
path angle γf and constant airspeed Vap (γf and Vap are defined by default).

1 2 3
1
4

1 2 3
4

1 2 3

2 3
1

(a) (b)

3 4
5

(c)

Figure 4: Vertical profile: a) initial phase; b) mission phases; c) final phase.

The set of free parameters is formed by the altitudes hi and the speeds Vi not fixed by the user. The
optimization is performed on this set of free parameters.
The fixed and free parameters are subject to constraints that, on one hand, guarantee that the UAV
is capable to fly the trajectory and, on the other hand, guarantee that the trajectory can be computed
by the TCI module. The constraints are: 1. The speeds and altitudes are inside the operational flight
envelope of the aircraft. 2. The distance between two waypoints is greater than the minimum necessary
distance to perform the turns associated to both waypoints. 3. The speed V1 is greater than V0 (this
constraint is added to ensure that the segment 3 of the initial phase is always an acceleration segment).
4. The speed Vn+1 is greater than Vap (this constraint is added to ensure that the segment 4 of the final
phase is always a deceleration segment). 5. The position of the TOC is between the waypoints WP0 and
WP1 . 6. The position of the TOD is between the waypoints WPn and WPn+1 . 7. The necessary distance
to accelerate/decelerate (segment 3 of the mission phase) and climb/descend (segment 4) between two
waypoints is smaller than the distance between both waypoints. 8. The absolute value of the path angle

ISBN-xxx-y-zz-yyyyyyy-x 33
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

in all climb/descent segments is less than 10 degrees (to avoid sharp changes in path angle). 9. The
heading changes are less than 120 degrees (to avoid too narrow inside turns which would take the actual
trajectory far from the nominal waypoint).

4.2 Optimizer

The parametric optimization is performed on the set of free parameters (x) once the objective function
(f ) and the constraints (g) are defined. The problem is formulated as follows:

minimize f (x)
(1)
subject to g(x) ≤ 0.

For the optimality criterium of minimum energy consumption considered in this work, the objective
function is given by Z tf
f= Pe dt, (2)
0
where Pe is the electric power generated on board and tf is the flight time. Both variables depend on
the value of the fixed and free parameters.
In this work, the optimization solver used is MATLAB’s fmincon, a sequential quadratic program-
ming (SQP) method (see Fletcher [7], for example).

5 TRAJECTORY COMPUTATION INFRASTRUCTURE

In this work the model adopted to describe the UAV motion is that of a point mass with three degrees
of freedom commonly used for trajectory prediction (Slattery and Zhao [6]); the equations then describe
the motion of the UAV center of mass.
Since the UAV is driven by an electric motor powered by fuel cells, its mass can be considered
constant. Since turns are not considered, in each segment the flight takes place in a vertical plane
defined by a heading angle χA . The scalar equations of motion are then the following
dV ηmp (Pe , V, h)Pe
m = − D(V, h, L) − mg sin γ,
dt V
dγ
mV = L − mg cos γ,
dt
dx
= V cos γ cos χA + wx , (3)
dt
dy
= V cos γ sin χA + wy ,
dt
dh
= V sin γ.
dt
In these equations V , γ are the modulus and the path angle of the aerodynamic speed (true aero-
dynamic speed TAS), x, y, h are the topocentric coordinates of the UAV center of mass, m is the
UAV mass, t the time, Pe the electric power supplied by the fuel cells, ηmp the overall efficiency of the
propulsion system, L the lift, D the aerodynamic drag, and wx , wy the components of the wind speed
on the x, y directions.

ISBN-xxx-y-zz-yyyyyyy-x 34
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

In this work the following simplifications are considered: parabolic drag polar of constant coefficients
dγ
CD = CD0 + CD2 CL2 , constant overall efficiency ηmp = constant, negligible normal acceleration = 0,
dt
and constant air density ρ in each flight segment (depending on the average altitude). The drag and
lift coefficients (CD and CL ) are defined by D = 12 ρV 2 SCD and L = 12 ρV 2 SCL , where S is the wing
surface area.
In the computation of the trajectory, the previous equations must be particularized for each flight
segment. To compute a given segment, the initial and stopping conditions must be defined. In this work
the following stopping conditions are considered: a) reach a given speed, b) reach a given altitude, and
c) reach a given waypoint.

6 RESULTS
In this section, results obtained for four different cases are shown. In the first one, a reference mission
is optimized and the results are compared to a non-optimized trajectory. In the second case, wind is
included and the results are compared with the reference mission. Next, a trajectory is analyzed in
which the waypoints provided by the user imply heading changes greater than 120 degrees. Finally,
another trajectory is analyzed in which the user fixes the altitudes of two waypoints but the distance
between them is too short to climb/descend.
The specifications of the UAV model considered in all the cases are: m =14 kg, CD0 =0.0229,
k =0.0174, ηmp =0.56, maximum electric power Pe,max =500 W, minimum electric power Pe,min =0 W.
The following values have been considered to define the trajectory: maximum altitude hmax =300
m, minimum altitude hmin =55 m, maximum airspeed Vmax =20.4 m/s, minimum airspeed Vmin =12.3
m/s, V0 =13.2 m/s, h0 =40 m, χ0 =270 degrees, γ0 =5 degrees, ∆h0 =15 m, Vap =13.2 m/s, hf =30
T −D
m, χf =270 degrees, γf = −2 degrees, ∆hf =15 m, acceleration=min( ), electric power for
m
climbs= Pe,max , electric power for descents= Pe,min .

6.1 Case 1

The horizontal profile of this mission is shown in Fig. 5. The total flown distance is 4578 m. In
this mission all the speeds are free parameters, whereas the following altitudes are considered as fixed
parameters: h2 = h3 =80 m, h4 =60 m.
The optimal trajectory obtained by the OTG is shown in Fig. 6. This trajectory is compared with
a non-optimized trajectory, where all the free parameters are fixed, with values h1 = hn+1 =60 m,
and V1 = V2 = V3 = V4 = Vn+1 =17 m/s. The total energy consumption of the optimal trajectory is
42.40 kJ, whereas the consumption of the non-optimized trajectory is 45.16 kJ (+6.51%).
The altitude profile is shown in Fig. 6a. It can be seen that, for the optimal trajectory, there is no
cruise segment between waypoints WP0 and WP1 . The values V1 and h1 are such that all the available
distance between the waypoints has been used to accelerate and climb. In this mission, the airspeed is

ISBN-xxx-y-zz-yyyyyyy-x 35
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

1000

800

600

x [m]
400

200

−600 −400 −200 0 200 400 600 800 1000

y [m]

Figure 5: Horizontal profile, case 1.

roughly constant, as it can be seen in Fig. 6b. The optimum value is about 14 m/s.
The optimization has been performed using MATLAB’s fmincon. The gradient option has been set
to central differences and all the tolerances are set to 10−9 . The simulation is performed in a 2.8GHz
machine running Linux Ubuntu and MATLAB 2009a. In this case, with 7 free parameters, the best
solution is found after 1300 trajectory computations and 24 seconds of CPU time.

90 18
Optimum trajectory Optimum trajectory
Non−optimized trajectory 17.5 Non−optimized trajectory
80
17

16.5
70
16
V [m/s]
h [m]

60 15.5

15
50
14.5

14
40
13.5

30 13
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
r [m] r [m]

(a) (b)

Figure 6: Vertical profile, case 1: a) altitude; b) airspeed.

6.2 Case 2

The mission in this case is the same that in the previous section, but now, a wind of 2 m/s coming from
the East (from the right side of the Fig. 5) is considered. The optimal trajectory obtained by the OTG
is shown in Fig. 7. In this case, the total energy consumption of the optimal trajectory is 42.88 kJ, 1%
greater than the consumption without wind.

ISBN-xxx-y-zz-yyyyyyy-x 36
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Since most of the altitudes are fixed, the altitude profile is analogous to the previous case. Again,
there is no cruise segment between waypoints WP0 and WP1 (all the available distance between the
waypoints has been used to accelerate and climb).
The main differences are found in the airspeed. Now, the airspeed varies according to the heading of
the segment. It goes from 13.5 to 14.5 m/s (lower with tailwinds and higher with headwinds). However,
the flight time is very similar with and without wind, 330.6 s and 329.1 s, respectively.

90 18

17.5
80
17

16.5
70
16

V [m/s]
h [m]

60 15.5

15
50
14.5

14
40
13.5

30 13
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
r [m] r [m]

(a) (b)

Figure 7: Vertical profile, case 2: a) altitude; b) airspeed.

6.3 Case 3

The horizontal profile of this mission is shown in Fig. 8a, which is a modification of the first mission
(Fig. 5). Now, the heading change at WP3 is 153 degrees, greater than 120 (the maximum allowed); all
the other heading changes are smaller than 120 degrees. The following altitudes are considered as fixed
parameters: h2 = h3 = h4 =80 m, and h5 =60 m.
Since the heading change at WP3 is greater than the maximum allowed, it is neccesary to modify
the turn at WP3 . One way to do this is to insert an auxiliary waypoint between WP3 y WP4 , called
WP3a . The speed and altitude of this waypoint are considered as free parameters.
The horizontal profile after the introduction of the auxiliary waypoint WP3a is shown in Fig. 8b.
Once this waypoint has been introduced, all the heading changes are smaller than 120 degrees.
The location of the new waypoint must meet the constraints described in Section 4.1: the heading
changes at WP3 , WP3a and WP4 , must be less than 120 degrees, and the distances between WP3 and
WP3a and between WP3a and WP4 must be greater than the minimum necessary distances to perform
the turns associated to the three waypoints. In this case, the location of the waypoint WP3a has been
chosen in such a way that the turn at WP3 is exactly 120 degrees and the distance between WP3 and
WP3a is 180 m.

ISBN-xxx-y-zz-yyyyyyy-x 37
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The vertical profile of the optimal trajectory obtained by the OTG is very similar to that obtained
in case 1, represented in Fig. 6.

1000 1000

800 800

600 600
x [m]

x [m]
400 400

200 200

0 0

−600 −400 −200 0 200 400 600 800 1000 −600 −400 −200 0 200 400 600 800 1000
y [m] y [m]

(a) (b)

Figure 8: Horizontal profile, case 3: a) original; b) modified.

6.4 Case 4

The horizontal profile of this mission is shown in Fig. 9a. The following parameters are considered as
fixed parameters: V4 =20 m/s, h1 =80 m, h2 =100 m, h3 = h5 =150 m, h6 =100 m, h7 =80 m.
In this case, it is not possible to compute a flyable trajectory due to the values of the fixed altitudes.
From WP7 to the runway threshold, the mission imposes a descent of 50 m in a horizontal distance of
760 m, which results in an average path angle of −3.8 degrees. However, if descents at constant airspeed
are considered, the minimum path angle found for the UAV used in this mission is −2.9 degrees, greater
than the necessary angle.
To be able to compute a flyable trajectory, a spiral descent has been introduced between WP7 and the
runway heading, see Fig. 9b. The effect of the spiral is to extend the available horizontal distance while
keeping the aircraft close to the original mission. Each turn of the spiral is modeled by 4 waypoints
forming a square. The altitudes and speeds associated to the new waypoints are considered as free
parameters. The distance between the new waypoints must be enough to perform the heading changes
associated to each waypoint.
In this mission, only one spiral turn has been considered. One of the vertices of the square (formed
by the waypoints WP8 , WP9 , WP10 and WP11 ) has been placed in the same coordinates as WP7 , as
shown in Fig. 9b. The length of the side of the square is 150 m, which extends the trajectory 600 m
and makes possible the descent.
The optimal trajectory obtained by the OTG is shown in Fig. 10. It can be seen that the descent is
almost a continuous descent and that the airspeed during the descent is about 14 m/s.
The size of the square is a parameter that must be chosen carefully. It can determine the number

10

ISBN-xxx-y-zz-yyyyyyy-x 38
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

of spiral turns and, therefore, the final consumption of the trajectory. For example, if in this mission
the length of the side of the square is 125 m, then 2 spiral turns are necessary to descend and the total
energy consumption is 6% greater.

1200 1200

1000 1000

800 800

600 600
x [m]

x [m]
400 400

200 200

0 0

−200 −200

−400 −400

−1500 −1000 −500 0 500 −1500 −1000 −500 0 500

y [m] y [m]

(a) (b)

Figure 9: Horizontal profile, case 4: a) original; b) modified.

180 21

160 20

140 19

120 18

100 17
V [m/s]
h [m]

80 16

60 15

40 14

20 13

0 12
0 1000 2000 3000 4000 5000 6000 7000 0 1000 2000 3000 4000 5000 6000 7000
r [m] r [m]

(a) (b)

Figure 10: Vertical profile, case 4: a) altitude; b) airspeed.

7 CONCLUSIONS
A strategic optimal trajectory generator for UAVs has been presented. The generation process is based
on the discretization of the aircraft intent and in the computation of the trajectory by a dynamic,
nonlinear, point-mass model. This process guarantees that the reference trajectory generated is flyable,
in the sense that it can be followed by the guidance and control system. The trajectory generator takes
into account the effects of constant horizontal winds.

11

ISBN-xxx-y-zz-yyyyyyy-x 39
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Four different cases have been considered. In the first one, results for a typical mission have been
presented and compared to a reference non-optimized trajectory. In the second case, the effects of a
constant wind on the optimal trajectory have been addressed. Finally, in the third and fourth cases,
two trajectories are considered in which heading changes greater than 120 degrees and deep descents
are introduced by the user, not complying with required constraints; in these cases auxiliary waypoints
are introduced to be able to calculate the optimal trajectories.
The optimal trajectory generator has been shown to be efficient and versatile.

8 ACKNOWLEDGEMENTS
This work has been funded by Aertec Ingeniería y Desarrollos S.L.U. (AERTEC) and the Spanish
Comisión para el Desarrollo Tecnológico e Industrial (CDTI), through the Project SINTONIA/Cenit
2009. The authors acknowledge the support given by the project managers Rafael Ortiz and Sara
Bellido.

REFERENCES
[1] A. Tsourdos, B. White, and M. Shanmugavel, Cooperative Path Planning of Unmanned Aerial
Vehicles, John Wiley & Sons, Ltd. (2011).

[2] A. Voelker, M. Dordelmann, K. Kouramas, et al., Three-layer Aircraft Control: Path-planning,

Guidance, Model Predictive Controller as Autopilot, Proceedings of the 18th IFAC Symposium on
Automatic Control in Aerospace (2010).

[3] W. Gill, and R. Maddock, EFMS Prediction of Optimal 4D Trajectories in the Presence of Time
and Altitude Constraints, EUROCONTROL PHARE DOC 97-70-09 (1997).

[4] A. Valenzuela, D. Rivas, and A. Franco, Cruise Optimization using Trajectory Patterns, AIAA
paper 2010-9140 (2010).

[5] A. Valenzuela, and D. Rivas, Conflict Detection and Resolution in Converging Air Traffic Using
Trajectory Patterns, Journal of Guidance, Control, and Dynamics, 34, pp. 1172-1189 (2011).

[6] R. Slattery, and Y. Zhao, Trajectory Synthesis for Air Traffic Automation, Journal of Guidance,
Control, and Dynamics, 20, pp. 232-238 (1997).

[7] R. Fletcher, Practical Methods of Optimization, John Wiley & Sons, Ltd. (1987).

12

ISBN-xxx-y-zz-yyyyyyy-x 40
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Design, modeling and stabilization of a multi-rotor helicopter.

M. SAMANO1, 2, R.CASTRO1, R.LOZANO2 and S. SALAZAR2
1
CINVESTAV, Department of Electrical Engineering, Mechatronics Section.
2
CINVESTAV, UMI-LAFMIA CNRS 3175.

[email protected], [email protected], [email protected],

[email protected].

Abstract

In this paper the analytical model, numerical simulation and experimental validation of an
eight rotors flying platform is being addressed. The opportunity to carry heavier payloads
and the possibility of continue operation despite the failure of a rotor represents the main
advantages of this configuration. The system is based on a set of low cost components
and its design and stabilization constitutes the cornerstone in the development of an
unmanned aircraft system.

Keywords: Aerial robot, aircraft dynamics, rotor, unmanned aircraft vehicle.

1. INTRODUCTION
On last decade the interest on the study and deployment of unmanned aerial vehicles has rapidly
increased, due to their proved application in a wide range of areas [1] such as aerial photograpy,
powerline inspection, sampling and analysis of atmosphere for forecasting, reconnaissance,
surveillance, etc. In the academic community, unmanned aerial vehicles (UAVs) constitute suitable
platforms to develop and test new generation of sensors and materials, implement control strategies,
multi-agent and multi-platform systems.

Although fixed wing long-endurance UAVs have had more success than rotatory wing UAVs, the
later has the not only the advantage of being able to take off and land vertically but also performs
stationary flight (hover) at a fixed point. A very successful design for UAV helicopter is the one that
has four horizontal rotors, since this configuration simplifies the design, maintenance and control of
the vehicle. However, like all mechanical systems have disadvantages, its carrying capacity and
flight endurance is limited by the power of their engines and power sources. Another disadvantage of
four rotors helicopters is that in case of a failure of any motor of the aircraft an uncontrolled
descend to earth occurs immediately.

On the other hand, the increase on the size and mass of the payloads, and the reliability of the
aircraft has leaded the community to explore new ideas so as to develop multirotor helicopters that
could be competitive in practical industrial applications. For example, in [7] a large quadrotor has
been designed and set-up, from the design of the quadrotor frame to the design and manufacturing of
efficient blades capable of carrying heavier loads [8].

ISBN-xxx-y-zz-yyyyyyy-x 41
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

In particular, we propose the design, construction and stabilization of an eight rotor helicopter with
the aim of lifting heavier loads (cameras, electronic boars, larger batteries, etc.) and as a first step to
build a reliable experimental platform for future extensive research on robust control, redundancy
and fault detection, in particular to explore the feasibility of the aircraft to continue its operation
despite a motor failure, under certain restrictions. Nevertheless, everything comes at a cost, large
amount of motors requires more energy sources, however, it is expected that motors require slower
angular speeds to generate enough thrust to lift the aircraft, so energy required by the system will not
increase too much compared to a tradition four rotor helicopter.

2. DYNAMIC MODELING
In this section the dynamic model of the eight rotors helicopter is presented; the platform is
considered as a moving rigid body with six degrees of freedom under the action of external forces
and torques.

2.1 Equations of motion

In order to simplify the modeling, some assumptions have been made, namely: the frame of the
helicopter is considered to be rigid, so that its own dynamics can be neglected; the structure of the
helicopter is symmetric and the helicopter propellers are considered to be stiff. Figure 1 shows the
reference frames used for the modeling of the helicopter. Let I denote an earth fixed reference frame.
This system is a right-handed orthogonal axis-system with the origin at the helicopter's center of
gravity (at the beginning of the motion). Under simplifying conditions I is considered to be the

inertial reference frame. On the other hand, B denotes the body-fixed reference frame; it is a right
handed orthogonal system with the origin at the helicopter's center of mass and it is assumed that the
axes of the B-frame coincide with the body principal axes of inertia.

Figure 1 Earth fixed reference frame and body fixed reference frame.

During several years, different approaches have been taken to derive the dynamic model of
multirotor helicopters; these include the Euler-Lagrange procedure [4], the Newton-Euler
methodology [9], the bond-graph analysis [10] and revisited quadrotor models including propeller

ISBN-xxx-y-zz-yyyyyyy-x 42
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

models near hovering [13]. Particularly, the former method is used in this work so as to derive the
equations that describe how the helicopter moves according to its inputs. Let us consider the

generalized coordinates vector q  [ , ]  R , where   [ x, y, z ] denotes the position of the

T 6 T

center of mass of the rotorcraft relative to frame I, and   [ ,  , ] are the Euler angles that
T

describe the orientation of the aerial robot. The kinetic energy of the helicopter is given by:

1 T
Ttranslational  m  (1)
2

where m represents the whole mass of the aircraft. On the other hand, it can be shown that the
rotational energy of the system can be expressed as a function of the aircraft angular velocities
expressed in body frame as [5],

1
Trotational   T J  (2)
2

In order to be able to track the movements of the helicopter it is necessary to project coordinates
back and forth from the reference frame to the helicopter body frame. The relationship between the
body frame and the inertial fixed frame is given by the Euler zyx rotational convention which leads
to the so called attitude matrix given by.

 c c  s c  c s s s s  c s c 
 
R ( , ,  )   s s
I
B c c  s s s  s c  s s c  (3)
 s c s c c 
  

where we have used c to denote cos  and s for sin 

Taking into account the projection of the angular rates from the body frame to the inertial frame,
given by   W , where
1 0  sin  
W  0 cos  sin  cos  
 (4)
0  sin  cos  cos  

The rotational kinetic energy can be rewritten as

1
Trotational   T J (5)
2

The potential energy comes due to the gravity force, is given by

V  mgz (6)

ISBN-xxx-y-zz-yyyyyyy-x 43
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

How, recalling the Lagrangian function L(q, q)  Ttraslation  Trotation  V the equations of motion

following are defined from,

d  L  L
   i (7)
dt  qi  qi

where  i  [ F , ] denote the generalized force and torque vector.

2.2 Forces

The force that the rotors produce is commonly called thrust. According to figure 2, the direction of
the vector thrust is always positive, since the position and orientation of the rotor disc for each rotor
relative to the body frame is invariant. So, it is possible to write the forces acting on helicopter’s

body, denoted FB as:

0
 
FB   0  (8)
u 
 
8
where u  f
i 1
i . From aerodynamics consideration, see for example [2], [3] and [13], forces and

torques produced by rotator wing are proportional to the squared propellers angular speed and are

then expressed as f i  ki i where ki  0 is a parameter that depends on the design of the blade and
2

the density of the air, and i is the angular speed of the i motor. Then, the forces expressed on the

inertial frame take the form

FI  RBI FB (9)

where we have used the attitude matrix given by (3).

Figure 2 Force inputs for eight rotors helicopter; detail of angle between helicopter arm.

ISBN-xxx-y-zz-yyyyyyy-x 44
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2.3 Moments

The torques causing the angular accelerations on the helicopter are generated by the thrust vectors
induced by the eight rotors. Considering figure 2, it can be shown that nonconservatives moments
acting about the x and the y axes of the body frame are given by

 3
   l[sin( )( f1  f 2  f5  f 6 )  sin( )( f3  f 4  f 7  f8 )] (10)
8 8

 3
   l[cos( )( f1  f 2  f5  f 6 )  cos( )( f3  f 4  f 7  f8 )] (11)
8 8

where l represents the distance from the center of mass to the rotor axes as depicted in figure 2 and

assuming that the arm lengths satisfy l1   l8 . Moreover, as a consequence of the torque applied
by the spinning rotors to overcome air resistance when rotating, the body experiences a torque in the
opposite direction of rotation around z axes given by

  112  222  332  442  552  662  7 72  882 (12)

There exists other forces and torques that could act on the helicopter's frame during flight [14] (for
example, gyroscopic effects due to propellers rotation, drag forces, blade-flapping effects etc.).
These are not considered in the modeling, since the main goal of the experiments is to achieve hover
state, and at low speeds, their influence can be neglected.

As a result of applying equation (7) for each generalized coordinate one obtains,

mx   s s  c s c  u
my   c s  s s c  u
mz  c c u  g
(13)

The rotational dynamics can be expressed on the standard form

M ( )  C(, )   (14)

1  T
where M ( )  J ( ) and C ( , )  J 
2 
 J  . Equations (13) and (14) are the
complete set of nonlinear equations describing the dynamics of the eight rotor helicopter. Even
though the helicopter has six degrees of freedom, and it is provided with eight rotors, it is still an
underactuated system, hence it is not possible to reach a desired point for all the DOF.

ISBN-xxx-y-zz-yyyyyyy-x 45
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3. A CONTROL SCHEME FOR HOVERING

It is well known, that a conventional linear control is capable of stabilizing an aircraft around a set-
point if it is not subjected to big disturbances such as wind [5]. In order to obtain a linear model from
equations (13)-(14),a state vector x is defined as

x   x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12    x x y y z z       

T T
(15)

For the sake of convenience we propose the input variables

u1 u2 u3 u4 
T
 u  mg      
T
(16)

Let x  0 be an equilibrium point, commonly known as hover state. Physically, hover state appears
when the thrust of the motors is just enough to overcome the gravitational force and the helicopter is
still in the air. The linearization of the system around hover produces the linear system
T
 u    
x   x2 g  x9 x4  g  x11 x6 g x8 x10 x12  (17)
 m I xx I yy I zz 
The output of the system is defined as

y  [ z    ]T (18)
From equation (17) it can be inferred that the previous linearization procedure helps to disregard the
coupled motion and to consider each rotational degree of freedom as an independent single-input
single-output system, under certain limits.

In order to stabilize the system we propose the classical PD control given by the equations

u1  k pz ( z  z d )  kd z ( z  z d )
u2  k p (   d )  kd (   d )
(19)
u3  k p (   d )  kd (   d )
u4  k p (  d )  kd (  d )

where z ,
d
 d ,  d ,  d , z d ,  d ,  d ,  d denote the desired positions and rates. Performance
specifications suggested in [11] have been observed in order to achieve enough stability so that the
helicopter could be driven by a pilot.

By means of a lever set-up, extensive experimental tests were carried on in order to determine the
non-dimensional thrust and drag coefficients using several propellers from different radius and pitch.
These experimental tests allowed us to verify that both force and torque are governed by a second
order degree equation as depicted in Figs. 3-4. According to figure 3, the helicopter would be able
to lift approximately a 1.2kg payload (in addition to its own weight -1.3kg including batteries) and
would be still maneuverable. The main drawback would be the power consumption that would lead

ISBN-xxx-y-zz-yyyyyyy-x 46
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

to a relatively short flight. On the other hand, mass was measured directly and inertia moments were
taken from a detailed 3D computer-assembly. The physical parameters are given in table 1.

Table 1 Experimental parameters

Parameter Value Units
M 1.3 kg
Ixx 6.28e-3 kg  m2
Iyy 6.28e-3 kg  m2
Izz 11.21e-3 kg  m2
g 9.779 m / s2
k 6.292e-6 -
 1.752e-7 -
Derivative and proportional gains between 1.5 and 2 were enough to achieve the control
specifications. Values of control inputs were further analyzed; simulations show that no saturations
on the actuators appear. From the results shown in figures 5 to 8 we can observe that given an initial
position, a classic PD controller stabilizes the eight rotor helicopter to the horizontal and the system
converges to the equilibrium point fast enough.

Figure 4 Rotor torque as a function of angular

Figure 3 Rotor thrust force as a function of
velocity.
angular velocity.

Figure 5 Closed-loop system response to an initial

Figure 6 Closed-loop system response to an
roll condition φ = 10 degrees.
initial pitch condition θ = 10

ISBN-xxx-y-zz-yyyyyyy-x 47
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 7 Closed-loop system response to an initial Figure 8 Closed-loop system response to an

altitude condition z = 1.5 meters initial yaw condition ψ = 10

4. EXPERIMENTAL PLATFORM
So as to verify the validity of the dynamic model and the effectiveness of the control scheme on
these new helicopter configuration, an experimental platform has been designed following
mechatronics design approaches such as the one described in [6] and specific methodologies for
UAV design [1], [12].

The body frame is constructed horizontally with the shape of an octagon with equally long arms. So
as to make it as stiff as possible, foam core carbon fiber laminates were chosen for the arms. These
are extremely lightweight and strong and also provide a good vibration insulation media that reduces
noise from the rotors. The arms are joined in the center of the structure with an octagon made of
aluminum as depicted in Fig 9.

Figure 9 Eight rotors helicopter; experimental platform.

Efforts have been made so that propeller axes of rotation remain always fixed and parallel. Off-the-
shelf brushless motors, propellers, RC-receptor, electronic speed controls and batteries have been
installed appropriately so as to keep the aircraft balanced. A Digital Signal Processor, an Attitude
and Heading Reference System (AHRS), an infrared range sensor and a Modem conform the
avionics of the helicopter. Main processor handles the control algorithm in response of the angular

ISBN-xxx-y-zz-yyyyyyy-x 48
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

positions, angular rates, and altitude provided by the sensors. By means of a RC-controller, altitude
and orientation commands are given to the helicopter and the whole data is transmitted via wireless
serial protocol to a remote computer for further analysis.

5. EXPERIMENTS
Several attitude control experiments have been conducted on the platform in order to validate the
model and to evaluate the performance of the whole architecture. Figures 10-12 shows the angular
position data, retrieved during an outdoors flight test. The helicopter is off by a pilot and left free at
some altitude point where the embedded control is responsible of complete angular stabilization.

Figure 10 Roll angle performance. Figure 11 Pitch angle performance.

Figure 12 Yaw angle performance.

As can be seen, the control strategy for hovering performs appropriately on the helicopter, angular
positions remain near the desired equilibrium point. Since any translational control is applied to the
helicopter, and due to low cost sensors, drift phenomenon has been observed on the helicopter; by
the time symbolic excursions have been achieved a remote command is given by the pilot to
minimize drift effect.

ISBN-xxx-y-zz-yyyyyyy-x 49
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

6. CONCLUSIONS
In this paper we have presented the development for hovering flight of an eight rotors helicopter.
Although the physical configuration and dynamical model appears to lead an over-actuated system
for each degree of freedom, the full system is indeed underactuated. Real-time experiments have
been performed to validate the control strategy proposed and to evaluate the whole system. It has
been demonstrated with experiments that a classic PD control is good enough to achieve hover in the
absence of disturbances. So as to achieve better performance it is expected that future experiments
take into account aerodynamic effects and control laws are based on nonlinear strategies.

7. REFERENCES
[1] R. Austin. Unmanned aircraft systems. UAVS Design, development and Deployment. Wiley,
2010.
[2] J. Seddon. Basic Helycopter Aerodynamics. BSP Professional Books, 1990.
[3] Stepniewski, W.Z., Keys C.N. Rotatory-Wing Aerodynamics. Dover Publications, 1984.
[4] R. Lozano, Objets volants miniatures. Modélisation et commande embarquée. Hermes Science
Publishing London , 2007.
[5] P. Castillo, R. Lozano, A. Dzul., Modelling and Control of Mini-Flying Machines. Springer-
Verlag, London, 2005.
[6] R. Bishop, Mechatronics: An Introduction. CRC Press, 2006.
[7] P. Pounds, R. Mahony, P. Corke, Modelling and Control of a Large Quadrotor Robot, Elsevier
Control Engineering Practice, Vol. 18, No. 7, pp 691-699, July 2010.
[8] P. Pounds, R. Mahony, P. Corke, “Design of a Static Thruster for Micro Air Vehicle Rotorcraft,”
Journal of Aerospace Engineering, Vol. 22, No. 1, pp 85-94, 2009.
[9] B. Erginer, E. Altug, Modeling and PD Control of a Quadrotor VTOL Vehicle, in Proc. 2007
IEEE Intelligent Vehicles Symposium, Istanbul, pp.894-899 ,2007.
[10] R. Hossain, G. Rideout, N. Krouglicof, Bond Graph Dynamic Modeling and Stabilization of a
Quad-Rotor Helicopter, in Proc. 2010 Spring Simulation Multiconference, Orlando, 2010.
[11] P. Pounds, R. Mahony, P. Hynes, J. Roberts, Design of a Four-Rotor Aerial Robot, in Proc.2002
Australasian Conference on Robotics and Automation, Auckland, pp.145-150, 2002.
[12] Cai G., Feng L., Chen B., Lee T. Systematic design methodology and construction of UAV
helicopters. Mechatonics 18, 2008, pp.545-558.
[13] P. Martin and E. Salaun. The true role of accelerometer feedback in quadrotor control. In Proc.
IEEE Int. Robotics and Automation (ICRA) Conf, 2010, pp. 1623-1629.
[14] G. M. Hoffmann, H. Huang, S. Waslander and C. Tomlin. Quadrotor Helicopter Flight
Dynamics and Control: Theory and Experiment. AIAA Guidance, Navigation and Control
Conference and Exhibit, South Carolina, August 2007.

ISBN-xxx-y-zz-yyyyyyy-x 50
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Towards an Indoor Testbed for

Mobile Networked Control Systems

Martin Larsson Jonas Lindberg Jens Lycke Karl Hansson

Aziz Khakulov Emil Ringh Fredrik Svensson Isak Tjernberg
Assad Alam Jose Araujo Farhad Farokhi Euhanna Gadhimi
Andre Teixeira Dimos V. Dimarogonas Karl H. Johansson

KTH Royal Institute of Technology, 100-44 Stockholm, Sweden.∗

Abstract

We consider the design of an indoor testbed composed of multiple aerial and ground unmanned
vehicles for experimentation in Mobile Networked Control Systems. Taking several motivational
aspects from both research and education into account, we propose an architecture to cope with
the scale and mobility aspects of the overall system. Currently, the testbed is composed of several
low-cost ARdrones quadrotors, small-scale heavy duty vehicles, wireless sensor nodes and a vision-
based localization system. As an example, the automatic control of an ARdrone is shown as well as
a platooning experiment with the HDVs.

keywords: Quadrotors, ARdrones, Heavy-duty vehicles (HDVs), Wireless Sensor Networks.

1 Introduction
Increasingly, control systems are operated over large-scale networked infrastructures. The control of
multiple Unmanned Aerial Vehicles (UAVs) and Intelligent Ground Vehicular Systems (IGVs) are an
example of that. Applications for these systems range from environmental monitoring [1], surveillance
and reconnaissance for security [2], to roadway traffic monitoring [3, 4] in the case of UAV systems.
Several From an IGV perspective, a higher interest has been paid to the coordinated control of vehicles
in roads in order to improving safety of drivers, increasing road capacities and ultimately save vehicle
fuel and reduce gas emissions. These features for the case of heavy-duty vehicles has been particularly
targeted by many researchers [5], where they propose vehicle platooning strategies to solve these issues.
Several examples of IGV systems were evaluated in the Grand Cooperative Driving Challenge (GCDC)
challenge in 2011 [6].
∗ Corresponding author: [email protected]

ISBN-xxx-y-zz-yyyyyyy-x 51
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

In all these applications, vehicles must rely on local sensing, computation and control and take benefit
of available communication links with other vehicles. Moreover, when taking into account the possible
distances between vehicles as well as their number, it may be infeasible to rely on solely centralized
computations to achieve mission goals and so, decentralized methods should be applied. This becomes a
problem of networked estimation and control, where the vehicle dynamics as well as network dynamics
should be taken into account at all times.
These research applications serve as motivation for our experimental testbed. Testbeds have been
developed within robotics, with ground vehicles [7], UAVs [8, 9] as well as the combination of ground
robots and wireless sensor networks [10]. The testbed proposed has the objective to join both ground and
aerial vehicles as well as wireless sensor networks. We use Commercial-Of-The-Shelf (COTS) small-scale
heavy-duty vehicles (HDVs) by Tamyia [11] and the ARdrones quadrotors by Parrot [12] and several
Telosb wireless nodes [13]. The localization of each vehicle is performed by a camera system composed
of Logitech Pro 9000 Web cameras.
We will begin by discussing the design requirements for the system based on the driving applications
and research performed in our lab as well as practical considerations. Ultimately we would like to have
a robust, flexible and scalable testbed with a rich environment of dynamical systems, sensors, actuators
as well as networking components. From this point of view, design considerations must be taken for the
choice of the hardware and software components. We start by presenting the overall system architecture
and proceed to detailed explanation of the main components of the testbed: the vehicles, software and
the proposed infrastructure. Finally we present a single experiment with the ARdrones.

2 Motivating Design Considerations

The design of the proposed experimental testbed was motivated by the interest on the control of multi-
agent systems with applications to area surveillance, roadway traffic monitoring and heavy-duty vehicle
platooning. These applications lead to fundamental research on networked control and estimation over
large-scale dynamical systems with resource limitations, and the impact of mobility in communication
networks.
Examples of some relevant research topics on these fields are:

• Control and estimation with network constraints (data loss, delays and bandwidth limitation)

• Vehicular routing algorithms and Roadway traffic modelling for network connectivity analysis

• Wireless Sensor Networks

We refer the reader to [14] for current research challenges in networked control systems, with special
emphasis on multi-agent systems. These are some of the topics that we expect to evaluate.
From the educational point of view, the Automatic Control Lab at KTH currently teaches undergrad-
uate courses on Modelling of Dynamical Systems, Hybrid and Embedded Control Systems, Nonlinear

ISBN-xxx-y-zz-yyyyyyy-x 52
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Control and a Project Course on Automatic Control. In all of them, there exist a laboratorial com-
ponent in which requires the students to perform experimentation on real dynamical systems. At the
moment, all the dynamical systems are static. It is our goal to integrate the proposed testbed to allow
the students to understand the issues of mobility when performing control and estimation of dynamical
systems.

3 System Requirements
Following the motivational aspects presented we can define the following requirements for the testbed:

• Robust, reliable and easy to use and maintain

• Modular so that each component can be used separately from the rest if required

• Scalable with ease integration of heterogeneous types of sensors, actuators and vehicles

• Data logging and real-time measurements capabilities

• Inexpensive

In order for the testbed to be used by researchers, undergraduate and master students, it must
provide a good degree of robustness, reliability as well as ease of usability. One important aspect is
that all the selected components are also used by other people so current development efforts could be
supported by manuals, online forums, etc. Moreover, all the software and hardware should be made in a
way that it is easy for any person to operate it in a matter of days. From a user point of view, we must be
able to provide real-time measurements as well as data logging capabilities so the developed algorithms
could be evaluated. In our case, various control and estimation algorithms are to be developed and
so, all the sensing and actuation values should be available at all times. Since all the communication
between components is done over a network, we must be able to have an idea of the delay introduced,
and network connectivity at all times. From a point of view of development, we should be able to
implement algorithms both in each vehicle as well as in a centralized mission controller, so decentralized
control could be implemented in a realistic way as well as in an emulated way. It is our goal to utilize as
much COTS equipment as possible and do not rely in highly advanced and costly systems. In this way
we are able to keep the development costs low as well as increase the number of systems components
and scale up in a faster/easier way.

4 Components for Multirobot Experiments

Our objective is to select suitable vehicles to target the motivational aspects in Sec. 2. For that, we
must choose ground and aerial vehicles with rich sensing, computational, actuation and communication
capabilities. Cost and ease of support and maintenance from the robotics communities should be also

ISBN-xxx-y-zz-yyyyyyy-x 53
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Central Computer
Mission Controller

Vision ARdrone HDV

Interface Interface Interface

WSN
Java ARdrone
Brigde

ARdrone Vision Ethernet HDV

Router Gateway WSN Gateway
Wi-fi Ethernet IEEE 802.15.4

ARdrones WSN Nodes

CPUs
USB

HDVs Sensors
Cameras Actuators

Figure 1: System architecture for the testbed.

considered. With the introduction of low-cost wireless sensors such as the IEEE 802.15.4 radio-enabled
Telosb node [13] it is now easy to integrate sensing, actuation, computation and communication under
a single platform. In the case of the Telosb node, its computational and memory capabilities are not
high, but the flexibility to customize the communication layers in the device are very appealing. In this
way we are able to implement our own networking protocols or use the ones available from the wireless
sensor network community using for e.g. TinyOS [15]. Moreover, there has been an increase interest
from companies to provide sensor and actuation boards with a plug and play feature that allows for the
fast development of many applications [16].
In order to provide measurements of a global position of each vehicle we must have a robust local-
ization system. Current commercially available systems to provide position measurements of robots are
generally highly accurate but very expensive. Due to this fact, we develop a custom localization and
tracking solution.
With respect to the software utilized in the testbed, we made the decision on relying on software that
would be commonly known by the students and researchers involved in the university and not integrate
any robotics-specific OS. Moreover, the software utilized should also be able to provide a simulation
environment for the experimental setups proposed.

5 Experimental Testbed Components

The testbed consists of many components that are interfaced together as is depicted in Fig. 1. In what
follows, we present the vehicles, software and infrastructure of the testbed.

ISBN-xxx-y-zz-yyyyyyy-x 54
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

5.1 Mobile Unmanned Vehicles

In the current tested we have deployed two types of vehicles for ground and aerial experimentation. For
ground experiments we chose a small scale version of a heavy-duty vehicle to perform realistic platooning
missions and for air we selected a commercial quadrotor which is a very flexibility platform with various
onboard sensors.

5.1.1 Small Scale Heavy-Duty Vehicles

Since we would like to stay as close as possible to the realistic HDVs used for our research, we selected
the Tamyia Scania HDV model as the ground vehicle [11]. Fig. 2(a) shows the Scania HDV. The HDV
with tractor and trailer has a cost of around 500 Euro and can be easily adapted to allow for being
controlled using Telosb nodes instead of the common RC controller. Moreover, it has the flexibility to
carry different sensors as e.g. Infrared sensors to detect the distance to the HDV in front of it performing
a platooning mission. Fig. 2(b) depicts the back of the HDV together with Telosb nodes. These wireless
nodes communicate with the central computer through a WSN Gateway and perform the HDV actuation
for turning and speed. The central computer is then responsible for computing the actuation inputs for
each HDV in the network. The actuation signals are sent to the HDV with a period defined by the user.
By using wireless sensor nodes we are able to have a good control of the delays and packet losses in the
network by designing a suitable Medium Access Control (MAC) mechanism. For our current setup we
are using the IEEE 802.15.4 standard MAC protocol.
Payload weight may be increased or decreased in order to have an impact on the vehicle dynamics
in a fairly realistic way. It also provides a geared motor with 3 different gear ratios, suspension with
metal springs and friction dampers.
The HDV is separated in the tractor and trailer. The tractor has a total length of 520 mm and the
trailer 930 mm. When connected the total length of the HDV is approximately 1150 mm.
In order to avoid errors when performing detection of the distance with the infrared sensors, we
designed the system with four infrared sensors. The system is composed by two short range (4 cm to
30 cm) and two long range (20 cm to 150 cm) sensors in each corner from Phidgets (3520 and 3522).
In this way, even if the HDV in front is not going completely straight when performing longitudinal
platooning, one of the sensors should give a good distance measurement. The sensors are then connected
to the wireless node which transmits the distance measurement in an event-based or periodic fashion,
as defined by the user.
The testbed has five HDV at the moment and we are able to implement control algorithms for
trajectory and speed tracking for single and multiple HDVs. With the acquired measurements for
actuation and sensing we generate dynamical models of the HDV and implement model-based estimation
and control algorithms.

ISBN-xxx-y-zz-yyyyyyy-x 55
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

IR Sensors and
Sensing node

Servo and motor

actua!on node

(a) ARdrone Quadrotor and Heavy-Duty (b) Heavy-Duty Vehicle

Vehicle with Telosb nodes

Figure 2: Image shows, 2(a) the ARdrones [12], the HDV including trailer [11], and 2(b) with the
changes made in the HDV to be controlled using Telosb nodes [13] and interface the Infrared sensors to
measure distances to truck ahead.
5.1.2 Quadrotors

The quadrotors selected for the testbed are the ARdrones from Parrot [12] which are commercially
available for around 300 Euro. These quadrotors are running an embedded Linux OS on a 468mhz
ARM processor and are able to communicate with a base station computer or smartphone over Wi-Fi.
As sensors, they have two cameras, one pointing the front and another to the bottom. Additionally,
the ARdrone has a 6DOF, MEMS-based inertial measurement unit IMU, which provides pitch, roll and
yaw measurements as well as an ultrasound for reporting the ARdrone height. These measurements are
used for the built-in stabilizing controller of pitch, yaw and roll. Supervisory control can be performed
using a base station or smartphone. The control commands are reference values for the control of roll,
pitch and yaw angles as well as lift throttle. The API was developed in Java which provides an interface
to the onboard sensors and actuate on the ARdrone. This API is connected to LabVIEW where the
mission controller is implemented. Our future goal is to control the ARdrone using Telosb nodes instead
of the Wi-Fi channel.
Several control objectives can be realized with this setup. At the moment we have four ARdrones in
the testbed and three Arducopters [17] as other air vehicles. The benefit of the Arducopter is that the
onboard sensing, actuation and stabilizing controller can be designed by the user which is not allowed
in the ARdrone.

5.2 Software

The main software used in the testbed is LabVIEW since it is a very intuitive and powerful programming
tool for performing real experiments, and it is currently used in many of the courses offered at KTH.
More specifically, we use the IMAQ Dx toolkit for image capture and processing and Control Design
and PID and Fuzzy Logic toolkit for control and estimation. In LabVIEW we are also able to simulate
our closed-loop control of the HDV and ARDrone based on their dynamical systems. An example of

ISBN-xxx-y-zz-yyyyyyy-x 56
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

the LabVIEW interface for the ARdrone control can be seen at [18]. Moreover, we develop a Java
application to perform the interface between the ARdrones and LabVIEW. The wireless nodes are all
running the TinyOS operating system, with the IEEE 802.15.4 standard MAC.

5.3 Localization Infrastructure and Design

The testbed is currently implemented in a 30 m long and 2.5 m wide corridor in the 8th floor at the EE
School building at KTH. A total of five HDV and four ARdrones as well as around 200 wireless nodes
Telosb are available for deployments. For vehicle localization we are using the Logitech 9000 Pro web
cameras that cover the whole corridor.
At the current deployment stage, the seven cameras are connected to three separate CPUs running
LabVIEW and communicating with the central computer over Ethernet, as depicted in Fig. 1. Each
camera overlaps the range of the following camera so each vehicle moving between cameras can be
correctly localized by at least one of the cameras.
At first we used diamond shapes of different color to quickly localize 4 (due to the color limitations)
vehicles. Later we developed an algorithm that used a QR-code like pattern to identify the agents and
determine orientation. This marker is a m by m matrix where the 4 corners are reserved to determine
the position and orientation of the marker. To create a hight contrast we use black background and
white regions. In the full setup we can read (m2 − 4)! different agents. By deciding on having exactly n
2

white regions (except the corners) in the array, we can have m n−4 different agents. This reduces the
computational complexity since we can discard point clusters with the wrong number of white regions.
In the real testbed we had m = 3 and n = 1.
When performing localization of the ARdrones in 3D with the camera system, we take advantage
of the correct measurement of the height of the ARdrone given by the onboard ultrasound sensors. For
both HDV and ARdrone localization the error is always below 5 cm.
The resolution of each acquired image is 320x240 (due to bandwidth limitations) and we can perform
the complete search with a period of 50 ms, but have to remark that localizing more vehicles will increase
computational time. The image processing algorithm running in LabVIEW takes the following steps:
1: Take pictures with all cameras
2: Do Background Subtraction (BGSub) to define foreground and update backgrounds for all cameras
3: In the areas defined as foreground: Do color threshold for Tag background, to define possible Tag areas
4: In the areas defined as Tag areas: Do color threshold for White regions and make a point cluster out of every possible
Tag
5: Discard all cluster with the wrong amount of points
6: In every cluster: Use the longest distance between any pair of point points to detect the main diagonal of the Tag
7: Use the main diagonal to find the orientation and try to read the Tag
8: If the Tag was successfully read; relate the position to the agent, else discard it
9: When all clusters are read, for every agent we use the position that is closest to the old position

ISBN-xxx-y-zz-yyyyyyy-x 57
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Leader vehicle

Velocity [cm/s]
50

0
0 5 10 15 20 25 30 35 40 45 50

Gap distance
Distance [cm]

50

0
0 5 10 15 20 25 30 35 40 45 50

Follower vehicle
Velocity [cm/s]

50

0
0 5 10 15 20 25 30 35 40 45 50
Time [s]

Figure 3: The above plots display the results from a platooning experiment with two vehicles. The
first and last plot shows the velocities of the leader and follower vehicles. The solid line represents the
reference velocity and the dash-dot line is the measured velocity. The dashed line is the control input
sent to the vehicle motor, shown in percent of the maximum. The second graph displays the distance
between the trucks, solid lines is the reference and the dashed line is the measured value. Vertical lines
denote where input steps are given.

6 Experiments

6.1 HDV platooning

The performance of the testbed was tested by an experiment with two vehicles platooning. Algorithms
for the vehicle platooning were developed in LabVIEW. Positioning data from the cameras was used to
estimate the vehicle speed and position with a Kalman filter. It was found that the cameras could track
the vehicles satisfactorily. Relevant plots from the experiment is shown in Figure 3.

6.2 Quadrotor

Due to space limitations we will only show a simple test with the ARdrone. This experimental test
was to make the ARdrone following an object. This object could for example be another UAV, but for
simplicity a red ball was used. The ARdrone faced the ball at all times while trying to keep a fixed
distance of one meter from the ball to itself. From the image stream of the camera, LabVIEW was used
to calculate the centroid of the red ball. The control algorithm then tried to keep the center of the ball
in the center of the camera image. This was done by yawing as a function of the error in x and increasing
lift speed as a function of the y-error (see Figure 4(a)). A Proportional Derivative (PD) controller was
used since the integrating part was removed since no steady state errors were encountered.

ISBN-xxx-y-zz-yyyyyyy-x 58
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(a) Camera view and error when tracking (b) Control of ARDrone to track a red ball
a red ball

Figure 4: Showing how the controller tries to keep the ball centered in the camera view in Fig. 4(a).
In Fig. 4(b) the blue dots show the calculated centroid pixel of the red ball. Light blue dots is later
in time then darker blue ones. As it can be seen, the UAV is able to keep the ball in the center of the
picture.

LabVIEW also calculated the area of the ball in the image and then used that to estimate the
distance of the ball to the camera. The viewed area A of an object at distance d decreases as the inverse
√
square of d. That is A ∝ d−2 , which gives d = 1/ kA, where k is a constant. Due to the oscillatory
behavior of the ARdrone, a low pass filter and an outlier filter had to be implemented.
As seen in Figure 4(b) the UAV was able to keep the ball in the center of its image. Also, it was
able to stay approximately one meter away from the ball, however, due to the relative high noise the
distance controller did not work as well as the one keeping the ball in the center of the image.

7 Conclusion
In this paper we propose an indoor testbed composed of unmanned aerial and ground vehicles. The
testbed is to be used for research in networked control systems as well as for educational purposes.
Even though the architecture and main components are defined, the testbed is still being developed
and several algorithms are being evaluated on it. These algorithms range from platooning of HDVs, to
coordinated control of the ARdrones to track HDVs. In the near future we plan to optimize the vision
system and install it for use in a larger test area to better handle flying agents. Moreover, we plan to
integrate Arducopters and other small low-cost holonomic and non-holonomic ground robots.

REFERENCES
[1] P. B. Sujit, J. B. de Sousa, and F. L. Pereira, “Coordination strategies between uav and auvs for
ocean exploration,” in European Control Conference, 2009.

[2] V. Kumar, D. Rus, and S. Singh, “Robot and sensor networks for first responders,” IEEE Pervasive
Computing, vol. 3, pp. 24–33, October 2004.

ISBN-xxx-y-zz-yyyyyyy-x 59
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[3] M. Boban, T. T. V. Vinhoza, J. Barros, M. Ferreira, and O. K. Tonguz, “Impact of Vehicles as

Obstacles in Vehicular Ad Hoc Networks,” IEEE Journal on Selected Areas in Communications,
vol. 29, no. 1, pp. 15–28, January 2011.

[4] K. Ro and J.-S. Oh, “Lessons learned: Application of small uav for urban highway traffic monitor-
ing,” in 45th AIAA Aerospace Sciences Meeting and Exhibit, Reno, NV, USA, 2007.

[5] A. Alam, A. Gattami, and K. Johansson, “An experimental study on the fuel reduction poten-
tial of heavy duty vehicle platooning,” in Intelligent Transportation Systems (ITSC), 2010 13th
International IEEE Conference on, sept. 2010, pp. 306 –311.

[6] Grand Cooperative Driving Challenge (GCDC), 2011. [Online]. Available: http://www.gcdc.net/

[7] N. Michael, J. Fink, and V. Kumar, “Experimental testbed for large multirobot teams,” Robotics
Automation Magazine, IEEE, vol. 15, no. 1, pp. 53 –61, march 2008.

[8] N. Michael, D. Mellinger, Q. Lindsey, and V. Kumar, “The grasp multiple micro-uav testbed,”
Robotics Automation Magazine, IEEE, vol. 17, no. 3, pp. 56 –65, sept. 2010.

[9] S. Bouabdallah, P. Murrieri, and R. Siegwart, “Design and control of an indoor micro quadrotor,”
in Robotics and Automation, 2004. Proceedings. ICRA ’04. 2004 IEEE International Conference
on, vol. 5, Apr. 2004, pp. 4393 – 4398 Vol.5.

[10] J. R. Martnez-de Dios, A. Jimnez, and A. Ollero, “Towards an integrating validation and simu-
lation tool for cooperating objects,” in Proceedings - 2nd International Workshop on Networks of
Cooperating Objects CONET2010. Stockholm, Sweden, 2010.

[11] Tamyia Scania R620 6X4 Highline Full Operation Kit. [Online]. Available: http://www.tamiya.
com/english/products/56322scania/index.htm

[12] Ardrone quadrotor. [Online]. Available: http://ardrone.parrot.com/

[13] J. Polastre, R. Szewczyk, and D. Culler, “Telos: enabling ultra-low power wireless research,” In-
formation Processing in Sensor Networks 2005., Apr. 2005.

[14] F. Bullo, J. Cortés, and S. Martı́nez, Distributed Control of Robotic Networks, ser. Applied Math-
ematics Series. Princeton University Press, 2009.

[15] D. Gay, P. Levis, and D. Culler, “Software design patterns for tinyos,” SIGPLAN Not., vol. 40,
no. 7, pp. 40–49, 2005.

[16] Products for USB Sensing and Control, Phidgets. [Online]. Available: http://www.phidgets.com/

[17] Arducopter. [Online]. Available: http://code.google.com/p/arducopter/

[18] NetCon, Networked Control Systems Group, Automatic Control Lab, KTH. [Online]. Available:
http://www2.ee.kth.se/web page/netcon/?p=811

10

ISBN-xxx-y-zz-yyyyyyy-x 60
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Aerial Object Following Using Visual Fuzzy Servoing

Miguel A. Olivares-Mendez, Iván F. Mondragón, Pascual Campoy, Luis Mejı́as*, Carol Martinez

Centro de Automática y Robótica (CAR) UPM-CSIC,C. José Gutiérrez Abascal 2, 28006 Madrid, Spain. *Australian
Research Centre for Aerospace Automation (ARCAA). Queensland University of Technology, Brisbane, Australia.

Email: [email protected] web: www.vision4uav.com

Abstract

This article presents a visual servoing system to follow a 3D moving object by a Micro Unmanned Aerial
Vehicle (MUAV). The presented control strategy is based only on the visual information given by an adaptive
tracking method based on the color information. A visual fuzzy system has been developed for servoing the
camera situated on a rotary wing MAUV, that also considers its own dynamics. This system is focused on
continuously following of an aerial moving target object, maintaining it with a fixed safe distance and centered
on the image plane. The algorithm is validated on real flights on outdoors scenarios, showing the robustness
of the proposed systems against winds perturbations, illumination and weather changes among others. The
obtained results indicate that the proposed algorithms is suitable for complex controls task, such object following
and pursuit, flying in formation, as well as their use for indoor navigation

Keywords- Visual Servoing, UAV, Object Following, Fuzzy Logic, Fuzzy Control, Soft Computing.

1 Introduction
Our research interest focuses on developing computer vision techniques to provide UAVs with an additional
source of information to perform visually guided tasks - this includes tracking and visual servoing, inspection,
pursuit and flying in formation among others.
Different works have been done where a vision system in conjunction with range sensor were used for Object
Following test. On [1] an omnidirectional visual systems is used as bearing sensor, while the distance to the target
is measured using a range sensor, for control a robotic wheelchair on indoors. Others systems have been proposed
only based on visual information for cooperative robotics [2]. Visual information also have been proposed on
aerial robotics for flying in formation [3]. Several approaches also have been proposed for fixed wind UAV flying
at constant altitude following circular paths, in order to pursuit a moving object on a ground planar surface [4],[5].
In the same way, several approaches have been proposed for rotary wind UAV following a terrestrial target [6],
[7].
Visual servoing also has been successfully implemented on aerial vehicles. Pose-based methods, in which is
necessary to estimate the 3D position, have been employed for applications like autonomous landing on moving

1
ISBN-xxx-y-zz-yyyyyyy-x 61
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

objects [8] or on fixed place without location information [9], while image-based methods have been used for po-
sitioning [10], generally assuming a fixed distance to the object, reducing the complexity of the derived controlled
and the necessity to estimate the reference depth.
The Soft Computing is getting more importance in the Robotics world in the last decade. These techniques are
used for a sort of uses in this researcher field. One of the most used Soft Computing techniques and proposal is the
Fuzzy Logic to create controllers. In contrast to conventional control, fuzzy logic control was initially introduced
as a model-free control design method based on a representation of the knowledge and the reasoning process
of a human operator. Fuzzy logic can capture the continuous nature of human decision processes and as such
is a definite improvement over methods based on binary logic (which are widely used in industrial controllers).
Hence, it is not surprising that practical applications of fuzzy control started to appear very quickly after the
method had been introduced in publications. Some of those applications of Fuzzy control and UAV are: for
visual servoing [11], for an autopilot using a Fuzzy PID control [12], an obstacle avoidance and path planning
system [13] and to manage a team of UAVs [14].
This paper presents a Fuzzy Servoing strategy using a real time flying objects following method based only
on visual information to generate commands in a Dynamic Look and Move control architecture. This work is
based on our previous visual control architecture developed for UAVs [15]. Section 2 presents the flying object
following problem statement. Section 3 explains how an adaptive color tracking method is used to identify and
track the target object on the image plane. Then this information is passed to the Fuzzy controllers to obtain the
commands to follow the flying objet from a safely distance as is presented on section 4. Section 5 shows the
UAV architecture and the fuzzy visual servoing system. Finally, section 6 shows the test results of the proposed
algorithm running onboard a UAV, validating our approach for an autonomous object flying following method
based on visual information.

2 Problem statement
Considering a flying object T moving with an unknown trajectory on the world space R3 , and a flying robot O
with an attached fixed calibrated pinhole camera, both having an idealized flying dynamics. The control goal
is to command the flying robot in order to track the target object, maintaining it always onto the camera FOV
with a fixed separation distance. Taking into account the Figure 1 and considering the target object as a 3D
spherical surface, it is projected on the camera image plane as a circular region that can be defined by its center
of projection xt = [xt , yt ]T and the circumference diameter øt . Because the target is an ideally spherical surface,
the projection point (xt ) can be considered as the image projection of target’s sphere centroid with coordinates on
the camera frame defines as XT c = [XT c ,YT c , ZT c ]T . The detected circumference (with a diameter of øt pixels) on
the image plane corresponds to the projections of the sphere perimeter (with a fixed diameter ØT ) that results of
the intersection of the plane which normal is parallel to the vector defined by the camera optical centre and the
sphere centroid, that divides the target in two hemispheres. The projected diameter also can be used to estimate
the distance to the target, because it is inversely proportional to the distance from the camera to the object.

2
ISBN-xxx-y-zz-yyyyyyy-x 62
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 1: UAV object Following setup in 2D. The control goal is to follow a target object (T ) with an unknown
trajectory using only the pose ((u, v)t ) and projection diameter (øt ) on the image plane. The objective is to
maintain the Target centered on the image plane with fixed distance or a proportional projection diameter on the
image

3 Detection Approach
Using cameras in outdoor environments is a challenging task. Sudden changes and inconsistencies with outdoor
illumination cause changes in the apparent colour as perceived by a camera. Here, it is described the details of
the tracking approach used for detect the target.
We approach the problem of tracking by exploiting the colour characteristic of the target. We define a basic
colour to the target by assuming a simple coloured mark to it and tracking this mark. Therefore, we rely on a
suitable and consistent colour representation that allows us to keep colour distributions derived from video image
sequences approximately constant (in outdoor settings). However, this process is not always perfect, and changes
still occur in colour distributions over time. An algorithm that has proven to deal with this issue by dynamically
adapting to changes in probability distributions is the Continuously Adaptive Mean Shift [16] (CamShift). This
algorithm is based in the mean shift originally introduced by Fukunaga and Hostetler [17].

(a) (b) (c) (d)

Figure 2: Camshift tracking of a colored red target on a image sequence. The white circle corresponds to the
boundaries of the tracked colored area

The Camshift algorithm is used to track a defined color on an image sequence, obtaining for each frame the

3
ISBN-xxx-y-zz-yyyyyyy-x 63
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

center of the color region and the the circumference that involves the tracked colored area. Figure 2 shows an
example of a color tracked sequence using the Camshift tracked of a red object.

4 Fuzzy Controllers
This Section presents the implementation of two Mamdani Fuzzy controllers which based on the visual informa-
tion previously described, generate yaw and pitch commands for the MUAV (velocity commands in degrees per
secods and milimeters per seconds).
The Fuzzy controllers were implemented using the MOFS (Miguel Olivares’ Fuzzy Software). This software
was used, previously, to implement fuzzy Controllers with other different platforms like a wheelchair [18] or on
an unmanned helicopter, where it was applied to control a pan and tilt visual platform onboard the UAV [19] and
for the autonomous landing of the aircraft [9]. With this software, it is possible to define easily fuzzy controllers
with the required number of inputs and select the type of membership functions, the defuzzification model and
the inference operator, from a sort of methods. A more detailed explanation of this software can be found in [20].
For this work two fuzzy controllers were defined. All the variables of the two controllers are defined using
triangular membership functions. The decision of this and the rest parts of the design of the controllers were
made based on the excellent results obtained in the previously mention works.
Both controllers have two inputs and one output. The desing of the controller of the yaw or heading of the
MUAV is shown in Figure 3. The first input of this controller is the angle estimation, in radians, between the
MUAV and the center of the image and the center of the object to follow (Figure 3(a)). The second input is the
difference between the last angle estimation and the actual angle (Figure 3(b)). This controller sends velocity
commands (degrees per seconds) for change the heading position of the aircraft (Figure 3(c)).

(a) First input. Estimation of the deviation of the object (b) Second input. Difference between the last two mea-
from the centre of the image capture from the MUAV. sures.

(c) Output. Velocity commands to change the heading of

the MUAV

Figure 3: Definition of the Yaw controller.

The second controller acts on the pitch state of the MUAV as is shown in Figure 4. The controller take the
data about the size of the object, in pixels, to follow and estimate the distance. Using as the first input the actual

4
ISBN-xxx-y-zz-yyyyyyy-x 64
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

size of the object (Figure 4(a)), and as the second one the difference between the last size measure and the actual
one (Figure 4(b)). The ouput of the controller consists in velocity commands to go ahead, in the case that the
object is far away (Figure 4(c)). Stay in the same position if the object is near at a predefined distance of segurity,
or go back if it is very close to the MUAV.

(a) First input. Estimation of the distance between the (b) Second input. Difference between the last two mea-
object and the MUAV. sures.

(c) Output. Velocity commands to keep the safe distance

to the object.

Figure 4: Definition of the Pitch controller.

The defuzzification process is made using the product model of inference and the height weight defuzification
model as is shown in the Eq. 1.

∑M l
l=1 y ∏ µB0 (y )
l
y=
(1)
∑M
l=1 ∏ µB0 (ȳ )
l

5 UAV platfom
We performed experimental tests using a Pelican quadrotor [21] and a moving colored target. The testbed shown
in Figure 5 has a low-level stability controller based on PID that uses information from GPS, IMU, pressure
altimeter and magnetometer fused using a Kalman filter. This controller is embedded, closed, unmodifiable but
gains are tunable. Onboard vision processing is achieved using a dual core Atom 1.6 GHz processor with 1 GB
RAM, wireless interface and support for several types of USB cameras (mono or stereo). This computer runs
Linux OS working in a multi-client wireless 802.11(a,b,g) ad-hoc network, allowing it to communicate with a
ground station PC used for monitoring and supervision.
The fuzzy controllers developed previously are implemented on the Pelican UAV using a dynamic look-and-
move servoing architecture as is presented on figure 6. In this scheme, the velocity references generated by the
controller (running onboard aircraft) are used as a input references for the Pelican Low Level controlled. This
low level controller, when is operated on position control, allows to get as input, velocity commands, as well as

5
ISBN-xxx-y-zz-yyyyyyy-x 65
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 5: CVG-UPM [22] Pelican QuadCopter testbed used for sense and avoid experiments

direct control actions. For this test, the velocity commands generated by the visual system are directly send as
input velocities for the Autopilot. This autopilot also allow to independently control the X, Y, Z and Yaw. Roll
and Pitch angles are no directly controlled by velocity command, but it is possible to control it by means of motor
direct control. So, in this control architecture, and considering that the camera is looking forward, the quadrotor
X and Z as well as the yaw angle will be controlled by the generated references (VXq = VZc ,VZq = VY c , ωZq = ωY c )
. The Y axis and Pitch, Roll angles are controlled by the low level autopilot.

Figure 6: Object Following Dynamic image-based look-and-move system architecture.

6 Test and results

Several tests have been done which different balloons used as the target object (in order to reduce the complexity
of an aerial moving object). These balloons are moved with a random trajectory on the 3D space. The visual
system composed of the Camshift color tracked and the Fuzzy Controler running onboard the Pelican quadcopter
(Figure 5) in a Intel Atom Board PC. The velocity commands are send to the low level autopilot through a serial
interface. Figure 7 shows a external view of a Object following tests and the onboard images used for the fuzzy
controller.
Here one of these flight-tests is presented. Figure 8 shows the trajectory that was made by the MUAV and
some captions made by the onboard camera. This trajectory was reconstructed using the GPS data of the aircraft.
In this case a red balloon was followed during almost two minutes.
The Figure 9(a) shows the estimation of the angle between the red balloon, the MUAV and the centre of

6
ISBN-xxx-y-zz-yyyyyyy-x 66
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(a) (b)

Figure 7: Object Following test using a fuzzy controller on a Pelican quadrotor:(a) External view of the tests using
a colored balloon. The balloon is manually moved with a random trajectory, (b). The onboard image captured,
and the projected diameter on the image plane identified using the Camshift color tracked.

Figure 8: Trajectory of the Pelican-MUAV during an object following test.

the image. This measure is obtained by the visual algorithm in radians, and was introduced to the first fuzzy
controller as the first input. The difference between the last measure of this angle and the actual one is introduced
as the second input. As is shown in this Figure, the controller keep the object in the centre of the image, despite
the effects of the wind and the random trajectory. Some captions of the most critical moments are include in the
Figure.
In Figure 9(b) is shown the output values of the yaw controller of the MUAV.
Figure 10(a) shown the measure of the balloon during the test. The size of the balloon gives the information
to the controller to go ahead, when the object it is far away or to go back when is so near. Some captions of the
most representing movements are include in this Figure. The first picture represents the begin of the test, in where
the four rotors were started. The second one shows the normal situation in where the object has the predefined
size. To test the reverse movements of the aircraft, the balloon was moved against the MUAV, that is shoed on the
third. The fourth picture shows how the MUAV was recovered from the previous situation.
Figure 10(b) shows the response of the Pitch controller to follow the object from a safe distance.

7
ISBN-xxx-y-zz-yyyyyyy-x 67
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(a) Measure of the angle between the MUAV, the object to (b) Output of the Yaw Fuzzy Controller.
follow and the centre of the image.

Figure 9: Measures for the orientation of the MUAV and response of the Yaw controller.

(a) Measure of the size of the red balloon during all the test. (b) Output of the Pitch Fuzzy Controller.

Figure 10: Measures for the orientation of the MUAV and response of the Yaw controller.

7 Conclusion
In this paper a Fuzzy visual servoing system for flying object following have been presented. The visual infor-
mation is provided by an adaptive color tracking algorithm based on CamShift. The fuzzy controllers acts on the
yaw and the pitch of a Micro-UAV, in order to keep the aircraft from a safe distance to the object to follow and
mantain it in the centre of the image.
Real tests on outdoors scenarios have been made to validate the proposed method and the fuzzy controllers.
The excellent results of these tests have demonstrated the robustness against the weather, wind and light variations
of the visual algorithm. The proposed visual algorithm is based on color detection and could be used to detects
objects with another color and different shape, not just red balloons. Also, these results prove the excellent
behavior of the fuzzy controllers performing the following action of the target objective from a safe distance.
In the future works, a height control of the MUAV is in the initial phase. Also, the uses of this algorithm and
controller for an obstacles avoidance tasks with fixed objects and other MUAVs have to be done.

8
ISBN-xxx-y-zz-yyyyyyy-x 68
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Acknowledgment
The work reported in this paper is the consecution of several research stages at the Computer Vision Group -
Universidad Politécnica de Madrid and at the Australian Reasearch Centre for Aerospace Automation - Queens-
land University of Technology (ARCAA-QUT). The authors would like to thank the Universidad Politécnica de
Madrid, the Consejerı́a de Educación de la Comunidad de Madrid and the Fondo Social Europeo (FSE) for some
of the Authors PhD Scholarship, the Australian Research Centre for Aerospace Automation and the European
Commission CORDIS. This work has been sponsored by the Spanish Science and Technology Ministry under the
grant CICYT DPI2010-20751-C02-01 and by the IRSES Program Marie Curie FP7 - PIRSES-GA-2009-230797
- ICPUAS (International Cooperation Program for Unmanned Aerial Systems Research and Development).

REFERENCES
[1] A. Argyros, P. Georgiadis, P. Trahanias, and D. Tsakiris, “Semi-autonomous navigation of a robotic
wheelchair,” Journal of Intelligent and Robotic Systems, vol. 34, pp. 315–329, 2002.

[2] A. Korodi, A. Codrean, L. Banita, and C. Volosencu, “Aspects regarding the object following control pro-
cedure for wheeled mobile robots,” WSEAS Trans. Sys. Ctrl., vol. 3, pp. 537–546, June 2008.

[3] A. Betser, P. Vela, G. Pryor, and A. Tannenbaum, “Flying in formation using a pursuit guidance algorithm,”
in American Control Conference, 2005. Proceedings of the 2005, june 2005, pp. 5085 – 5090 vol. 7.

[4] A. Savkin and H. Teimoori, “Bearings-only guidance of an autonomous vehicle following a moving target
with a smaller minimum turning radius,” in Decision and Control, 2008. CDC 2008. 47th IEEE Conference
on, dec. 2008, pp. 4239 –4243.

[5] K. S. Fahd Rafi, Saad Khan and M. Shah, “Autonomous target following by unmanned aerial vehicles,” in
Proc. SPIE, Orlando (Kissimmee), FL, USA, April 2006.

[6] H. Deng, X. Zhao, and Z. How, “A vision-based ground target tracking system for a small-scale autonomous
helicopter,” Image and Graphics, International Conference on, vol. 0, pp. 739–743, 2009.

[7] F. Lin, B. Chen, K. Y. Lum, and T. Lee, “A robust vision system on an unmanned helicopter for ground
target seeking and following,” in Intelligent Control and Automation (WCICA), 2010 8th World Congress
on, july 2010, pp. 276 –281.

[8] S. Saripalli and G. S. Sukhatme, “Landing a helicopter on a moving target,” in Proceedings of IEEE Inter-
national Conference on Robotics and Automation, Rome, Italy, April 2007, pp. 2030–2035.

[9] M. Olivares-Mendez, I. Mondragon, P. Campoy, and C. Martinez, “Fuzzy controller for uav-landing task
using 3d-position visual estimation,” in Fuzzy Systems (FUZZ), 2010 IEEE International Conference on,
2010, pp. 1 –8.

9
ISBN-xxx-y-zz-yyyyyyy-x 69
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[10] O. Bourquardez, R. Mahony, N. Guenard, F. Chaumette, T. Hamel, and L. Eck, “Image-based visual servo
control of the translation kinematics of a quadrotor aerial vehicle,” Robotics, IEEE Transactions on, june
2009.

[11] M. A. Olivares-Mendez, P. Campoy, C. Martinez, and I. Mondragon, “Visual servoing using fuzzy con-
trollers on an unmanned aerial vehicle,” Eurofuse workshop 09, Preference modelling and decision analysis,
Sept. 2009.

[12] Y. Shengyi, L. Kunqin, and S. Jiao, “Design and simulation of the longitudinal autopilot of uav based on
self-adaptive fuzzy pid control,” in Computational Intelligence and Security, 2009. CIS ’09. International
Conference on, vol. 1, Dec. 2009, pp. 634–638.

[13] T. Dong, X. Liao, R. Zhang, Z. Sun, and Y. Song, “Path tracking and obstacles avoidance of uavs - fuzzy
logic approach,” in Fuzzy Systems, 2005. FUZZ ’05. The 14th IEEE International Conference on, May 2005.

[14] J. Smith and T. Nguyen, “Fuzzy logic based resource manager for a team of uavs,” in Fuzzy Information
Processing Society, 2006. NAFIPS 2006. Annual meeting of the North American, June 2006, pp. 463–470.

[15] P. Campoy, J. F. Correa, I. Mondragón, C. Martı́nez, M. Olivares, L. Mejı́as, and J. Artieda, “Computer
vision onboard UAVs for civilian tasks,” Journal of Intelligent and Robotic Systems., no. 1-3, 2009.

[16] G. R. Bradski, “Computer vision face tracking for use in a perceptual user interface,” Intel Technology
Journal, no. Q2, 1998. [Online]. Available: http://citeseer.ist.psu.edu/585206.html

[17] K. Fukunaga and L. Hostetler, “The estimation of the gradient of a density function, with applications in
pattern recognition,” Information Theory, IEEE Transactions on, vol. 21, no. 1, pp. 32 – 40, Jan. 1975.

[18] M. Olivares and J. Madrigal, “Fuzzy logic user adaptive navigation control system for mobile robots in
unknown environments,” Intelligent Signal Processing, 2007. WISP. IEEE International Symposium on.

[19] M. A. Olivares-Mendez, P. Campoy, C. Martinez, and I. Mondragon, “A pan-tilt camera fuzzy vision con-
troller on an unmanned aerial vehicle,” IEEE/RSJ International Conference on Intelligent Robots and Sys-
tems 2009. IROS09, October 2009.

[20] I. F. Mondragón, M. A. Olivares-Mendez, P. Campoy, and C. Martinez, “Unmanned aerial vehicles uavs
attitude, height, motion estimation and control using visual systems,” Autonomous Robots, vol. In Press,
Corrected Proof, pp. –, 2010.

[21] AscendingTechnologies, “Ascending technologies,” http://www.asctec.de, 2010. [Online]. Available:

http://www.asctec.de

[22] CVG-UPM, “Universidad Politécnica de Madrid. Computer Vision Group. Vision for UAV Project,”
http://www.vision4uav.com, 2010.

10
ISBN-xxx-y-zz-yyyyyyy-x 70
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

On the Use of UAVs for Vessel Inspection Assistance∗

Alberto Ortiz, Francisco Bonnin-Pascual and Emilio Garcia-Fidalgo

Dep. of Mathematics and Computer Science, University of Balearic Islands

Cra. Valldemossa, km 7.5, 07122 Palma de Mallorca, Spain
email: [email protected], [email protected], [email protected]

Abstract

Vessel maintenance entails periodic visual inspections of internal and external parts of the vessel
hull in order to detect the typical defective situations affecting metallic structures, such as cracks,
coating breakdown, corrosion, etc. The main goal of the EU-FP7 project MINOAS is the automation
of the inspection process, currently undertaken by human surveyors, by means of a fleet of robotic
agents. This paper overviews a UAV to be used as part of this fleet, and succinctly describes its
control architecture as well as a self-localization solution for this vehicle. Promising experimental
results are discussed in the experimental results section.

keywords: Vessels maintenance, UAV, visual inspection, visual odometry

1 INTRODUCTION
The movement of goods by ships is today one of the most time and cost effective methods of transporta-
tion. The safety of these vessels is overseen by the classification societies, who are continually seeking
to improve standards and reduce the risk of maritime accidents. Structural failures are a major cause of
accidents, and can usually be prevented through timely maintenance. As such, vessels undergo annual
inspections, with intensive Special and Docking Surveys every five years, which ensures that the hull
structure and related piping are all in satisfactory condition and are fit for the intended use over the
next five years.
To illustrate the enormity of the inspection task, the surveying of a central cargo tank on a very large
crude carrier (VLCC), involves checking over 860m of web frames (primary stiffening members) and
approximately 3.2km of longitudinal stiffeners. Furthermore, this surveying is performed in a potentially
hazardous environment with both flammable and toxic gases and significant heights involved. As a
result, although accidents are extremely rare, when they do arise they can have serious consequences.
Due to these complications, the total cost of a single surveying can exceed $1M once you factor in the
∗ This work is partially supported by projects SCP8-GA-2009-233715 (EU FP7 MINOAS), AAEE 0108/09 (Conselleria
d’Innovacio, Interior i Justicia, Govern Balear), and FEDER funding.

1
ISBN-xxx-y-zz-yyyyyyy-x 71
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 1: (left) Staging required during a vessel inspection. (right) Oil tanker in shipyard during
construction.

vessel’s preparation, use of yard’s facilities, cleaning, ventilation, and provision of access arrangements
(see figure 1[left]). In addition, the owners experience significant lost opportunity costs while the ship
is inoperable.
1
The main objective of the EU FP7 project MINOAS is the effective virtual teleportation of the
surveyor to the different areas of the vessel hull that need inspection, so that a reduction in the inspection
time and the costs involved, as well as an increase in the safety of the operation, can be effectively
achieved (see [1] for a detailed discussion).
Contrary to similar past projects (ROTIS and ROTIS-II), the scope of MINOAS comprises both dry
and wet areas of the vessel, and not only flooded ballast tanks or the external hull. The MINOAS project
is neither limited to tele-operated floating tethered vehicles, but considers a varied set of robotic tech-
nologies with different locomotion capabilities, including magnetic crawlers, remotely operated vehicles
(ROV) and unmanned aerial vehicles (UAV).
This paper presents the aerial platform adopted within the context of MINOAS, as well as discusses
on the self-localization requirements of the application and how they are planned to be solved. As
part of the development of the navigation strategy, an almost-closed-form solution to visual odometry
is described, together with a set of experimental results which have been obtained from the different
experiments performed.
The rest of the paper is structured as follows: section 2 discusses on the requirements imposed by
the application and describes the UAV solution finally adopted, section 3 describes the intended control
architecture and, particularly, focuses on the navigation strategy, section 4 describes a monocoluar
odometer that has been developed as part of the localization strategy, and, finally, section 5 provides
experimental results about the performance of this odometer.
1 Marine INspection rObotic Assistant System, http://www.minoasproject.eu

2
ISBN-xxx-y-zz-yyyyyyy-x 72
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2 PLATFORM DESCRIPTION
As part of the MINOAS concept, the aerial platforms are intended to provide detailed surveys of the
vertical structures that can be found in vessel holds (see figure 1). Therefore, the main requirements stem
directly from the very nature of the inspection process: the vehicle must be able to perform vertical,
stationary and low speed flight, as well as permit indoor flight. These requirements rapidly discard
fixed-wing aircrafts and focus the search on helicopter-type UAVs, naturally capable of manoeuvres
such as hovering and vertical take-off and landing (VTOL).
Among the different kinds of helicopter designs that have been proposed, multi-rotor configurations
present several advantages over comparably scale helicopters (see, for instance, [2]): (1) they do not
require mechanical linkages to vary rotor angle of attack as they spin, what simplifies the design of the
vehicle and reduces maintenance time and cost; (2) the use of several rotors allows each individual rotor
to have a smaller diameter than the equivalent helicopter rotor, for a given vehicle size; and (3) flight is
safer than for other helicopters because the small rotor size make them store less kinetic energy during
flight, what reduces the damage in case the rotors hit any object.
Among other multi-rotor UAVs, the four-rotor, or quadrotor, is emerging as the most popular multi-
rotor configuration. This kind of vehicle consists of four rotors in total, with two pairs of counter-rotating,
fixed-pitch blades located at the four corners of the aircraft. In this platform, the control of the vehicle
motion is achieved by varying the relative speed of each rotor. Moreover, because each pair of rotor
blades spin in opposite directions, they cancel out any torque, keeping the helicopter flying straight. As
a result, precise flight and stable hovering can be achieved. Finally, counter rotating propellers increase
efficiency and flight times, as no extra thrust is needed to compensate for unwanted rotation.
A Pelican quadrotor from Ascending Technologies is the platform finally chosen to implement the
aerial vehicle for the MINOAS project. This is a 50cm-diameter micro aerial vehicle (MAV) with 10-
inch propellers, able to carry a payload of 500g, and fitted with an inertial measuring unit (IMU) and
a barometric pressure sensor. The vehicle is provided with attitude stabilization running on an ARM7
microprocessor [3].
Furthermore, for the application at hand, the MAV has been outfitted with the lightweight Hokuyo
URG-04LX-UG01 laser scanner, a front-looking stereo rig comprising two lightweight uEye 1226-LE-
C cameras and a third uEye 1226-LE-C bottom looking unit. For onboard processing, the vehicle
carries a Kontron pITX-SP board equipped with an Intel Atom 1.6GHz processor and 2GB RAM. See
figure 2(left) for a picture of the MAV in its current configuration. As can be observed, some of the
beams of the laser scanner are deflected by two mirrors on the right and the left sides of the quadrotor
to provide two height estimates.

3
ISBN-xxx-y-zz-yyyyyyy-x 73
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 2: (left) MAV to be used in project MINOAS, in its current configuration. (right) The UAV con-
trol architecture expressed by means of GLOC3 generic components. Although not explicitly indicated,
all the components are connected to the communication bus. Dotted lines represent logical connections
between components. (FDM stands for feature detection and matching.)

3 ABOUT THE CONTROL ARCHITECTURE

As part of the inspection process, the UAV is expected to achieve a set of waypoints for acquiring images
of the underlying metallic structures to provide the surveyor with an overall view of the state of the
vessel. Those images must as well be tagged with pose information, so that the same UAV, or any
other member of the robot fleet, can, on demand of the surveyor, re-visit the area for acquiring close-up
images or else taking other sensor data, e.g. measure the thickness of the hull.
Hence, the motion capabilities of the MINOAS flying robot essentially comprise waypoint navigation
and automatic take-off and landing. To this end, for localization purposes, the platform is intended to
implement a navigation strategy based on combining, by selection or by fusion, visual odometers using
the front-looking stereo vision system (SVS) and the ground-looking camera (GLC), as well as a laser
scan matching-based odometer (LS). The robot pose estimation produced is finally complemented with
the estimation provided by an external optical follower able to track a vehicle fitted with an ultra-bright
LED, and comprising a camera and a laser pointer (see [4] for a first version of this device, which is
currently being used to track a magnetic crawler). This redundant strategy is intended to provide robust
positioning information able to tolerate the failure of any of the positioning subsystems in case of the
vehicle getting out of the line of sight of the optical tracker or in case the vehicle motion cannot be
estimated from the available sensor data, i.e. lack of features when imaging non-textured surfaces or
lack of distinctive structures within laser scans in long straight walls. Although the vehicle is supported
by a base station, due to the lossy nature of wireless communications, images are not transmitted, but
are processed on-board; subsequent calculations can, however, be performed at the base station.
The control software is implemented following the Generic LOosely-Coupled Component-based Con-
trol software architecture (GLOC3) [5], while the Robot Operating System (ROS) is used as the mid-
dleware for software distribution. Figure 2(right) shows the control architecture for the whole system
expressed in terms of the generic components of GLOC3.

4
ISBN-xxx-y-zz-yyyyyyy-x 74
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 3: Coordinate frames assumed and symbols used by the visual odometer.

4 A VISUAL ODOMETRY SOLUTION

As part of the effort on the development of the UAV localization strategy for project MINOAS, this
section describes a fast visual odometer for motion estimation using local image features for image
point matching between camera frames and the M-estimator SAmple and Consensus (MSAC) [6], a
RANSAC variant, for rejecting matchings disagreeing with an inter-frame homography transformation.
Due to its robustness to motion blur and large displacements, we use the pyramidal implementation of
the Kanade-Lucas-Tomasi (KLT) optical flow tracker [7] for feature tracking across images. Features
are first located by means of the good-features-to-track detector (OpenCV implementation), replacing
lost features as needed. The whole feature detection and tracking solution provides an adequate level
of performance at reduced execution times, compatible with the computational power of the onboard
processor.
Briefly speaking, once a set of matched features is available, the method backprojects features into
the world using previous knowledge of the camera position with respect to the center of the vehicle
(l1 , l2 , l3 ), the camera focal distance f , the camera tilt τ , and the vehicle height zv and pitch φ and
roll γ angles, all three supplied by onboard sensors. The odometer finally becomes into a least squares
problem in almost-closed-form solution that allows estimating motion in x, y and yaw, ∆x, ∆y and ∆ψ.
Figure 3 illustrates the different symbols involved in the derivation of the visual odometer.
(2
According to figure 3, equation 1 transforms from world (x, y, z) to image coordinates (xp , zp ) :
C
TR
z }| {
      
xp 1 0 0 0 1 0 0 0 1 0 0 0 1 0 0 −l1
      
yp  0 1 0 0 0 1 0 −(l3 + f ) 0 cτ sτ 0 0 1 0 −l2 
      
 =    
 zp  0 0 1 0   0 −sτ 0  0 
    0 0 1 0  cτ  0 0 1 
1
1 0 f
0 1 0 0 0 1 0 0 0 1 0 0 0 1
| {z } | {z }
P Rx (−τ )

2 In the following, cα and sα stand for, respectively, cos α and sin α.

5
ISBN-xxx-y-zz-yyyyyyy-x 75
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

R R
TW b TW a
z }| {z }| {
     
cγ 0 sγ 0 1 0 0 0 cψ sψ 0 0 1 0 0 −xv x
     
 0 0  0  0  −yv   
 1 0  0 cφ sφ  −sψ cψ 0  0 1 0  y 
      (1)
−sγ 0 cγ 0  −sφ 0  0  −zv   
  0 cφ  0 0 1  0 0 1  z 
0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 1 1
| {z }| {z }| {z }
Ry (−γ) Rx (−φ) Rz (−ψ)

where yp turns out to be a free parameter that determines the plane where the 2D point (xp , zp ) is
located, so that yp = −∞ and yp = 0 correspond to, respectively, the lens center and the sensor plane.
The corresponding inverse perspective transformation is thus:
 −1  −1  −1
(x, y, z, 1)T = TW
R
a
R
TW b
C
TR P −1 (xp , yp , zp , 1)T (2)

Equation 2 next becomes into the following 3D line in world coordinates:

       
x x − M2 f cψ + M6 f sψ (M x + M2 f + M3 zp + M4 )cψ − (M5 xp + M6 f + M7 zp + M8 )sψ
   v    1 p 
       
y  =yv  + − M6 f cψ − M2 f sψ  + λ (M5 xp + M6 f + M7 zp + M8 )cψ + (M1 xp + M2 f + M3 zp + M4 )sψ (3)
       
z zv − M10 f M9 xp + M10 f + M11 zp + M12
 
where: M1 M2 M3 M4
 
M5 M6 M7 M8   −1  −1
 
  = TW R
b
C
TR P −1
M9 M10 M11 M12 
 
0 − f1 0 1

For image features corresponding to points lying on the floor, z = 0 and consequently:
zv − M10 f
λ=− (4)
M9 xp + M10 f + M11 zp + M12
Accordingly, at time k, equation 3 can be written as follows:
   (k)  (k)  (k)   (k)  (k)  (k)  (k)
x x  C −D A −B cψ xv C +λA −(D + λ B) cψ
  =  v +   + λ(k)      =  +   
y yv D C B A sψ yv D + λB C + λA sψ

(k) (k)
(5)
zv − M10 f
with: λ(k) = − (k) (k) (k) (k)
M9 xp + M10 f + M11 zp + M12
(k) (k) (k) (k) (k) (k)
A = M1 xp + M2 f+ M 3 zp + M4 C (k) = − M2 f
(k) (k) (k) (k) (k) (k) (k)
B = M5 xp + M6 f+ M 7 zp + M8 D = − M6 f
(k) (k) (k+1)
At time k + 1, if the vehicle has rotated in yaw ∆ψ, has moved at (xv + ∆xv , yv + ∆yv , zv ),
and changed from pitch φ(k) to φ(k+1) and from roll γ (k) to γ (k+1) , for the same scene point at world
coordinates (x, y, z = 0), we have:
     (k+1)  (k+1)   
(k) (k) + ∆ψ)
x x + ∆xv C −D (k+1)  A −B  c(ψ
 = v +    +λ    (6)
(k)
y yv + ∆yv D C B A s(ψ(k) + ∆ψ)
   (k+1)  (k+1)   
(k) (k) (k)

xv + ∆xv
 C −D (k+1) 
A −B
  cψ c∆ψ − sψ s∆ψ
= (k)
+ +λ 
yv + ∆yv D C B A sψ(k) c∆ψ + cψ(k) s∆ψ
    
(k)
xv + ∆xv (C + λ A)(k+1) cψ(k) − (D + λ B)(k+1) sψ(k) − (D + λ B)(k+1) cψ(k) − (C + λ A)(k+1) sψ(k) c∆ψ
=  (k) +  .
yv + ∆yv (D + λ B)(k+1) cψ(k) + (C + λ A)(k+1) sψ(k) (C + λ A)(k+1) cψ(k) − (D + λ B)(k+1) sψ(k) s∆ψ
Combining equations 5 and 6 we can write:
 (k)
(C + λ A)cψ − (D + λ B)sψ
  =
(D + λ B)cψ + (C + λ A)sψ
    
∆x (C + λ A)(k+1) cψ(k) − (D + λ B)(k+1) sψ(k) − (D + λ B)(k+1) cψ(k) − (C + λ A)(k+1) sψ(k) c∆ψ
 v +    (7)
∆yv (D + λ B)(k+1) cψ(k) + (C + λ A)(k+1) sψ(k) (C + λ A)(k+1) cψ(k) − (D + λ B)(k+1) sψ(k) s∆ψ

6
ISBN-xxx-y-zz-yyyyyyy-x 76
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Equation 8 can be next obtained rearranging equation 7:

 
∆xv
  
1 0 (C + λ A)(k+1) cψ(k) − (D + λ B)(k+1) sψ(k) − (D + λ B)(k+1) cψ(k) − (C + λ A)(k+1) sψ(k)  ∆yv 
 


=
0 1 (D + λ B)(k+1) cψ(k) + (C + λ A)(k+1) sψ(k) (C + λ A)(k+1) cψ(k) − (D + λ B)(k+1) sψ(k) c∆ψ 


s∆ψ
 (k)
(C + λ A)cψ − (D + λ B)sψ
  (8)
(D + λ B)cψ + (C + λ A)sψ

A least squares framework can be devised now from equation 8 to determine ∆xv , ∆yv , c∆ψ and
s∆ψ if for a number of ground points (xi , yi , zi = 0), i = 1, . . . , n, their image projections (xp,i , zp,i )
have been matched between frames k and k + 1 (3 :
   2
∆xv
   
  ∆yv  
   
min U  −V (9)
∆xv  c∆ψ  
   
∆yv
c∆ψ, s∆ψ
s∆ψ

where:
 (k+1)  (k)
1 0 P1 −Q1 P1
   
0 1 Q P1  Q  (k+1)


1 

 1
  Pi = (Ci + λi Ai )(k+1) cψ(k) − (Di + λi Bi )(k+1) sψ(k)
1 0 P −Q2  P 
 2   2 (k+1)
Qi = (Di + λi Bi )(k+1) cψ(k) + (Ci + λi Ai )(k+1) sψ(k)
   
U =
0 1 Q2 P2   , V =
 Q2 
 , (10)
    (k)
 . . . . . . . . . . . . . . . . . . . . . Pi = (Ci + λi Ai )(k) cψ(k) − (Di + λi Bi )(k) sψ(k)
   
    (k)
1 0 Pn −Qn   Pn  Qi = (Di + λi Bi )(k) cψ(k) + (Ci + λi Ai )(k) sψ(k)
   
0 1 Qn Pn Qn

By means of the pseudo-inverse U + = (U T U )−1 U T , the solution to the least squares problem is:
X (k+1) X
    (k)
A= Pi D= Pi
∆xv AF − CD − BG X
    (k+1) X (k)
 ∆yv   AG + BF − CE  B= Qi E= Qi
   
 =  /(A2 + B2 − nC) X  (k+1) 2  (k+1) 2 X
c∆ψ  AD − nF + BE  (k+1) (k) (k+1) (k)
    C= Pi + Qi F = Pi Pi + Qi Qi
s∆ψ AE − nG − BD X (k+1) (k) (k+1) (k)
G= Pi Qi − Qi Pi
(11)

A further non-linear refinement step follows next to ensure c∆ψ 2 +s∆ψ 2 = 1, which is not guaranteed
by the least squares solution. To this end, equation 9 is re-arranged as a non-linear optimization
problem aiming at minimizing for ∆xv , ∆yv , ∆Ψ and the Levenberg-Marquardt algorithm is used [8].
The odometer can be globally regarded as an almost-closed-form solution because experiments have
shown that a few iterations (2-5) are enough to attain the minimum.
A standard Kalman filter assuming constant acceleration constitutes the final stage of the odometer.
The state is given by (ẋ, ẍ)T = (ẋv , ẏv , ψ̇, ẍv , ÿv , ψ̈)T and the measure by ∆x = (∆xv , ∆yv , ∆ψ)T :

A
 t z }| {  t−1 H  t
ẋ I ∆t I ẋ h it hz }| i{ ẋ
  =   +Q, ∆x = ∆t I 0   + R (12)
ẍ 0 I ẍ ẍ
3 Observe that two matchings are enough to estimate ∆xv , ∆yv , c∆ψ and s∆ψ.

7
ISBN-xxx-y-zz-yyyyyyy-x 77
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Table 1: Average absolute error in vehicle pose: xv (2nd col), yv (3rd col), ψ (4th col), execution time
(5th col) and frames processed per second (6th col).
# frames ǫxv (cm) ǫyv (cm) ǫψ (o ) ms fps
DS1 641 6.17 8.43 0.66 11.61 86.15
12.10 5.91 2.75 10.82 92.38
DS2 1681 8.56 8.61 0.56 10.99 90.98
11.32 19.22 7.11 10.12 98.78
DS3 2401 15.37 7.58 0.57 10.77 92.82
11.65 30.09 10.39 9.98 100.21
DS4 2041 6.50 4.41 0.25 12.16 82.21
6.91 7.84 3.94 9.30 107.56

To finish, figure 4 summarizes the structure of the visual odometer.

5 EXPERIMENTAL RESULTS AND CONCLUSIONS

This section reports the performance results obtained for the visual odometry approach that has been
described. To this end, the publicly available datasets 1LoopDown (DS1), 2LoopsDown (DS2), 3Loops-
4
Down (DS3) and hoveringDown (DS4) made public by the EU-FP7 project sFly have been used. The
datasets were collected from a quadrotor flying 1, 2 and 3 loop sequences and hovering within a space of
approximately 1m × 1m × 1m. These datasets comprise ground truth from a Vicon system and images
from a ground-looking camera, as well as height and IMU data (accelerations, attitude rates, absolute
angles and absolute headings). See [9] for a detailed description of the datasets.
Figure 5 and table 1 provide quantitative data from the performance evaluation. For every dataset
DSi, the true height provided by the Vicon was used, and the experiment was run twice: first feeding
the odometer with the roll, pitch and yaw angles from the ground truth (each first row of the table), and
second emulating more realistic conditions, in which the pitch and roll angles came from the IMU, while
the yaw angle, although available, was not used (each second row of the table). As can be observed,
best results are in general obtained if the odometer is provided with the yaw angle, although the rule is
4 http://projects.asl.ethz.ch/sfly/doku.php?id=mav datasets

Figure 4: Summary of the visual odometer.

8
ISBN-xxx-y-zz-yyyyyyy-x 78
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

200 200
visual odometer visual odometer
ground truth ground truth

150 150

DS1 100 100

Y (cm)

Y (cm)
50 50

0 0

−50 −50

−100 −100
−200 −100 0 100 200 300 400 500 −200 −100 0 100 200 300 400 500
X (cm) X (cm)

DS1 DS2 DS3 DS4

500 500 500 200

400 400 400

150
X (cm)

300 300 300

100

200 200 200

50

100 100 100

0
0 0 0

−50
−100 −100 −100

−200 −200 −200 −100

0 100 200 300 400 500 600 700 0 200 400 600 800 1000 1200 1400 1600 1800 0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500

200 200 200 100

150 150 150

50
Y (cm)

100 100 100

50 50 50

−50

0 0 0

−100
−50 −50 −50

−100 −100 −100 −150

0 100 200 300 400 500 600 700 0 200 400 600 800 1000 1200 1400 1600 1800 0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500

500 500 500 200

400 400 400

150
X (cm)

300 300 300

100

200 200 200

50

100 100 100

0
0 0 0

−50
−100 −100 −100

−200 −200 −200 −100

0 100 200 300 400 500 600 700 0 200 400 600 800 1000 1200 1400 1600 1800 0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500

200 200 200 100

150 150
150
50
Y (cm)

100 100
100

0
50 50

50

0 0
−50

0
−50 −50

−100
−50
−100 −100

−100 −150 −150 −150

0 100 200 300 400 500 600 700 0 200 400 600 800 1000 1200 1400 1600 1800 0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500

Figure 5: [top] Plots of xy estimations for DS1, feeding (left) and not feeding (right) the VO with true
yaw. [bottom] Plots of the x and y estimations against the respective ground truth for the different
datasets, feeding (upper) and not feeding (lower) the VO with true yaw. The green lines correspond to
the ground truth and the red lines to the estimation.

9
ISBN-xxx-y-zz-yyyyyyy-x 79
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

not always met. In any case, the error in x is never larger than 15 cm, the error in y is never larger than
30 cm and, finally, the error in ψ is less than 10◦ . Regarding the execution time, on average it is around
10.5 ms per frame, or approximately 95 frames per second. These times correspond to a dual-core Intel
Core i5 @2.53GHz processor.

6 ACKNOWLEDGEMENTS
The authors are very thankful to Kontron and Quality By Measurement (QBM) for supplying the
onboard Intel Atom board.

REFERENCES
[1] A. Ortiz, F. Bonnin, A. Gibbins, P. Apostolopoulou, W. Bateman, M. Eich, F. Spadoni, M. Caccia,
and L. Drikos. First steps towards a roboticized visual inspection system for vessels. In Proc. of
IEEE Conference on Emerging Technologies and Factory Automation, pages 1–6, 2010.

[2] P. Pounds, R. Mahony, and P. Cork. Modelling and control of a quad-rotor robot. In Proceedings of
the Australasian Conference on Robotics and Automation, 2006.

[3] D. Gurdan, J. Stumpf, M. Achtelik, K.-M. Doth, G. Hirzinger, and D. Rus. Energy-efficient au-
tonomous four-rotor flying robot controlled at 1 khz. In Proc. of IEEE International Conference on
Robotcis and Automation, pages 361–366, 2007.

[4] Markus Eich and Thomas Vögele. Design and control of a lightweight magnetic climbing robot for
vessel inspection. In Proc. of IEEE Mediterranean Conference on Control and Automation, pages
1200–1205, 2011.

[5] A. Ortiz, F. Bonnin-Pascual, E. Garcia-Fidalgo, and J.P. Beltran. A control software architecture
for autonomous unmanned vehicles inspired in generic components. In Proc. of IEEE Mediterranean
Conference on Control and Automation, pages 1217–1222, 2011.

[6] P. Torr and A. Zisserman. MLESAC: A new robust estimator with application to estimating image
geometry. Computer Vision and Image Understanding, 78:138–156, 2000.

[7] Jean-Yves Bouguet. Pyramidal implementation of the lucas kanade feature tracker. Description of
the algorithm, 2000.

[8] M. Lourakis. levmar: Levenberg-marquardt nonlinear least squares algorithms in C/C++. [web
page] http://www.ics.forth.gr/~lourakis/levmar/, Jul. 2004. [Accessed on 24 July 2011.].

[9] G.H. Lee, M. Achtelik, F. Fraundorfer, M. Pollefeys, and R. Siegwart. A benchmarking tool for
mav visual pose estimation. In Proceedings of the International Conference on Control, Automation,
Robotics and Vision, 2010.

10
ISBN-xxx-y-zz-yyyyyyy-x 80
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Present Perspective of the Airworthiness

Regulatory Initiatives for Unmanned Aerial Systems

CUERNO-REJADO, C. and DOMINGO-LEAL, P.J.

E.T.S.I. Aeronáuticos, Universidad Politécnica de Madrid
Plaza del Cardenal Cisneros 3, 28040 Madrid, SPAIN. Email:[email protected]

Abstract
During the last years several worldwide civil and military initiatives have appeared devoted to
assess the safety levels of the unmanned aircraft. These initiatives are very heterogeneous; some
of them are centered on airworthiness aspects while others focus on operations. From the point
of view of a potential UAS manufacturer the actual situation has plenty of uncertainties in
relation to the regulations to be applied for certifying the design, manufacturing and
maintenance, and from the point of view of the potential operator the situation is analogous
with respect to operational procedures. The objective of this work is to clarify the present
airworthiness regulatory scene by summarizing all the regulatory efforts up to date and
preparing a comparative analysis of them.

Keywords: unmanned air system, airworthiness, regulations.

1. INTRODUCTION
The UAS (Unmanned Aircraft Systems) are complete systems designed for flying without a human
pilot on board and include, following EASA [1, 2], not only the UAV (Unmanned Aircraft Vehicle)
but also the systems that allow the proper operation, such as the control station, the communications
link and the launch and recovery systems.
From 1917, when the first controlled flight of an unmanned aircraft (Curtiss N-9 US NAVY) took
place [3], the development of the UAS has been notable especially in the military field and during
recent war periods. This great evolution of the UAS has caused an important diffusion of these
systems nowadays, making it necessary to establish an adequate regulatory frame which allows the
safe operation of these aircrafts in military, civil or commercial missions.
Nowadays, the problem for civil operations is partially solved, the general solution being to require
operating the UAS outside of the non-segregated airspace or, sometimes to book timer fractions inside
it in such a way that the UAS could be safely operated.
The question that now arises is how the operation of UAS could be allowed in the non-segregated
airspace, at the same time as the conventional aircraft, helping UAS to develop all their potential. To
this end it is necessary to convince the airworthiness and operations authorities that the unmanned

ISBN-xxx-y-zz-yyyyyyy-x 81
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

aircraft are able to develop the same level of safety as the manned ones. So the UAS are not a hazard
for the other traffic and for the people and material on ground.
For reaching such scenario it is necessary to solve not only the lack of regulations but also the
development of highly reliable technologies that allow UAS to reach equivalent levels of safety to the
manned aircraft. Keeping in mind the options stated by several regulators [1, 4, 5, 6], it is possible to
enumerate the premises that all integration regulation initiatives should consider:

 UAS should reach equivalent levels of safety to conventional manned aircraft.

 UAS operations should not increase the risks for the users of the non-segregated airspace, nor
third parties.

 UAS should use the same ATM procedures and the same flight rules as the other users of the
airspace.

 The air traffic services given to the UAS should be transparent to the other users and the air
traffic controllers.
These integration premises can be summarized as four problems to be solved: airworthiness
certification, interaction with ATM, interaction with other aircraft and flight crew certification. In this
work an overview of the present existing, both civil and military, regulatory initiatives are going to be
presented focusing, mainly, on the airworthiness certification initiatives although not exclusively.
Previous comparative studies, in which not only the airworthiness regulatory initiatives have been
summarized but also quantitatively analyzed, have been published by the authors elsewhere [7, 8].

2. RULEMAKING ACTIVITIES IN EUROPE

2.1. Organizations

From the civil point of view, the regulatory competence in Europe for unmanned aircraft with an
operative mass over 150 kg relies on EASA, as is established in its Basic Regulation [9], art. 4,
section 4. For that reason, in 2005, the Agency began the rulemaking process issuing an A-NPA
(Advance-Notice of Proposed Amendment), A-NPA 16-2005 [1], centered on the establishment of a
policy (and later on a Certification Specification code) on UAS certification. Before 2005 and almost
at the time of appearing the older Basic Regulation [10] which created EASA, a joint
JAA-EUROCONTROL initiative appeared: “The Joint JAA/EUROCONTROL Initiative on UAVs –
UAV Task Force”. That project aimed at the development of a model for regulating UAS. The
initiative was fruitful and its Final Report [4] appeared on May 2004, which is essential to understand
the actual point of view of EASA on UAS integration. This Report presents a deep multidisciplinary
analysis on the future establishment of a regulatory frame for UAS, including safety/security,
airworthiness, operational approvals, maintenance and licenses, but not ATM.

ISBN-xxx-y-zz-yyyyyyy-x 82
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The philosophy of the A-NPA 16-2005 [1] follows the conclusions of the JAA/EUROCONTROL Task
Force, maintaining the two possible approaches for the new regulation:

1. Conventional: tailoring to UAS, in specified conditions, an existing code for manned aircraft.

2. Safety objectives: creating completely new regulations based on complying total safety
objectives and centered on the most relevant hazards.

For selecting the appropriate manned airworthiness EASA’s CS code used as the design standard for
every UAV system, the A-NPA document presents two alternatives too that in future regulations will
be reduced to one:

 Alternative 1: Impact energy method. The method compares the hazard presented by a UAV
with that of existing conventional aircraft to obtain an indication of the appropriate level of
requirements which should be applied. The most significant feature of this proposal is that it
relies on a comparison with existing conventional aircraft design requirements which
contribute to a currently accepted level of safety, and avoids controversial assumptions about
future contributions to that level of safety from operational, environmental or design factors.
The comparison criterion is based on the fact that the capability of a vehicle to harm any third
parties is broadly proportional to its kinetic energy on impact. For the purposes of the
comparison method it is assumed that there are only two kinds of impact: either the impact
arises as a result of an attempted emergency landing under control (unpremeditated descent
scenario), or it results from complete loss of control (loss of control scenario). Once the kinetic
energy for each scenario is computed it is possible to determine the appropriate CS code to be
applied. Then it would be necessary to construct a certification basis which addresses the same
aspects of the design as the existing codes, and to the level indicated by the kinetic energy
comparison. Special conditions would be required for any novel feature of the design not
addressed by the existing codes.

 Alternative 2: Method based on UAV safety objectives. Safety objectives have been used as a
means to define and justify the civil aircraft characteristics. These safety objectives are oriented
to on board people protection and are defined by the FAR/CS-25/23 [11-14] regulations. As
there are no people on board of UAV, safety objectives criteria for UAV must be redefined and
oriented to on ground people protection. By comparing current aircraft safety objectives (as
they are defined in the regulations) to the UAV proposed safety objective, a correspondence
between CS-23/25 [13-14] and UAV categories is established. Three main parameters have
been considered: statistical aircraft fatal losses, technical aircraft losses defined in the
regulations and catastrophic failures.

Following its rulemaking procedure, EASA published the A-NPA 16/2005 [1] as a “policy document”
in order to receiving proposals from other institutions and completing the regulation before arriving at

ISBN-xxx-y-zz-yyyyyyy-x 83
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

a consolidated version. During this process, the Agency and interested parties (National Authorities,
manufacturers, etc.) interacted, and the results were compiled in the CRD 16/2005 document [2]. This
document was issued on December 2007 and was opened for comments until February 2008. It
compiled 320 comments from 45 different sources. These comments or suggestions, after being
studied by EASA, can be identified as accepted, partially accepted, considered (accepted but meaning
no changes in the text) or rejected.

After a public discussion in the EU after the A-NPA and CRD publication, the Agency finally stepped
forward in the rulemaking procedure for UAS and issued a new Policy Statement document [15] in
September 2009. This policy establishes general principles for type-certification (including
environmental protection) of an Unmanned Aircraft System (UAS). The policy complies with the
current provisions of the Basic Regulation [9], Regulation (EC) Nº 1702/2003 [16] and all
Management Board Decisions relating to product certification. This policy shall be used by the
Agency’s staff when certificating UAS. This policy statement is therefore an interim solution to aid
acceptance and standardization of UAS certification procedures and will be replaced in due course by
AMC and guidance material to Part-21 when more experience has been gained. The main topics
covered by the policy statement are the following:

 Routine case: The issue by the Agency of a type-certificate (TC) will be based upon the
applicant demonstrating compliance with a defined type-certification basis and a certificate of
airworthiness (CofA) is granted to an individual UAS when compliance with the approved type
design has been shown. Any applicant applying for UAS type-certificate is required to
demonstrate their capability by holding a design organization approval (DOA), issued in
accordance with Part-21 subpart J.

 Alternate approach (within the scope of Part-21): to facilitate an early introduction of civil
UAS operations, it will be possible to apply for an airworthiness approval. This approach
recognizes that some UAS may benefit from a stepwise approach in conjunction with the issue
of a restricted TC and/or restricted CofA. This alternative may be based on the safety target
approach, using an overall target level of safety defined by the Agency, in lieu of a specified
airworthiness code.

 UAS control stations and other remote equipment performing functions that can prejudice
take-off, continued flight, landing or environmental protection, shall be considered as part of
the aircraft and included in the type-certification basis.

 The applicable airworthiness code or codes to be used as reference for establishing the
type-certification basis will be proposed by the applicant using a methodology for selecting the
applicable manned aircraft code. A tailoring of the code should be proposed by the applicant, in
a justified manner. Nevertheless EASA recommends the “Alternative 1” [1].

ISBN-xxx-y-zz-yyyyyyy-x 84
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

 The Agency acknowledges that USAR [17], developed by the French Military Authorities, and
later updated by NATO FINAS1 group to STANAG 4671 [18], has been developed using a
methodology closely related to the one described in the EASA’s policy. At an applicant’s
request, the Agency may accept USAR version 3, STANAG 4671, or later updates, as the
reference airworthiness code used in setting the type-certification basis provided that the code
identified by the methodology of the Policy does not indicate that safety standards in excess of
CS-23 [14] are required, and the safety targets included in the system safety assessment reflect
values resulting from the application of this policy.

 There are several special conditions covered in the Policy, for which the Agency includes some
guidance. These conditions are the following: emergency recovery capability, command and
control link, level of autonomy, human machine interface, control station, types of operation
and system safety assessment.

 Other issues considered are the application of Part-21 subpart I for noise certificates, and
Annex I of Part-M of Regulation (EC) Nº 2042/2003 [19] for the continuing airworthiness, and
the Agency’s point of view on the certification of “detect and avoid” systems.

From the civil point of view, the regulatory competence in Europe for unmanned aircraft with an
operative mass under 150 kg relies on the National Authorities, as it is established in its Basic
Regulation [9], art. 4, section 4. By the date, if one person/organization wants to operate an UAS
needs to obtain an approval in the State where the UAS is going to operate. The requirements that the
UAS must comply could be different from one Authority to another as there is no existing common
regulation. Nevertheless although EASA is not competent in this matter, it agrees in the high interest
of having a harmonized opinion and also in the collaboration of EUROCAE WG-73 as a developer of
a set of certification guidelines. Currently it is JARUS2 who is in charge of harmonizing and
developing the draft versions of a new European rule for light UAS category (LUAS) taking as basis
the CS-VLA [20] and CS-VLR [21] airworthiness European rules. Recently (March 2011) a new
European initiative has emerged in order to create a new working group inside EUROCAE WG73
SG4 called “Small RPA3” who will became, coordinately with JARUS, in charge of developing the
proposals for a new regulatory frame for LUAS/SRPA (UAS with MTOW<150 kg, and out of EASA).

Parallel to EASA, other organizations have also been working for the future integration of UAS in
non-segregated airspace. The EDA, for example, has been working in several projects. One of them

1
FINAS: Flight In Non-segregated Airspace.
2
JARUS: Joint Authorities for Rulemaking on UAS; an international coordination group initiated and
chaired by CAA The Netherlands, formed by eleven European National Authorities and coordinated with
Australia, Canada and USA Authorities.
3
Small RPA stands for Small Remotely Piloted Aircraft (ICAO preferred term for UAS).

ISBN-xxx-y-zz-yyyyyyy-x 85
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[22] was published in February 2007 and its scope and key results comprise technical requirements
and architecture for LE-UAV4 communications. Another remarkable project published in December
2007 [23] demonstrated the feasibility of a mid-air collision avoidance system in the near future, not
being necessary the development of new technologies. Besides, in 2007, the EDA Steering Board in
Defence Ministers formation tasked EDA to develop a roadmap for the integration of UAS into
non-segregated airspace by 2015, a study that was carried out by the Air4All Industry Consortium.
One of the identified basic points to solve for the future has been the traffic sense & avoid issue. Thus,
regarding this topic, one of the most important strategic developments is the MIDCAS5 project. In
November 2008, in relation to the harmonization and certification standards, EDA established a
MAWA6 to develop a comprehensive regulatory framework in the EU (EMARS7) for military and
state aircraft, including UAS. MAWA is divided into four Task Forces addressing not only a common
regulatory framework, but certification processes, maintenance organization approval and certification
codes and standards. The MAWA Task Force 1 edited a document entitled European Harmonized
Military Airworthiness Basic Framework [24], where the basic characteristics and baselines of the
process of harmonization of the military airworthiness arrangements across Europe are established.
Frequency spectrum allocation is being studied and analyzed too. In August 2008, the EDA
participating Member States created the task “Air4All Frequency Group” to work in identifying
appropriate spectrum requirements to define, consolidate and promote a common European position
on the regulatory and operational requirements for UAS “2012 World Radio Conference”. In
December 2008, EDA signed a contract for a study on military frequency spectrum allocations
required for the insertion into the general air traffic of the unmanned aircraft systems (SIGAT8). To
end with EDA, we can find other activities in relation to the UAS that have been carried out in the last
years or that are being developed nowadays, like the first high level European Conference on UAS,
which was co-organized by EDA and EC in 2010, the Network Enabled Capability (NEC) in support
of UAS insertion into GAT, the FUAS9 project, the establishment of the role of UAS in EU crisis

4
LE-UAV: Long-Endurance Unmanned Aerial Vehicle.
5
MIDCAS: Mid Air Collision Avoidance System. The MIDCAS mission is to demonstrate the baseline of
solutions for the UAS Midair Collision Avoidance Function (including separation), acceptable by the
manned aviation community and being compatible with UAS operations in non-segregated airspace by
2015.
6
MAWA: Forum for Military Airworthiness Certification.
7
EMARs: European Military Airworthiness Requirements.
8
SIGAT: Consortium to support a study on military spectrum requirements for the insertion into the
general air traffic for the UAS and identification of an appropriate spectrum for the integration of UAS in
non-segregated airspace.
9
FUAS: Future Tactical Unmanned Aerial Systems.

ISBN-xxx-y-zz-yyyyyyy-x 86
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

management operations, and several tasks about UAS engines.

Another relevant European stakeholder is EUROCONTROL. This organization has created two Task
Forces during the last years, in association with the JAA/EASA, one of them for civil application and
another one for military application. The civil task force and its results have been presented in
previous lines. In military air traffic, the EUROCONTOL UAV-OAT-TF report was published in July
2007 [5] and deals with the relations between UAS and ATM of the military air traffic. One of the
recent works of EUROCONTROL is the first phase of the CAUSE project [25], which was released in
May 2010. This document is one of the studies that comprise the ACAS Program. It arises from the
limitations that sense & avoid function has. Some of the key points to consider in the full CAUSE
study is the ability of an UAS to detect other aircraft despite the ability of other aircraft to avoid
collisions, the ability of an UAS to detect and act avoiding collisions against other aerial objects, the
ability of ACAS in manned aircraft to detect UAS, the interoperability of the UAS anti-collision
system and the ACAS, and the differences in the performances of the UAS anti-collision system when
it is equipped with a remote pilot. As it has been said, CAUSE considers the anti-collision
requirements in two phases. The first one identifies the potential safety issues and determines, in high
level terms, the collision avoidance equipage requirements of UAS. The second one will develop a
methodology to quantify UAS collision avoidance performance, will develop metrics and targets for
UAS collision avoidance performance and will promulgate results and conclusions to ICAO.

Another document has been published recently. It is entitled EUROCONTROL Air Traffic
Management Guidelines for Global Hawk in European Airspace [26]. It defines the minimal ATM
requirements for GH/EH10 flights in the European airspace to allow to the GH/EH operators to use
them as a basis for negotiating access to the national airspace concerned.

In the same direction as EDA and EUROCONTOL, EUROCAE is also an important part for the
future integration of UAS into non-segregated airspace. Within EUROCAE there is a Working Group
focused on research issues related to the operation of UAS in the context of European ATM, the
WG-73.This Working Group, chaired by a representative of EUROCONTROL and another one of the
FAA, in coordination with the RTCA SC-20311, is currently preparing performance specifications for
UAS.WG-73 participates in the ICAO UAS Study Group and cooperates with other bodies dealing
with UAS. It is comprised of four subgroups: UAS Operations and Sense & Avoid, Airworthiness and
Continued Airworthiness, Communication, Command and Control, Spectrum and Security (C3SS)
and Light UAS & Operations with Visual Management of Separation. In this last subgroup CAA UK
regulation on Light UAS, CAP 722 [27] has been selected as the basis for the development of a

10
GH/EH: Global Hawk / Euro Hawk.
11
RTCA SC-203: Radio Technical Commission for Aeronautics Special Committee 203.

ISBN-xxx-y-zz-yyyyyyy-x 87
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

generic recommendation for regulation. Also, in this last subgroup, a task force has recently been
created, as has been referred to in a previous section of this paper. In short, this organization is
working in the creation of six deliverables. The Deliverable 1 [28] consists of an initial set of
airworthiness certification and operational approval items. Its report was finished in January 2007.
The Deliverable 2 [29] sets out the actions to be taken by this Working Group and a delivery schedule.
The Deliverable 3 [30] aims to provide a document in order to collaborate with the development of
recommendations and a requirements framework for civil UAS. This deliverable has been the latest
achievement of WG-73 and it is composed of four volumes. The Deliverable 4 will define the
requirements for command, control and communication systems and should be coordinated with the
MASPS12 to be developed by RTCA SC-203. The Deliverable 5 will define the requirements for UAS
Separation Assurance & Collision Avoidance Systems and should be also coordinated with the
MASPS to be developed by RTCA SC-203. Finally, the Deliverable 6 will identify those aspects that
would require special ATM considerations. This last activity depends on other activities, so periodic
summary reports will be edited every fourth quarter.

In reference to the most important European stakeholders and as part of its General Studies
Programme [31], ESA has initiated some activities to evaluate the state of the art for cooperative
satellite-UAS missions, and analyze future missions that can be supported by improved BLOS13
communications. In February 2010 EDA and ESA signed two contracts for feasibility studies on
“Satellite Services for the Integration of UAS into European Airspace”. The signature event was in the
EDA´s conference “Bridging Efforts-Connecting Civilian Security and Military Capability
Development”. These studies address the use of satellites to enable insertion of UAS into
non-segregated airspace, using satellite communication for command and control, sense and avoid
and ATC. Another application is to downlink payload data to the end-user in real time. Finally,
command and control of UAS is in many ways similar to command and control of satellites or robotic
control. ESA intends to start an activity to find their common points. With regards to frequencies, ESA
is actively following the ITU14 and CEPT15 activities.

2.2. Individual States

From the point of view of the individual initiatives developed by European States, France and United
Kingdom have been the most active countries. In France there has been a remarkable contribution by
means of the French code, USAR [17], which has been developed by the Délégation Générale pour

12
MASPS: Minimum System Performance Standards.
13
BLOS: Beyond Line Of Sight.
14
ITU: International Telecommunication Union.
15
CEPT: European Conference on Postal and Telecommunications Administrations.

ISBN-xxx-y-zz-yyyyyyy-x 88
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

l’Armament (French Defence Ministry) and it is compulsory for all military French UAS. It is based
on the EASA’s CS-23 code [14], tailored to fixed wing UAS (tactical, MALE, HALE and UCAV).
This code is the basis for the military code NATO STANAG 4671 [18], ratified in 2007. This
document contains a set of airworthiness regulations addressed to certify military fixed wing UAS
with a MTOW between 150-20000 kg which will operate in the non-segregated airspace. The purpose
of this code is to obtain ELOS for the affected UAS similar to the fixed wing aircraft certified with
FAR Part 23 [12] and EASA’s CS-23 [14] codes (from which it comes from). At the same time it
includes some special features of the UAS that are recognized through new subparts, so Subparts A to
G come from CS-23 and from H to I, there are new parts devoted to communications, command and
control data link and ground control station, all of them specific UAS topics. From the point of view
of UAS with MTOW<150 kg, under the National Authorities responsibility, the French Civil Aviation
Authority (DGAC) is submitting for consultation two draft decrees providing a framework for the
specific activities carried out using remotely piloted aircraft. One of the decrees include provisions
concerning the design of civil aircraft without anyone on board, their conditions of use and the skills
required of their users relate to airworthiness, piloting, operations and the use of electromagnetic
frequencies. The other decree is related to airspace and it amends the Decree of 21 December 2009
relating to the conditions for integrating into, and movements in, French airspace of unmanned civil or
military aircraft.

On the other hand, in the United Kingdom, there are two regulatory regimes. Military and civil UAS
must comply with the applicable regulation. The DEF STAN 00-970 [32] is applicable to UAS
procured by the Ministry of Defence (MoD). It includes certification requirements for UAS including
design, development and testing topics. This code has been used to complement
JAR-23/CS-23[33,14] codes in the codification of NATO STANAG 4671 [18]. On the other hand, the
main civil requirements are set out in the ANO16 2009 [34]. The national requirements about
airworthiness certification or continuing airworthiness will only apply to non-military aircrafts which
come within one of the exempt categories defined in EASA Regulation. It can be observed that there
are no defined requirements about the design and construction of civil UAS within United Kingdom.
This situation may change with the future issue of Chapter A8-22 of BCAR17 section A [35].
Nowadays, the CAP 722: Unmanned Aircraft System Operations in UK Airspace, Guidance [27] is
available. It is an advisory document issued by the Civil Aviation Authority (CAA). It provides
guidance on operating UAS in UK airspace. The fourth (2010) edition includes the changes
introduced in ANO 2009 [34] with respect to the requirements for sUAS operators to obtain a CAA
Permission when its aircrafts are being used for aerial works, and also in some cases in surveillance

16
ANO: Air Navigation Order.
17
BCAR: British Civil Airworthiness Requirements.

ISBN-xxx-y-zz-yyyyyyy-x 89
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

works or for data acquisition (small Unmanned Surveillance Aircraft). Besides, the UAS below 7 Kg
are now considered in this new legislation, with the purpose of guaranteeing the public security,
implementing appropriate operational restrictions. Finally, some guidance about additional details that
must be supplied in the license application to operate sUAS has been included too (Appendix 1 to
Section 3, Chapter 1).

Finally there are other States in the UE which are taking actions in regulating UAS under 150 kg. For
instance, Czech Republic has recently published an UAS Regulation which comprises the national
requirements for design, production, maintenance, modifications and operations of UAS that meet the
criteria of Annex II of the Regulation (EC) nº 216/2008 as amended and it is deemed to be a
recommended practice for operation of model aircraft with a maximum mass lower than 20 kg. On the
other hand, the situation in Spain is not so developed although some efforts are being done. One area
of work is the future amendment of the Spanish Regulation of Air Navigation in order to include UAS
operations. By the moment, the operation of UAS with a maximum mass lower than 25 kg, in BLOS
mode is being authorized in restricted areas. In relation to airworthiness certification, there are several
UAS certified in Spain in the experimental category using a simple basis of certification inspired in
STANAG 4671, meanwhile a new standard for licenses and operations is being developed based on a
draft document prepared by JARUS.

Before proceeding to the next section, it is necessary to mention the work of some association such as
UAVNET which is also working for the future integration of UAS in non-segregated airspace,
preparing workshops and developing relevant documents.

3. RULEMAKING ACTIVITIES IN THE UNITED STATES

3.1. Organizations

In the United States of America, the FAA (Federal Aviation Administration) has established the
UAPO18, which is in charge of the UAS integration process within the U.S. NAS19. For the moment,
until a new FAR for UAS appears (expected by the end of 2011), UAPO has developed a set of
certifications and authorizations for UAS. In the case of civil operations, it will be necessary to ask for
a special airworthiness certification in the experimental category. As the experimental category is very
wide and comprises a great variety of aircrafts, the case of UAS is specially considered in FAA Order
8130.34 [36]. It should be pointed out that FAR Part 21 [37] prevents the use of the UAS in
commercial missions. Finally it is important to note that FAA Order 8130.34 [36] is not a detailed
airworthiness code, but a set of operational limitations.

18
UAPO: Unmanned Aircraft Programme Office.
19
NAS: National Airspace.

ISBN-xxx-y-zz-yyyyyyy-x 90
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

In the case of public operations, FAA will issue a Certificate of Authorization (COA) or Waiver which
represents a temporary procedural interim mechanism authorizing UAS flight for a special purpose
and into a limited area. The objective of the FAA together with the public organization operating the
UAS is to obtain the same ELOS with the COA as the manned aircraft. This process is adequate for
enabling limited number of flights, but does not provide the level of airspace access necessary to
accomplish all missions. Finally, for the issue of the COA, the FAA has published some
recommendations [38], in which the military code MIL-HDBK-516-B [39] is strongly recommended
as a guidance code.

On the other hand, model aircraft are regulated on a voluntary basis by means of the FAA’s Advisory
Circular AC 91-57 [40]. To clarify the applicability of this AC for the operation of UASs, the FAA has
published its opinion [41] and has created the Aviation Rulemaking Committee (ARC) for Small UAS
(sUAS) [42], which has regulated the commercial applications of small UAS for which the AC 91-57
[40] is not applicable. The recommendations of the sUAS ARC for creating a new regulatory
framework for these aircrafts have been published elsewhere [41] and look like a sFAR (special FAR)
code. In principle, the philosophy followed by the sUAS ARC was similar to the criteria for the
light-sport aircraft category: the manufacturer had to show compliance with some identified
consensus standards to obtain the airworthiness certificate. Nevertheless, the sUAS ARC states that all
sUAS should comply with the essential requirements published in Appendix B of [41]. Presently, it
exists under progress the sUAS Part 107 Regulation [43]. It should be noted that within FAA there is
also a FAA UAS System Safety Study Group (1309) which works in coordination with EASA and
EUROCAE WG-73.

Based on the present, it can be stated that the requirements for the future access to NAS make
necessary an improvement of the current situation. One point to consider is the “DoD-FAA
Memorandum of Agreement” [44], that allows certain DoD operations and that provides an
opportunity for DoD to operate easier in exchange for providing operational data to the FAA to
support UAS related activities. In the future, elements of the DoD-FAA Memorandum of Agreement
[44] must be explored to adapt them to other federal agencies in order to promote cooperation,
operational data collection and expanded access. Another important issue for the future will be the
safety case methodology that ensures that a UAS is safe to fly in the airspace required to complete its
mission, and includes any operational or technical implementation where the risk to the NAS has not
been quantified. On the other hand, currently, DoD and FAA are collaborating on preliminary
activities to establish criteria for UAS lost-link NAS Access Plan procedures. Also, between
December 2008 and March 2009, matter experts from government, industry and academia concluded
that a Target Level of Safety (TLS) approach is recommended because it quantifies the total risk of the
system. In relation to the standards, FAA, DoD and NASA are developing UAS certification
requirements in collaboration with other government agencies and industry. Nowadays, UAS

ISBN-xxx-y-zz-yyyyyyy-x 91
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

developers comply with both military and civil existing standards. Civil standards organizations such
as ASTM, SAE and RTCA are developing many specific standards to UAS. Later, in this paper some
of these organizations will be analyzed. In relation to the technology, UAS integration activities need
to include NextGen20 technology, due to compatibility issues and performance and safety benefits.
In April 2010 the UAS Task Force Charter was created. This organization has been able to identify
joint requirements, capability gaps and issues that need to be solved to increase access of UAS in the
NAS. Within the UAS TF, the UAS ExCom21 working groups aim to optimize the FAA´s review and
approval process for UAS flights under a COA and to develop a Federal UAS NAS Access Plan. Thus,
there are several IPTs that perform certain tasks. The AI IPT22, for instance, has developed the DoD
UAS Airspace Integration Plan [45] that supersedes the 2004 OSD Airspace Integration Plan [46].
Better capabilities in some points like Ground Based Sense and Avoid, Airborne Sense and Avoid,
airworthiness certification, operational policies and procedures and development of technical
standards, are described in this document. Moreover, this Airspace Integration Plan establishes
specific near, mid and long term activities to achieve improved UAS NAS access. On the other hand,
the UCS-WG23 aims to develop a common, open and scalable architecture for command and control
of UAS with a maximum gross take-off weight in excess of 20 lbs. In June 2010, the UCS-WG
released the version 1.0 [47] of the core architecture, focused on standardizing data structure and
information flow. There is also an F&B IPT 24 that provides technical analysis and regulatory
guidance to help to achieve more efficient and effective utilization of spectrum for UAS operations.
This IPT has some recent accomplishments, like the support to the initial development of the Small
UAS Digital Data Link, the support to the Joint UAS Center of Excellence in defining spectrum usage
options for inclusion within their UAS CONOPS 25 documents, and the analysis on satellite
communications interference bounds which helped establish UAS BLOS CONOPS. Another IPT, the
UAS R&E IPT26, listed sense and avoid needs and established a logical progression of technology to
meet needs over time as part of its activities, and is currently working on developing an UAS
Propulsion and Power roadmap.
In reference to the available documentation about UAS in the United States, it should be noted that an
Unmanned Systems Roadmap is launched every two years with the purpose of supplying a vision in

20
NextGen: Next Generation Air Transportation System.
21
UAS ExCom: UAS Executive Committee is a joint committee comprised of senior executives from
DoD, FAA, DHS and NASA.
22
AI IPT: Airspace Integration Integrated Product Team.
23
UCS-WG: Unmanned Control Segment – Working Group.
24
F&B IPT: Frequency and Bandwidth Integrated Product Team.
25
UAS CONOPS: UAS Concept of Operations.
26
R&E IPT: Research and Engineering Integrated Product Team.

ISBN-xxx-y-zz-yyyyyyy-x 92
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

terms of, for example, the technology advancements needed. Last edition was published in 2011 [48].

3.2. Organizations of standardization

Since 2003, the committee F38 of ASTM27 collaborates with the FAA in the drafting of airworthiness
legislation for UAS for its integration in the NAS. The mission of the F38 Committee is to produce
practical, consensus standards that facilitate UAS operations at an acceptable level of safety for use by
industry, academia, government organizations and regulatory authorities. It is divided into three
subcommittees related to the UAS integration: Airworthiness, Flight Operations and Personnel. In
August 2009 the FAA asked standards development organizations to create a set of standards for
sUAS and ASTM Committee F38 was approved for it. After a meeting held in February 2010, where
FAA representatives expressed their objectives, the members of the F38 committee built a framework
and determined an implementation deadline to develop a set of specific standards.

Another association involved is RTCA28. This association develops recommendations for the FAA
basically, in relation with the communications, navigation, surveillance and air traffic management
system. In 2004, SC-203 was created for the development of standards in relation with the UAS. This
committee is divided into three groups. WG1 is responsible for developing the MASPS and is
subdivided into three subgroups: Document Integration, Operations and Systems. WG2 is responsible
for the sense & avoid issue and is subdivided into four subgroups: System Safety, Sensors, Algorithms
and Human Factors. Finally, WG3 works in command, control and communication, data link and
avionics. The deliveries already finished and those that are planned are:

- Guidance Material for Unmanned Aircraft System (UAS): It provides key definitions and
assumptions [49].
- Operational Services & Environment Definition (OSED): It provides baseline descriptions of
UAS operational functions and performance characteristics, ATC services, NAS environment and
procedures. It was published in March 2010 [50].
- UAS MASPS: It is scheduled for December 2012.

- Sense and Avoid MASPS: It is scheduled for December 2013.

- Control and Communication MASPS: It is scheduled for December 2013.

The results of the special committee will be developed considering the methodology that is described
in DO-264, Guidelines for Approval of The Provision and Use of Air Traffic Services [51] and they
are not expected before of 2015-2017.

27
ASTM: American Section of the International Association for Testing Materials.
28
RTCA: Radio Technical Commission for Aeronautics.

ISBN-xxx-y-zz-yyyyyyy-x 93
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

4. COMMON INITIATIVES
In 2007 ICAO established the Unmanned Aircraft Systems Study Group (UASSG) with several targets:
to guarantee the global interoperability and harmonization, to develop a regulatory concept for UAS
and an associated guidance material and to review the SARPs29 of ICAO. Also, UASSG aims to
contribute to the development of the technical specifications and to coordinate with ICAO ACP30. The
sixth meeting of the ICAO UASSG met in Montreal to produce a guidance document in time for the
ICAO UAS Symposium to be held in April 2013. The UASSG has published Circular 328 [52] that is
divided into three areas (operations, equipment and personnel) and that will serve as an interim
reference document for States and until the UAV SARPs are finalized. Besides, the UASSG is
developing new terminology and definitions, analyzing the ICAO Annexes and suggesting new
SARPs. A new Part to Annex 6, Part 4, will be also created to cover UAS.

On the other hand, the NATO is the responsible for the military regulations in most European
countries. This organization includes several groups. One of them, JCGUAV31, created the FINAS
initiative which aims to recommend and to document the NATO guidelines to allow the UAS border
operations in non-segregated airspace. Some of the main works of this group are several practices.
The STANAG 4670 [53] deals with the UAS operator training (pilot) and it lists the required abilities.
The STANAG 4671 [10] has been mentioned previously and it is related to EASA CS-23. This
STANAG has been ratified by United States, United Kingdom, France and Netherlands with some
reservations. Besides, we can comment other tasks like the Study 4685, “Application of
Human-Interface Engineering for UAV System Design, Mission Effectiveness and Operational
System Safety”, which will analyze the accidents of UAS, and the human factors that have
contributed to them, and the ongoing “NIAG Study SG 134”, about sense and avoid. Other JCGUAV
teams have produced different practices too. The STANAG 4586 [54] defines architectures,
interfaces, communication protocols, data elements and message formats. This standard identifies five
levels of interoperability to accommodate operational requirements. The respective operational
requirements and approved CONOPS will determine or drive the required Level of Interoperability
that the specific UAV System will achieve. The last remarkable STANAG that we will mention is the
STANAG 4660 [55]. It deals with a study about command and control data link for NATO UAV
interoperability. Other studies are the “Study 7192” about UAV Operator Medical Standards and the
“NATMC” Working Paper, about ATM. On the other hand, there is also a Frequency Management
Sub-Committee for the Spectrum Management.

29
SARPs: Standards and Recommended Practices.
30
ICAO ACP: ICAO Aeronautical Communications Panel.
31
JCGUAV: Joint Capability Group on Unmanned Aerial Vehicles.

ISBN-xxx-y-zz-yyyyyyy-x 94
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Finally, an important Centre of Excellence is the “Joint Airpower Competence Centre (JAPCC)”. This
entity edited in March 2008 the “Flight Plan for Unmanned Aircraft Systems (UAS) in NATO” [56],
and also published “Strategic Concept of Employment for UAS in NATO” [57] in January 2010.

Next table summarizes all the regulations that have been presented in previous lines, showing the year
when they were created and the main topic that they deal:

Table1: Summary table

ORGANIZATION REGULATION TOPIC

EUROPE
1994 JAA JAR 23, Normal, Utility, Aerobatic and Commuter Airworthiness
Category Aeroplanes
2002 EC Regulation (EC) Nº 1592/2002 of the European Common rules
Parliament and of the Council
2003 EC Regulation (EC) Nº 1702/2003 for the airworthiness and Airworthiness,
environmental certification of aircraft and related certification
products, parts, appliances as well as for the certification
of design and product organizations
2003 JARUS CS-VLA, European Aviation Safety Agency certification Airworthiness
Specification for Very Light Aeroplanes
2003 JARUS CS-VLR, European Aviation Safety Agency Certification Airworthiness
Specification for Very Light Rotorcraft.
2003 EASA CS-23, Normal Utility, Aerobatic and Commuter Certification
Aeroplanes. Specifications
2003 EASA CS-25, Large Aeroplanes Certification
Specifications
2004 JAA-EUROCONT UAV Task-Force Final Report: A Concept for European Safety/security,
ROL Regulations for Civil Unmanned Aerial Vehicles (UAVs) airworthiness,
operational approvals,
maintenance, licenses
2005 EASA Advance-Notice of Proposed Amendment (NPA) Nº Certification
16/2005, "Policy for Unmanned Aerial Vehicles (UAV)
Certification". Cologne (Germany)
2005 DGA-CEV UAV Systems Airworthiness Requirements (USAR). Airworthiness
Versión 3.0

ISBN-xxx-y-zz-yyyyyyy-x 95
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2007 EASA Comment Response Document (CRD) to Notice of Certification

Proposed Amendment NPA Nº 16/2005 "Policy for
Unmanned Aerial Vehicles (UAV) Certification".
Cologne (Germany)
2007 EDA Technology demonstration Study on Digital Line of Sight Arquitecture for UAS
and Beyond Line of Sight Data Links for communications
Long-Endurance Unmanned Aerial vehicles (LE-UAVS)
2007 EUROCONTROL Eurocontrol Specifications for the Use of Military ATM
Unmanned Aerial Vehicles as Operational Air Traffic
outside Segregated Airspace, Ed. 1
2007 EUROCAE Deliverable 1 Airworthiness,
WG-73 operational approval
items
2008 EC Regulation (EC) Nº 216/2008 of the European Common rules
Parliament and of the Council
2008 EDA Technology Demonstration Study on Sense and Avoid Sense and Avoid
Technologies for Long Endurance Unmanned Aerial Technologies
Vehicles (LE-UAVS)
2009 EASA Policy Statement Airworthiness Certification of Airworthiness
Unmanned Aircraft Systems (UAS), E.Y013-01
2009 EDA European Harmonized Military Airworthiness Basic Airworthiness
Framework
2009 CAA Air Navigation Order -
- EUROCAE Deliverable 2 Airworthiness
WG-73 Certification,
Operational Approval
items
2010 EUROCONTROL Unmanned Aircraft Systems-ATM Collision Avoidance ATM Collision
Requirements Avoidance
2010 EUROCONTROL EUROCONTROL Air Traffic Management Guidelines for ATM
Global Hawk in European Airspace
2010 CAA CAP 722 Unmanned Aircraft System Operations in UK UAS operation in UK
Airspace - Guidance - Ed.4 airspace
2010 EUROCAE Deliverable 3 Airworthiness
WG-73 Certification, Operat.
Approval items

ISBN-xxx-y-zz-yyyyyyy-x 96
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2011 CAA DEF STAN 00-970 Design and Airworthiness Design, Airworthiness
Requirements for Service Aircraft, Part9, UAV Systems
- CAA British Civil Airworthiness Requirements (BCAR) Airworthiness
UNITED STATES
1981 FAA Advisory Circular AC 91-57, Model Aircraft Operating Operating Standards
Standards

2004 DoD Airspace Integration Plan Airworthiness

certification,
policies…
2007 FAA Unmanned Aircraft Operations in the National Airspace Operations
System

2007 RTCA SC-203 Guidance Material for Unmanned Aircraft System (UAS) Guidance Material

2007 RTCA SC-203 Operational Services & Environment Definition (OSED) Operational Services

2008 FAA Order 8130.34, Airworthiness Certification of Unmanned Airworthiness

Aircraft Systems
2008 DoD MI-HDBK-516-B, Airworthiness Certification Criteria Airworthiness
2008 FAA Order 1110-150, Small Unmanned Aircraft System sUAS
Aviation Rulemaking Committee
2009 FAA Comprehensive Set of Recommendations for sUAS sUAS
Regulatory Development.

2010 DoD Arquitecture Version 1.0. Command & control

- DoD UAS Airspace Integration Plan. Sense and Avoid,

airworth., certific…
2011 FAA sUAS Part 107 Regulation sUAS

2011 DoD Unmanned Systems Roadmap. 2011. Roadmap

COMMON INITIATIVES
2007 NATO STANAG 4671-UAV Systems Airworthiness Requirements Airworthiness
(USAR) for North Atlantic Treaty Organization (NATO)
Military UAV Systems. Ed.1
2007 NATO STANAG 4586, Standard Interfaces of UAV Control Interfaces of UCS
Systems (UCS) for NATO UAV Interoperability, Ed.2

ISBN-xxx-y-zz-yyyyyyy-x 97
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2008 NATO Flight Plan for Unmanned Aircraft Systems (UAS) in Flight Plan
NATO
2009 NATO STANAG 4670, Recommended Guidance for the Training Training of UAS
of Designated Unmanned Aerial Vehicle Operator Operators
(DUO)
2010 NATO Strategic Concept of Employment for UAS in NATO Guidance,
considerations,
concepts for optimum
UAS employment
2011 ICAO Circular 328, Unmanned Aircraft Systems Operations,
equipment, personnel
- NATO STANAG 4660,on Harmonised Requirements for an Command and
Interoperable Command and Control Data link (IC2DL) Control Data Link
for UAVs

5. CONCLUSIONS
A comprehensive outlook of the regulatory framework for UAS has been presented. If the actual
market for civil and military UAS wants to grow, it will be necessary to integrate the operations of
such vehicles into the same airspace (no segregated) as the manned aircraft. To reach such scenario, a
set of harmonized airworthiness codes or standards will have to be elaborated in order to ensure the
same safety levels as the conventional manned aircraft. Although the present initiatives are somewhat
confusing and disperse, the EASA Policy Statement, for MTOW greater than 150 kg, and the future
FAR Part 107, for small UAS, would be a remarkable basis for the new regulatory process.

REFERENCES
1. European Aviation Safety Agency (EASA). Advance-Notice of Proposed Amendment (NPA) Nº
16/2005, "Policy for Unmanned Aerial Vehicles (UAV) Certification". Cologne (Germany) : s.n.,
2005.
2. European Aviation Safety Agency (EASA). Comment Response Document (CRD) to Notice of
Proposed Amendment NPA Nº 16/2005 "Policy for Unmanned Aerial Vehicles (UAV) Certification".
Cologne (Germany) : s.n., 2007.
3. Newcombe, L.R. Umanned Aviation: A Brief History of Unmanned Aerial Vehicles. Reston, VA
(USA) : American Institute of Aeronautics and Astronautics, 2004.
4. Joint Aviation Authorities (JAA). UAV Task-Force Final Report: A Concept for European
Regulations for Civil Unmanned Aerial Vehicles (UAVs). 2004.
5. Eurocontrol. Eurocontrol Specifications for the Use of Military Unmanned Aerial Vehicles as

ISBN-xxx-y-zz-yyyyyyy-x 98
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Operational Air Traffic outside Segregated Airspace, Ed. 1. 2007.

6. Department of Defense (USA). Airspace Integration Plan for Unmanned Aviation. 2004.
7. Integration of UAS in the civil airworthiness regulatory systems: present and future.
Cuerno-Rejado, C., Martínez-Val, R. and García-Juliá, E. Bristol (Great Britain) : s.n., 2010. 25th
Bristol International Unmanned Air Vehicle Systems Conference.
8. Unmanned Aircraft Systems in the Civil Airworthiness Regulatory System: A Case Study.
Cuerno-Rejado, C. and Martínez-Val, R. 2011, Journal of Aircraft, Vol. In Press.
9. European Commission . Regulation (EC) Nº 216/2008 of the European Parliament and of the
Council. 2008.
10. European Commission. Regulation (EC) Nº 1592/2002 of the European Parliament and of the
Council. 2002.
11. Federal Aviation Administration (FAA). Federal Aviation Regulations Part 25.
12. Federal Aviation Administration. Federal Aviation Regulations Part 23.
13. European Aviation Safety Agency (EASA). CS-25, Large Aeroplanes.
14. European Aviation Safety Agency (EASA). CS-23, Normal Utility, Aerobatic and Commuter
Aeroplanes.
15. European Aviation Safety Agency (EASA). Policy Statement Airworthiness Certification of
Unmanned Aircraft Systems (UAS), E.Y013-01. 2009.
16. European Commission. Regulation (EC) Nº 1702/2003 for the airworthiness and environmental
certification of aircraft and related products, parts, appliances as well as for the certification of
design and product organizations. 2003.
17. Délégation Génerale pour l´Ármament (DGA)-Centre dÉssais en Vol (CEV). UAV Systems
Airworthiness Requirements (USAR). Versión 3.0. 2005.
18. North Atlantic Treaty Organization (NATO). STANAG 4671-UAV Systems Airworthiness
Requirements (USAR) for North Atlantic Treaty Organization (NATO) Military UAV Systems. Ed.1.
2007.
19. European Aviation Safety Agency (EASA). Regulation Nº 2042/2003 on the continuing
airworthiness of aircraft and aeronautical products, parts and appliances and on the approval of
organizations and personnel involved in these tasks. 2003.
20. European Aviation Safety Agency (EASA). CS-VLA, European Aviation Safety Agency
Certification Specification for Very Light Aeroplanes.
21. European Aviation Safety Agency (EASA). CS-VLR, European Aviation Safety Agency
Certification Specification for Very Light Rotorcraft.
22. European Defence Agency (EDA). Technology demonstration Study on Digital Line of Sight and
Beyond Line of Sight Data Links for Long-Endurance Unmanned Aerial vehicles (LE-UAVS). 2007.
23. European Defence Agency (EDA). Technology Demonstration Study on Sense and Avoid
Technologies for Long Endurance Unmanned Aerial Vehicles (LE-UAVS). 2008.

ISBN-xxx-y-zz-yyyyyyy-x 99
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

24. European Defence Agency (EDA). European Harmonized Military Airworthiness Basic
Framework. 2009.
25. EUROCONTROL. Unmanned Aircraft Systems-ATM Collision Avoidance Requirements. 2010.
26. EUROCONTROL EUROCONTROL Air Traffic Management Guidelines for Global Hawk in
European Airspace. 2010.
27. Civil Aviation Authority. CAP 722 Unmanned Aircraft System Operations in UK Airspace -
Guidance - Ed.4. 2010.
28. EUROCAE WG-73. Deliverable 1. 2007.
29. EUROCAE WG-73. Deliverable 3. 2010.
30. EUROCAE WG-73. Deliverable 2.
31. European Space Agency (ESA). General Studies Programme (GSP). 2006.
32. Civil Aviation Authority (CAA). DEF STAN 00-970 Design and Airworthiness Requirements for
Service Aircraft, Part9, UAV Systems. 2011.
33. Joint Aviation Authorities (JAA). JAR 23, Normal, Utility, Aerobatic and Commuter Category
Aeroplanes. 1994.
34. Civil Aviation Authority (CAA). Air Navigation Order. 2009.
35. Civil Aviation Authority (CAA). British Civil Airworthiness Requirements (BCAR).
36. Federal Aviation Administration (FAA). Order 8130.34, Airworthiness Certification of
Unmanned Aircraft Systems. 2008.
37. Federal Aviation Administration (FAA). Federal Aviation Regulations, Part 21.
38. Federal Aviation Administration (FAA). Comprehensive Set of Recommendations for sUAS
Regulatory Development. 2009.
39. Department of Defence (DoD). MI-HDBK-516-B, Airworthiness Certification Criteria. 2008.
40. Federal Aviation Administration (FAA). Advisory Circular AC 91-57, Model Aircraft Operating
Standards. 1981.
41. Federal Register Notice. “Unmanned Aircraft Operations in the National Airspace System”, 13rd
February, 2007.
42. Federal Aviation Administration (FAA). Order 1110-150, Small Unmanned Aircraft System
Aviation Rulemaking Committee. 2008.
43. Federal Aviation Administration (FAA). sUAS Part 107 Regulation. 2011.
44. Department of Defence (DoD) / Federal Aviation Administration (FAA). DoD-FAA
Memorandum Agreement. 2007.
45. Department of Defence (DoD). DoD UAS Airspace Integration Plan.
46. Department of Defence (DoD) - Federal Aviation Administration (FAA). OSD Airspace
Integration Plan. 2004.
47. Department of Defence (DoD). Arquitecture Version 1.0.
48. Departmet of Defence (DoD). Unmanned Systems Roadmap. 2011.

ISBN-xxx-y-zz-yyyyyyy-x 100
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

49. Radio Technical Commission for Aeronautics, SC-203. Guidance Material for Unmanned
Aircraft System (UAS). 2007.
50. Radio Technical Commission for Aeronautics, SC-203. Operational Services & Environment
Definition (OSED). 2007.
51. Radio Technical Commission for Aeronautics, RTCA SC-203. DO-264, Guidelines For
Approval Of The Provision And Use Of Air Traffic Services. 2000.
52. International Civil Aviation Organization (ICAO). Circular 328, Unmanned Aircraft Systems.
2011.
53. North Atlantic Treaty Organization (NATO). STANAG 4670, Recommended Guidance for the
Training of Designated Unmanned Aerial Vehicle Operator (DUO). 2009.
54. North Atlantic Treaty Organization (NATO). STANAG 4586, Standard Interfaces of UAV
Control Systems (UCS) for NATO UAV Interoperability, Ed.2. 2007.
55. North Atlantic Treaty Organization (NATO). STANAG 4660,on Harmonised Requirements for
an Interoperable Command and Control Data link (IC2DL) for UAVs.
56. North Atlantic Treaty Organization (NATO). Flight Plan for Unmanned Aircraft Systems (UAS)
in NATO. 2008.
57. North Atlantic Treaty Organization (NATO). Strategic Concept of Employment for UAS in
NATO. 2010.

ISBN-xxx-y-zz-yyyyyyy-x 101
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

UAS Integration into Non-segregated airspace in the short term:

SINUE Project

JAVIER CAINA (Indra), PABLO GONZALEZ (Indra),

SARA LUIS (Ineco), LOURDES GARCIA (Ineco)

Abstract
Several sectors are interested in spreading the use of UAS and pushing their operations into
non-segregated airspace. The most feasible application in the short term heading to this
integration is a maritime surveillance mission. In this framework, SINUE “Satellites for the
Integration in Non-segregated airspace of UAS in Europe”, a project financed by ESA,
aims to analyze and demonstrate by simulation the feasibility of integrating UAS in civil
airspace relying on satellite-based systems. This paper describes the current barriers UAS
have, the contribution of the space segment to the UAS community and the benefits of the
SINUE project in the short term.

Keywords: UAS, SINUE,CBA

1. INTRODUCTION
Unmanned Aircraft Systems (UAS) have had a boost in technologies and development during the last
few years thanks to enhancements in electronics, RF systems and improvements in embedded
components.
Several sectors are interested in spreading the use of UAS and pushing their operations into
non-segregated airspace. For this initiatives from several working groups and organizations have
assumed the challenge to overcome the current barriers in line with some other efforts made by
European agencies like ESA and EDA, and more recently the European Commission.

2. CIVILIAN APPLICATIONS
Previous to SINUE, a study held by ESA [1], included a survey conducted among UAS stakeholders
in order to have an overview of those missions that have more interest for the unmanned community.
The set of stakeholders included UAS and payload manufacturers, Satellite service providers,
Regulatory and Standardization bodies, and UAS-related working groups and associations.

-1-

ISBN-xxx-y-zz-yyyyyyy-x 102
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Fig. 1 : UAS missions sorted according to interest of stakeholders.

The result of this survey is shown in Figure 1. During the study, it was confirmed that UAS missions
requiring satellite services for Beyond Line of Sight (BLOS) operations are of high interest mainly
because of the valuable use of the space sector for navigation and communications (command/control
and payload) services.

As part of the results, Sense &Avoid was pointed out as the most important challenge in the short term
for integration into non-segregated airspace, an issue that is being developed by specific projects such
as the MID Air Collision Avoidance System (MIDCAS - EDA).

The most feasible missions for a short-term demo would be:

2.1. Maritime surveillance and coastguard

UAS are an interesting option for maritime missions due to their often long-range and dull nature.
Additionally they are being deployed over water and not highly populated areas, make them even
more attractive in the short term for introducing civil UAS applications.

For these kind of missions, some civilian applications will overlap military ones facilitating the
utilisation of technology already in use with minimal modification. In this case whether the platform
comes under civil or military jurisdiction will affect flight rules and certification.

Some of the coastguard applications include search and rescue, fisheries protection and surveillance,
and monitoring of maritime traffic, pollution, sea ice and coastal erosion.

2.2. Border control

Land borders are considered suitable for UAS operations (coastal operations are included in the
previous item) especially for patrolling above sparsely populated areas. These missions generally
favour the use of MALE and tactical UAS rather than short-range, portable systems. Advances in
payload miniaturisation, navigation systems and sensor technology let surveillance be carried out
regardless of the weather during day and night which is a crucial requirement for this kind of security
operations.

-2-

ISBN-xxx-y-zz-yyyyyyy-x 103
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Although in some regions the use of these platforms would not be competitive with ground-based
solutions or radar, in remote or mountainous regions they are likely to be used.

3. CURRENT BARRIERS AND REGULATIONS

In order to successfully implement civil applications, several barriers have first to be overcome.
According to Air4All [2], the following challenges are identified including technical, regulatory,
procedural and transversal issues.

Fig. 2 : Challenges identified by Air4All.

In particular, for the short-term, the key regulatory issues include:

 Internationally harmonised regulatory and standardisation framework for UAS

 Airspace and ATM system evolution to cope with the increasing demand of airspace users,
among them the Unmanned Aviation community

 Certification: reliability of UAS and the safety of their operations

 Insurance liability costs & responsibility

As part of the current regulatory framework, ICAO UASSG has recently published “ICAO Circular
328 – Unmanned Aircraft Systems (UAS)” [4] that, although not binding, was issued as a guideline
for harmonization and basis for future regulations. It highlights the following aspects:

 Unmanned aircraft (UA) are, indeed, aircraft; therefore, existing Standards And

-3-

ISBN-xxx-y-zz-yyyyyyy-x 104
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Recommended Practices (SARPs) apply to a very great extent.

 All aircraft, whether manned or unmanned, share a large degree of commonality with regard
to airworthiness. A majority of UAS assessments will likely rely on what is already prescribed for
manned aviation.

 Many existing SARPs are applicable to UAS; others may require interpretive or innovative
solutions.

EASA has also published a policy draft in 2005 in order to guide industry on UAS
certification process. From this, it came in 2009 the first edition of “E.Y01301 – Policy Statement:
Airworthiness Certification of Unmanned Aircraft Systems (UAS)”. Throughout this document it is
shown that existing procedures for manned aircraft apply (Part 21).

From ICAO and EASA positions, it is clear that large UAS are suitable for gaining the
confidence of relevant authorities for UAS missions in the short-term.

4. SPACE SEGMENT BENEFITS

Among the main services that satellites can provide today to UAS are (1) Navigation and Positioning,
(2) Communications (safety and payload), and (3) cooperative surveillance (e.g. use of satellite
images for UAS mission planning and execution). Although the first and third services are not
actually UAS specific, the second one is crucial for BLOS operations.

4.1. Navigation/Positioning

In an increasingly complex and sophisticated aviation environment precise positioning and navigation
capabilities are essential for airspace users thus as a future civilian airspace user UAS require such
capabilities. Operations using automatic take-off and landing (ATOL) capability are expected to be
much more frequent in UAS (even for the smallest types) than in manned aircraft, and such a
capability relies on very precise positioning and navigation systems.

4.2. Safety/Payload communications

Beyond Line Of Sight (BLOS) data links are key enablers for UAS operations requiring long ranges
(> 200 Km), typically performed for long endurance UAS, which are usually within MALE or HALE
categories. So far missions requiring BLOS operations have been mainly performed in military
missions using a segregated airspace or secured corridors. However, to perform BLOS operations in
the non-segregated airspace, UAS will have to be integrated into the ATM system which requires
overcoming the issues previously mentioned, and should be “transparent” to the Air Traffic Control
(ATC).

-4-

ISBN-xxx-y-zz-yyyyyyy-x 105
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

4.3. Cooperative surveillance

Many UAS being operated today, mainly in the military field, have software for mission planning,
execution and post-processing, which is based on digitalized raster maps. The information included in
these maps has to be up-to-date, sufficient and accurate. Therefore the provision of images of the
interest area taken by satellites is of great value as proven in current military missions, together with
their integration into the UAS software.

It is important to mention that although presented here as a satellite service for UAS it can be also
considered as a potential application since satellites have a number of limitations (see Table 1 where
UAS can assist on the provision of imagery for other UAS or other users).

The integration of satellites and UAS has the potential for unique civil and security global missions,
including time-critical and life-critical operations. It is worth noting that the level of benefit of
satellite services are UAS-category dependent as can be seen in Table 2.

Table 1 : Complementary capabilities of Satellites and UAS.

The UAS-Satellite synergy stems from their complementary characteristics with regards to the
capability to provide data to the operators or users. Strengths of one system can balance weaknesses
of the other, as shown in Table 2.

Table. 2 : Satellite services for UAS according to UAS categories.

-5-

ISBN-xxx-y-zz-yyyyyyy-x 106
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

5. SINUE PROJECT
SINUE (Satellites for the Integration in Non-segregated airspace of UAS in Europe) project [3] aims
to analyze and demonstrate by simulation the feasibility of integrating UAS into non-segregated
airspace relying on satellite-based systems for safety of flight communications as well as the use of
satellite communications for the provisioning of high data rate mission payload data link (Fig.3). The
goals of the mission simulations performed during the project were two-fold: on the one hand, to
demonstrate from a technical point of view the feasibility of using satellite systems for the
provisioning of such services; on the other hand to illustrate to a non-UAS aware audience how the
integration issues are handled.

Fig. 3 : SINUE project architecture.

This project studied different UAS mission types and the technology available for the short-term
keeping the main stakeholders in the loop for designing a suitable demonstration.

The Consortium consisted of a group of companies (Fig.4) from different countries of the European
Union (Spain, Luxembourg, Portugal and Germany), bringing together the required expertise areas on
UAS, ATM, Simulation, and Satellite technologies.

Fig. 4 : SINUE Consortium.

The SINUE project was structured into three phases: (1) Mission selection, where user needs,
state-of-the-art technology and scenarios were analysed for a suitable mission definition; (2) Mission
definition, where a detailed mission definition was depicted and associated simulations were
performed, and (3) Mission plan, where a mission plan together with an evaluation framework was
defined, including viability analysis and roadmap.

-6-

ISBN-xxx-y-zz-yyyyyyy-x 107
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Throughout the project several aspects were considered including specific mission architecture,
LOS/BLOS coverage, contingency plans, and ATC integration. ATC simulations were performed
using real traffic data in the operational area implementing different types of scenarios. These
simulations involved (Fig.5) real controllers, pseudo-pilots handling other traffic, UAV pilot, and
supervisors for the whole implementation and satcom analysis. This was a successful activity that
showed that normal air traffic operations in the area were not affected by the insertion of UAS as an
airspace user.

Fig. 5 : Air traffic and ATC simulation architecture.

6. SINUE ANALISYS AND VIABILITY OF THE SELECTED UAS

MISSION
6.1. Selected mission: Maritime Surveillance and Coastguard

A maritime surveillance and coastguard mission represents the best balance between demonstrated
challenges on the one hand and generated costs on the other, making it the selected mission for the
demonstration. This selection is in line with results shown in Fig. 1.

Some of the main objectives of this mission are detection / recognition / identification of objectives
along their path, and integration of the UAS into the end-user surveillance system / command &
control centre.

According to the mission definition the end user will work directly with the SINUE Service Provider,
who is responsible, in coordination with the military authorities, for providing the services and

-7-

ISBN-xxx-y-zz-yyyyyyy-x 108
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

infrastructure required to carry out the mission.

The SINUE Service Provider is in charge of obtaining a temporary/experimental airworthiness

certification, and handling other services such as satellite communications and insurance together with
the UAS manufacturer.

The location of the mission is a confined maritime area within a TMA (Terminal Manoeuvring Area).
Surveillance services should be capable of being performed 24/7 and, for a single UAS operation,
there should be a maximum endurance of not less than 15 hours in the maritime surveillance
configuration.

The airspace infrastructure consists of a set of airports and en-route nav-aid stations (VOR – DVOR,
DME, NDB or ILS) distributed in the TMA in order to get full coverage by either radar or satellite at
the area. This infrastructure is supported by a set of services, such as meteorological information, air
traffic control, rescue or fire fighting services, which facilitate air navigation in the region.

Instrument departures and arrivals in the airport are required to be as transparent as possible on
current ATC operations. VOR, DME, ILS, NDB or TACAN systems are available. The UAS shall fly
to restricted airspace as part of its mission, following ATC departure procedures. The most risky
problems arise due to sharing airspace with conventional aviation. Collision avoidance systems for
UAS are not yet available, therefore civil and military ATC controllers have to be coordinated in order
to temporarily segregate corridors as needed.

In addition there could be other aircraft operating in VFR in the mission area flying at higher altitudes
than those contemplated in the UAS and they should be taken into account.

According to the mission, the UAS will follow standard procedures:

 Standard Instrument Departure

- UAS will coordinate their departures with the normal operations of a civil airport by
following standard procedures.

 Mission route

- The UAS starts from its base and flies its predefined 3D waypoint surveillance pattern
through the area of interest. Due to other traffic, it is foreseen that this route will be changed
and/or left and re-entered according to the voice commands of the air traffic controllers to the
UAS operator. There are also FIR standard routes that the UAS may have to cross within the
area.

 Arrival phase

- The UAS shall start its approach without affecting other flight plans, and, if it does, it shall be
directed to a holding racetrack before starting the final approach and landing.

-8-

ISBN-xxx-y-zz-yyyyyyy-x 109
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

During all phases of flight there will be contingency routes following by predefined waypoints agreed
by ATC. These routes ensure a safe operation by letting the ATC know where the aircraft is at all
times.

6.2. Economic Viability of the Selected Mission

The economic study developed during the SINUE project showed the viability of the implementation
and operation of SINUE in the geographic scenario described above. A Cost Benefit Analysis (CBA)
model has been developed taking into account Operational Costs and Benefits considering several
alternatives including the use of combine unmanned manned aircraft operations in the mid-term.

Among the most important operational benefits identified and monetized during the project are:

 Maintenance Costs Savings

 Fuel Consumption Savings

 Personnel Costs Savings

Among the operational costs:

 Aircraft Insurance Costs

 Remote Support Costs

 Satellite Communication Costs

Although only operational costs and benefits have been realized in the CBA, there are other costs and
benefits that were studied. With respect to these variables, the study included expert judgment within
the project for Infrastructure and One-off costs (Cost of the SINUE system, training, etc…).

CBA analysis identified aircraft insurance as one of the most important economic drivers for mission
deployment. This factor has to be taken into account as one of the critical aspects during mission
planning (see current barriers in Sec.3)

Some qualitative benefits were also analyzed:

 No crew onboard, so no additional risk for losing human lives

 Operational coverage is highly extended by using BLOS capabilities due to the use of
SATCOM systems. Real-time collection and data transmission is achieved by using these technologies
that are not available in the current manned service.

 There is an important increase in endurance for end-user missions without risking the crew
due to tiredness or excess workload.

 Modern technology payload configurable for each mission (EO/IR and/or Radar) is
available.

-9-

ISBN-xxx-y-zz-yyyyyyy-x 110
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

 End-user will gain leading experience with unmanned systems improving technical
capabilities.

 Experience with UAS will let the end-user go further when planning future missions in dull
environments.

7. CONCLUSIONS
UAS integration into non-segregated airspace has several challenges that require a joint UAS
community effort towards wide civil applications. As a summary it is worth noting the following key
points:

 The most interesting civilian applications for UAS stakeholders are security and emergency
management
 A maritime patrol demonstration operating from a dual-use airport and flying the mission in
segregated airspace is possible in the short-term
 The SINUE demonstration will be useful for:
- Demonstrating added value for end-users
- Testing progressive ATC integration
- Acquiring operational experience
- Demonstrating economic viability for end-users
 The SINUE demonstration will be in line with the current initiatives and guidelines from the
UAS community (EC, ICAO, EASA).

REFERENCES
[1] Feasibility Study for Unmanned Aerial System supported by Integrated Space Systems (SINUE).
SINUE Consortium. 2010.
[2] Air4All Study Final report. UAS Insertion into General Air Traffic. Ref.: 07-arm-001, Issue: 1.0,
June 2008.
[3] Satellite-Unmanned Aerial Vehicle (UAV) Cooperative Missions: Status and Outlook.
Indra-MDA. 2009
[4] ICAO Unmanned Aircraft Systems (UAS) Circular (Cir 328 AN/190), Montreal, March 2011.
[5] EASA policy statement on airworthiness certification of Unmanned Aircraft Systems (UAS)
(E.Y013-01), Cologne, 25 August 2009.
[6] SINUE website: http://iap.esa.int/activities/SINUE

BIOGRAPHIES
Javier Caina is electronics engineer and physicist (2004) and has a Master degree in Aerospace

- 10 -

ISBN-xxx-y-zz-yyyyyyy-x 111
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

engineering by the Polytechnic University of Madrid (UPM) and the Technical University of Munich
(TUM). Unmanned Aircraft Systems Engineer in Indra since 2008, he has participated in several UAS
programmes such as PASI tactical system, MIDCAS Sense & Avoid, and has been working for
ESA/EDA initiatives first as part of the SUAV feasibility study and later as technical manager of
SINUE project.

Pablo González is Director of Unmanned Systems in Indra since 2007, his business unit is currently
developing and operating several kind of UAS and terrestrial and naval robotic systems (UGV, UUV
and SUV), including in particular the PASI tactical system deployed in Afghanistan by the Spanish
Army since 2008. His previous experience was mainly in the space field. From 2000 to 2007 he was
managing the European Programs of Indra Espacio, working for customers as the European Space
Agency (ESA), EUMETSAT and EUTELSAT. Master degree in Aeronautical engineering by the
Polytechnic University of Madrid (UPM) in 1989, during the university period he was working in
Aerodynamic research with different grants in NASA (Lewis and Langley Research Centers) and
INTA (Torrejón Air Base). He lectured Mathematics in the Aeronautical Engineering School (UPM)
from 1997 to 1999.

Sara M. Luis is MSC in Aeronautics Engineer (2009), with a specialization in Air Navigation, in the
Polytechnic University of Madrid, Spain. She works as a project engineer at INECO and currently
participates in coordination and technical management in ANS European Projects, such as TITAN, an
EC 7th FP, and SESAR. She has worked in the EC R&D projects OPTIMI, RESET and EPISODE 3
collaborating in IT and CBAs, and previously did a fellowship in the Polytechnic University of
Madrid in Air Navigation.

Lourdes García Bartolomé has a BSc degree in Economics by the Complutense University of Madrid.
After working for three years in the Insurance and Banking sectors, she joined INECO in 2000. She
worked for six years as consultant at AENA where she contributed to the development of the
MEDINA project in accordance with the EMOSIA methodology, as well as in different economic
analysis using this tool. Currently, she is a Cost-Benefit expert in the Business Development and
Planning Division, where she participates in economic analysis activities of international research
projects (SJU, ICAO, EC, Eurocontrol). In addition, she is working in SESAR (Single European
Sky ATM Research) Programme as Airport CBA leader, WP16.6.6 (Business Case Maintenance,
Support and Coordination).

- 11 -

ISBN-xxx-y-zz-yyyyyyy-x 112
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Potential UAS Applications vs. Air Power Security Trends

ALBERTO P. MARTINEZ BORJA

CASSIDIAN, Pº John Lennon, 2 - Getafe (Spain), [email protected]

Abstract
This paper discusses potential UAS applications in the security domain beyond the
traditional Air Power perimeter linked to some key UAS capabilities at system level. Three
areas have been outlined for such a power landscape. These areas cover:

Air policing and air space security in the context of homeland security.
Power projection, where the basic counter land and counter sea activities are
expanded as to comprise missions related to peace support and stabilisation
operations.
Intelligence, surveillance and reconnaissance as the currently dominating field for
UAS applications.

Keywords: Security, UAV, ISR, UAS Capabilities, Air Power.

ACRONYMS, ABBREVIATIONS, SYMBOLS

AEW – Airborne Early Warning IS(TA)R - Intelligence, Surveillance, (Target Acquisition) &

AI/BAI – Air/Battlefield Air Interdiction Reconnaissance

AIS - Automatic Identification System LTAs - Lighter Than Air
ASW - Anti Submarine Warfare MALE – Medium Altitude Long Endurance
ASuW - Anti Surface Warfare MTBF - Mean Time Between Failures
BM – Ballistic Missile MTM - Maritime Traffic Management
2
C – Command & Control MOUT - Military Operations in Urban Terrain
CAS - Close Air Support MPA – Maritime Patrol Aircraft
CCF - Conventional Counter Force NRBC - Nuclear, Radiological, Biological, Chemical
COMAO - Composite Air Operations PSO - Peace Support Operations
COMINT - Communications Intelligence SAR/MTI – Synthetic Aperture Radar/Moving Target Indicator
EEZ - Exclusive Economic Zones SCAR - Strike Coordination and Reconnaissance
EO – Electro Optical SIGINT – Signals Intelligence
ESM – Electronic Support Measures UAV – Unmanned Air Vehicle
FFTTEA - Find-Fix-Track-Target-Engage-Assess UCAV – Unmanned Combat Air Vehicle

HALE – High Altitude Long Endurance URAV – Unmanned Reconnaissance Air Vehicle
IR – Infra Red VTOL – Vertical Take-off & landing
-1-

ISBN-xxx-y-zz-yyyyyyy-x 113
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

1. Introduction

Being Cassidian the Defense branch of EADS and the owner of the UAS, (since they were
incorporated to our portfolio more than ten years ago), it has been our constant task to carefully
analyze, describe and update the potential applications of such unmanned systems, (specially
for security purposes).

This task is accomplished by well-established Company teams where contributors come from
various disciplines inside Cassidian as Strategy, Operational Analysis, Preliminary System
Design, R&D, etc always filtered with the customer feedback gathered trough the multiple events,
working groups and official institutions where EADS is present as the recognized industrial
leader in Europe.

This paper summarizes our findings on this field during the last decade, having its content full
validity as today although for obvious reasons all company confidential aspects have been
removed.

2. Air Power Landscape

2.1. - Peacetime Air Sovereignty (From Air Policing to Air Space Security)

Air Policing has been a major task for air forces during peacetime. Standard procedures can be
characterized as reactive upon an alert from air traffic control, military or civilian, by dedicated
fighter aircraft held in alert status. Three trends can be identified in this area:

a) First, an extension of the basically reactive scheme from ground alert to pre-emptive
provisioning of fighter aircraft for extended time periods to secure a limited air space against
intruders, e.g. in case of a special events as a summit. Due to the extension of potential threats
to low and slow flying smaller objects and the consequently shortened timelines, Air Patrolling,
also together with Air Surveillance by on-board sensors, is then the preferred operational
scheme.

This asks for fighter aircraft with extended endurance / time on station, short turn around,
increased MTBF for relevant systems and short repair times, in order to keep required fleet sizes
at a minimum. Although this might indicate the utility of UAVs, the specific demands of such an
environment with highly critical rules of engagements, visual identification, interaction with the

-2-

ISBN-xxx-y-zz-yyyyyyy-x 114
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

object and prescribed procedures make it unlikely for UAS application.

b) Second, further to the above trend in Air Policing, the need for securing the airspace in
general arises. This means an extension to permanent operation in enlarged areas, ideally the
whole of air space when threat launches within the territory must be accounted for. Prior to this
action, Air Surveillance has to be accomplished, necessarily (supported) by airborne sensors
because of the nature of some critical objects (low and slow and small). It is essential that false
alarms be kept at a minimum, this putting new and peculiar demands on sensors and alert
identification schemes (e.g. behavioural). UAVs or even LTAs could be an option to fill up Air
Surveillance, but for radar as the prime all weather sensor, acceptable coverage come with
weight and power off-take. In the light of these determinants, AEW functions appear currently as
out of reach for UAVs, although this is a desirable UAS application. When novel hybrid, e.g.
bi-static technology, evolves for air surveillance sensors, this could change and Air Surveillance
be made accessible for UAVs.

c) Third, in analogy to the above, the need for homeland security and defence in peacetime,
crisis and war, is manifest. Here, the population, the infrastructure and the territory have to be
defended against all sorts of air threats from outside, land or sea, including missiles and small
planes, also without pilot. These could carry NRBC loads and thus should preferably be kept
away from the territory. This asks for detection, alert and action as forward as possible. UAS
contributions can be viewed like for air space security on the one hand. On the other, UAVs for
air-to-air action might be attractive when positive identification is already assured. However, it
has to be kept in mind that action would still require specific sensors and weaponry. Besides this,
significant numbers of fighter aircraft weapon systems are already in the inventory, ready to
cover such tasks, at least when potential threat directions can be determined or timelines favour
ground alert.

Homeland Defence against a ballistic missile threat is a specific topic, ranking high on the NATO
agenda since the Prague summit. In the context of Theatre Ballistic Missile Defence, potential
contributions from UAVs have been extensively analyzed. The findings can be briefly
summarized as follows: for terminal defence, combat aircraft are, in many respects, inferior to
land-/sea-based systems as actors, and are not suited for mid-course interception. They can be
used for forward (boost-ascent phase) interception, when UAVs might be utilized as well as
sensor/fire control and as interceptor carriers. Because of the BM ranges encountered in
homeland defence (except for sea launched shorter range missiles), however, forward defence

-3-

ISBN-xxx-y-zz-yyyyyyy-x 115
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

does not appear as an option here. The only application of UAS could be to support Early
Warning and Fire Control for terminal defence with IR sensors stationed at altitude in MALE or
HALE orbits.

2.2. - Power Projection

This subsumes the core air and space activities "Counter Land", "Counter Sea" as well as
1
aspects of PSO (Peace Support Operations), Stabilisation Operations, Petersberg Tasks and
Coercive Action. In terms of operating conditions, the challenges posed and the means involved,
Conventional Counter Force (CCF) is included here as well.

Whereas during the Cold War the support to land forces, second to conducting the air war, had
its focus on traditional AI/BAI (Air Interdiction / Battlefield Air Interdiction), and CAS (Close Air
Support) was largely transferred to organic army means, a remarkable shift has since occurred,
and the immediate contribution to activities on ground has gained prominence. However, it has to
be kept in mind that this is also due to the fact that recent conflicts, from the second Gulf War, did
not see an opponent Air Power matching Western capabilities.

The established trend in land warfare can be characterized by the dissolution of the "linear
battlefield", with smaller units of land forces also operating in the depth of hostile territory. They
may not have a physical link to neighbouring units and operate on a more and more "empty
battlefield", also because the opponent is forced to disperse and hide when facing an
overwhelming air threat. Such agile, lighter troops carry less firepower and protection than
armoured mechanized units. In the extreme, we see Special Forces as well as small contingents
of non-combat troops in PSO operating "on their own". In parallel, following an analogous
rationale, asymmetric response by the enemy in particular, an ever increasing importance of
MOUT (Military Operations in Urban Terrain) is obvious. Here, dispersion and concealment of
enemy forces occurs in the civilian population and a built-up environment. Third, the operating
conditions and asymmetric response result in a permanent threat to supply and communication
lines.

In consequence, the demand on Air Power can be characterized by: information gathering for
operational pictures with monitoring and surveillance of extended areas or lines with minimum

1
Humanitarian and rescue tasks, peace keeping tasks and tasks of combat forces in crisis
management, including peace making
-4-

ISBN-xxx-y-zz-yyyyyyy-x 116
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

latency up to targeting (ISTAR), in particular for increasingly difficult (concealed, camouflaged)

mobile targets with short move-dwell cycles, provision of "fire power on demand" in a
sensor-to-shooter connexion for Time Sensitive Targeting in a Joint Fire Support context. Effects
Based Operations ask to emphasize accurate (time and location), tailored/graded effects, in
particular in the vicinity of civilians or civil installations. In terms of doctrinal missions, this
evolution affects the Air Interdiction (AI), Close Air Support (CAS) in particular, Strike
Coordination and Reconnaissance (SCAR), and is at the heart of Air Power contributions in PSO
and Petersberg Tasks.

UAS are expected to provide for significant contributions across these missions and core combat
roles of UCAV are to address the above spectrum as far as technology allows.

The recently created SCAR mission type deserves special attention as it explicitly mentions
"unmanned armed reconnaissance", particularly when projected into low intensity warfare. Whilst
Offensive Air in COMAO (Composite Air Operations) or Sensor-to-Shooter via Recce-Attack
Interface or even Armed Recce ask for URAV or a UCAV recce derivative, a low threat
environment allows for "the employment of non-specialized ISR/C² aircraft, typically armed with
munitions and system that better enhance target designations, to locate, coordinate, attack or
target mark for AI missions away from the presence of friendly forces". A weaponized MALE falls
into this class.

In the maritime domain - Counter Sea - , similar considerations apply for the ASuW (Anti Surface
Warfare) mission. Whereas attacking naval vessels corresponds to the attack on defended
ground targets, small maritime targets like fast boats, fishery boats, etc. can be compared to
transient ground targets, at least with respect to demands on identification and reactivity. Again,
UCAV are viewed with a potential in response to such missions, complementing current manned
combat aircraft. This can also be extended to carrier based operations, as envisioned by the US
Navy.

At the time, a judgement towards the suitability of UAS for ASW (Anti Submarine Warfare)
appears as premature – no concepts are known. Even for the MPA task to detect submarines, a
number of questions remains open, e.g. with regard to the suitability of UAV for sonobuoy
operations. Aerial Mining, the third mission class in Counter Sea, is not seen fit for UAS
application.

-5-

ISBN-xxx-y-zz-yyyyyyy-x 117
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

For Coercive Action, especially at long distance, a potential for UAS could be identified, in
particular when a final target confirmation cannot be trusted to a weapon, but again no concepts
are known.

For completeness, Counter Space is to be mentioned, but, e.g. in anti-satellite, this is more an
issue of appropriate effectors than of carriers. All of the application areas addressed above rely
heavily on the information provided by ISR means. This is addressed in the sequel.

2.3. - Intelligence-Surveillance-Reconnaissance (ISR)

ISR is pre-requisite to the concept of Information Dominance and central to all sorts of military
operations discussed above. At the same time, its notion and importance extends into the
civil-military security domain and even beyond into purely civilian (e.g. industrial) demands. In
addition to this vast spectrum, ISR is the main UAS application domain, as of to-day and with
continuing growth, and a huge manifold of systems and concepts are around.

Because of the many facets involved, the following can only attempt to highlight some primary
aspects with special regard to UAVs and without trying to systemize or cover the whole. Micro- or
Mini-UAVs as used with small land forces units will be excluded from these considerations.

In Strategic Reconnaissance, UAS are already well established for extended area imaging
ground surveillance as well as for signal intelligence (although for peacetime global intelligence
gathering and COMINT, manned mission aircraft remain the means of choice). This relates to
HALE-, partly to MALE-, based systems, and because of the mutual limitations, they will rather
complement than compete against satellites.

Because of their basic range-endurance characteristic and the routine character of the missions,
it is easy to predict further capability and application growth, mainly through their sensors
(resolution, tracking, etc.). It may even be postulated that UAVs will supersede manned mission
aircraft in this area (with the exception of worldwide SIGINT). However, it is clear that such UAVs
can only operate in a benign environment. As soon as Long or Medium Range SAMs or even
fighters arise as a threat, survivability cannot be kept to an acceptable level. Stand-Off operation
is generally at the expense of covered area of interest and tends to limit the mission strongly.

-6-

ISBN-xxx-y-zz-yyyyyyy-x 118
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The area of Operative Reconnaissance possesses some analogy to the above, with more
demanding requirements on sensors and persistence because Target Acquisition is included.
The (capability) requirements leading to the Cassidian MALE “Talarion” cover the mission needs
in this area. However, it is questionable whether a concept based on a MALE UAV can fulfil the
full demand. Operation is within a theatre, thus under threat, and for difficult targets, e.g. urban,
concealed, etc., other flight characteristics like dwell may be required. Therefore, complementary
systems, preferably UAVs, can find their place. Such complements are, for part of the spectrum,
already envisioned for the organic recce of army and navy. VTOL UAV falls into this category, but
it should be clear that the "vertical" extends beyond take-off and landing to sensor dwell over
specific spots.

Beyond traditional naval mission needs, a strong growth in the demand for Maritime Surveillance
can be foreseen. In a general security notion, military and civilian, the maritime area is gaining
importance. Further drivers are increasing regulations towards a Maritime Traffic Management
as well as a closer control of Exclusive Economic Zones (EEZ).

Tasks to be performed also in peacetime comprise cueing (from intelligence, MTM, wide area
monitoring including AIS) for specific activities, surveillance/observation of pre-determined areas
with detection and tracking of objects afloat, classification of such objects, identification –
non-cooperative in specific cases, finally alert and action, even up to weapon employment. Thus
the tasks can broadly be categorized into Surveillance-Reconnaissance-Action.

A large number of different systems, i.e. coastal radars, satellites, ships and aircraft linked to
2
various C instances provide information and contribute to these tasks, and the first challenge is
to interface these multi-national and multi-agency inputs for the production of maritime pictures,
building upon Surveillance and Reconnaissance.

The means involved contribute in different domains with their specific characteristics, and to
arrive at suitable roles for UAS, a segregation of the maritime domain together with a projection
of means onto the tasks per sub-domain has to be done. Such segregation may lead to a
differentiation of high sea, passages/shipping lines, approaches, EEZ/shelf, straits/choke points,
littoral and harbours.

-7-

ISBN-xxx-y-zz-yyyyyyy-x 119
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

UAVs may be envisioned across these areas, to different degrees/tasks. This refers in particular
to a naval HALE (with maritime radar, SAR/MTI, AIS, ESM), a naval MALE (ditto, plus EO/IR
sensors and further sensing equipment), possibly a weaponized MALE.

Drones, e.g. for harbour protection are outside our scope, and ship based means will be
addressed under Tactical Recce below.

As already addressed in the preceding sections, persistent Air Surveillance is a core capability
across Air Power applications and beyond into the peacetime security domain. Because of its
operating environment, all weather in particular, and because of the challenge posed by
increasingly low signatures of flying objects, together with low and slow flight profiles, powerful
radar capabilities enabling large area / air space coverage are key in the mid-term time horizon.
As of to-day, this excludes UAVs from the AEW domain. In an extended / far-term horizon, it can
be imagined that either novel technology could allow for a smaller, reduced power off-take,
receiving component to be installed on a HALE/MALE type UAV or that the emergence of large
UAVs, e.g. for air transport, could provide for an Off-The-Shelf UAV platform on which a big radar
can be installed.

Whereas the above areas are primarily related to joint (in the armed forces context) and
cross-service (in the security context) aspects, there is a continuing need for Tactical
Reconnaissance organic to army, navy and air force. Legacy systems responding to this need
from the air are drones and tactical UAVs with army, (manned) helicopters for (smaller) navies
and (manned) reconnaissance aircraft in the air forces.

Whilst the general need progresses continually into the future, the demand is changing in quality
and asks for novel sensing and platforms, besides other drivers like Time Sensitive Targeting
and a Sensor-to-Shooter connexion.

For organic army purposes, this demand can be characterized firstly by detection,
reconnaissance, targeting of concealed and camouflaged targets, including the urban
environment with a reach requirement up to and beyond 150 km. This may ask for dwell above/in
the vicinity of suspected targets, and such a dwell capability is also an enabler for specific tasks
like CBRNE (Chemical, Biological, Radiological, Nuclear or Explosive) detection, including mines
and IEDs (Improvised Explosive Devices). Clearly, UAVs are the means of choice here.

-8-

ISBN-xxx-y-zz-yyyyyyy-x 120
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The trend is from current fixed wing drones towards VTOL, not only facilitating the operation by
army units but also providing the desired dwell capability. Then the "VTOL" notion falls short, and
a hover capability beyond take-off and landing is essential.

For organic naval purposes, the above basic tasks arise again (in part), especially when littoral
operations are taken into account. Beyond this, additional features emerge. Naval operations on
board vessels from Corvette class are a main driver, and object detection, classification,
recognition and identification has to cover ships of all sizes. Reach extends up to 200 NM,
comprising some 150 NM into the shore region. This enforces a BLOS (Beyond Line Of Sight)
capability. As of to-day, helicopters are used for such purposes, based on Frigates (or bigger,
with hangars), but there is a clear demand for UAVs of smaller size, compatible with Corvettes
and requesting only small space aboard the ship. Actually, unmanned helicopters are being tried,
but other concepts may prove better suited.

Concerning the air force, the situation looks different, and no clear trend can be acknowledged.
Currently, the "Network Enabled Operations" still dominate, embracing concepts like sensor grid
and shooter grid. Here, the ISTAR (Intelligence, Surveillance, Target Acquisition and
Reconnaissance) functions are believed to be covered by persistent monitoring of the area of
operations through (joint, but air force operated) ISR means as addressed above, so that no air
force organic tactical recce is needed. But, in our view, this will not be maintained into the future.
When current legacy manned reconnaissance aircraft like Tornado Recce or ECR (Electronic
Combat and Reconnaissance) go out of service, a gap can be foreseen which arises within the
missions addressed above under Strategic Attack, Offensive Counter Air, Air Interdiction,
Coercive Action into the depth of hostile airspace.

Because of distances and threat, besides other drivers like information transfer, essential
functions within the Find-Fix-Track-Target-Engage-Assess (FFTTEA) cycle model are not
covered with joint ISR for a good portion of operational conditions, in particular when Time
Sensitive Targeting is required.

To resolve this emerging gap, three tracks can be envisioned:

-9-

ISBN-xxx-y-zz-yyyyyyy-x 121
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

- Reliance on a better on-board air-to-surface sensing for fighter/combat aircraft, with

(more) on-board processing and sensor/information fusion plus intra-attack-package
information exchange,

- Creation of a dedicated manned recce derivate – integration of a recce pod together

with on-board processing and information exploitation plus the above mentioned
information exchange with "shooters",

- Development of a dedicated UAV based system for tactical reconnaissance to operate

within a sensor-to-shooter connexion in an integrated way. Options are:

 UAVs conceptually related to recce

 A "flying pod" URAV, air launched and controlled by a combat aircraft (manned
derivate)

 A UCAV Recce derivative as addressed above in the Power Projection section.

As previously emphasized, the ISR applications extend beyond pure military tasks to civil-military
ones where a growing demand emerges due to the increase of the possibilities given by the UAS
in terms of persistence and cost. In the Security related ISR (civil-military), the monitoring of local
and time-limited events (e.g. sport championships, political summits…) could further develop the
UAS applications allowing a better control. This could also contribute to crowd control when riots
emerge as well as support disaster relief activities.

Another need in this field regards the infrastructure surveillance, especially for traffic monitoring.
Insertion into civil air traffic is mandatory for such applications, but this is already recognized for
MALE/HALE and related projects are under way. Lastly, as mentioned above, the maritime
surveillance, in particular of the EEZ (Exclusive Economic Zone), is gaining importance and
more and more viewed as a vital part of the territory.

Another Security related ISR task is the constant surveillance of some larger areas, for which
UAVs are particularly suited via their reach-endurance characteristics when they keep fleet sizes
at a minimum. For these applications, the availability of some automatic detection algorithms as
well as stable and weather-proof sensors is key. The surveillance of the border or of a pipeline
requires a permanent surveillance of a large area and the detection of small details whereas the
sea traffic surveillance does not need such a precision. For such applications, specific demands
on a UAV can arise, in analogy to those addressed above for army organic recce.

- 10 -

ISBN-xxx-y-zz-yyyyyyy-x 122
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3. Conclusions

Potential UAS applications in the security domain have been presented. A clear link with
technology evolution and maturity, the various manned solutions available today, the importance
of manned-unmanned synergies and our belief that the use of existing assets must be extended
and optimized has been highlighted.

Trying to rank these broad areas in terms of relevance is difficult and depends on the time
horizon, on perceived evolution in the security domain and on means already in the inventory.

For a shorter horizon, say, up to 2020, it can be expected that the area of power projection is the
driver, including the ISR integral to it. If not boosted by disrupting events, homeland security can
be expected to gradually grow further in relevance, also extending into airspace aspects, again
together with related ISR. Communication needs, e.g. back-up to the civilian infrastructure, may
become an issue.

Further and derivate applications arise when UAS upon which to draw are already in place.

References

1. “Allied Joint Doctrine for Air and Space Operations” AJP-3.3 Ch2 SD2, NATO, January 2006

2. Hezlett, A. “Aircraft and Sea Power” New York, Stein and Day, 1970

3. Laite, B.C. “Maritime Air Operations” London, Brassey´s 1991

4. McCormick, Gordon H. “Terrorist Decision Making” Annual review of Political Science

5. “National Strategy for Homeland Security” Washington, D.C. The White House, July 2002

6. Pardesi, Manjeet Singh “Unmanned Aerial Vehicles / Unmanned Combat Aerial vehicles: Likely

Missions and Challenges for the Policy-Relevant Future” Air and Space Power Journal XIX, no. 3.

7. Tedder, R. “Air Power in War” London, Hodder & Stoughton, 1957

8. Tirpak, John A. “Find, Fix, Track, Target, Engage, Assess” Air Force Magazine 83 no. 7.

- 11 -

ISBN-xxx-y-zz-yyyyyyy-x 123
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

OPARUS - OPEN ARCHITECTURE FOR BORDER

SURVEILLANCE UAS

JORGE BUENO*, JUAN ALBERTO HERRERIA**

and OLIVIER REICHERT***

*Ingenierı́a de Sistemas para la Defensa de España, S.A., Spain, [email protected]

**Ingenierı́a de Sistemas para la Defensa de España, S.A., Spain, [email protected]

***SAGEM Group SAFRAN, France, [email protected]

Abstract

Illegal immigration and smuggling in Europe is a hot issue that is jeopardising the security of
the European Union and the safety of those who illegally attempt to reach European territory under
precarious conditions with the hope of a better life that might never occur. European authorities
know the problem and are concerned about the better methods to deter smugglers and illegal immi-
grants from such practices by using the appropriate technology and the legal framework established
by the European Parliament and the respective Member States. One technology that is revealing
as key to detect, identify, and track smugglers and illegal immigrants is Unmanned Aircraft Sys-
tems. OPARUS (Open Architecture for UAS based Surveillance System) Project, co-funded by the
European Commission under the 7th Framework Programme aims at defining a technological and
cost-efficient Open Architecture for UAS engaged in border surveillance within European Territory,
either over land or sea territory, that is compliant with the ever evolving and not already finished
regulatory and standardisation framework. The present paper gives an overview of the project, the
scenarios developed and the results achieved so far.

keywords: UAS, border surveillance, Framework Programme, open architecture

1 INTRODUCTION TO THE OPARUS PROJECT

The OPARUS (Open Architecture for UAS based Surveillance System) Project is a Coordination and
Support Action co-funded by the European Commission under the 7th Framework Programme. The
project started in September 2010 and it will run for 18 months. Its overall objective is to define an open
architecture for the operation of unmanned air-to-ground wide area land and sea border surveillance
systems in the European Union.
The OPARUS project comprises six work packages namely. WP1 Concepts and Scenarios, WP2
Legislation Analysis, WP3 Technical Analysis, which is split into WP31 Surveillance Sensors, WP32

1
ISBN-xxx-y-zz-yyyyyyy-x 124
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Platforms, WP33 Secure Data Link and Network, WP34 Independent Control Station, WP4 Open
Architecture Definition, WP5 Information Exchange and Dissemination and WP6 Project Management.
Project partners are amongst the key actors in the field of UAS and security, including SAGEM (as
project Coordinator), Polish Air Force Technology Institute, BAE Systems, Dassault Aviation, DLR,
CASSIDIAN (EADS-CASA), IAI, INTA, ISDEFE, ONERA, SELEX GALILEO, THALES Systemes
Aeroportes, THALES Communications and Tony Henley Consulting Ltd. In addition a number of end
users participate actively in the project activities through the End-user Advisory Board supporting the
definition of the scenarios and the selection of the open architecture. Today, these are FRONTEX,
Guardia Civil, Polish Border Guards and Polish Army, Guardia di Finanza and South Coast Border
Partnership.
The concept of Open Architecture has a six-fold meaning in the framework of OPARUS.

• It is based on non-proprietary standards

• It is based on a regulatory framework that enables a coordinated operation

• It implies that there exist common requirements for sensors integration on the unmanned aerial
platform to enable platform adaptation to the specificity of the mission

• It allows that unmanned aerial platforms are capable of perform the same mission regardless of
the country or operator

• It operates in an environment where there is cooperation amongst nations and bodies to coordinate
and manage the missions and assets to be protected.

• Coordination amongst nations does not only imply political coordination, but also effective coor-
dination when an unmanned aerial platform that needs to cross borders between nations is allowed
to be piloted from another control station through handover procedures.

In addition an open architecture does not only implies that it is only open in the sense described
above, but also it allows to operate together with other open architectures, in other words, UA based
in open architecture 1 is capable to operate in the same physical environment carrying out the same
mission as UA based in open architecture 2, etcetera. This concept relies in what is usually known as
Open Systems Interconnection.
The open architectures to be defined within the framework of the OPARUS project is driven by
the scenarios and missions developed by WP1 based on the requirements provided by the End User
Advisory Board and in agreement with them through the interaction in the first Workshop that took
place at FRONTEX facilities in Warsaw. Three main scenarios have been developed in WP1.

1.1 North Poland Land Borders

Poland due to its geographical situation in the East of Europe is one of the natural borders with non-
Member States. The fact that Poland is a place where average salaries are much higher than those across

2
ISBN-xxx-y-zz-yyyyyyy-x 125
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

the eastern border, together with not solid democratic eastern states, makes Member States like Poland
an attractive destination. Illegal immigration as well as smuggling and organised crime is becoming a
serious concern for Polish Authorities who are concerned about the creation of a security line between
Poland and Ukraine, Belarus and the Russian Federation whose citizens are able to cross into Poland
easily by Land.
Illegal immigration occurs both during the day or night and during the four seasons. Surveillance
has to be carried out across an area of 80x600 km in a daily basis in order to look for individual persons
(or group or persons), cars and trucks. The responsibility for border surveillance lies within the Polish
Border Guards who undertake actions for surveillance, identification and interception. The total border
extension is split into smaller areas covered by different patrols separately.

1.2 South Mediterranean

Illegal immigration in the Mediterranean area occurs mostly in summer during night due to the good
weather conditions. Surveillance has to be carried out across an area of 100x50 NM during the first
surveillance phase and of 100 NM radius during the second search phase in a daily basis to look for low
speed boats full of illegal immigrants. The responsibility for border surveillance depends on national
authorities of the countries involved. In addition, Morocco and Spain operate joint naval patrols to
detain boat immigrants. Italy and Libya carry out joint maritime patrols too.

1.3 Canary Islands

The proximity of the Canary Islands to the West African coasts makes them an attractive and accessible
point of entry to European territory. Illegal immigrants arrive to Canary Islands coasts in small and
unsafe boats, risking their lives most of the times.
The EU has agreements with some sub-Saharan African countries for cooperation against illegal
immigration. Patrols from both EU countries and sub-Saharan countries cooperate in the interdiction
activities of vessels engaged in illegal immigration.
In some cases the hazardous embarkations used do not allow immigrants coming from western Africa
to make the entire journey to the Canary Islands. Therefore they are first transported in larger boats
(mainly old tuna ships) during the major part of the trip and then they transfer to smaller boats when
approaching the final destination.
Illegal immigration occurs mostly during the night in the summer. Surveillance has to be carried
out across an area of 100x50 NM during the first surveillance phase and of 100 NM radius during the
second search phase in a daily basis to look for low speed boats full of illegal immigrants. In addition
surveillance is also performed upon request based on intelligence information.
Besides these three scenarios, a theoretical approach to two more scenarios has been followed although
they are not considered for follow up activities in the project. These two scenarios are

• Greek and Turkish land border

3
ISBN-xxx-y-zz-yyyyyyy-x 126
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

• English Channel

2 DEFINITION OF THE OPEN ARCHITECTURE

The technological core of the open architecture is based on four areas:

• Surveillance sensors: active, passive, imaging, and non-imaging.

• Platforms: fixed wing, vertical take-off and landing, tilt rotor, airship.

• Secure data link: line of sight, beyond line of sight, networks.

• Independent Control Station: command and control of the aircraft, control and process of sensors.

The surveillance sensors analysis is based on four aspects, namely situation awareness provided by
sensors, available sensors performances, sensor collaboration and technical solution issues to perform
surveillance mission.
Situation awareness analysis consists on the identification of the principles, cuing, fusion, planning
and scheduling requirements. The analysis is performed to identify all the properties that sensors have.
Due to no sensor type can alone fulfil all the needs of a single mission in one scenario, the use of different
airborne sensors combined (sensor collaboration) is a value to perform surveillance operations. Sensor
collaboration will complement coverage of one single sensor alone and will increase mission performance
in terms of gain, data rate, and capacity and detection capabilities as well as reduce the limitations due
to adverse weather conditions.
There is a wide variety of sensors with different performances:

• passive electro-optical (EO) and infrared (IR) sensors

• synthetic aperture radars (SAR) inverse synthetic aperture radars (ISAR)

• real aperture or side-looking airborne radars (RAR/SLAR)

• active optical sensors (LIDAR/LADAR light detection and ranging/laser radar)

• special usage sensors like gamma-ray & neutrons detectors

Choosing an appropriate platform class strongly depends on the missions and scenarios and requires
many different types of data. Every platform has its own characteristics (e.g. operating height, speed)
but must also be capable of carrying and supporting other sub-systems (e.g. sensors, data link). With
regard to this, there are many types of platforms that might be suitable to perform border surveillance
missions, these include fixed wing UA (small, medium and large), rotary wing UA (helicopters, tilt
rotors, hybrid helicopters, etc.), lighter than air (aerostats, airships, hybrid airships). In order to select
the appropriate platforms types, the information was gathered and grouped in eight sections: vehicle
manufacturer and programme; vehicle type and physical attributes; propulsion and power generation;

4
ISBN-xxx-y-zz-yyyyyyy-x 127
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

launch recovery and deployment; air vehicle performance; safety, command and control, already inte-
grated mission sensors and costs. The data gathered serve to propose initial options to operate the
platforms in the selected scenarios.

• Small UA can fit in the back of a 4x4 vehicle and can be operated by small police or army units.

• Lighter than air UA can operate from a farmers field, can give very long endurance missions and
can gather information silently

• Vertical take-off and landing UA can operate from small patrol boats on the sea

• Medium UA can operate from small airfields and can cover sizeable areas of land or sea borders
rapidly

• Large UA can operate a long way from their bases and can stay on station for many hour

From the point of view of communications and networking, OPARUS technical analysis is split
into data links and networks. It is important to mention that OPARUS is not dealing with analogue
data links and not addressing ATC data links as they can be purchased from Commercial Off-The-Self
(COTS) products.
Table 1 shows the three main types of data links addressed to define the open architecture.

Table 1: Types of data links addressed.

Radio Line of Sight Beyond Radio Line of Sight

Cellular Mobile Service
(RLOS) (BRLOS)
Command and control:
2G technologies
L-DACS1 DVB-S/S2/SH
(EDGE)
L-DACS2
High data/rate payload:
WIMAX
3G technologies
DVB-S/S2 Ku-band dedicated
(CDMA)
DVB-T/T2
DVB-SH
4G technologies
Inmarsat
(LTE/LTE advanced)
Iridium

With regards to networks, Internet Protocol (IP) is considered without questioning it for dissemina-
tion purposes. There is a set of desired features for the network that are intended to enhance the border
surveillance missions performed by UAS.

5
ISBN-xxx-y-zz-yyyyyyy-x 128
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

• Quality of Service: voice and video transmission needs low latency

• Mobility: There might be nodes to be moved in the network

• Transport: Real time, multimedia, multicast

• Session/Signalling: finding the requested information

• Presentation: a common representation of information is needed for interoperability

It is important to mention that the open architecture and interfaces proposed in OPARUS will be
compatible with the ones proposed by FRONTEX in the framework of the EUROSUR Programme to
be able to share information between Member States.
Control Stations are not marketed as stand-alone, not even assigned a brand name. The degree of
sophistication and complexity is related to the associated UA and payloads capabilities Control station
has three main functions:

• Mission planning.

• Mission control.

• Data manipulation.

They can be located in or near area of operation but can also be controlled from very distant
locations. UA Pilot can control aircraft (high level command) with joystick or Control points (pointers)
and mission plans are pre-loaded into the UA during the pre-take-off phase, but they can be re-planned
in real time.
In the case of medium and large UA Pilot generally operates from 2 mission displays - normal pilot
cockpit display and a route or map display. Sensor operator has 2 displays also - one showing sensor
status and one showing route of flight.
For small, man-portable UA, simple, easy to use systems are converted from a system packed in a
rucksack on operators back to an operable system in five minutes. The control station, without the
laptop, weighs less than 4 kg. The operators of these units usually do not need pilot training.
Increased use of UAS recommends commonality among Control Stations to save money in training
and development costs.
Based on the survey and classification of the above technology, a set of potential architectures will
be defined. Although at the moment of writing these lines, not all the potential architectures were
defined in detail, an example is shown in figures 1 and 2 below. These architectures are still pending of
being assessed and thus no more information can be given on whether they are suitable ones for being
implemented in the short term or may need further research for a long term implementation.
In the possible alternative shown in Figure 1 for the North Poland Land Borders, the following
components are considered:

• UA and sensors

6
ISBN-xxx-y-zz-yyyyyyy-x 129
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 1: One open architecture alternative for the Polish Land Border scenario

– a mini (20 kg MTOW) or small UA (60 kg MTOW) equipped with one low end IR sensor,

– or a medium tactical fixed wing (300 kg MTOW) or large tactical fixed wing (500 kg MTOW)
UA.

• Control Station (CS)

– The mini UA is commanded and controlled from a mobile CS through a RLOS data link.

– The small, medium tactical fixed wing and large tactical fixed wing are controlled from a
fixed CS through a RLOS link.

• Data link and network

– There are ground data link stations (antennae) deployed along the border to provide RLOS
coverage.

– The network between the CS and the control centre (local and regional) is established through
a radio link due to existing infrastructure. It is recommended to implement a wired Ethernet
or optical fibre to increase bandwidth.

– An Air Traffic Control (ATC) link with the UA pilot exists for coordination. It can be either
a direct telephone link or a UA relay.

In the possible alternative shown in Figure 2 for the Canary Islands, the following components are
considered:

• UA and sensors

– a light MALE BRLOS fixed wing UA (1000 kg MTOW) equipped with a maritime radar in
the X band and medium end EO/IR sensors,

7
ISBN-xxx-y-zz-yyyyyyy-x 130
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 2: One open architecture alternative for the Canary Islands scenario

– or a medium MALE BRLOS fixed wing UA (3500 kg MTOW) equipped with a maritime
radar in the X band and medium end EO/IR sensors.

• Control Station (CS)

– UA are commanded and controlled from a fixed CS located in the Canary Islands through a
BRLOS link

• Data link and network

– The network between the CS and the control centre (local and regional) is established through
a radio link due to existing infrastructure. It is recommended to implement a wired Ethernet
or optical fibre to increase bandwidth.

– An Air Traffic Control (ATC) link with the UA pilot exists for coordination. It can be either
a direct telephone link or a UA relay.

At the same time a maritime patrol equipped with maritime radar is patrolling the Atlantic Ocean
area between the Canary Islands and the African coasts. Once the maritime radar detects a point of
interest, a medium VTOL (200 kg MTOW) or mini fixed wing (20kg MTOW) UA is launched from the
ship to identify and track the illegal immigrants or smugglers.

3 ASSESSMENT OF THE OPEN ARCHITECTURES

This section intends to give an overview of the assessment process to be followed in order to select the
most suitable open architecture to perform border surveillance missions with UAS in the three scenarios
selected.

8
ISBN-xxx-y-zz-yyyyyyy-x 131
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The assessment and selection of the Open Architecture method is performed within OPARUS WP4,
although the method to be used in the process implies collaboration with the rest of OPARUS WPs.
Such assessment implies to consider the different alternatives from four perspectives.

3.1 Operational efficiency perspective

To assess operational efficiency of the proposed open architectures, a trade-off between operational
scenarios and missions, and technology selected surveillance sensors, platforms, data links and control
station. The assessment will specifically consist in determining if the technology selected matches the
expectations of the operations in an efficient way, or in other words, the technology selected matches
the expectations of the end users. Therefore technology selected will be cross-checked with the missions
and operational requirements previously captured.

3.2 Legal perspective

To assess the proposed open architectures from a legal assessments, the trade-off have to be in this case
between operational scenarios and missions, legislation references and technology selected surveillance
sensors, platforms, data links and control station. The trade-off should result in requirements for the
selection of the open architecture where the legislation issues, technology and operations are all in
agreement. Opposite, the trade-off should result in constraints, and therefore a gap that would need
further research, where the legislation issues, technology and operations are not in agreement.

3.3 Technology performance perspective

The assessment of the technology proposed for the open architectures will be made through the assess-
ment of their performance. Aspects such as reliability, maturity (e.g. compared to NASA TRL) will be
addressed to determine if the technologies considered are appropriate to be included in one or another
alternative of the open architecture.

3.4 Costs perspective

The cost assessment will be based on the purchase cost that every technology selected has when inte-
grated. The combination of platforms, sensors, data link and control station (i.e. alternatives to open
architecture) costs will be then compared with other alternatives to open architectures to assess whether
the costs of the architecture related to operational capabilities and efficiency is cost-effective or not.

4 SUMMARY
Throughout this paper we have presented the on-going work of the OPARUS project aimed at defining an
open architecture for border surveillance missions performed by UAS in three different scenarios, namely

9
ISBN-xxx-y-zz-yyyyyyy-x 132
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

north Poland land border, south Mediterranean and Canary Islands. The work done by OPARUS part-
ners does not imply only to define the open architecture, but to define the scenarios and to collaborate
with other on-going projects such as PERSEUS or SEABILLA in such a definition to avoid duplication
of work and to achieve an homogeneous view of the border surveillance problem across the European
Union. Although the work presented here is a preliminary one, as complete results are not ready yet,
it gives an overall view of what it is expected to be presented once the project is finished. Additional
information is available on website: http://www.oparus.eu

5 ACKNOWLEDGMENTS
The authors want to thank the European Commission for funding this interesting project (Grant Agree-
ment No: 242491) and to the rest of the OPARUS partners, SAGEM, Polish Air Force Technology
Institute, BAE Systems, Dassault Aviation, DLR, CASSIDIAN (EADS-CASA), IAI, INTA, ISDEFE,
ONERA, SELEX GALILEO, THALES Systemes Aeroportes, THALES Communications and Tony Hen-
ley Consulting Ltd for interesting discussions in order to give content to the work being done within
the project as well as to the end users today involved in the project, FRONTEX, Guardia Civil, Polish
Border Guards and Polish Army, Guardia di Finanza and South Coast Border Partnership, for giving
us advise and guiding us in the definition of the scenarios and open architecture.

6 REFERENCES

REFERENCES
[1] OPARUS Consortium Annex I Description of Work, (2010).

[2] OPARUS Consortium D11 Concepts, Scenarios and Missions for Border Surveillance with UAS,
(2011).

[3] OPARUS Consortium D41 Open Architecture for the Operation of UAS in Border Surveillance,
(2011).

10
ISBN-xxx-y-zz-yyyyyyy-x 133
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Rapid Development of Integrated 3D Terrain Maps using Light

Unmanned Aerial Vehicles

G. Heredia1, J. A. Cobano1, D. Pérez1, G. Núñez1, L. Durán1, F. Caballero1, V. Vega1, F.

Real1 and A. Ollero1,2
1
University of Seville, Camino de los Descubrimientos s/n,41092 Seville, Spain
[email protected]
2
Center for Advanced Aerospace Technologies (CATEC) Aeropolis, Aerospace Technological Park of
Andalusia, 41309, La Rinconada (Seville), Spain

Abstract
This paper presents a new system for the rapid development of integrated 3D terrain maps
using a lightweight Airborne Laser Scanning (ALS) mounted on a light UAV. One of the
ALS is the generation of digital terrain models and digital surface models. ALS is usually
done with manned aircraft or medium to large size Unmanned Aerial Vehicles (UAVs).
Also, a camera is mounted on the UAV and an image mosaic is built from aerial images.
The system can be transported in a small van and deployed and operated by two persons in
short time, and therefore it can be used in time-critical event response systems for
emergency situations. Experiments have been performed in a Spanish airfield.

Keywords: UAV, 3D terrain maps, Airborne Laser Scanning, LIDAR, GPS

1. INTRODUCTION
Airborne Laser Scanning (ALS), also known as LIDAR (LIght Detection And Ranging), is a very
effective and accurate method for the generation of detailed terrain maps from aerial platforms [1][2].
The availability of Global Positioning Systems (GPSs) and inertial measurement units (IMUs), which
are a combination of accelerometers and gyroscopes, has made mobile mapping systems possible,
particularly for aerial surveys [3]. This technology allows the highly automated generation of digital
terrain models (DTM) and digital surface models (DSM). Recently, ALS has been developing quickly
and it is capable of generating a point cloud which samples the target surface using laser pulses.
The applications of ALS are very diverse including: mapping (road, electrical transmission lines),
DTM and DSM generation in forested areas and urban areas, derivation of vegetation parameter, etc.
In particular, ALS or LIDAR has been employed in biomass estimation [4], in automatic roof plane
detection for solar potential assessment [5], for fine-scale mapping [6], to generate three dimensional
city models [7] or in topographic applications [3][8]. Other works perform fusion of LIDAR data and

-1-

ISBN-xxx-y-zz-yyyyyyy-x 134
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

aerial imagery [9][10].

The application of LIDAR presents a significant advantage with respect to other ones because the
data can be acquired in atmospheric conditions (reduced visibility, cloudy, etc.) where aerial
photography cannot be used.
ALS sensors are usually heavy (typically from 15 to 50 kg or more [11]), and therefore they are
mounted for operation on manned airplanes and medium-size unmanned aerial vehicles (UAV)
including helicopter [6] or fixed-wing [12]. The cost of these solutions is usually high. In the last
years lightweight ALS for automotive applications, from manufacturers such as Sick and Ibeo, have
become available. They have been mounted on a small helicopter UAV for close aerial mapping [6].
The factors affecting laser-target position accuracy in these systems are numerous, but the main
contributions come from laser scanner accuracy and from attitude and position accuracy of the aerial
platform [13]. In some applications the requirements on point density and DTM accuracy can be as
high as several points per square meter and 0.1m, respectively.
This paper presents the fast automated generation of 3D terrain maps from ALS data acquired with
a lightweight laser mounted on a small fixed-wing UAV (see Figure 1) and of image mosaic from
aerial imageries obtained by a video camera mounted on UAV. The map describes the surface
including all the objects (trees, buildings, cars, etc.) on the ground and the mosaic shows an overall
image of the overflow terrain. The ALS system consists of: a GPS, an IMU and a laser scanner. In this
paper, a precise kinematic positioning of the aerial platform by differential GPS and inertial attitude
determination by IMU provide the accurate reference to an external co-ordinate system. The system
furnishes geometric results in terms of distance, position, attitude, and coordinates.

Laser and camera

Figure 1: Laser and camera mounted on Megastar UAV.

The advantage of this system is its rapid deployment and application in emergency missions like
natural disasters or catastrophes where the terrain is inaccessible. Moreover, the cost is minimized.
For each shot, the spatial vector from the laser platform to the point of reflection is established,
thus providing the XYZ co-ordinates of the footprint. The overall vertical system accuracy is usually
in the decimeters order and most systems presently operate at flying heights of up to about 100 m

-2-

ISBN-xxx-y-zz-yyyyyyy-x 135
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

above ground. Data point density depends on the number of pulses transmitted per unit time, the scan
angle of the instrument, the altitude of the UAV above ground level, and the forward speed of the
UAV.
The paper is organized as follows. Section 2 presents the hardware system used. Section 3 presents
the sensorial integration of the system on a small unmanned aerial vehicle. Section 4 presents the
experimental data and the obtained results. Finally, the conclusions are presented in Section 5.

2. HARDWARE SYSTEM
The overall system consists of the Megastar UAV and two ground stations. The Megastar is a small
fixed wing UAV that weighs 5.9 kg, has a wingspan of 2.4 m and a payload of 1.5 kg. The first station
computes the 3D terrain map from the ALS data and the image mosaic from the aerial imagery (see
Figure 2). The link with the UAV is through of an 900 MHz Wireless Ethernet Bridge (IP922). The
second station is the base station of the autopilot, which is the AP04 in the application presented in
this paper (see Figure 2). This autopilot has an integrated radio link that allows reliable data
communications up to more than 100 km away. It includes sensors (accelerometers, gyroscopes,
magnetometers, and static and dynamic pressures), a GPS, dual redundant processors, an interface to
control servos and other peripherals, and a long-range radio link.

Figure 2: Overall hardware system.

3. SENSORIAL INTEGRATION
The sensorial system mounted on the Megastar UAV consists of a Differential GPS, IMU and air
sensors. Moreover, a laser scanner and a video camera are mounted in the UAV (see Figure 3).
The laser scanner used is a SICK LD-MRS400001. It weighs 1 kg and the dimensions are 164.5mm

-3-

ISBN-xxx-y-zz-yyyyyyy-x 136
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

x 88mm x 93.2mm. The detection range is up to 130 m.

Figure 3: Sensors mounted on the UAV.

The video camera is a small standard resolution camera that weighs 131 g. The camera sends the
images to the ground station using an analog transmitter. The image mosaic is built in the ground
station.
LIDARs together with the Global Positioning System (GPS), and the Inertial Navigation Systems
(INS) can be used to generate accurate 3D terrain map. A Kalman filter is implemented to estimate the
UAV position and attitude from the GPS, IMU and air sensors data. The combination of all the sensors
allows the positioning of a point cloud to a high degree of accuracy. This computation is carried out
offline. The measurements of each sensor should be synchronized to generate the map. First, the raw
ALS data are obtained and then a linear interpolation is applied using Delaunay triangulation.
A global reference system is considered to compute the point cloud (see Figure 4). The system
furnishes geometric results in terms of distance, position, attitude, and coordinates.
Each measured point is transformed from:
[P] = [L] ⋅ [T ]GPS ⋅ [R ]IMU (1)
where [P] is the 3D point on the global reference system, [L] is the 3D point on the LASER
Reference System, [T]GPS is the translation matrix and [R] IMU is the rotation matrix. This
transformation allows to obtain the raw point cloud, and from these data the DTM is computed.
The algorithm implemented builds the mosaic from the aerial imagery by using the homographic
relation produced by the terrain [14]. The mosaic stores all the information obtained by the camera
and defines the relation in the map between the images and the planar scenario.
The algorithm can detect the relative displacement of the UAV. However, the position estimation
deteriorates along the sequence of images. This is because of the accumulative errors in the
homographic computation. The building of the mosaic can help to reduce the errors and so to increase
the accuracy to estimate the position.

-4-

ISBN-xxx-y-zz-yyyyyyy-x 137
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 4: Reference systems considered in the Megastar UAV.

4. EXPERIMENTS
Experiments presented have been performed in the airfield of Alcazarén in the Spanish province of
Valladolid in the framework of the ROBAUCO Spanish project. This project is focused on how the
mobile robots could help in emergency situations or catastrophes. Ground and aerial robots are
considered and they cooperate to support the emergency professionals in order to increase personnel
security from a better information of the environment (detect dangerous materials, survey of an area,
etc.).
The considered scenario in the experiments emulates an emergency zone (see Figure 5). The UAV
should survey the zone and generate a 3D terrain map and the corresponding image mosaic to support
the emergency mission.

Emergency zone (trees)

Figure 5: The emergency zone considered in ROBAUCO project.

-5-

ISBN-xxx-y-zz-yyyyyyy-x 138
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The flight altitude is 100 m, the angular range of the camera is 60º (from -30º to 30º) and the
scanning frequency is 25 Hz. Figure 6 shows the map generated from the raw data obtained from the
ALS data. The raw data are processed by applying linear interpolation to compute another map (see
Figure 7). This map describes better the emergency zone. The vertical accuracy obtained from the
ALS data is approximately 5 cm.

754

752

750

748

746
750
745
Z (m)

740
735 744

40
742
20
0
0 -20
740
-20 -40
-60
-40
-80 738
-60 -100
-80 -120
736
-140
-100
-160
Y (m) -120 734
-180 X (m)

Figure 6: 3D terrain map computed from ALS system considering raw data.

Figure 8 presents the mosaic built from aerial imagery. Comparing Figure 7 with Figure 8 it can be
observed the emergency zone. In this case, the zone of trees was considered the emergency zone. The
building of the mosaic helps to understand and better describe the integrated 3D map obtained from
the ALS data, as well as to increase the accuracy in the position estimation by means of a different
software which is not described in this paper.

-6-

ISBN-xxx-y-zz-yyyyyyy-x 139
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 7: 3D terrain map computed from ALS system by applying the triangles for linear
interpolation.

Figure 8: Mosaic built from aerial imagery.

-7-

ISBN-xxx-y-zz-yyyyyyy-x 140
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

5. CONCLUSIONS
This paper has presented the fast automated generation of integrated 3D terrain maps from ALS data
acquired with a lightweight laser mounted on a small UAV, and of image mosaic from aerial imageries
obtained by a video camera mounted on the UAV. The ALS system consists of: a GPS, an IMU and a
SICK laser scanner. A Kalman filter is implemented to estimate the aerial platform position from GPS
and IMU data with respect to a global reference system. The main advantage of the presented system
is its rapid deployment and application in emergency missions.
Several experiments have been performed in the airfield of Alcazarén in the Spanish province of
Valladolid in the framework of the ROBAUCO project. Future efforts will be focused on performing
the fusion of ALS data and aerial imagery to obtain information in detail of the emergency zone.

6. ACKNOWLEDGMENTS
This work has been partially supported by the ROBAUCO project, the PLANET European project
(European Commission, FP7 ICT257649) and the Andalusian Government project P09-TEP-5120.
Author G. Heredia has been partially supported by the NAFTGAR (DPI2010-21762-C02-01)
Spanish National Research project, and other authors have been partially supported by the ROBAIR
(DPI2008-023847) Spanish National Research project.

REFERENCES
[1] A. Wehr and U. Lohr, “Airborne laser scanning-an introduction and overview”, ISPRS Journal of
Photogrammetry & Remote Sensing, vol. 54. pp. 68–82, 1999.
[2] E.P. Baltsavias, “Airborne laser scanning: basic relations and formulas”, ISPRS Journal of
Photogrammetry & Remote Sensing, vol. 54. pp. 199–214, 1999.
[3] J. C. Suárez, C. Ontiveros, S. Smith and S. Snape, “Use of airborne LiDAR and aerial
photography in the estimation of individual tree heights in forestry”, Computer and Geosciences, Vol.
1, Issue 2, pp. 253-262, March, 2005.
[4] D. Riaño, E. Chuvieco, S. Condes, J. Gonzalez-Matesanz and S. L. Ustin, “Generation of crown
bulk density for Pinus sylvestris L. from lidar”, Remote Sensing of Environment 92, pp. 345–352,
2004.
[5] A. Jochem, B. Höfle, M .Rutzinger and P. Pfeifer, “Automatic Roof Plane Detection and Analysis
in Airborne Lidar Point Clouds for Solar Potential Assessment”, Sensors, 9, pp. 5241-5262, 2009.
[6] Y. Lin, J. Hyyppä and A. Jaakkola, “Mini-UAV-Borne LIDAR for Fine-Scale Mapping”, IEEE
Geoscience and Remote Sensing Letters, vol. 8, pp. 426-430, 2011.
[7] P. Dorninger and N. Pfeifer, “A comprehensive automated 3D approach for building extraction,
reconstruction, and regularization from airborne laser scanning point clouds” Sensors, 8 (11), pp.
7323-7344, 2008.

-8-

ISBN-xxx-y-zz-yyyyyyy-x 141
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[8] J. Shan and C. Toth, “Topographic laser ranging and scanning: Principles and processing”, CRC
Press, London, UK, 2008.
[9] M. Salah, J. Trinder, A. Shaker, M. Hamed, and A. Elsagheer, “Aerial Images and Lidar Data
Fusion for Automatic Feature Extraction using the Self-Organizing Map (SOM) Classifier”, in
Proceedings of Laser scanning 2009, IAPRS, Vol. XXXVIII, Part 3/W8, Paris (France), pp.317-322,
September 1-2, 2009.
[10] M. Huber, W. Schickler, S. Hinz and A. Baumgartner, “Fusion of LIDAR Data and Aerial
Imagery for Automatic Reconstruction of Building Surfaces”, 2nd GRSS/ISPRS Joint Workshop on
Remote Sensing and Data Fusion over Urban Areas, Berlin (Germany), May 22-23, pp. 82-86, 2003.
[11] M. Lemmens, “Airborne LIDAR sensors product survey”, GIM International, pp. 16-19, Feb.
2009.
[12] M. Nagai, T. Chen, R. Shibasaki, H. Kumagai and A. Ahmed, “UAV-borne 3-D mapping system
by multisensor integration,” IEEE Transaction on Geoscience and Remote Sensing, vol. 47, no. 3, pp.
701–708, Mar. 2009.
[13] C. Glennie, “Rigorous 3D error analysis of kinematic scanning LIDAR systems”, Journal of
Applied Geodesy, vol. 1, pp. 147–157, 2007.
[14] F. Caballero, L. Merino, J. Ferruz and A. Ollero, “Homography Based Kalman Filter for Mosaic
Building. Applications to UAV position estimation”, IEEE International Conference on Robotics and
Automation (ICRA2007), pp. 2004-2005 Roma, Italy, 10-14 April, 2007.

-9-

ISBN-xxx-y-zz-yyyyyyy-x 142
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Scalable UAV for Flight Mechanics Education

JESÚS GONZALO DE GRADO

University of León, Engineering School
Campus de Vegazana 24071 León (Spain)
[email protected]

Abstract
A scalable airplane has been depicted to ease the practical education on flight mechanics.
This paper introduces the concept and explains how to setup a practical experiment to teach
mechanics of flight. Apart from the catalogue of modules, a software tool for the correct
configuration of the airplane and the preparation of the flight plan are detailed. Among the
huge variety of test envisaged, some of the most basic ones are described. The status of the
program and the proposal for future activities are already discussed.

Keywords: Flight mechanics, modular, scalable, UAV

1. INTRODUCTION
The University of León offers, included in the Aerospace Graduate plan, a Flight Mechanics subject.
The philosophy of new European Higher Education Area, moving from knowledge acquisition
towards skill acquisition, motivates the work of this paper: a practical approach to the education on
flight dynamics by means of a small, unmanned, scalable airplane and a companion support tool.
Although in other subjects an equilibrated load between theoretical and practical contents is easily
achievable, flight mechanics faces safety, availability and economical problems that are often
insurmountable unless one relies on simulation. Practical work is reduced to calculus and flight
simulations, sometimes with some hardware installed in the loop [1].
But the know-how is best retained when real exercises are carried-out. This way, a facility for real
flight mechanics experiment is demanded.
When there are enough funds, the work could include real flights in a small airplane that allow the
passenger/student to see and feel what stall, adverse yaw, shear wind or Dutch roll is. If not, remotely
controlled airmodels can be used to show certain flight characteristics. In general, there is not
sufficient time for the student to play a significant role in the whole process, which together with the
fact that he is not able/allowed to pilot the model, make the experience quite poor; however, the
potential of the method is confirmed [2].

-1-

ISBN-xxx-y-zz-yyyyyyy-x 143
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

What it is here proposed is the development of a facility for flight mechanics education based on a
modular model aircraft equipped with the appropriate sensors to readily provide the student with
comprehensive information of onboard parameters and relate them with his design and en-route
decisions. The model, being safe, needs to be easy to fly, maintain and repair; and of course
affordable.

2. THE AIRPLANE
The airplane is a scalable structure that can be built and configured in real-time, just in the airfield.
What is being proposed here is a catalogue of aerodynamic-shaped modules that can easily be
assembled into a full aerial vehicle, able to fly in automatic or manual modes. The selection of
modules and their configuration permits the student to build, without construction effort, a wide
variety of airplanes, depending on the requested exercise.
The catalogue of modules includes pure and aileron-mounted lifting surfaces, propulsion, vertical
surfaces, longitudinal extensors, landing supports and other smart modules (onboard processor and
avionics), all with common structural and electrical interfaces.
Figure 1 shows how easily two control configurations can be implemented for the same wingload and
propulsion subsystems, as an example of the concept envisaged.

X
X

Y
Y 17
17 18
18 19
19
20
20
21
21

24

22
22
23
23

24

Figure 1: Examples of similar configuration with different lateral control system

Basic lifting module (Figure 2)

The basic module is a wing section with two mechanical interfaces, corresponding to streamlined
profiles that provide the thickness, the chord and taper of the module and possibly, certain sweep,
torsion and dihedral angles. A covering skin encloses a space to install other pieces of equipment
inside and defines the final aerodynamic form of the module, which can be kitted with more internal
ribs for stiffness. The along-wing dimension, from initial to end interfaces, contributes to the
wingspan. Port and starboard versions of the module need to be available for symmetric
configurations.

-2-

ISBN-xxx-y-zz-yyyyyyy-x 144
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The module provides mechanical interfaces that are compatible with the rest of the modules, always
in a male-female fashion so that other modules can be added to the chain. The simplest interface, used
to attach wings to fuselages in airplane models, is based on a bayonet-hole coupling; inserting the
modules in a common carbon fibre tube or rail is also possible. There are some attachment solutions
being considered at this moment. In general, the idea is to minimize the loads transferred among the
modules, trying to maintain a constant wing load along the wingspan; this is possible because no
bulky payload is required for the pursued objective and hence, batteries, electronic hardware, motors,
etc. can be accommodated quite uniformly inside the airplane.
Although there are no electrical components inside, the module has to be able to transfer power and
data from one side to the other, so cables, male-female connectors and passages need to be provided.

X Y 5

6 2
Z a

λ
3

7
c
8
4

Figure 2: Basic Lifting module

Lift Control Element (Figure 3)

Similar to the basic module, this also provides a moveable structure, able to change the air flow
around the airfoil, hence providing control over the aerodynamic actions. The model can be used as
aileron, rudder, flap, airbrake or their combination.
Now, the internal servomechanisms require electrical power and commands, which are provided by
the power bar (cable) and the common data bus, replicated inversely in both sides.
More sophisticated versions of this module could implement leading edge flow control elements,
dihedral and torsion angle control and other mechanisms to morph the shape of the wing, always
under control of the command bus.

X Y

2
Z a

1
9

10

-3-

ISBN-xxx-y-zz-yyyyyyy-x 145
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 3: Lift control module

Wing terminators
This module provides a blunt wing tip that favours the aerodynamic characteristics and encloses the
internal structure of the wing. This item does not further allow the assembly of new modules to the
wing.
The module can include lights, powered from the electrical interface. Should the data bus request a
terminator, this module provides it.

15

X Y
16
Z

Figure 4: Wing terminator or Vertical stabilizer module or Rudder

Propulsive module (Figure 4)

The propulsive module is the thrust provider, by means of any of the known technologies applicable
to this size of airplanes: propellers (electrically or piston driven) and jets.
The simplest case is the electrical motor, normally brushless and with an integrated DC regulator/
controller. The power is taken from the main bar and the thrust command from the data bus. The
propeller may be installed in a pushing or pulling position.
In the case of the reciprocating engine, a fuel tank is necessary; this can be installed in a neighbouring
basic module or a further ad-hoc module. Also, a servomechanism controls throttle position from the
commands received by the common bus. This kind of engines requires electrical filters to be
implemented in order not to export noise to the rest of the lines.
Jet engines are more powerful and complicated, but still conceptually allowed by the system.

X Y
12
Z
11

Figure 5: Propulsive module

Other structural elements

A bunch of standard or custom structural elements can be envisaged. For example, vertical stabilizers

-4-

ISBN-xxx-y-zz-yyyyyyy-x 146
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

or winglets can be achieved with the element shown in Figure 6, where a special element exposes
three mechanical interfaces, one to attach itself and two more to continue the chain horizontally and
vertically. The vertical stabilizer is able to provide lateral aerodynamic forces with respect to the
mechanical interface. It may or may not contain elements for aerodynamic control.
A longitudinal extension (Figure 5) is a mast that replicates the mechanical and electrical interface in
a backward position (or forward) of the wing, allowing the joining of new elements to support flight
control (elevators, rudders, canard, etc.) or alternative airplane configurations.
Other examples of structural elements are fuselages, landing gears, catapult attachments, etc., always
under the same mechanical and electrical chainable scheme.

c
2

12

1
11

14
X Y

13

Figure 6: Longitudinal extension module

Avionics and electrical configuration

Avionics is treated as a non structural module, that needs to be attached to structural ones where there
is room to do so. Besides, this flexibility is often used to meet center of gravity constraints.
Thus, this module provides electrical power, onboard processing capabilities, data bus control,
navigation avionics, communication management and other nonstructural elements, including possible
payloads. The architecture is depicted in Figure 10.
The onboard processor is and open hardware platform with ad-hoc software, essentially in charge of
configuring the distribution of commands across the modules. The processor has plenty of space to
allow future experiences for students interested in flight controller programming.

-5-

ISBN-xxx-y-zz-yyyyyyy-x 147
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

T
Data Power Power supply
bus bar (solar, alternator,
fuel-cell, etc.)

Navigation Power storage

avionics (batteries)

Onboard data
handling
(processor
+memory+I/F)

Module 1

Module 2

Module N

Figure 7: Electrical architecture onboard

3. CONFIGURATION TOOL
A modeling tool estimates if the selected configuration is feasible and, if so, uploads the flight
parameters to the onboard computer so that the system is ready to carry out the test under real
atmospheric conditions. A mission planning task is necessary previous to the test, to translate the
mission requirements into aircraft commands and flight control definition. These data are transferred
to the ground segment, which is in charge of managing communications with the aircraft, sending
commands and showing its telemetry in real-time, storing the full state vector for future playback. The
concept can be completed with a weather mini-station, which is affordable and easy to install in the
surroundings of the operation area.
Prior to the experiment, the student should be able to understand the basics of airplane aerodynamic
design, the impact of the selected configuration on performance and the methods to evaluate the flight
stability and control [3][4].
Besides the theoretical classes on thin airfoil aerodynamics and foundations of flight mechanics,
available computational methods are reviewed, with special attention to the Vortex Lattice Method
(VLM), widely used in preliminary design phases for subsonic airplanes.
The grounds of the VLM is the solution of the Laplace’s equation by a combination of vortex
singularities. Although good VLM open codes [5][6] are available and well validated for the

-6-

ISBN-xxx-y-zz-yyyyyyy-x 148
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

assessment of aircraft performance, a customized user-friendly software tool has been developed to
easily apply the computational method to the educational aircraft model and to help with the task of
configuring it. This way, the user is allowed to quickly select the modules to be included in the
requested mission and their distribution throughout the vehicle.
The system imports many available databases with aerodynamic profile information, allowing custom
designs and alternative flight configurations. The tool has been tested against Tornado [6] code,
showing good matching for lift and induced drag coefficients in most of the conventional designs.
The application is also able to estimate the total centre of mass and the inertia tensor, to quickly assess
the feasibility of flight; some recommendation on mass centering are automatically issued. The
adimensional aerodynamic coefficients are calculated and plotted together with their stability
derivatives for all the desired positions of control commands.
Once the airplane is built, aerodynamic parameters and flight path can be uploaded onboard and tests
executed. Real time and playback monitoring allow the student to check if selected configuration
performs appropriately with respect to the requirements.

4. PRACTICAL EXERCISES
After the familiarization process, a pack of requirements are issued and allocated to student groups.
These include payload mass, gliding performance, cruise speed, ceiling, top speed, manoeuvrability
constraints, ascent path angle, stall, etc. The evaluation is carried out on the basis of the performance
parameters estimated by each configuration and a report with the reasoning behind them. A couple of
the best configurations can actually be built and flown. Thus, students check and post-process the
telemetry files to validate the initial assessment.
Initially, the proposed exercises are the following:
 Cruise at maximum speed: two full-throttle, hippodrome circuit, level flights are prepared,
sharing all elements and parameters (including wing load) except from the wingspan.
 Cruise at minimum consumption speed: different models are forced to fly straight at a set of
various airspeeds, whereas the electrical consumption history is stored.
 Maximum range and autonomy in gliding mode: the models are flown without propulsion.
The students can program tentative angles of attack to achieve the longest range of a
minimum descent rate.
 Minimum speed and recovery from stall: the air speed is reduced progressively whereby the
angle of attack is monitored. Several trials allow the estimation of the stall angle and its
consequences. When different aerodynamic profiles are available, the exercise can be
repeated to observe the diversity of behaviors.
 Response to step control variables: theoretical steps cannot be achieved in real
experimentation. However, quick action on the command surface or the throttle can emulate

-7-

ISBN-xxx-y-zz-yyyyyyy-x 149
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

quite well such functions. The reaction of the airplane is stored for later analysis.
 Phugoid development and analysis: the term phugoid refers to the long period of exchange
between kinetic and potential energy that presents an aircraft during its stable flight when a
perturbation is introduced [6].
 Coordinated and uncoordinated turns: uncoordinated turns (lateral acceleration present
relative to the body reference) can be programmed and monitored from the on-board inertial
unit. The rate of the mix of roll and yaw commands can be calculated and tested until
coordinated turns are achieved.
The advantage of the system is that the exercises can be run in a very rapid manner. The student is
encouraged to test alternative configurations, which can be shared around the class as simple files in
common directories.
The project is currently moving from the conceptual ideas to the first prototypes: hardware modules
have been machined in foam, on board electronics is already available and the configuration tool is
well advanced.

5. CONCLUSIONS
Real flight experience is difficult to obtain for university students. A concept based on an autonomous
(safe, and no need to care about the pilot), configurable (many options with the same hardware),
scalable (many practical missions available), affordable (small replaceable modules, technology
borrowed from amateur radio models) airplane can meet most of the needs identified in the flight
mechanics education.
A number of examples have been shown, proposing practical exercises to groups of students, that are
very attractive and with immediate technical feedback. The real flight exercises allow comparison of
estimated and real datasets with many possibilities of post-processing for value-added results.

6. REFERENCES
[1] Amireh N., Kang P., Strom D., Chui C.K., Modular Toy Aircraft, Patent US7811150B2, 2010
[2] Chartier, B.J., Gibson B. A., Project-Based Learning: A Search and Rescue UAV – Perceptions of
an Undergraduate Engineering Design Team: A Preliminary Study, Proceedings of the 2007
Australasian Association for Engineering Education (AaeE) Conference, Melbourne, Australia
[3] NASA Astronomy and Astrophysics Advisory Committee (AAAC) Annual Report, FY 2002–2003
[4] Roskam, J., Airplane Design: Parts I-VIII, DARCorporation, Lawrence, KS, 1997
[5] Athena Vortex Lattice, http://web.mit.edu/drela/Public/web/avl, active in 2011
[6] Tornado Vortex Lattice Method, http://www.redhammer.se/tornado/index.html, 2011
[7] Pradeep S., A century of phugoid approximations, Aircraft Design Vol 1, Iss 2, 89-104, 1998
Richards, J., Suleman, A., Aarons, T, Canfield, R., Multidisciplinary Design for Flight Test of a Scaled

-8-

ISBN-xxx-y-zz-yyyyyyy-x 150
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Joined Wing SensorCraft, CSME Forum 2010, Victoria, British Columbia, Canada, 2010

-9-

ISBN-xxx-y-zz-yyyyyyy-x 151
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Fuel Cells: Alternative Propulsion for Unmanned Aerial Vehicles

1∗
OSCAR GONZALEZ-ESPASANDIN, 2TERESA J. LEO and 3EMILIO NAVARRO
1
E.T.S.I. Aeronáuticos, Universidad Politécnica de Madrid, Pza. Cardenal Cisneros, 3, 28040 Madrid,
Spain, E-mail: [email protected]
2
E.T.S.I Navales, Universidad Politécnica de Madrid, Avda. Arco de la Victoria,s/n, 28040 Madrid,
Spain, E-mail: [email protected]
3
E.T.S.I. Aeronáuticos, Universidad Politécnica de Madrid, Pza. Cardenal Cisneros, 3, 28040 Madrid,
Spain, E-mail: [email protected]

Abstract
In this work, the possibility of implementing the fuel cell technology in the Unmanned
Aerial Vehicle (UAV) propulsion systems is considered. The potential advantages of the
Proton Exchange Membrane (PEM) fuel cells and Direct Methanol Fuel Cells (DMFC), their
fuels (hydrogen and methanol) and their storage systems are revised from technical and
environmental points of view. Some already operative commercial applications are
described. The main constraints of this kind of fuel cells are analyzed in order to elucidate
the viability of future developments. Since the low power density is the main problem of fuel
cells it would be necessary hybridization with electric batteries in most cases.

Keywords: Fuel Cell, Unmanned Aerial Vehicles, Propulsion, Hydrogen, Methanol

1. INTRODUCTION
The Unmanned Aerial Vehicles (UAVs) have seen a dramatic boost in recent years, encouraged by
recent technological developments, so that they are already considered as integral and indispensable
part of modern armed forces with an increasing number of civil applications and dual
use. Main countries have already acquired UAVs or plan to do it soon. At present their propulsion
systems are based on different types of internal combustion engines, but the global energy situation,
which was preceded by the energy crises of the 70's last century, and strategic reasons make alternative
propulsion systems begin to be introduced. Fuel cells have advantages in terms of endurance, stealth
properties, emissions and efficiency that make them interesting for use in UAVs. Starting from the
premise that there are important energy and environmental problems in our society, the advantages and
disadvantages of fuel cells as an option for propulsion systems for UAVs are going to be analyzed.

2. WORLD ENERGY OUTLOOK

The world energy situation, with its economic, strategic and environmental issues, highlights the need
to develop alternative systems to fossil fuels for propulsion and power generation. Today's society has a
great dependence on such fuels, especially oil and its derivatives, because of its easy production, low
cost until recent years, logistic advantages, high energy density, and versatility, which allows
obtaining many products such as plastics and fertilizers.

∗
Corresponding author.

-1-

ISBN-xxx-y-zz-yyyyyyy-x 152
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The progressive oil depletion was modelled in 1956 by the American geophysicist Marion King
Hubbert, whose prediction for the United States peak production was ratified in 1975 by the National
Academy of Sciences, confirming the validity of Hubbert's calculations. Hubbert's predictions for the
planet expect a world peak production for early twenty-first century. This means that, unlike the energy
crisis of the 70s of last century, motivated by political constraints, the current economic and financial
crisis, with high energy prices (Fig. 1) and raw materials, seems to obey the geological conditions
exposed by Hubbert, being at least aggravated by other causes, including geopolitical conflicts [1]. It is
worth mentioning that the steady increase in world population and energy thirst of the emerging powers
can not but worsen the situation. China contributes 36% of the projected growth in global energy use,
and India 18% [2].

Figure 1: Historical world oil prices 1861-2011.

Moreover, fossil fuels pose a serious environmental problem. They are the main causes of the
increasing CO2 presence in the atmosphere, hence one of the most important culprits of global warming
and the presence of atmospheric emissions of other pollutants [3].
Therefore there is an urgent search for new energy policies based on the diversification of energy
sources and their origin, energy saving policies and the use of efficient energy conversion systems.
Of course, the world of aviation is not an exception to the above considerations. The UAVs are at
nascent stage of development (in fact their regulations are in process of generation), but the
implementation of fuel cell propulsion systems are more advanced than in conventional aircrafts. The
reason is the fact of not being manned, so its weight is comparatively smaller and they do not need the
life support systems for the crew and passengers. In the military case, there are other operational
advantages such as a less noisy operation and lower thermal signature.

3. UAV PROPULSION SYSTEMS

The propulsion system of an UAV consists of the following elements:

ƒ Energy source: chemical fuels (fossil fuels, biofuels, chemicals), electricity, solar energy (in
conjunction with photovoltaic cells), hydrogen, methanol, energy mechanics.

ƒ Storage media: tanks, batteries, capacitors, flywheels, metal hydrides, etc.

-2-

ISBN-xxx-y-zz-yyyyyyy-x 153
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ƒ Converter to mechanical energy: internal combustion engine, fuel cell + electric motor
(or electric motor only, Fig. 2).

ƒ Converter to lift/thrust: rotor, fan, propeller, jet engine, etc.

In addition, any propulsion system includes a power control or engine speed system, heat management
system and auxiliary electrical power generation.

Figure 2: Diagram of a propulsion system with a converter to mechanical energy based on a

combination fuel cell + electric motor.

Despite the recent boom in greener propulsion systems (electric, solar, hybrids, hydrogen internal
combustion engine, etc.) the vast majority of current UAV engines are still driven by conventional
internal combustion piston, normally fed with fossil fuels. As alternative to conventional propulsion,
three types of propulsion systems can globally be considered in UAVs:

ƒ Alternative thermal systems: where different thermodynamic cycles, fuel or engine types can
be used (e.g. spark ignition reciprocating engines that are fuelled by gasoline).

ƒ Electrical systems: where the power required is obtained by an electric motor and electric
power is generated or stored in different ways.

ƒ Hybrid systems: combining any of the systems listed above, even the same type (e.g.
combination of fuel cell and battery, or Regenerative Fuel Cell Systems, RFC, which combine
fuel cell, battery and photovoltaic cells).

Figure 3: Diagram examples of (A) serial and (B) parallel hybrid propulsion systems.

Hybridization can be achieved by combining heat engine/electric motor with batteries or fuel
cell/electric motor with batteries. In either case the need to install motors, batteries, inverters, control
units and so on make the system heavier, bulkier and more expensive than others. It still has
advantages, especially from the standpoint of a definitive implementation of alternative systems and the
compensation for the low power density of the fuel cells. There are two possibilities: series and parallel
hybrid systems (Fig. 3).

-3-

ISBN-xxx-y-zz-yyyyyyy-x 154
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The conversion to lift/thrust systems are closely linked to the type of aircraft, either fixed wing, rotary,
lighter than air, etc. The UAV propulsion systems can be classified by the systems of conversion to
mechanical energy. Since the internal combustion is the usual method, it is separated into reciprocating
engines according to the fuel used (petrol or diesel), and jet engines.

ƒ Reciprocating engines: There is a great variety attending to the combustion process (spark-ignition
engines, compression ignition engines, etc.), cycle (two-stroke and four-stroke engines), intake
manifold pressure (engine naturally aspirated and supercharged engines), air or water-cooled, etc. In
this case the main classification is determined by the type of working cycle and its related fuel, e.g.,
aviation gasoline (Piston-AvGas) or diesel (Piston-Diesel). Among its weaknesses compared to other
types are the vibrations that do not deliver the torque smoothly. They are suitable for small and
medium size UAVs and for short range, since the jet engines work well at full power, but with partial
loads there is a significant increase in their consumption.
ƒ Jet engines: They produce thrust and can be classified into jet engines (Jet-turbine, typical of
Unmanned Aerial Combat Vehicles; UCAVs) and turbofans (typical in high subsonic, more used in
commercial aviation than in UAVs, although there are examples like Global Hawk HALE
UAV). Instead of thrust, that could deliver power to a shaft that drives a propeller or rotor. In this
case they are turboprop engines (turbo-propeller), where a turbine engine is connected to a
traditional propeller (e.g. Predator B UAV MALE), and turboshaft engines, typical of rotary wing
aircraft. They deliver torque more smoothly and with less vibration than reciprocating engines, but
are not suitable at low speeds due to its high consumption.
ƒ Electric motors: They convert electrical into mechanical energy by moving a propeller, a fan or a
rotor. Electrical energy is supplied by a battery, a photovoltaic cell or a fuel cell. They have the
advantage of being the quietest and one of those with lower thermal signature. Currently only micro
and mini UAVs are powered by batteries and electric motors [4]. They are undergoing continuous
improvement, but electricity demand comes not only from the engine, but the payload and
communication systems, which limits the endurance or speed. Fuel cells and photovoltaic cells have
already been tested in UAVs, but they are far from being mature technologies. In 2009 there were
111 electric-powered UAV while in 2011 there are around 160.
ƒ Other types of engines: Like the Wankel rotary engine, of simple and smooth operation, which is
gaining acceptance as their problems of sealing and torsional vibration are being solved. Currently
Israeli Hermes 180 and Hermes 450 UAV use this type of engines, 28 and 38 kW respectively, with
high durability and a specific consumption of 0.35 kg/kWh. The need of a geared motor and the high
cooling requirements produce an increase in weight compared to the reciprocating engines.

The breakdown of 2010 UAVs by size and type of propulsion system is shown in Figure 4. As can be
seen, there are many factors to be considered when choosing a power plant for an UAV, and there is no
reference book or guide. It is necessary to make an exhaustive assessment of the requirements in each
particular UAV.

-4-

ISBN-xxx-y-zz-yyyyyyy-x 155
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 4: UAVs histogram according to sizes and powerplants.

In any case, the engine is only one component of the propulsion system. It must be mounted on the
UAV and it must be provided with means of starting, fuel supply with the required cooling, control, and
gas exhaust if required. All this facts will influence the final choice. When comparing power and
energy densities not only the weight of the engine must be taken into account, but also the weight of the
energy storage system and the engine auxiliary systems. In large UAVs, the engine will therefore have a
great size and will probably come with its associated subsystems. However, in a small UAV there will
probably be a need of a subsequent selection of those subsystems. In order to compare different types
of propulsion systems, Table 1 is shown.

Table 1: Specific power and power density of different systems [5].

System Specific Power Power Density
[W/kg] [W/L]
Combustion Engine 90-1100 40-1200
PEMFC 2-105 10-120
DMFC 2-60 2-70

With regard to electric propulsion systems, Lithium Ion batteries have a specific power of 450-1800
W/kg, but the weight of the electric motor and other auxiliary systems is not included, therefore more
information is required to compare these systems. The weight of storage tanks or auxiliary systems
such as control systems is generally unknown. Most of authors or manufacturers do not usually give
information about these systems.
UAVs are not manned and they carry on board typically light payloads (e.g. surveillance and
communications), so the weight sum of the propulsion system and the fuel usually exceed one third of
the total UAV weight, a proportion higher than in conventional aviation [4]. Therefore, any reduction in
the propulsion system weight or in specific consumption can have a significant effect on the endurance
increase or the downsizing of the UAV. The current research focuses on the improvement of internal
combustion engines and in the use of new energy sources, by means of fuel cells, for example.
The breakthroughs about the internal combustion engines are on the side of downsizing and HCCI
technology (Homogeneous Charge Compression Ignition). The latter is based on producing auto-
ignition of a lean and homogeneous mixture (air/fuel ratio> 25) at multiple points in the combustion

-5-

ISBN-xxx-y-zz-yyyyyyy-x 156
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

chamber. Another research field is related to a new piston engine concept, the stepped piston, which
uses the movement of a piston to provide compression to a second piston.
Recent advances in the achievement of lighter electric motors and batteries have allowed making the
electric propulsion more competitive. The main drawback is the low specific energy density of
batteries, resulting in large volumes (around four times the equivalent volume of fossil fuels). Thus, the
problem relies on the endurance.

4. FUEL CELLS AS ALTERNATIVE FOR UAV PROPULSION

Among the most interesting alternative propulsion systems for UAVs are those based on fuel cells. It is
a technology still immature but growing, with scope for improvement in reducing weight, volume and
costs. They are electrochemical systems that convert the chemical energy contained in fuels directly
into electric energy. When fed with hydrogen, they produce no greenhouse gases, being water the only
waste, and the level of noise generation is low. Unlike batteries, a fuel cell does not need to be
recharged; it operates while fuel and oxidizer are supplied from outside. The fuel cell consists of an
anode where the fuel is injected (usually hydrogen, ammonia or methanol) and a cathode where an
oxidant is introduced (usually air or oxygen), separated by an electrolyte ionic conductor [6]. In those
fuel cells having an acid electrolyte the protons travel across of it from the anode to the cathode and in
those having an alkaline electrolyte, the negative ions cross from the cathode to the anode. In both
cases, the electrons are forced to travel through an external circuit (producing energy in the form of
electrical current) because the membrane is electrically isolated. Usually fuel cells produce low
voltages and they must be assembled into a fuel cell stack to reach the power required in usual
applications. Then, fuel cells are by their nature, modular devices and their power can go from
microwatts to megawatts, which makes them useful in a variety of applications.
The main advantages of fuel cells are their low emissions, high efficiency, modularity, reversibility (this
property is exploited in RFC systems), fuel flexibility, range of applications, low noise and infrared
signature. However, its disadvantages include cost, sensitivity of the electrode catalyst to poisons, their
experimental state and the lack of availability of H2.
There are different types of fuel cells, which differ in the temperature range of operation and the
electrolyte used. According to the method of combining these types of fuel cells with other elements,
new concepts might emerge. RFC Systems, where the stack uses its reversibility mode as electrolyzer
or as fuel cell (e.g. Helios UAV/NASA solar power combined with day/night cycles); or hybrid
systems, relying on batteries when more maneuvering power is required [7]. The different
characteristics of each type of fuel cell make them more appropriate for a particular application. High
temperature fuel cells are not the most appropriate for UAVs due to their large size and weight, large
start time, and the auxiliary systems required to manage the dissipated heat. There have been some
attempts at UAV miniaturization, which might have an initial application in RFC combined with solar
power, for long endurance flights (up to 5 years project Boeing/DARPA with Solid Oxide Fuel Cell,
SOFC [8]).
Airships are also a potential new application for fuel cells. An example is a Lockheed Martin Project,
High Altitude Airship that will operate above the jet stream (an advantage over combustion engines) in

-6-

ISBN-xxx-y-zz-yyyyyyy-x 157
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

a geostationary position to deliver persistent station keeping as a surveillance platform. It would be an

alternative to satellites. It will be a 137 m length airship with a RFC solar/ H2/O2 propulsion system and
will be able to remain 10 years flying [9].
PEM fuel cells operate at low temperature (80-90 ºC), the use of polymeric electrolyte provides high
current densities, have quick start and high specific energy density and low specific power density.
Therefore they offer potential advantages in low maneuverability and high endurance UAVs, which is
the type of UAV most developed up today (e.g. surveillance applications). The reason is that these
properties allow obtaining relatively light, low volume and cost systems compared with other types of
fuel cells [10]. As well, PEM fuel cells are those that have reached a more mature market and there are
several commercial applications. In addition to the efficiency and environmental benefits discussed
above, the low noise and thermal signature are invaluable advantages in the military field. The fuel
used is hydrogen, stored by different means (compression, liquefaction, metal or chemical
hydrides). Hydrogen disadvantages include the costs of generation (is an energy carrier, not an energy
source), the compression/liquefaction for storage and distribution and the safety issues like the risk of
explosion. However, it has a high heating value and the reaction product is only water.
DMFC stacks (a special category of PEM fuel cells), fuelled directly with methanol, have about half the
efficiency (higher heat losses) and power density than PEM fuel cells, but higher energy density
(greater endurance) and simplicity. In addition, the storage system is lighter than in the case of
hydrogen and the logistics are simpler, being profitable the fossil fuels distribution net. Besides, the
handling and storage of methanol is much more advantageous than that of hydrogen, which requires
heavy deposits that must withstand high pressures. Methanol is toxic, but this is not a problem if
following the appropriate protocols. Furthermore, it could be obtained from biomass (from wood
distillation) and therefore it could be considered a renewable fuel. It emits only water and CO2, but this
is reabsorbed by the biomass produced to generate the new methanol, thus closing the cycle. Its cost is
lower than that of hydrogen. There are already DMFC stacks for terrestrial applications and they could
be applied in very light, low maneuverability UAVs. The market trend for UAVs is precisely towards
miniaturization. Given its low power density, probably DMFC will need to be combined with hybrid
battery systems (except perhaps in systems with low power demand as airships), in case of specific
requirements of high power demand. Batteries have high conversion efficiencies and higher power
densities than any fuel cell. However, their energy density is clearly lower, so that their endurance is
small. The Lithium-Ion batteries are currently the most advanced.
For small UAVs, compressed hydrogen systems are not practical, since they are not scalable down. In
this case it is preferable non-cryogenic liquid fuels such as methanol or chemical hydrides, where the
fuel storage and delivery is even more critical than the battery performance itself.
There are already developments in which a fuel cell delivers power to an electric motor, which in turn
drives a propeller or a rotor [11]. Recent Protonex developments in compressed hydrogen-powered
PEM fuel cells achieve power densities up to 1 kW/kg, although power up to 1 kW are commercially
available for models not exceeding 250 W/kg. These figures worsen around 20% by adding the weight
of the electric motor, which provokes obtained values worse than in other systems. Moreover, the mass
of hydrogen, added to its storage system worsens the figures compared to those of fossil fuels; the data

-7-

ISBN-xxx-y-zz-yyyyyyy-x 158
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

of methanol are not so unfavorable regarding the storage system. However, the energy conversion
process is very efficient compared to internal combustion engines and also great endurance is achieved.
This is due to its high energy density, which makes them very suitable for long endurance surveillance
missions. As an example, the Puma UAV equipped with a Protonex PEM stack flew 9 hours improving
the 2 hours of the same battery-electric UAV [10].
Thus, the advantages of a fuel cell-powered electric-motor are:
ƒ More efficient than fossil fuel technologies.
ƒ High energy density, which means great endurance.
ƒ Reliability: few moving parts and easy automation.
ƒ Flexibility of operation: they are reversible, working at high performance without interruption on
a wide range of powers, which can also change rapidly
ƒ Modular and easy to implement.
ƒ Negligible noise and low vibration.
ƒ Low or zero emissions.
ƒ Variety of applications: in addition to propulsion systems in UAVs, they might be used in APU's,
auxiliary power systems, ground control stations, etc.
The main disadvantages are:
ƒ High cost, because it is not a mature technology and in the use of expensive materials like
platinum, used as a catalyst.
ƒ Sensitivity to fuel contamination, necessitating expensive filtering systems.
ƒ The need for qualified personnel for maintenance.
ƒ Low power density compared to other systems, especially in DMFC stacks.
ƒ Unavailability of hydrogen: H2, one of the fuels used in fuel cells, is not free in nature. It must be
obtained by water electrolysis or hydrocarbons reforming, which refers to it as an energy carrier
rather than an energy source. There is also no distribution infrastructure.
ƒ Unproven reliability for commercial use: there is little ‘real’ commercial fuel cell UAVs, and
where they exist, their implementation is very recent, and therefore technology is not mature
enough. Some still unresolved issues are affecting the performance of fuel cells, especially in
regard to life cycle, which results in their marketing. There are few manufacturers.

Due to the reasons given above, the advantages and disadvantages of an UAV propulsion system based
on fuel cells are practically the same as those of the fuel cell stack itself, slightly worsened by the need
to add the weight of the electric motor.
The main problem of fuel cells is their low power density, so that the UAV performance would have
restrictions and would be very dependent on its aerodynamic design, weight and the fuel cell
performance [12]. This makes interesting the complementary use of batteries to form hybrid systems,
especially in the case of direct methanol fuel cells, having these lower power densities. There are not
examples found of UAV powerplant based on DMFC stack (not even hybrid); only a few experimental
developments at the early stages (Quantum Sphere Inc. [13], [14]). There are examples of hydrogen-
based PEM UAVs, as already seen, and UAV hybrid projects based on compressed hydrogen PEMs [7],

-8-

ISBN-xxx-y-zz-yyyyyyy-x 159
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

but all are developments. In Fig. 5 power of commercial fuel cells is plotted versus its weight and the
two UAV application cases are located in the left lower corner of the graph. In both cases the fuel is
hydrogen, stored through chemical hydrides. The graph shows that with the current technology a fuel
cell can be only implemented in small UAVs. Direct methanol fuel cells found in the market have very
low power.

UAV

Figure 5: Power vs. weight of PEM and DMFC commercial fuel cells.

Attention should also be paid to the potential loss of performance of fuel cells depending on the
environmental conditions of aircraft flight (changes in pressure/altitude, temperature, vibration,
humidity, salinity, radiation, shock, etc.). Most fuel cells are designed to work on land, statically, with
environmental conditions relatively stable. However, in flight, these conditions change more or less
sharply depending on the UAV ceiling, its speed, airfield location, and so on. The influence on the
propulsion system performance is clear, since parameters such as humidity will affect the operation of
the fuel cell membrane. There is little literature on the influence of environmental conditions on the
performance of a UAV, although a manufacturer such as Horizon Energy Systems has published data on
its technical specification sheets, (UAV Aeropak fuel cell, 200 W, cartridges hydrides [15]).

4. CONCLUSIONS
Could be asserted the feasibility to implement a fuel cell in a UAV? The answer is yes although,
obviously, not in all cases of fuel cells or UAVs. Conclusions:
ƒ Fuel cells technology is still immature but growing, with clear room for improvement in reducing
weight, volume and costs. Compared with conventional systems, fuel cells offer higher energy
density and lower specific power density.
ƒ Fuel cells offer potential advantages in low maneuverability UAVs, which are currently the most
manufactured type (e.g. surveillance applications), and strong endurance requirements.
ƒ PEM fuel cells are those that having reached a more mature market. The fact of being low
temperature and fast start up are features consistent with its implementation in UAVs. It is not
interesting the option of reforming other fuels (natural gas, gasoline, etc.) because it would add the
weight and volume of the reformer.
ƒ It is important to take into account the fuel storage system, given the existing number of cases for
each type of fuel cell. In the case of a UAV, it is essential the overall weight fuel cell/storage

-9-

ISBN-xxx-y-zz-yyyyyyy-x 160
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

system. The weight of auxiliaries, such as thermal control systems or water management must be
added.
ƒ There are already several commercial UAV applications of PEM hydrogen fuel cells. In all cases the
target is an UAV of small size. There are fuel cell manufacturers offering fuel cells designed for a
general type of UAV and UAV manufacturers that have developed an UAV equipped with a
powerplant based on a fuel cell specifically designed for that UAV.
ƒ Methanol, compared to hydrogen, has advantages in terms of its storage systems, especially in
logistics, but also in the energy density and cost, and even in safety issues, despite its toxicity, which
is not an important problem following the proper protocols. DMFC fuel cells have less power
density in comparison to other types of fuel cells, but higher energy density.
ƒ Since the low power density is the main problem of fuel cells it would be necessary hybridization
with electric batteries in most cases.

REFERENCES
[1] K. S. Deffeyes, Hubbert's Peak: The Impending World Oil Shortage, Princeton Univ. Press (2009)
[2] International Energy Agency (EIA) Wold Energy Outlook, Executive Summary (2010)
[3] Intergovernmental Panel on Climate Change (IPCC), http://www.ipcc.ch (2011)
[4] R. Austin, Unmanned Aircraft Systems: UAVS Design, Development and Deployment, Wiley (2010)
[5] Hu, Zhiyu “Jerry”. Nano Patents and Innovations (Blog). Oak Ridge Scientist Uses Nanomaterials
to Make Thermoelectric Converter That Is Much More Efficient Than Fuel Cells, Batteries or Gas
Engines, 13-04-2010. http://nanopatentsandinnovations.blogspot.com/2010/04/oak-ridge-scientist-uses-
nanomaterials.html.
[6] J. Larminie and A. Dicks, Fuel Cell Systems Explained (Second edition), Wiley (2003)
[7] M. K. Fürruter and J. Meyer, Small Fuel Cell Powering an Unmanned Aerial Vehicle. IEEE
AFRICON (2009)
[8] Fuel Cell Today, 03-12-2010, http://www.fuelcelltoday.com/news-events/news-
archive/2010/december/versa-power-wins-boeingdarpa-contract-to-supply-sofc-for-uav
[9] Lockheed Martin, 05-06-2011,
http://www.lockheedmartin.com/products/HighAltitudeAirship/index.html#
[10] A. Léon, Hydrogen Technology. Mobile and Portable Applications. Springer-Verlag (2008)
[11] D. E. Parekh, Fuel Cell Revolution. Aerospace Testing International, March 2010, pp.26-30.
[12] T. H. Bradley, B. A. Moffitt, D. N. Mavris and D. E. Parekh, Development and experimental
characterization of a fuel cell powered aircraft. Journal of Power Sources 171, pp.793-801 (2007)
[13] QuantumSphere Inc., http://www.qsinano.com/apps_fuelcell.php (2011)
[14] K. McGrath, J. Chrostowski, Direct Methanol Fuel Cells for UAV Propulsion, TechConnect World
Conference & Expo 2010, Anaheim, CA. (2010)
[15] Horizon Energy Systems, 17-10-2011
http://www.hes.sg/files/AEROPAK_Technical_Data_Sheet.pdf

-10-

ISBN-xxx-y-zz-yyyyyyy-x 161
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Tutorial for the application of Kane’s Method to model a

small-size helicopter

Luis A. Sandino1 , Manuel Bejar2 and Anibal Ollero1,3

1 University of Seville, 2 University Pablo de Olavide, 3 Center for Advanced Aerospace Technologies

Seville, SPAIN

[email protected], [email protected], [email protected]

Abstract

This paper presents a tutorial for applying Kane’s method to derive a mechanical model of a small-size
helicopter. Throughout this development main concepts underlying Kane’s methodology are explained in detail.
Furthermore, references to other approaches in the field of classical mechanics are included to analyze the
advantages provided by Kane’s method when compared to other alternatives. Since this work considers the most
general case of two rigid bodies, fuselage and main rotor, the resulting model will account for most significant
modelling issues in the mechanics behaviour of a small-size helicopter, such as gyroscopic effects. Finally it is
emphasized that the adopted pedagogical approach should contribute to illustrate the method to beginners.

Keywords: UAS, Helicopter, Mechanical modelling, Kane’s method

1 INTRODUCTION
The various approaches to obtain equations of motion for mechanical systems can be classified as either vector
approaches based on the direct use of D’Alembert’s or Jourdain’s principles such as Newton-Euler equations or
Kane’s equations, or scalar approaches based on Hamilton’s or Gauss’ variational principles such as Lagrange’s
equations or the Gibbs-Appell equations [3].
Although the equations obtained using the various approaches are equivalent in the sense that they generate the
same numerical results in simulation, Kane’s method holds some unique advantages. Both Lagrange’s equations
and Kane’s equations use generalized coordinates and thus embed configuration constraints. Therefore, multi-
body mechanisms yet possess few degrees of freedom, are better served by either one of these methods than by
standard application of Newton-Euler equations. Kane’s method, because it uses generalized coordinates yet is
based on a vector approach, is sometimes referred to as Lagrange’s form of D’Alembert’s Principle. But Kane’s
method also uses generalized speeds, which allows motion constraints to be embedded. So Kane’s method has
been compared to the Jourdain’s Principle [10] and Kane’s equations have been likened to the Gibbs-Appell
equations [2, 9] and Maggi’s equations [4]. The use of generalized speeds in Kane’s method enables the selection

ISBN-xxx-y-zz-yyyyyyy-x 162
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

of an independent set of motion variables for non-holonomic constrained systems which allows to obtain a model
in first order differential equations form. In Lagrangian approach, where only generalized velocities are available,
the method of Undetermined Multipliers should be used to handle motion constraints.
As mentioned above, the use of generalized speeds can lead to more compact equations of motion. Indeed,
a convenient choice of generalized speeds can result in equations that are diagonal in the generalized speed
derivatives; that is, uncoupled dynamical equations. A further advantage of Kane’s method over Lagrange’s
method derives from its membership in the vector approach classification. The projection of constraint forces
which are known to be non-working forces, allows such forces to be disregarded from the outset of the analysis,
which leads to savings in the operations required during formulation.
The paper is organized as follows. Section 2 introduces variables used in Kane’s equations for deriving a
kinematic model, and compares them to those used in Lagrange’s equations. After this, an application example
given by the kinematic model of a small-size helicopter is presented according to the pedagogical nature of this
work. In section 3, dynamics corresponding to Kane’s equations for a generic mechanical system are presented,
highlighting the relationship with the projection of Newton-Euler’s equations into a space tangent to the configu-
ration space. Again, these general concepts are better illustrated by presenting their application to a small-size
helicopter. Finally, section 4 is devoted to the conclusions.

2 DERIVATION OF KINEMATICS AND REFERENCES TO

LAGRANGE’S APPROACH
Let S be a mechanical system with n degrees of freedom moving in an inertial reference frame N subjected to m
motion constraints. For the sake of generality, the system consists of Pl (l = 1, · · · , λ ) particles (points possessing
mass) and Bk (k = 1, · · · , ν) rigid bodies which centres of mass are denoted by BO
k . Points of external force

application on rigid bodies Bk are denoted by Pf ( f = 1, · · · , φ ). Thus, the union of the former reference points is
Pj ( j = 1, · · · , µ) = Pl ∪ BO
k ∪ Pf where µ = λ + ν + φ .

The first step in order to build a kinematic model for S is the definition of a set of configuration variables that
describe location of reference points Pj∗ = Pl ∪ BO
k and orientation of reference frames fixed within each body Bk .

In Kane’s method as well as in Lagrange’s method, generalized coordinates qi (i = 1, · · · , n) are used as configu-
ration variables. Generalized coordinates specify point locations and reference frame orientations as relative to
each other rather than all relative to inertial reference frame N. In this way, they describe only the allowed confi-
gurations for S and therefore encapsulate some configuration constraints. For example, if a revolute joint connects
two bodies, only the joint angle is needed to describe the orientation of the second body if orientation of the first
is known. In general, instead of the set of 3λ + 6ν variables that specifies the location and orientation of all com-
ponents of S when defining them relative to a common origin in N, a set of generalized coordinates only consists
of n = 3λ + 6ν − c independent variables, where c is the number of configuration constraints encapsulated.
In order to characterize motion of S, an additional set of variables called motion variables is also required.
In Lagrange’s equations these are given by the generalized coordinate derivatives, also called generalized ve-
locities q̇i (i = 1, · · · , n). However, motion variables in Kane’s method are defined as functions generally linear

ISBN-xxx-y-zz-yyyyyyy-x 163
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

in the generalized coordinate derivatives and non-linear in the generalized coordinates. The use of such func-
tions, called generalized speeds ur (r = 1, · · · , n), provides significantly more compact and thus computationally
efficient equations:
h iT h iT
u1 ··· un = Yn×n · q̇1 ··· q̇n + Zn×1 (1)

where matrices Y and Z are functions of generalized coordinates qi (i = 1, · · · , n) and possibly of time t. The
reciprocal expressions of the generalized coordinate derivatives in terms of the generalized speeds are called
kinematic differential equations:
h iT h iT
q̇1 ··· q̇n = Wn×n · u1 ··· un + Xn×1 (2)

where W = Y−1 and X = Y−1 Z.

In Kane’s method, in the same way that configuration constraints are encapsulated within the definition of
a set of generalized coordinates, any motion constraint of the system that may arise for physical reasons (e.g.,
rolling constraints) can be fully embedded by a wise choice of generalized speeds. This will lead to manageable
first order dynamical differential equations when describing the entire motion of the system. Additionally, it
is often convenient the consideration of differentiated configuration constraints also as motion constraints1 . In
contrast to Kane’s method, if motion constraints are present when applying Lagrangian approach, they must be
carried along as part of the system description and handled using Lagrange’s Method of Undetermined Multipliers
[2].
Concerning the steps required to embed motion constraints, firstly they should be expressed as linear relation-
ships among the generalized speeds [5]:
h iT
A· u1 ··· un +B = 0 (3)

where matrices A and B are functions of generalized coordinates qi (i = 1, · · · , n) and possibly of time t. Given m
motion constraints, there will be m dependent generalized speeds in the set ur (r = 1, · · · , n) that can be expressed
in terms of the remaining p = n − m independent generalized speeds. After selecting these p independent motion
variables, equation (3) can be transformed into the following expression between dependent and independent
generalized speeds:
h iT h iT
u p+1 ··· un = C· u1 ··· up +D (4)

Thus, the embedding process is achieved substituting (4) into every kinematic expression containing depen-
dent generalized speeds before formulating dynamical equations. As pointed out before, the definition of the
generalized speeds, that is up to the analyst, can have a considerable impact on the compactness of motion equa-
tions. However these effects will not become evident until dynamical equations are formulated in a subsequent
section of this paper.
This section concludes with Table 1 summarizing the differences between Kane’s and Lagrange’s methods
when deriving a kinematic model for the general system previously analyzed.
1 The terms configuration constraint and motion constraint have been used in favour of the roughly equivalent terms holonomic constraint
and non-holonomic constraint so that a differentiated holonomic constraint, which is still properly referred to as a holonomic constraint, can be
grouped with non-holonomic constraints. In simple words, configuration constraints are expressed as equations involving only configuration
variables and motion constraints are expressed as equations involving both motion and configuration variables.

ISBN-xxx-y-zz-yyyyyyy-x 164
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Lagrange’s method Kane’s method

generalized coordinates generalized coordinates

Configuration variables
qi (i = 1, · · · , n) qi (i = 1, · · · , n)

generalized velocities generalized speeds

Motion variables
q̇i (i = 1, · · · , n) ur (r = 1, · · · , p)

Constraints encapsulated c c+m

Constraint equations m -

Table 1: Summary of differences between Kane’s and Lagrange’s method when defining a kinematic model

Application example for tutorial purposes: small-size helicopter

As advanced in the beginning, this paper intends to improve the understanding of Kane’s methodology by pre-
senting an application example: a small-size helicopter. The system is depicted in Fig. 1 and consists of 3
components: fuselage, main rotor and tail rotor. As pointed out in [7, 8], for most commercially available small-
size helicopters the inertial effects of the main rotor (gyroscopic effects) become the main component influencing
the rotational dynamics of the whole mechanical system whereas the tail rotor inertial influence is negligible.
This means that main rotor and fuselage will be considered as rigid solids (ν = 2) and tail rotor will only act as an
application point of force on the fuselage (φ = 1). Since there are no particles in the system, λ can be considered
nil.

Figure 1: Helicopter description: reference frames, centres of mass and dimensions of interest for modelling purposes

Helicopter motion will be described in an inertial reference frame N where a dextral set of orthogonal unit
vectors ni (i = 1, 2, 3) is fixed. Fuselage is denoted by F, whereas its mass and centre of mass are given by mF and
F O respectively. A dextral set of orthogonal unit vectors fi (i = 1, 2, 3) is fixed in F and then the central inertia
O
dyadic of fuselage can be expressed as IF/F = IF11 f1 f1 + IF22 f2 f2 + IF33 f3 f3 . Main rotor is denoted by MR and
is modelled as a thin solid disk with constant angular speed ωMR . The same nomenclature criterion for fuselage
stands for this case, that is mass and centre of mass are denoted respectively by mMR and MRO . A dextral set
of orthogonal unit vectors mri (i = 1, 2, 3) is also fixed in MR, which allows the definition of the central inertia

ISBN-xxx-y-zz-yyyyyyy-x 165
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

O
dyadic of main rotor as IMR/MR = IMR11 mr1 mr1 + IMR11 mr2 mr2 + 2 · IMR11 mr3 mr3 with mr3 = f3 being the
rotation axis. Scalar constants IF11 , IF22 , IF33 and IMR11 are the principal moments of inertia of the corresponding
bodies.
An additional reference point O fixed in F is used to describe the location of F O , MRO and tail rotor T RO by
means of the corresponding dimensions defined in Fig. 1:
O O O
pO→F = dO−F O ,3 f3 pO→MR = dO−MRO ,3 f3 pO→T R = dO−T RO ,1 f1 (5)

Assumption that F O is located on the rotation axis of the main rotor will be achieved in the real system by an
appropriate placement of the equipment on the fuselage. Under all these definitions and assumptions, the centre
of mass H O of the whole system relative to O is given by:
O O
O mF pO→F + mMR pO→MR mF · dO−F O ,3 + mMR · dO−MRO ,3
pO→H = = f3 = dO−H O ,3 f3 (6)
mF + mMR mF + mMR

Once system geometry and mass distribution have been specified, next step is the definition of the generalized
coordinates that have been introduced previously. Since rotation axis and centre of mass position for MR are
O
totally fixed in F for simplicity (mr3 = f3 and pO→MR = dO−MRO ,3 f3 ), orientation of MR relative to F can be
described by only one generalized coordinate. Therefore the two rigid solid compound that models the small-size
helicopter possesses c = 5 configuration constraints and consequently the number of degrees of freedom is given
by n = 6ν − c = 7.
The position of centre of mass H O in the inertial reference frame N is described by generalized coordinates
qi (i = 1, 2, 3):
O →H O
pN = q1 n1 + q2 n2 + q3 n3 (7)

Generalized coordinates qi (i = 4, 5, 6) are the Euler-angles (roll, pitch and yaw) corresponding to successive
rotations (body123 order, see [6]) that describe the orientation of F in the inertial reference frame N. Furthermore,
generalized coordinate q7 is the angle corresponding to the rotation of MR relative to F. Thus, unit vectors ni , fi
and mri are geometrically related by the direction cosine matrices shown in Table 2.

f1 f2 f3 mr1 mr2 mr3

n1 c5 c6 −c5 s6 s5 f1 c7 −s7 0

n2 c4 s6 + s4 s5 c6 c4 c6 − s4 s5 s6 −s4 c5 f2 s7 c7 0

n3 s4 s6 − c4 s5 c6 s4 c6 + c4 s5 s6 c4 c5 f3 0 0 1

where si = sin(qi ) and ci = cos(qi ).

Table 2: Direction cosine matrices

Once generalized coordinates are defined, kinematic equations can be formulated as follows:
N dpO→H O
N HO
v , = q̇1 n1 + q̇2 n2 + q̇3 n3 (8)
dt
N df N df N df
N 2 3 1
ω F , f1 · f3 + f2 · f1 + f3 · f2
dt dt dt
= (s6 q̇5 + c5 c6 q̇4 )f1 + (c6 q̇5 − s6 c5 q̇4 )f2 + (q̇6 + s5 q̇4 )f3 (9)
F
ω MR = q̇7 f3 (10)

Generalized speeds ui (i = 1, · · · , 7) are defined in such a way that (8), (9) and (10) can be written in a more

ISBN-xxx-y-zz-yyyyyyy-x 166
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

compact way:
N vH O = u1 n1 + u2 n2 + u3 n3 (11)
ui , q̇i (i = 1, 2, 3) ⇒


u4 , s6 q̇5 + c5 c6 q̇4 



NωF = u4 f1 + u5 f2 + u6 f3 (12)
u5 , c6 q̇5 − s6 c5 q̇4 ⇒




u6 , q̇6 + s5 q̇4 

F ω MR = u7 f3 (13)
u7 , q̇7 ⇒

The former leads to the following kinematic differential equations:

q̇i = ui (i = 1, 2, 3)

q̇4 = − (s6 u5 − c6 u4 ) /c5

q̇5 = s6 u4 + c6 u5

q̇6 = u6 + s5 (s6 u5 − c6 u4 ) /c5

q̇7 = u7 (14)

As was explained before, configuration constraints are embedded in the definition of generalized coordinates
itself. In contrast, in order to embed motion constrains, their mathematical formulation and subsequent replace-
ment into equations of motion is required. In this case, the only motion constraint is given by the fact that the
main rotor is assumed to be rotating with constant angular speed ωMR :
u7 = ωMR (15)

Therefore, there is m = 1 dependent generalized speeds and p = n − m = 6 independent generalized speeds.

Hence, last equation from (14) can be removed from the analysis and the remaining six equations constitute the
kinematical differential equations and form the first half of the equations of motion for the system under study.

3 DERIVATION OF DYNAMICS AND REFERENCES TO

NEWTON-EULER’S APPROACH
After deriving a kinematic model that describes the configuration and motion of the target system, next step in
Kane’s method is the definition of some auxiliary kinetic expressions: partial velocities, partial angular velocities,
generalized active forces, and generalized inertia forces. Subsequently these variables will be related to each other
by applying mechanics principles to obtain the final dynamic model.

3.1 Partial velocities and partial angular velocities

Position of reference points Pj ( j = 1, · · · , µ) with respect to N are given by vectors pPj = p(q1 , · · · , qn ,t). An
expression for the velocity vector N vPj in terms of q̇i (i = 1, · · · , n) is obtained by applying the chain rule to these
vector functions:
N dpPj h N ∂ pP j N ∂ pP j
i h iT N ∂ pPj
N Pj
v = = ··· · q̇1 ··· q̇n + ( j = 1, · · · , µ) (16)
dt ∂ q1 ∂ qn ∂t
where the left superscript indicates that differentiation is performed in frame N. Substituting (2) into (16) yields:
h N ∂ pP j N ∂ pP j
i h iT h N ∂ pPj N ∂ pPj
i N ∂ pPj
N Pj
v = ··· ·W· u1 ··· un + ··· ·X+ (17)
∂ q1 ∂ qn ∂ q1 ∂ qn ∂t
Finally, (4) must be used to remove dependent generalized speeds from (17). Once this last replacement is
P P N Pj
done, the r-th partial velocity vr j of Pj is then defined as the coefficient for ur (r = 1, · · · , p), i.e., vr j = ∂ ∂ uvr ,

ISBN-xxx-y-zz-yyyyyyy-x 167
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

P
in the resultant expression. The portion of N vPj not containing any ur is denoted by vt j and called the velocity
remainder of Pj in N. Rewriting (17) according to this nomenclature yields:
h i h iT P
N Pj P P
v = v1 j ··· v pj · u1 ··· up + vt j (18)

By a similar reasoning, the angular velocity of rigid bodies Bk (k = 1, · · · ν) can always be expressed in terms
of the independent generalized speeds [5]:
h i h iT
N B B B
ω Bk = ω1k ··· ω pk · u1 ··· up + ωt k (19)

B
where the r-th partial angular velocity ω r k of Bk is again defined as the coefficient for ur (r = 1, · · · , p), i.e.,
B N B B
ω r k = ∂ ∂ωur k . The portion of N ω Bk not containing any ur is denoted ω t k and called the angular velocity remainder
of Bk in N.
Partial velocities will be defined only for those points Pj subjected to applied forces and/or possessing mass,
whereas partial angular velocities will be defined only for those rigid bodies Bk subjected to applied torques
and/or possessing inertia.

Application example for tutorial purposes: small-size helicopter (continuation 1)

Considering (11), (12), (13) and using the expressions:

N Fi N HO O →F
v = v + N ω F × pH i

N
ω MR = N
ω F + F ω MR (20)

where Fi is any point fixed in F, Table 3 with partial velocities and partial angular velocities for points Pj ( j =
F O , MRO , T RO ) and rigid bodies Bk (k = F, MR) of the system under study can be constructed by inspection.

r 1 2 3 4 5 6

vrFO n1 n2 n3 (dO−H O ,3 − dO−F O ,3 )f2 (dO−F O ,3 − dO−H O ,3 )f1 0

O
vMR
r n1 n2 n3 (dO−H O ,3 − dO−MRO ,3 )f2 (dO−MRO ,3 − dO−H O ,3 )f1 0
O
vTr R n1 n2 n3 dO−H O ,3 f2 −dO−H O ,3 f1 − dO−T RO ,1 f3 dO−T RO ,1 f2

ω Fr 0 0 0 f1 f2 f3

ω MR
r 0 0 0 f1 f2 f3

Table 3: partial velocities and partial angular velocities

3.2 Generalized active forces and generalized inertia forces

According to [5], generalized active forces are defined as dot product of partial velocities and active (i.e. applied)
forces and dot product of partial angular velocities and active torques. Then, for each reference point Pj subjected
to applied forces and for each rigid body Bk subjected to applied torques, generalized active forces are given by:
P B
(Fr )Pj , vr j · RPj (Fr )Bk , ω r k · TBk (r = 1, · · · , p) (21)
P B
where vr j
is the r-th partial velocity of Pj , RPj is the resultant of all active forces acting on is the r-th Pj , ω r k
partial angular velocity of Bk and TBk is the resultant of all active torques acting on Bk . The r-th generalized
active force Fr can then be determined by summing the results over all reference points Pj and all rigid bodies Bk :
µ ν
Fr = ∑ (Fr )Pj + ∑ (Fr )Bk (r = 1, · · · , p) (22)
j=1 k=1

ISBN-xxx-y-zz-yyyyyyy-x 168
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Inertia force R∗Pj and inertia torque T∗Bk are defined by [5] as:
R∗Pj , −mPj N aPj
O O
T∗Bk , −IBk /Bk · N α Bk − N ω Bk × IBk /Bk · N ω Bk (23)
N d N vPj Bk /BO
where mPj is the mass of Pj , N aPj , dt is the acceleration of Pj in N, I k is the inertia dyadic of Bk about
N N B
its centre of mass BO N Bk is the angular velocity of B in N and N α Bk , d ω k
k (i.e. the central inertia dyadic), ω k dt
is the angular acceleration of Bk in N. Note that RPj is equal to the time derivative in N of linear momentum
∗
O
LPj = mPj N vPj of Pj and T∗Bk is equal the time derivative in N of angular momentum HBk = IBk /Bk · N ω Bk of Bk ,
both with negative sign. Therefore, generalized inertia forces are defined as dot product of partial velocities and
inertia forces and dot product of partial angular velocities and inertia torques. Then, for each reference point Pj
possessing mass and for each rigid body Bk possessing inertia, generalized inertia forces are given by:
P B
(Fr∗ )Pj , vr j · R∗Pj (Fr∗ )Bk , ω r k · T∗Bk (r = 1, · · · , p) (24)
P B
where vr j is the r-th partial velocity of Pj , R∗Pj is the inertia force for Pj , ω r k is the r-th partial angular velocity of
Bk and T∗Bk is the inertia torque of Bk . The r-th generalized inertia force Fr∗ can then be determined by summing
the results over all points Pj and all rigid bodies Bk :
µ ν
Fr∗ = ∑ (Fr∗ )Pj + ∑ (Fr∗ )Bk (r = 1, · · · , p) (25)
j=1 k=1

Application example for tutorial purposes: small-size helicopter (continuation 2)

Concerning forces and torques applied to the system (See Fig. 2), main rotor generates a force FMR = fMR,3 f3
applied at point MRO and torques MMR,i = tMR,i fi (i = 1, 2, 3) applied to MR, whereas tail rotor generates a force
FT R = fT R,2 f2 applied at point T RO and a torque MF = tT R,2 f2 applied to F. Force of gravity W j = −m j gn3 ( j =
F, MR) applied at centres of mass F O and MRO is also considered, where g is the acceleration of gravity.

Figure 2: Forces and torques applied to the helicopter

Now, using equations (21) and (22) the generalized active forces for the first and fourth partial velocities can
be obtained as follows:
O O O
F1 = vF1 · RF O + vMR
1 · RMRO + vT1 R · RT RO + ω F1 · TF + ω MR
1 · TMR
3
= n1 · WF + n1 · (FMR + WMR ) + n1 · FT R + 0 · MF + 0 · ∑ MMR,i
i=1

= fMR,3 s5 − fT R,2 c5 s6
O O O
F4 = vF4 · RF O + vMR
4 · RMRO + vT4 R · RT RO + ω F4 · TF + ω MR
4 · TMR

= (dO−H O ,3 − dO−F O ,3 )f2 · WF + (dO−H O ,3 − dO−MRO ,3 )f2 · (FMR + WMR )+

3
+ dO−H O ,3 f2 · FT R + f1 · MF + f1 · ∑ MMR,i = tMR,1 + dO−H O ,3 fT R,2 (26)
i=1

ISBN-xxx-y-zz-yyyyyyy-x 169
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Following the same procedure, the remaining generalized active forces are given by:
F2 = fT R,2 (c4 c6 − s4 s5 s6 ) − fMR,3 s4 c5

F3 = fMR,3 c4 c5 + fT R,2 (s4 c6 + c4 s5 s6 ) − (mF + mMR )g

F5 = tMR,2 + tT R,2

F6 = tMR,3 + dO−T RO ,1 fT R,2 (27)

Regarding generalized inertia forces for the first and fourth partial velocities, using equations (23) to (25)
yields:
O O
F1∗ = vF1 · R∗F O + vMR
1 · R∗MRO + ω F1 · T∗F + ω MR ∗
1 · TMR
O O O O
= n1 · (−mF N aF ) + n1 · (−mMR N aMR ) + 0 · (−IF/F · N α F − N ω F × IF/F · N ω F )+
MR/MRO MR/MRO
+ 0 · (−I · N α MR − N ω MR × I · N ω MR ) = −(mF + mMR )u̇1
O O
F4∗ = vF4 · R∗F O + vMR
4 · R∗MRO + ω F4 · T∗F + ω MR ∗
4 · TMR
O O
= (dO−H O ,3 − dO−F O ,3 )f2 · (−mF N aF ) + (dO−H O ,3 − dO−MRO ,3 )f2 · (−mMR N aMR )+
O O O O
+ f1 · (−IF/F · N α F − N ω F × IF/F · N ω F ) + f1 · (−IMR/MR · N α MR − N ω MR × IMR/MR · N ω MR )

= K456 u5 u6 − K4p4 u̇4 + K45 u5 (28)

Following the same procedure, the remaining generalized inertia forces are given by:
F2∗ = −(mF + mMR )u̇2

F3∗ = −(mF + mMR )u̇3

F5∗ = K546 u4 u6 − K5p5 u̇5 + K54 u4

F6∗ = K645 u4 u5 − K6p6 u̇6 (29)

where parameters Kxxx have the following values:

mF · mMR · (dO−F O ,3 − dO−MRO ,3 )2
K456 = IF22 − IF33 − IMR11 +
mF + mMR
mF · mMR · (dO−F O ,3 − dO−MRO ,3 )2
K4p4 = IF11 + IMR11 +
mF + mMR
K45 = −2IMR11 ωMR
mF · mMR · (dO−F O ,3 − dO−MRO ,3 )2
K546 = IF33 − IF11 + IMR11 −
mF + mMR
mF · mMR · (dO−F O ,3 − dO−MRO ,3 )2
K5p5 = IF22 + IMR11 +
mF + mMR
K54 = 2IMR11 ωMR

K645 = IF11 − IF22

K6p6 = IF33 + 2 · IMR11 (30)

Note that all operations in (26) and (28) are performed after expressing all vectors in the reference frame N
using the corresponding relationships of Table 2.

3.3 Equations of motion

Classical Newton-Euler’s equations for each reference point Pj and each rigid body Bk of system S can be ex-
pressed as:
RPj − L̇Pj = 0; TBk − ḢBk = 0 (31)

where RPj is the resultant of forces acting on Pj , LPj is the linear momentum of Pj , TBk is the resultant of all
torques acting on Bk and HBk is the angular momentum of Bk . Projecting the first and the second equations in

ISBN-xxx-y-zz-yyyyyyy-x 170
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

P B
(31) into the vector space formed by vr j and ω r k respectively (which is, in fact, equivalent to apply Jourdain’s
principle) and summing both results over all bodies of system S yields:
µ ν µ ν
P P
∑ vr j · RPj + ∑ ω r k · TBk + ∑ vr j · (−L̇Pj ) + ∑ ω r k · (−ḢBk ) = 0
B B
(r = 1, · · · p) (32)
j=1 k=1 j=1 k=1

Thus, comparing (32) with equations (21) to (25) leads to:

Fr + Fr∗ = 0 (r = 1, · · · , p) (33)

which are known as Kane’s equations. This set of first order differential equations in the generalized speeds is
called dynamical differential equations and constitutes the second half of the equations of motion of system S.
These equations together with the kinematic differential equations in (2) will govern the behaviour of system S.

Application example for tutorial purposes: small-size helicopter (continuation 3)

Finally, equation (33) is applied to obtain the dynamical differential equations, resulting in:

(mF + mMR )u̇1 = fMR,3 s5 − fT R,2 c5 s6

(mF + mMR )u̇2 = fT R,2 (c4 c6 − s4 s5 s6 ) − fMR,3 s4 c5

(mF + mMR )u̇3 = fMR,3 c4 c5 + fT R,2 (s4 c6 + c4 s5 s6 ) − (mF + mMR )g

K4p4 u̇4 = tMR,1 + dO−H O ,3 fT R,2 + (K456 u6 + K45 ) u5

K5p5 u̇5 = tMR,2 + tT R,2 + (K546 u6 + K54 ) u4

K6p6 u̇6 = tMR,3 + dO−T RO ,1 fT R,2 + K645 u4 u5 (34)

where parameters Kxxx are those listed in (30). Equations (14) together with (34) form the equations of motion
for the helicopter. It is remarkable that these equations are uncoupled in the generalized speed derivatives thank
to the choice of generalized speeds made in (11), (12) and (13). This fact makes the model suitable for numerical
integration. Note also that constraint forces and torques (i.e. interaction forces and torques between main rotor
and fuselage) were not present during the formulation. The reason is that the vector space used to project Newton-
Euler’s equations to derive Kane’s equations is orthogonal to those forces and torques, which allows to disregard
them from the outset of the analysis.

3.4 Comparison of computational efficiency between Kane’s and Newton-Euler’s

methods

The procedure presented in this paper to obtain the equations of motion of a small-size helicopter can be easily
performed with the help of MotionGenesis software [1], a symbolic manipulator for the analysis of mecha-
nical systems that implements Kane’s method. In order to compare computational efficiency between Kane’s
and Newton-Euler’s methods, equations (14)-(34) and the model corresponding to direct application of Newton-
Euler’s equations (31) were generated using Autolev2 scripts. Each script was executed several times in a laptop
with an Intel Core i5 M480 processor at 2.67 GHz. The average CPU time was 1.58 seconds for the script corres-
ponding to Kane’s method and 2.35 seconds for the one corresponding to Newton-Euler’s method. The size of
MATLAB code generated by Autolev was 16 kBytes in the first case and 24 kBytes in the second one. This code
2 version 4 of MotionGenesis software.

10

ISBN-xxx-y-zz-yyyyyyy-x 171
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

was used to implement the helicopter model together with the controller proposed by [8] in MATLAB-Simulink
with a sample time of 0.01 seconds. The resulting system was executed in a PC/104 Single Board Computer with
MATLAB-Simulink’s xPC real-time operating system. The simulation consisted of the application of a sequence
of reference step signals to the variables q1 , q2 , q3 and q6 during 60 seconds. The average Task Execution Time
(TET)3 was 2.24 × 10−4 seconds for the model obtained with Kane’s method and 4.85 × 10−4 seconds for the
model obtained with Newton-Euler’s method. These results are summarized in Table 4a. Regarding accuracy, Ta-
ble 4b shows the maximum absolute difference between variables qr (r = 1, · · · 6) of both models during real-time
simulation. As can be seen the difference between the two models is very small and the computational efficiency
of the Kane’s method is better.

r max |qKane
r − qN−E
r |
Newton-Euler’s method Kane’s method
1 3.3398 × 10−4
Average CPU time
2.35 s 1.58 s 2 3.3527 × 10−4
(Autolev scripts)
3 4.2196 × 10−5
Generated MATLAB code
24 kB 16 kB
4 3.9635 × 10−5
size

5 4.0048 × 10−5
Average TET (Model
4.85 × 10−4 s 2.24 × 10−4 s
real-time execution) 6 2.4435 × 10−6

(a) Computational efficiency (b) Accuracy

Table 4: Comparison between Kane’s and Newton-Euler’s methods

4 CONCLUSIONS
This paper presents a pedagogical development of Kane’s method through its application to model a small-size
helicopter. To this end, the formulation of the equations of motion emphasizes the interpretation of the concepts
underlying in Kane’s methodology in order to ease its understanding for beginners. Concerning the model itself,
this work considers the general case of two rigid bodies, fuselage and main rotor. Hence, the resulting modelling
structure accounts for most significant modelling issues in the mechanics behaviour of small-size helicopters.
The method has proved to hold some unique advantages when compared to other traditional approaches.
On the one hand, the use of generalized coordinates allows to embed configuration constraints. On the other
hand, the adoption of generalized speeds enables the derivation of a compact model in first order differential
equations form. Indeed, an adequate choice for generalized speeds can result in equations that are uncoupled in
the generalized speed derivatives. Another remarkable advantage is that constraint forces are disregarded from
the outset of the analysis.
Finally it should be also noted that equations of motion generated by Kane’s method are characterized by their
easy implementation and computational efficiency. This efficiency is two-fold: on the one hand, equations are
3 This value is the measured CPU time to run the model equations and post outputs during each sample interval.

11

ISBN-xxx-y-zz-yyyyyyy-x 172
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

obtained with the fewest symbolic operations, on the other hand, obtained equations require the fewest numerical
calculations for their solution.

ACKNOWLEDGEMENTS

This work have been partially supported by Junta de Andalucia excellence project P09-TEP-5120 and European
Commission projects PLANET (FP7-ICT-2009-5) and CONET (FP7-ICT-2007-7).

References
[1] Motiongenesis Kane 5.x. http://www.motiongenesis.com/, 2011.

[2] H. Baruh. Analytical Dynamics. McGraw Hill, 1998.

[3] H. Baruh. Another look at the describing equations of dynamics. Journal of the Chinese Society of Mecha-
nical Engineers, Transactions of the Chinese Institute of Engineers, Series C/Chung-Kuo Chi Hsueh Kung
Ch’eng Hsuebo Pao, 21(1):15–23, 2000.

[4] M. Borri, C. Bottasso, and P. Mantegazza. Equivalence of kane’s and maggi’s equations. Meccanica,
25(4):272–274, 1990.

[5] T. R. Kane and D. A. Levinson. Dynamics, Theory and Applications. McGraw Hill, 1985.

[6] T. R. Kane, P. W. Likins, and D. A. Levinson. Spacecraft Dynamics. McGraw Hill, 1983.

[7] K. Kondak, M. Bernard, N. Losse, and G. Hommel. Elaborated modeling and control for autonomous small
size helicopters. VDI Berichte, (1956):207, 2006.

[8] K. Kondak, M. Bernard, N. Meyer, and G. Hommel. Autonomously flying vtol-robots: Modeling and
control. In IEEE International Conference on Robotics and Automation, pages 736–741, 2007.

[9] J.G. Papastavridis. A panoramic overview of the principles and equations of motion of advanced engineering
dynamics. Applied Mechanics Reviews, 51(4):239–265, 1998.

[10] J. Piedboeuf. Kane’s equations or jourdain’s principle? In IEEE Midwest Symposium on Circuits and
Systems, volume 2, pages 1471–1474, 1993.

12

ISBN-xxx-y-zz-yyyyyyy-x 173
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

UAV PISTON ENGINE TEST BENCH

JUAN CARLOS CELADA FERNÁNDEZ and PABLO FERNÁNDEZ RODRÍGUEZ

Área de Ingeniería Aeroespacial. Escuela de Ingenierías Industrial, Informática y

Aeronáutica. Campus de Vegazana. Universidad de León. 24071 León (Spain)

[email protected] [email protected]

Abstract

This article describes an educational test facility for mid-power UAV piston engines
that could be used in different kinds of UAS. Some recent designs try to join the
benefits of helicopters and planes in one single UAV. These kinds of hybrid
aircrafts require rapid adjustments in engine torque when actuation changes
between helicopter and airplane modes. This variation can be easily set in the bench
using an Eddy currents dynamometer by changing the main coil excitation. The
external mounting case of the dynamometer acts on a load cell and measures the
applied torque. Rpm, pressure and temperature sensors provide info about the
engine state through an analog-digital converter card. While using this, the student
can manage the software implemented on Labview and obtain graphical view of
parameter variations. This development makes easy the comprehension of pressure-
volume diagram which is one of the main difficulties in the conceptual study of a
propulsion plant. The design of the test bench is modular in order to allow future
improvements like the addiction of sensors or actuators, with hardware in the loop
philosophy, and the use of different engines by an adjustable cradle.

Keywords: test bed, piston engine, dynamometer.

1. INTRODUCTION
Various kinds of engines can be selected to be used on UAV’s. Vehicle mass and speed of operation
are the most important parameters on this selection. Up to middle weight and low speed piston
engines can be used with the advantage of low consumption.

Fixed and rotary wing combined aircrafts, used to obtain VSTOL and speed advantages, as can be
necessary for remote sensing on situations with big areas of coverage from little points of operation.
This situation is usual in offshore fishing boat.

The engine power needed by these kinds of aircrafts depends on the flight mode and has to change
quickly between stationary flight and forward flight. To study the engine response in these cases, a
test bench can be quite useful. The test bed, equipped with a dynamometer acting as a brake could be
managed to quickly set the required torque. During this state some parameters, like engine speed and
pressure or temperature of fluids or parts of the engine could be read and saved to be analyzed latter.

The “Instituto Nacional de Tecnología Aeroespacial”, INTA, is involved on design an UAV with
characteristics as related before called “helicóptero adaptativo avión” HADA. The engine selected
for this project is one of the most usual piston engines for amateur aircrafts in the world, the
ROTAX 912ULS, that can be tested and optimized using the test bench.

As an educational resource, the test bench could be used to do laboratory practice on subjects as
aircraft engines or automation in university engineering studies. Similar cases using micro-jets had
been done by Propulsión and Fluid Mechanics Department, School of Aeronautics, Universidad
Politécnica de Madrid [1].

ISBN-xxx-y-zz-yyyyyyy-x 174
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2. SELECTED ENGINE
This engine [2] is a 4-stroke S.I. engine specially developed for recreational aircraft with 4
horizontally opposed cylinders in boxer configuration. This engine is similar to the ones used on
UAV Predator RQ-1K.

Table 1: Main characteristics of Rotax 912 ULS.

Bore x Stroke 84.0 mm x 61.0 mm

Displacement 1,352 cc

Power output Max: 69 kW @ 5500 rpm

Torque max 128 Nm @ 5100 rpm

rpm Max. 5800 (5 min)

Cylinder head 4 separate cylinder heads

Compression ratio 10.5:1

Cooling Liquid-cooled cylinder heads and RAM air cooled cylinders

Cyl. Head temp Max. 135 deg. C

Coolant temp Max. 120 deg. C

Lubrication: Dry sump lub. with trochoid pump, oil return by BLOW-BY gas

Oil quantity 3 liters SAE 20W50 or SAE 30 high performance automotive oil

Oil pressure NORMAL 4 bar @ 5500 rpm, MAX 7 bar. MIN 1.5 bar @ 2800 rpm

Oil temperature NORMAL 90-110 deg. C MIN 50 deg. C MAX 130 deg. C

Ignition Double ignition system, CDI, magneto with electronic advance

Fuel Mechanical diaphragm pump. Double constant depression (CD) carb.

Fuel pressure Max. 0.4 bar Min. 0.15 bar

Fuel consumption 4.5 gal./hr. @ 75% cruise

Starter Integrated electric starter

Gear Integrated reduction gear 2.43:1 with torsional shock absorption

3. TEST BENCH
The test bench has a modular design in order to make expansions easier, like adding new
instrumentation or auxiliary equipment [3]. It has the following parts:

ISBN-xxx-y-zz-yyyyyyy-x 175
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

 Metallic bedplate, sub-bedplate for the brake and 3D adjustable engine attachment:

 Adjustable engine attachment.

 Transfer case attachment.

 Eddy current brake.

 Brake-engine junction with protection.

 Auxiliary starting system.

 Electronic box for sensors.

Fig 1; General view of main bed showing dynamometer and engine.

 Electronic parameter acquisition/control data systems by means of a control display:

 Magnetic and thermal protectors for system components.

 Throttle, starting and stop control unit.

 Signal converters:

 Thermometry: 6 channels, 4 Pt-100 (0 - 220 ºC) and 2 K-thermocouple (0 -

1000 ºC).

 Manometric: 2 fluid pressure channels (0 – 25 bar) and 1 combustion chamber

pressure channel (max 35 MPa).

 Torque: 1 extensiometric load cell 250kg

 rpm: magnetic sensor (0-10.000 rpm)

 Brake regulation unit by means of a PID for its modes

ISBN-xxx-y-zz-yyyyyyy-x 176
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

 PC with TFT monitor and printer

Fig 2; General view control console.

The most important elements (dynamometric brake and sensors) and those which are necessary for
PV diagram are described below.

3.1. Dynamometric brake

Operation principles are based on Eddy currents or Foucault currents that take place in electric
machine operations. Not as the electric engines where they want to reduce to the maximum these
Eddy currents to elevate its yield, a brake of Eddy currents takes advantage of the parasitic currents
that take place in the stator, to offer resistance to the axis turn.

The resistant torque is proportional to the intensity that flows by the coil and the braking power is
the product of that torque times the speed of turn.

The control of the brake resistant torque is made electronically by means of a P.I.D. controller
(Proportional, Integral, Derivative) that applied to the power stage feeds the coils of the brake
allowing to stabilize the set brake-engine in conditions of predetermined torque and regime,
manually or of automatic commanded mode from the computer according to a work cycle
programmed by the user.

The electric dynamometers are easily adjustable and very stable, but highly dependent on engine
torque and rpm. That is, the operation is stable when the resistant torque curve cuts the engine power
curve at some specific regime. Obviously, the resistant torque available at any instant must be
superior to the engine torque. Maximum allowable values are:

 Max. Power: 135 kW

 Max. Torque: 400 Nm

 Max. RPM: 10.000 rpm

Multiple operating modes are available, trying to cover all engine operating ranges:

ISBN-xxx-y-zz-yyyyyyy-x 177
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Table 2: Dynamometer models used.

Constant torque. Feedback torque is used in this Constant RPM. Excitation is varied in order to
mode. maintain rpm constant.

Proportional torque. Excitation follows the M ≈ n2. Excitation increases resistant torque in
following expression. function of this expression.

M M n  n 
2
a  M
 K  a   
M MAX n nMAX M MAX  nMAX 

ΔM
Δn

Constant RPM mode is very useful because it permits the simulation of variable pitch propellers.
UK’s UAV, the BAE Systems HERTI uses such propellers that improve performance.

3.2. Sensors
Apart from load measurement cell and rpm transducer, there are some sensors in order to
measure pressure and temperature at several locations. These sensors are analogical, so they
produce a voltage that must be converted into a digital signal. This mission is accomplished by
the Data Acquisition Card which has 16/2 analogical input/output channels and 16/16 digital
input/output channels. Once the analog signal is converted into a digital one, it can be employed
by the control software. Analogical outputs are used to regulate the throttle and the brake load.
Digital outputs are mainly intended for condition detectors and safety mechanisms.

ISBN-xxx-y-zz-yyyyyyy-x 178
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2 x Temp

2 x Temp

2 x Press
Test bed Control
1 x rpm
IN
Test engine Power shaft Dynamometer 1 x Load cell

OUT
1 x Brake Coil

1 x Throttle

Fig 3; Main parameters flow chart.

3.3. PV diagram
In order to plot the indicated pressure-volume diagram, a pressure sensor has been installed inside
the combustion chamber and an encoder in the crankshaft, which allows knowing its current
position. The pressure sensor is a piezoelectric device which is able to endure the high pressures and
temperatures reached in the combustion chamber. Also, the measurement rate must be high enough
to grant great accuracy. The sensor is mounted by means of a largely modified spark plug. The
engine dimensions (diameter and stroke) must be provided to calculate chamber volume as a
function of crankshaft position.

4. SOFTWARE DESCRIPTION
This software is aimed at measuring several physical parameters UAV engine performance (regime,
torque, power, temperature, pressure,…), for what it does a series of tests resulting in values which
allow to determine whether the engine meets some stated requirements or not. A Generic and
upgradeable software is used, just requiring to add new engine definitions to the database, choosing
the pertinent engine description prior to the test.

The Main menu has the following options:

 Engine configurator.

 Power test.

 PVPA test.

 Test results.

 Terminate.

Furthermore, the software’s database is used to configure engine’s make, model, and version and a
series of parameters such as alarms, piston diameter, stroke and number of cylinders (necessary for
PVPA test), end regime for power curve, and time length of power curve.

4.1. Test configuration

During test configuration, visualization parameters are set for the engine, as well as which
validations are taken into account for considering the test valid, or not valid.

ISBN-xxx-y-zz-yyyyyyy-x 179
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Validation limits are set for “Automatic validation” operation, in which the software checks whether
the measured values fall into range or not, validating the test in accordance with. In case a value goes
out of automatic validation range, the test goes on, not as with alarms, which terminate the assay.

4.2. Testing
At the beginning of each test, some reference data is recorded as to identify each assay. Additionally,
brake control is established in open or closed lace.

A maximum of sixteen parameters are visualized during each test. Data acquisition for these
parameters can be “continuous” o “point by point”. This can be changed during operation. Aside the
graphics and bars, the main screen displays basic engine controls such as throttle and ignition.

At the end of each assay a .log file is saved in the system and a screen pops-up showing validation
information for the test.

4.3. Study cases

As the software offers the possibility of visualizing multiple plots with different physical parameters,
even from different engines, users can compare performance under many variable conditions.
Furthermore, the program offers the possibility of exporting numeric values to a plane text file which
can be later treated with any software such as MATLAB or excel.

Following advice from Ferguson, [4] a couple of study cases are proposed:

4.4. Pressure-volume measurement and combustion analysis

Measurements of cylinder pressure can be used to determine not only the location of peak pressure,
but also instantaneous heat release, burn fraction, and gas temperature [5], reviews various models
used to analyze the pressure data.

The software plots representative pressure versus crank angle data and pressure versus cylinder
volume. These plots can be used to determine the polytropic exponent n from the slope of the curve
of nonreacting portions of the compression and expansion strokes. Instantaneous heat release and
burn fraction curve can also be calculated from these plots.

Fig 4; P-alfa and PV diagrams.

ISBN-xxx-y-zz-yyyyyyy-x 180
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

4.5. Effect of ambient pressure and temperature. Power curve

Engines are designated to operate over large ranges of ambient temperature, pressure, and humidity.
With a test bench, alumni can study bmep, power or torque performance variations under such
variable conditions.

Fig 5; Graphical screen showing data obtained.

5. CONCLUSION
Piston engines are very popular among UAS when costs are considered. The test bed is designed to
fit any model under 135 kW, enough for any current UAV operating nowadays. Further
developments of the test bed consider the addition of a gondola in order to simulate internal UAS air
flow, allowing to a better understanding of refrigerating properties using hardware in the loop
philosophy. In addition, the ability to adjust torque almost instantly makes this test bed suitable for
the simulation of particular situations such as stationary flight. A better understanding of
performance, consumption and reliability is the key to the development of projects such as INTA’s
HADA hybrid UAV concept.

In an educational basis, this test bed is suitable to strengthen concepts by simulating different
operating situations. The possibility of comparing different engines previously recorded in the
database can lead to a better understanding of the concept of similarity in engines. What is more,
teacher can take advantage of the exporting features of the software for the design of homework
tasks.

REFERENCES
[1]G. L. Juste, J. L. Montañes and A. Velazquez. Micro-Jet Test Facility for Aerospace Propulsión
Engineering Education. Madrid, ETSI – UPM, 2009.

[2]Operators Manual For Rotax Engyne Type 912 Series, BPR Rotax GmbH&Co, 2007.

[3]Banco de Ensayo de Motores Térmicos con Freno Eléctrico U-135, Sidilab, 2010.

[4]C. R. Ferguson, A. T. Kirkpatrick. Internal Combustion Engines. Second Edition, John Wiley &
Sons, Inc. 2001.

[5]D. Foster. An Overview of Zero-dimensional Thermodynamic Models for IC Engine Data

Analysis. SAE paper 852070

ISBN-xxx-y-zz-yyyyyyy-x 181
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Alternatives for programming a low-cost UAV in robotics

undergraduate course

Mario Castro de Lera(1) Vicente Matellán(2)

(1) Vega Space GmbH. Darmstadt (Germany)

(2) Grupo de Robótica - Escuela de Ingenierı́as Industrial e Informática
Campus de Vegazana. Universidad de León. 24071 Leon (Spain)

e-mail: [email protected], [email protected]

Abstract

The aim of this paper is to present different alternatives to program a commercially available
flying platform, and to discuss what we consider more convenient to be used by our under-graduate
students in a regular robotics course to teach them basic vision based control. Due to the background
of our students, we will focus more on the description of the software application interface, than
on the hardware, although we will describe also both. We will describe three different alternatives
to program the ARDrone UAV manufactured by Parrot. This environments are based on the SDK
provided by the manufacturer, and two ones based on free software robotics SDKs.

Keywords: Teaching, SDK, ARDrone, ROS, URBI

1 INTRODUCTION
Attracting students to computer and electronics related studies is becoming a problem in developed
countries. Robotics has been identified as a good tool for increasing the enrolling in technical studies
[1, 2]. Our group has been using different platforms [3, 4] in the last years, but these classic wheeled
platforms are becoming old-fashioned. Fortunately, the miniaturization of inertial sensors chipsets,
accelerometer and gyroscopes, have created a wide variety of very light IMUs (Inertial Measurements
Units) which are letting manufacturers create very cheap UAVs, that we can use to teach the basic of
robotic control.
The quick development of these low-cost flying machines is allowing the introduction of these plat-
forms in universities as teaching tools. Some researching groups have built their own platform them-
1
selves, ENAC (Toulouse, France) and ETH (Zurich, Switzerland)2 . We are not focused on building
our own hardware, so we decided to use commercially available ones.
1 http://www.enac.fr
2 http://www.eth.ch

1
ISBN-xxx-y-zz-yyyyyyy-x 182
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

In particular, we have chosen the ARDrone from Parrot3 . It is a commercial quadroptor helicopter
which is sold for around 300 euros. The manufacturer allows its use as a programmable platform by
providing an open Software Development Kit (SDK). Using this SDK an its specifications, third-party
manufacturers have also create alternative development environments, as for instance Gostai (offering
Urbi support), or Willow Garage (providing ROS support). In the next sections we will analyze this
alternatives.
The rest of the paper is organized as follows. In the next section we summarize the hardware of
the system, and the basic software provided by the manufacturer. In the following two sections we are
describing third-parties alternatives for programming the drone, so in third section we analyze Urbi and
in fourth section the ROS-based approach. Fifth section will be the comparison, and in the last one we
discuss the alternatives a justified our decision.

2 ARDrone Hardware and software

The platform ARDrone has been developed to provide a flying toy device controlled from a remote
device, like a smartphone or a computer, using a WiFi ”ad-hoc” communication. Although it has been
designed to be controlled off-board, it has an on-board Linux system running on an ARM processor. The
main mission of the embedded systems is processing data from IMU, propulsion engines, and managing
communications. The processor is also in charge to control USB, I2C and WiFi comunication.

2.1 Hardware

The hardware of the ARDrone can be described as a traditional robotic platform made of sensors,
control, and actuators. We briefly describe these elements in this section.
As sensors ARDrone has integrated a IMU composed of three accelerators and three gyroscopes, one
of them is a yaw precision gyrometer. This IMU provided raw accelerations and angular rotation data.
From the frontal camera a stream of images is sent to be processed to the off-board processor. The
bottom camera release a stremming of black and white of images which by using visual computation
are used to provided horizontal speed, they can also be sent to the off-board computer system, but only
one video stream can be transmitted, so a switching command is provided. Finally, a ultrasound sensor
give the altitude information with respect to the floor.
The control system computes the IMU, the bottom camera, and the ultrasound sensor information
to generate stabilization on board commands to the propulsion engines.
The quadroptor is propelled by four engines dedicated to stabilize the platform.
3 http://www.parrot.com

2
ISBN-xxx-y-zz-yyyyyyy-x 183
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2.2 Software

The ARDrone embedded operating system is based on a Linux kernel. The embedded software is in
charge of providing services to the platform. WiFi ad-hoc connection is created during the boot of the
system allowing communication between the off-board and on-board computers. The on-board DHCP
server gives the IP address to the connected device. Telnet port is available to accept connections with
the ARDrone. It also uses UDP ports to send telemetry and receive commands.

Figure 1: Description of ARDrone Architecture, courtesy of Parrot

3 Control Client Application

The ARDrone was designed to implement a human control through the UDP connection. A Control
Client Application (CCA) is an application developed to receive, decode, and analyze the navigation
data and the pipeline video information and to send back control commands. In the commercial version,
the CCA simply shows the data to a human operator using a mobile phone or a computer, and sends
back the commands issued by the operator.
As we have previously stated, the goal of our courses is training students to design autonomous
control systems, so using this platform, the goal will be to design a CCA running in a PC. This article
analyze three programming environments that currently allow to create Control Client Applications.

3
ISBN-xxx-y-zz-yyyyyyy-x 184
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3.1 SDK by Parrot

The ARDrone SDK provides tools and resources needed to create control applications for computers,
smartphones, and in general, for any hardware device supporting the WiFi ad-hoc mode. The ARDrone
SDK allows customize and extend the functionalities of the reference platform.
The ARDrone library has been developed in C++. This library is currently released like an open-
source library with high level APIs to access to the drone. It provide all the hardware raw values issue
from the IMU components. It manages the communications and the video pipeline.
The basic control interface shows a very complete status of the ARDrone internal control system as
well as the streaming of the video and allows to store about 284 telemetry parameters collected by the
processor.

Figure 2: Control interface, courtesy of Parrot

3.2 Urbi

Urbi is a open source software application for robotics developed by Gostai. Gostai Standard Robotics
API allows to developer creating Urbi Engines for standard devices and components implemented as
UObject and attibuting methods/attributes/events to access them. The developement of UObject with
C++ allows to access to low-level hardware details.
Urbi includes a runtime enviroment, called Gostai Runtime. Its development environment includes
a set of graphical development tools called Gostai Suite, a toolsuite containing Gostai Studio, an IDE
to observe robotic behavior using final state machines, and Gostai Lab to easily create Graphical User
Interfaces (GUIs).
Urbi is designed to create components and drivers, called UObjects, which can run on top of Gostai
Runtime and manage the hardware devices of the robotic platform. It provides interfaces with other
platforms, notably with ROS and Matlab. Urbiscript is the programming language used to manage the

4
ISBN-xxx-y-zz-yyyyyyy-x 185
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

components of the robot. It is a script language object-oriented. Its main feature is the parallel and
event-driven for process orchestration.

Figure 3: URBI Architecture by Gostai (TM), courtesy of Gostai

The module developed [8] for URBI has been created in C++ to give easy access to navigation data
and video pipeline and create friendly commands.
URBI proposes Urbiscript as programming language. It is a easy language allowing parallel task
execution. As script language it isn’t needed to be compiling with each new modification.
The program in Urbiscript can analyze navigation data and perform visual computing to generate
the commands. An urbiscript CCA can be executed without compilation.
With the following basic control Urbiscript we can control the drone tracking a red ball.

var Global.ball = colormap.new("drone.cameraVal", 200, 255, 0, 120, 0, 255, 0.0015, true);

myTag:every(0.1s)
{
if (ball.visible)
{
drone.speedX = -(ball.y * 2);
drone.speedY = (ball.x * 2.5);
}
else
{
drone.speedYaw = 0;
drone.speedZ = 0;
};
},

5
ISBN-xxx-y-zz-yyyyyyy-x 186
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3.3 ROS

ROS (Robot Operating System)[6] provides libraries and tools to help software developers create robot
applications. It provides hardware abstraction, device drivers, libraries, visualizers, message-passing,
package management, and more. ROS was originally developed in 2007 under the name switchyard by
the Stanford Artificial Intelligence Laboratory in support of the Stanford AI Robot (STAIR[?])
The University of Brown [7] has developed a specific stack for ARDrone to have access to navigation
data and pipeline video from the frontal camera. This driver provides also the commands to be sent
from the CCA to control the ARDrone.
The programming languages used to created CCAs in ROS are C++ and Python. It is necessary to
generate a package and to compile it before using.

4 Software comparative
The three different control programming systems proposed in this paper allow to develop CCAs.
The SDK by Parrot provides a very high access to all the hardware of the ARDrone. However
programming using the SDK is reserve to advanced developers with a high C++ skills. The libraries
release with the SDK can be used as source code to obtain drivers for other software environments.
ROS is an Open source platform that maintains thanks to a big community which has created an
abundant amount of libraries. ROS is an extended platform for robots programming, and also for
learning basic robotic skills, so using it is an advantage for robotics students.
On the other hand, URBI is the easiest programming environment when we use Urbiscript to control
the ARDrone. We can produce a control application with a MMI with a low programming skill. Unfor-
tunately there is not, at this moment, many libraries to develop control application but the bridge to
use ROS module open de possibility to perform advanced control application by using both platform.
For example, in Urbiscript with 22 code lines we can controlled the following tracking a red ball by
visual computing.

5 Conclusion
We propose to use ARDrone as an accessible, cheap, flying platform for robot programming due to the
low cost and easy programming.
In this article three software programming solution are proposed to provided different level of hard-
ware access, programming complexity and visualization as tool for robot teaching at the University.
From the point of view of teaching drivers development and advanced hardware access, we consider the
original SDK as the best solution. This is also the recommended software environment when we want
to teach flying stabilization control techniques.
From the point of view of robotic control teaching, Urbiscript is a programming language accessible
for beginner. The extensive variety of libraries, the graphic visualization and 3D track of attitude data

6
ISBN-xxx-y-zz-yyyyyyy-x 187
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

allows to consider ROS also a very friendly platform to teach.

6 Acknowledge
We thank Fablab Artilect and IRIT from Toulouse for hosting and lending the ARDrone equipment
during our stay in Toulouse.

REFERENCES
[1] Retention 101: Where Robots Go . . . Students Follow. Journal of Engineering Education, pp. 85-89,
(2003).

[2] Sünderhauf, N., Krause, T. & Protzel, P. (2005). RoboKing - Bringing Robotics closer to Pupils.
Proc. of IEEE International Conference on Robotics and Automation ICRA05, Barcelona, Spain,
pp. 4265-4270.

[3] Different robotic platforms for different teaching needs. Vicente Matellán y José Marı́a Cañas Plaza.
Seminario Hispabot 2003. Alcalá de Henares (Madrid), 29 y 30 de Abril de 2003. ISBN 84-607-
7238-1.

[4] Programming commercial robots. José Marı́a Cañas, Vicente Matellán Bruce MacDonald and Geof-
frey Biggs. Software Engineering for Experimental Robotics, Series: Springer Tracts in Advanced
Robotics , Vol. 30, pp. 125-132. Brugali, Davide (Ed.). ISBN: 978-3-540-68949-2,

[5] STAIR: Hardware and Software Architecture. Morgan Quigley, Eric Berger, Andrew Y. Ng (2007),
AAAI 2007 Robotics Workshop.

[6] ROS: an open-source Robot Operating System. Morgan Quigley, Brian Gerkey, Ken Conley, Josh
Faust, Tully Foote, Jeremy Leibs, Eric Berger, Rob Wheeler, Andrew Ng. International Conference
on Robotics and Automation (ICRA 2009).

[7] ROS driver for the Parrot AR.Drone. ROS package of the University of Brown.
http://code.google.com/p/brown-ros-pkg/wiki/ardrone brown/ (2010).

[8] AR. Drone et Urbi. Jean Charles MAMMANA. http://www.psykokwak.com/blog/index.php/2010/10/08/58-

ar-drone-et-urbi-en-remote/ (2010).

7
ISBN-xxx-y-zz-yyyyyyy-x 188
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

DESIGN AND EXPERIMENTAL VALIDATION OF A

PROTOTYPE OF DUCTED-FAN AERIAL SERVICE ROBOT∗

Roberto Naldi, Fabio Brué, Francesco Forte and Lorenzo Marconi

CASY-DEIS, Università di Bologna, Bologna, 40133, Italy.

Abstract

This work presents the design and experimental validation of an innovative prototype of ducted-
fan aerial robot specifically realized for tasks that may require physical interaction between the
vehicle and the environment. The prototype is characterized by a redundant number of aerodynamic
control surfaces that can be governed, though suitable control allocation policies, in order to obtain
the desired resultant control torque vector to be employed by the autopilot to stabilize the system
dynamics. Experiments are then presented in order both to derive the aerodynamic characteristics
of the control surfaces and to validate the performances of the proposed design.

keywords: aerial robots, modeling, ducted-fan, flight tests

1 INTRODUCTION
Recent applications of aerial robotics have shown the capability of miniature unmanned vehicles to
accomplish tasks requiring physical interaction between the vehicle and the surrounding environment.
These operations include, amon others, cooperative grasping and transportation [9], cleaning [2], docking
[5] and represent a new research direction for both the design and control of aerial vehicles. The vision,
which inspired also the European Project AIRobots [1], is indeed to allow airborne systems to succeed
in many service robotics operations [12] that are usually achieved only by robots fixed on the ground.
Goal of this paper is then to present the modelling and the design principles of an innovative ducted-
fan aerial robot that has been specifically realized to accomplish operations requiring physical interaction
with the environment. The choice of the ducted-fan configuration, see among others [8], [10] and [15],
derives precisely from the fact that, for this class of systems, the propeller is protected by a shroud, and
then the system can safely come into contact with the surrounding environment.
With respect to the prototype proposed in [10], one of the main feature of the prototype presented
in this work is given by the presence of a redundant number of aerodynamic control surfaces. The
∗ This research is framed within the collaborative project AIRobots (Innovative aerial service robots for remote inspec-
tions by contact, ICT 248669) supported by the European Community under the 7th Framework Programme.

1
ISBN-xxx-y-zz-yyyyyyy-x 189
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Fb Y
X

Z L6

D6

Fb
Y T X c2
L3
L2
L4 γ3 γ2
Z Fb
Y γ4 γ 1 L1
c.g. a1
L5 γ5 γ8
b
γ6 γ7
a2
L8
L6 c3
d L7
c1

dT
D5
D1
dT

(a) Control vanes. (b) Control vanes. (c) 3D Model.

Figure 1: The mechanical layout of the ducted-fan MAV.

aerodynamic forces and torques produced by the vanes have to be properly combined in order to obtain
a desired final control action useful to govern the system dynamic. In this respect this work focuses both
on the experimental investigation of the aerodynamic characteristics of the control vanes subsystem and
then also on the design of an effective control allocation policy able to minimize the aerodynamic drag
resistance and also to successfully produce the desired virtual control vector, despite the presence of
possible aerodynamic disturbances such as interferences among the vanes.
The paper is organized as follows. Section 2 presents the modeling and the aero-mechanical char-
acteristics of the prototype. Section 3 describes the control allocation policies synthesized to properly
govern the control surfaces. The experimental validation of the proposed algorithms and of the overall
design is then given in Section 4. Finally, in Section 5 some considerations are postponed.

2 DUCTED-FAN DESIGN AND MODELING

The ducted-fan prototype is composed of three different subsystems (see also [10]). The first one consists
of a fixed-pitch propeller driven by an electric motor (see respectively (a2) and (a1) in Figure 1(c)). This
subsystem generates the main thrust required to counteract the gravity force. The second subsystem
consists of a set of control vanes that are positioned below the propeller (see (b) in Figure 1(c) and Figure
1(b)) in order to properly deviate its airflow. The vanes produce a certain number of aerodynamic lift
and drag forces by which the system dynamics are controlled. This fact is investigated in the following
by analyzing the force / torque generation mechanism. The third and last subsystem is composed of
the duct and the fuselage that contains all the avionics and application-purpose hardware ((c1) and (c2)
in Figure 1(c) respectively); the role of the duct, in particular, is to protect the environment from the
propeller and to improve the efficiency of the thrust generation, by thus increasing the flight endurance or
the payload. To detect possible contacts with the surrounding environment, the duct has been equipped
with a set of contact probes, labelled as (c3) in Figure 1(c).

2
ISBN-xxx-y-zz-yyyyyyy-x 190
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

To derive a mathematical model of the system, the Newton-Euler equations of motion of a rigid
−
→ − → − →
body have been employed. By considering the inertial coordinate frame Fi = {Oi , i i , j i , k i } and the
−
→ − → − →
coordinate frame Fb = {Ob , i b , j b , k b } attached to the body, the dynamical model of the aircraft with
respect to the inertial frame is described by

M p̈ = Rf b
(1)
J ω̇ = −ω × Jω + τ b

where f b and τ b represent, respectively, the vector of forces and torques applied to the vehicle expressed
in the body frame, M the vehicle total mass, J the inertia matrix, p = col(x, y, z) the position of the
center of mass, ω the angular velocity expressed in the body frame and R the rotation matrix relating the
two reference frames. The vector product can be defined as ω× := Skew(ω) where Skew(col(x1 , x2 , x3 ))
denotes the skew-symmetric matrix with the first, second and third row respectively given by [0, −x3 , x2 ],
[x3 , 0, −x1 ] and [−x2 , x1 , 0].

Table 1: Ducted-Fan first prototype: features and Hw components.

Weight, payload, diameter 1100 g, 1100 g, 320 mm

Airframe Mechanics APC Electric 11x55E propeller, carbon fiber airframe
Servo Motors 8 HG-D202HB, BLDC-Motor 1 x Scorpion SII 3020
Flight Time approx. 7 minutes (with half payload)

2.1 FORCES AND TORQUES GENERATION

A detailed modeling of the ducted-fan’s force and torque generation mechanisms requires advanced
numerical and experimental tools for aerodynamic analysis [8]. In order to simplify the treatment,
several approximations are introduced, assuming that the flight envelope is limited only to the hover or
stationary flight. The interested reader can refer to [8], [15] and references therein for further details in
this direction. Let us start by considering the first subsystem. Following [13], the propeller generates
both a thrust force T and a resistance aerodynamic torque N given by

2
T (wP ) = kT wP , N (T ) = kN T (2)

with wP the angular speed velocity of the propeller and where kT and kN are constant coefficients
collecting aerodynamical parameters. According to Froude’s momentum theory, see [14], the propeller
generates also an airflow inside the duct. In particular, the induced air velocity directed along the body
z-axis can be approximated as s
T
Vind =
2ρSdisk
in which Sdisk denotes the area of the propeller’s disk and ρ the air density. The angular speed of
the propeller also influences the airflow’s motion by inducing a rotation around the propeller spin axis.

3
ISBN-xxx-y-zz-yyyyyyy-x 191
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The airflow is then employed by the second subsystem in order to generate the aerodynamic control
forces. This subsystem consists of a stator, realized with eight different fixed profiles, connected to a
set of eight independent control vanes each one driven by a servo controller (see Figure 1(b)). Two
different functionalities characterize the stator. From one side it helps to compensate for the torque
generated by the motor along the propeller’s spin axis. On the other side, due to its particular layout, it
reduces the airflow disturbances on the control vanes. In practice the size and the aerodynamic profile
of the stator have been tuned in order to compensate for the rotational component of the propeller
airflow velocity: as a result the control vanes, which are placed precisely below the stator forming a
continuous aerodynamic profile, are immersed into a relative wind that can be assumed to be directed
along body z axis as depicted in Figure 1(b). By changing the angle of attack of each vane, both lift
and drag aerodynamic forces are produced. In particular, by considering each i − th control vane as a
wing immersed into a relative wind of magnitude Vind (see among others [13]), the aerodynamic lift and
drag forces, Li and Di , can be computed as functions of the aerodynamic pressure
1 2
q(Vind ) = ρV (3)
2 ind
obtaining
Li (αi ) = SCL (αi )q(Vind ) , Di (αi ) = SCD (αi )q(Vind ) (4)

with i ∈ {1, 2, ...8} and where S is the vane’s surface (by constructions all the eight control vanes share
the same aerodynamic characteristics) and CL (αi ), CD (αi ) are respectively the lift and drag coefficients.
Under the stationary flow assumption [13], the lift and drag coefficients are nonlinear functions of
the angle of attack which depend on the specific aerodynamic characteristics of the profile and then
to be properly estimated experimentally (see among others [11]). For the prototype described in this
work, the aerodynamic characteristics of the control vanes have been directly obtained experimentally
by using a 6-axis load cell and measuring the forces and torques produced by tilting the control surface
from −90 degrees to 90 degrees. The aerodynamic forces, obtained considering the propeller thrust fixed
at the hovering value, have been depicted in Figures 2(a) and 2(b) showing also a possible analytical
approximation. In particular, for small angles of attacks (see also [13]), it turns out that a reasonable
approximation of the coefficients is given by

CL (αi ) = cL αi , CD (αi ) = cD αi2 + cD0 , i ∈ {1, 2, ..8} (5)

with cL , cD , cD0 constant parameters collecting the characteristics of the given airfoil profile and αi the
vane’s angle of attack.
With an eye on the lift force in Figure 2(a) and on the drag force in Figure 2(b), observe that the
analytical approximations (5) are valid approximately for the angle of attack belonging to the interval
[−ᾱL , ᾱL ], in which the value α = 0, at which the lift is zero and the drag is minimum, is obtained tilting
the vane at −10. With an eye on Figure 2(a), observe also that the aerodynamic side force, namely the
component perpendicular to both the lift and drag forces, is not perfectly zero as expected. This fact
is caused mainly by the aerodynamic interferences deriving from the fact that the control vanes, in the

4
ISBN-xxx-y-zz-yyyyyyy-x 192
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

specific prototype, are quite close to each others. Observe that the magnitude of the side force can be
reasonably neglected at small angle of attack values.

1.5
3.5
α = −α L α =0 α=αL
α = −α L α =0 α=αL
L
3
1.0

2.5
0.5
2
Lift [N]

Lift
0 1.5 Drag
(small AoA

[N]
approximation) Drag Force (small AoA
approximation)
1
−0.5

0.5 Side
(Expected Value)
−1.0
0
Side Force
−1.5 −0.5
−90 −80 −70 −60 −50 −40 −30 −20 −10 0 10 20 30 40 50 60 70 80 90 −90 −80 −70 −60 −50 −40 −30 −20 −10 0 10 20 30 40 50 60 70 80 90
Tilt Angle [°] Tilt Angle [°]

(a) Lift force. (b) Drag and Side forces.

Figure 2: The aerodynamic lift and drag forces produced by an aerodynamic surface at hover compared
to the analytical approximation (5). The pictures also show the side force resulting from the aerodynamic
interference of the vanes.

2.2 LIFT AND DRAG FORCES PRODUCED BY THE VANES

T
Let us denote with ib , j b and k b the unit vectors in the body frame Fb , namely ib := [1 0 0] , j b :=
T T
[0 1 0] , and k b := [0 0 1] . With an eye on Figure 1(b), we denote with Li (αi ) ∈ IR3 , Di (αi ) ∈ IR3
the lift and drag force vectors in the body frame
 
cos(γi − π/4)
 
 
Li (αi ) := Li (αi )  − sin(γi − π/4) , Di (αi ) := Di (αi )k b (6)
 
0
with i ∈ {1, 2, ...8} and where γi := (π/4)i, see also in Figure 1(b), denote the relative orientation of
each control vane with respect to the body y axis. In the following, for sake of compactness, we will
also denote the set of all the vanes angles of attack as α := {α1 , α2 , ..., α8 }.
With the above definitions at hand, the resultant force vector produced by the eight control vanes
can be computed as  P 
8 T b
P8 T b
 i=1 Li (αi ) i + i=1 Di (αi ) i 
 P 8 P 8 
FV (α) :=  L (α )T j b + i=1 Di (αi )T j b  (7)
 P i=1 i i P8 
8 T b T b
i=1 Li (αi ) k + i=1 Di (αi ) k
and the resultant torque vector is given by
8
X 8
X
τV (α) := ri × Li (αi ) + ri × Di (αi ) (8)
i=1 i=1

where, for each i ∈ {1, 2, ...8},

 1 
dT sin(γi − π/4)
 2 
 1 
ri :=  dT cos(γi − π/4) 
 2 
d

5
ISBN-xxx-y-zz-yyyyyyy-x 193
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

denotes the point of application of each aerodynamic pair of lift and drag forces with respect to the
center of gravity of the system which coincides precisely with the origin of the body fixed reference
frame. In particular the coefficient d denotes the distance between the center of pressure of each vane
and the center of mass projected along the body z axis, and the coefficient dT the distance between the
center of pressure of two opposite vanes (see also Figure 1(b)). The values of d and dT indeed represent
the length of level arms by which the aerodynamic lift and drag forces can be employed also to produce
a torque vector to govern the attitude of the vehicle.

3 CONTROL ALLOCATION
Goal of the control allocation policy is to obtain a control torque vector which can be used as a virtual
control input by the controller - see in particular [6], [4], [3]. The control allocation scheme proposed in
this paper assumes that the control law is synthesized using vectored-thrust control paradigms, such as
the ones proposed for instance in [10] and [7]. In this control allocation problem, denoted as vectored-
thrust control allocation problem (VT-CAP), the 8 independent control vanes are governed to obtain the
three virtual torque control inputs required to obtain full controllability of the attitude dynamics of the
vehicle. The redundancy of the number of control vanes, eight vanes to generate only three independent
components, is employed to minimize a given index cost. In particular, the control allocation problem
consists of the following optimization problem
8
X
min αi2 s.t.
i=1 (9)
τV (α) = v,

in which v := [vx , vy , vz ]T ∈ IR3 is the desired virtual control vector for the attitude dynamics.
The index cost has been chosen in order to minimize the overall drag force contribution, see the
approximated expression in (5), and also to minimize the undesired aerodynamic side force resulting
from the aerodynamic interference of the vanes depicted in Figure 2(b).
In order to design a control allocation policy actually employable on a real physical system, it is also
necessary to take into account for constraints on the control input vector α. In particular we consider
the following constraint

αi ∈ Ωi , i ∈ {1, 2, ..., 8}, Ωi := {αi ∈ IR : |αi | ≤ ᾱL } (10)

where ᾱL > 0 is the value obtained graphically in Figures 2(a) and 2(b). The above choice allows to
employ the approximated expression in (5) to model the lift and drag forces.
By defining
(i − 1)π
hi := , i ∈ {1, 2, ..., 8}
4

6
ISBN-xxx-y-zz-yyyyyyy-x 194
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

equations (7) and (8) can be rewritten as

 8

X
 cL Sq(Vind ) cos (hi ) αi 
 
 i=1 
 X8 
 
FV = cL Sq(Vind ) − sin (hi ) αi  (11)
 
 i=1 
 X 8 
 
Di (αi )
i=1

and  
8
X  8

d sin (hi ) αi  X
1 2
 i=1   2 dT cos (hi ) αi 
   
 X 8   i=1 
   X8 
τV = cL Sq(Vind ) d cos (hi ) αi  + cD Sq(Vind )  1 2 . (12)
   2 dT − sin (hi ) αi 
 i=1   
 X8   i=1 
 
− 21 dT αi 0
i=1

Then by choosing each angle of attack αi , i ∈ {1, 2, ..., 8}, as

 0 
1 vx vy0 vz
ᾱi (v, k) := sink (hi ) + cosk (hi ) + (13)
cL Sq(Vind ) λ(k)d λ(k)d 4dT

where k := 2j + 1 with j ∈ N ∪ {0} and

2
λ(k) := 2 + ,
2(k − 1)/2

since
8
X 8
X 8
X 8
X
sin (hi ) cosj (hi ) = cos (hi ) sinj (hi ) = 0, sink (hi ) = cosk (hi ) = 0,
i=1 i=1 i=1 i=1
8
X 8
X
k+1 k+1
sin (hi ) = cos (hi ) = λ(k) ,
i=1 i=1

we obtain
h iT h iT
τV = vx0 vy0 vz + µ(Vind ) vy0 vz −vx0 vz 0


with µ(Vind ) := cD / 4dc2L Sq(Vind ) . Finally, choosing
   −1  
vx0 1 µ(Vind )vz vx
 =    (14)
vy0 −µ(Vind )vz 1 vy
h iT
we are able to obtain the desired control torque vector τV = vx vy vz . Observe that the policy
(13) defines a feasible solution to (9) parameterized by k. The latter in practice allows to choose how
the desired control action is distributed among the different vanes. In fact, from (13), for large values
of k only the vanes aligned along the x and y body axis contribute to generate respectively the virtual
torque inputs vx and vy , while for small values of k, all the vanes cooperate in obtaining all components
of v.

7
ISBN-xxx-y-zz-yyyyyyy-x 195
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

P8
The index cost J for the above solution is given by J(α) := i=1 ᾱi2 (v, k) where, by considering
(13),
8
X  
1 1 1 1
ᾱi2 (v, k) = N (k)(vx0 )2 + 2 N (k)(vy0 )2 + 2 vz2
i=1
(cL Sq(Vind ))2 d2 d 2dT

having defined N (k) := (2 + (2/2k−1 ))/λ(k)2 . It can be verified that the minimum value for N (k)
corresponds to 1/4 and it is reached for k = 1. For larger values of k, the function N (k) asymptotically
converge to 1/2. Since for small angle of attack values the coefficient µ(Vind ) << 1 and then, from (14),
vx ≈ vx0 , vy ≈ vy0 , the value k = 1 corresponds to the optimal solution to (9) under the choice (13).

4 EXPERIMENTAL VALIDATION
The effectiveness of the proposed control allocation has been also verified through experiments. In
particular Figure 3(a) shows how the desired control torque is obtained with an average error, in term
of the ratio of the undesired components with respect to the desired one, that is maintained lower than
10 %. Figure 3(b) shows the position of the center of pressure by computing the ratio between the force
and torque measured by the load cell during the previous experiment. Observe that the small variations
are less than 5 % of the nominal value (to avoid numerical issues, the ratio has not been computed when
the commanded force is close to zero).

0.6 vy reference
v 0.28
x
0.4 v 0.26
y
v
z
0.24
0.2
0.22
d [m]
[Nm]

0 0.2

0.18
−0.2
0.16

−0.4 0.14

0.12
−0.6
0.1
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
time [sec] time [sec]

(a) Control allocation. (b) Position of the center of pressure (d).

Figure 3: The allocation of one of the torque component (vy ) and the position of the center of pressure
with respect to the center of mass during the experiment.

The overall aero-mechanical design has been finally validated also by means of a hovering experiment
(see Figure 5). By using a motion tracking system based upon infrared cameras, both the position
and the orientation of vehicle have been employed for feedback. The controller has been then realized
employing standard optimal linear control techniques considering the linear approximation of the system
dynamics (1) around the hovering equilibrium point. Figures 4(a) and 4(b) show respectively the position
and the attitude of the vehicle during 100 seconds of hovering flight. As expected the vehicle is able to
maintain the constant desired position with a reasonably small tracking error, which is approximately
of about 10 − 15 cm. Future works will consider the employment of an onboard inertial measurement
unit to improve the performances of the control design.

8
ISBN-xxx-y-zz-yyyyyyy-x 196
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

1 0.2
[m]

[rad]
0 0
x reference φ reference
x φ
−1 −0.2
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180

1 0.2
[m]

[rad]
0 0
y reference θ reference
y θ
−1 −0.2
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180

0
0.2

[rad]
[m]

−1 0
z reference ψ reference
z −0.2
ψ
−2 −0.4
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
time [sec] time [sec]

(a) Position. (b) Attitude.

Figure 4: The position and the orientation (Euler angles) of the prototype during a hovering experiment.

Figure 5: The ducted-fan prototype during a flight-test.

5 CONCLUSIONS
In this work the design and the experimental validation of a prototype of miniature ducted-fan aerial
robot has been presented. The aerodynamic characteristics of the attitude control vanes subsystem have
been identified experimentally and the obtained results have been employed to synthesize suitable control
allocation policies in which the aerodynamic drag forces and the undesired aerodynamic disturbances
have been minimized. The proposed design has been validate both measuring the actual forces and
torques produced by the vehicle’s actuators and also by means of preliminary flight tests.

REFERENCES
[1] Airobots project website. http://www.airobots.eu.

[2] A. Albers, S. Trautmann, T. Howard, T.A. Nguyen, M. Frietsch, and C. Sauter. Semi-autonomous
flying robot for physical interaction with environment. In Proceedings of the IEEE Conference on
Robotics Automation and Mechatronics (RAM), pages 441–446, Singapore, 2010.

9
ISBN-xxx-y-zz-yyyyyyy-x 197
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[3] K.A. Bordignon. Constrained Control Allocation for Systems with Redundant Control Effectors.
PhD. Thesis, Virginia Polytechnic Institute and State University, 1996.

[4] A. Casavola and E. Garone. Adaptive fault tolerant actuator allocation for overactuated plants. In
Proceedings of the 2007 American Control Conference, New York City, USA, 2007.

[5] L. Gentili, R. Naldi, and L. Marconi. Modelling and control of vtol uavs interacting with the
environment. In Proceedings of the IEEE Conference on Decision and Control, pages 1231–1236,
Cancun, Mexico, 2008.

[6] O. Härkegård. Backstepping and Control Allocation with Applications to Flight Control. PhD.
Thesis, Department of Electrical Engineering, Linkping University, 2003.

[7] M.D. Hua, T. Hamel, P. Morin, and C. Samson. A control approach for thrust-propelled under-
actuated vehicles and its applications to vtol drones. IEEE Transsactions on Automatic Control,
54(8):1837–1853, 2009.

[8] O.J. Ohanian III, P.A. Gelhausen, and J. Inman. A compact method for modeling the aerodynamics
of ducted fan vehicles. In Proceedings of the 48th AIAA Aerospace Sciences Meeting Including the
New Horizons Forum and Aerospace Exposition, Orlando,Florida, USA, 2010.

[9] D. Mellinger, M. Shomin, N. Michael, and V. Kumar. Cooperative grasping and transport using
multiple quadrotors. In Proceedings of the International Symposium on Distributed Autonomous
Systems, Lausanne, Switzerland, 2010.

[10] R. Naldi, L. Gentili, L. Marconi, and A. Sala. Design and experimental validation of a nonlinear
control law for a ducted-fan miniature aerial vehicle. Control Engineering Practice, 18(7):747–760,
2010.

[11] R.E. Sheldahl and P.C. Kilmas. Aerodynamic characteristics of seven symmetrical airfoil sections
through 180-degree angle of attack for use in aerodynamic analysis of vertical axis wind turbines.
Sandia National Laboratories Energy Report, 1981.

[12] B. Siciliano and O. Khatib, editors. Springer Handbook of Robotics. Springer, 2008.

[13] R.F. Stengel. Flight Dynamics. Princeton University Press, 2004.

[14] B. Thwaites. Incompressible Aerodynamics. Oxford University Press, 1960.

[15] M.W. Zhao and C. Bill. Aerodynamic design and analysis of a vtol ducted-fan uav. In Proceeding
of the 26th AIAA Applied Aerodynamics Conference, Honolulu, Hawaii, USA, 2008.

10
ISBN-xxx-y-zz-yyyyyyy-x 198
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

On the Use of Calculus of Variations to Determine Optimum Rotor

Shapes for Micro Air Vehicles
GREGOR THURLOW, ALVARO CUERVA and OSCAR LOPEZ-GARCIA
Department of Aerospace Engineering, University of Glasgow, Glasgow, Scotland,
[email protected]

Abstract
This report uses Calculus of Variations and Blade Element (BET) and Momentum Theories to
investigate the design of low-Reynolds number, hovering rotors typical of Micro Air Vehicles. The
mathematical model is derived and, using the same fundamental process, the traditionally known
Ideal Twist Rotor (ITR) and Optimum Rotor (OR) are found as well as a third rotor, called the
Minimum Power Rotor (MPR) being explored. The rotors are derived using linear and non-linear
versions of the BET including root and tip losses. It is found that for low aerodynamic efficiencies,
the MPR exhibits a significant performance benefit over the OR. It is also shown that the MPR
design exhibits a non-uniform induced velocity, a feature which contradicts the widely accepted
view that for a rotor to consume minimal power, there must be uniform induced velocity profile
across the disk. An asymptotic analysis is used to show that when the aerodynamic efficiency is
increased, the solution of the MPR tends towards that of the OR. The result obtained can be
exploited to improve the performance of rotary-wing Micro Air Vehicles.

Keywords: Micro Air Vehicle, Rotor, Optimization, Low-Reynolds Number

1. INTRODUCTION
With recent advances in nano-scale technology, electronic devices have become increasingly
smaller, making the development of sub-scale data sensors and autonomous flight control
systems a reality. This has spawned interest in the production of flying machines whose tiny
dimensions and operating environment allow an unprecedented number of practical applications.
Fixed wing designs, although providing long range and endurance times, are not ideal for
surveillance missions as their low maneuverability and inability to hover mean they can only loiter
above a target. The solution therefore, lies with rotary and flapping wing designs which have the
advantage of being able to hover, allowing the ability to 'perch and stare'. Despite recent
success, the development of flapping wing designs has been slow due to the little-understood
and complex nature of the unsteady aerodynamics associated with this method of flight. The
rotary wing concept however, has provided the most breakthroughs in MAV design.
MAVs may serve a wide range of applications including damage assessment, detection
of mines and artillery fire as well as relaying communications. Outside warfare applications, the
use of quad-rotor UAVs are already in use for surveillance purposes by UK police forces [13] and
power companies who need to inspect wind turbines and pylons.
Operating on such a micro-scale presents several problems, not least the degradation in
aerodynamic performance associated with ultra-low Reynolds numbers. As the scale of the rotor
decreases, the viscosity of the air becomes more influential and the power required to hover
increases. This causes the ratio of ideal power to actual power (known as the Figure of Merit) to
reduce from around 0.8 for the most efficient helicopters down to around 0.3 – 0.4 for typical
MAVs (Leishman, 2006, pp.280, 412). The experimental and computational work for
1

ISBN-xxx-y-zz-yyyyyyy-x 199
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

aerodynamic surfaces operating under such conditions mostly covers 2D airfoil design for fixed
wing aircraft. Works such as that carried out by Sunada et al. (1997), Mueller (1999) and Kunz
(2003) examine the effect of airfoil camber, thickness, surface roughness and edge shape
measured at Reynolds numbers as low as 4 × 103. There is questionable validity, however, in
simply extrapolating fixed-wing results to the rotor case due to the 3D flow conditions caused by
the sustained blade proximity of each blade tip vortex and wake interactions (Leishman, 2006,
pp.584). There are only several works where attempts to optimize rotating-wing geometry for
ultra-low Reynolds number regimes have been carried out. Notably, Bohorquez [8] optimizes
blade geometry of rotors operating between Re = 5000 – 60,000 by taking numerical and
experimental values of thrust and power coefficients and using an algorithm to find the optimum
power loading for a specified thrust.
If the capabilities of MAVs are to extend beyond their current level, then it is important to
continue to improve the efficiency of small-scale rotors. By minimizing the power for a given
thrust in the low Reynolds number regime, the benefits for the rotary-wing MAV include extended
endurance times, increased payloads as well as reducing the dimensions of the aircraft for a
given thrust. Work, such as that presented here is the key to putting rotary-wing MAVs at the
forefront of a modern and rapidly developing industry.

2. REVIEW OF LOW REYNOLDS NUMBER AERODYNAMICS

The main problem in MAV rotor design from an engineering perspective is the degraded
aerodynamic performance of airfoils at low Reynolds numbers. Two of the most important factors
affecting airfoil performance are the compressibility and viscosity of air. For typical MAV
airspeeds, the pressure variations around the lifting surfaces are small compared with the
absolute pressure (low Mach number), so compressibility effects may be ignored. The viscosity
of the air, therefore, may be considered independently. For decreasing Reynolds numbers, the
viscous forces become increasingly effective in relation to the inertial forces, essentially making
the fluid more 'sticky' as the scale of the airfoil is reduced. This has implications on the condition
of the thin layer of fluid near the airfoil surface known as the boundary layer.
For Reynolds numbers above 106 (from light aircraft upwards), natural flow disturbances
such as those caused by surface roughness occur even at small angles of attack which cause
the boundary layer to transition from a laminar to a turbulent state. Below this value, the
behaviour of the boundary layer is very changeable and this range, known as ‘low-Reynolds
numbers’ will be the focus of the following discussion. Typically, MAV aircraft operate at
Reynolds numbers in the region of 30,000 to 60,000 (Bohorquez, 2010). Below Reynolds
numbers of 105 the laminar boundary layer separates more readily increasing the profile drag of
airfoil and the results obtained from experiments grow increasingly inconsistent. Towards ultra-
low Reynolds numbers (below 10,000) small modifications in airfoil shape can significantly affect
the performance. In general, the overall lift to drag ratio (the aerodynamic efficiency) continues to
decrease well below 10. Lissaman [6] notes that there is a “critical” Reynolds number around
70,000 at which a smooth airfoil will experience a dramatic decrease in aerodynamic efficiency of
more than an order of magnitude over a very short Reynolds number range due to the separation
of the boundary layer from the surface of the airfoil. This can be seen in Figure 1 where the
2

ISBN-xxx-y-zz-yyyyyyy-x 200
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

aerodynamic efficiency of smooth airfoils can be seen to fall below that of rough airfoils –
associated with insects’ wings – as Re is reduced.

Figure 1 – Reduction in Aerodynamic Efficiency with Reynolds Number

This is due to the fact that at low Reynolds numbers, transition of the boundary layer from
laminar to turbulent has not yet taken place and at the onset of even a small pressure gradient,
the relatively poor separation resistance of the laminar boundary layer causes the flow to detach.
The separated layer very rapidly undergoes transition to a turbulent flow which in turn increases
the entrainment making it possible for reattachment to take place. This is known as a laminar
separation bubble. Carmichael, (1981) as cited by Lissaman (1983) and Mueller (1999), notes
that a rough rule is the distance of separation to reattachment can be expressed as a bubble
length corresponding to a chord Reynolds number of 50,000.
Thus for Reynolds numbers lower than this, the airfoil is physically too short for
reattachment to occur and the airfoil is essentially stalled. This explains the rapid drop in smooth
airfoil performance. The aim for MAV designers is to reduce the critical Reynolds number by
delaying the onset of separation as the scale of the airfoil is reduced. This is a delicate procedure
as it has been seen that tiny changes in airfoil shape can significantly affect the performance
[10]. It is known that the maximum efficiency and optimum lift coefficient of an airfoil are strongly
influenced by the local Reynolds number. Therefore, for MAV rotor design, it is desirable to
maximize the efficiency of every span-wise blade element. In order to do this, it is necessary to
have accurate and reliable data on the airfoil, and so for this investigation, an extensively tested
airfoil, the NACA 0012, is used.

3. VARIATIONAL FRAMEWORK
The use of Calculus of Variations is used to determine a set of equations for which the necessary
parameters of the rotor are computed. It is carried out in accordance with Section 2 of the book,
‘Applied Optimal Control’ by Bryson and Ho (1975) which uses the standard method of
Functional Optimization for dynamical systems. The rotor may be represented as a dynamical
system where the parameters vary over the span of the blade (as oppose to time-varying
dynamical systems). A functional is introduced which consists of a set of control functions ,
which vary with , the non-dimensional blade position. These are used to represent a

ISBN-xxx-y-zz-yyyyyyy-x 201
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

performance index consisting of an objective function which is minimized over the span of the
rotor. It therefore remains to determine the set of control functions that minimize the functional:

, , 1… (1)

For the case of the rotor, the objective function is equal to the induced power, the profile power
or a combination of the two. The aim is to minimize the performance index subject to a given
thrust coefficient. There are two constraints; the first is an algebraic equality constraint presented
by the Blade Element and Momentum Theories (see later). The equality constraint is represented
simply as:

, 0, 1… (2)

The second constraint is a path constraint presented by the total thrust coefficient given by the
Blade Element Theory. This is a predetermined value where the boundary conditions are known
and the integral must have a specified value along the optimal path:

̅ , , 0 0, 1 ̅ , 1… (3)

By manipulating Equation (3), the constraint is transformed into a differential form which again, is
a function of the control functions and non-dimensional blade position:

, (4)

These are adjoined to the performance index by way of two Lagrange multipliers and to
obtain the following the augmented problem:

∗
, , , (5)

Equation (5) represents the complete mathematical model of the rotor. It now remains to
determine each of the control functions and the two Lagrange multipliers in order to compute
the rotor parameters. This is achieved by employing Calculus of Variations whereby the first
variation of the augmented functional is computed and set to zero in order to find a minimum for
the objective function. It is necessary to simplify the augmented problem by defining the following
scalar function:

, , , , , , (6)

The equation above (called the Hamiltonian) combines the two constraints with the objective
function. Note that the Hamiltonian is not a function of the thrust coefficient.
The augmented problem can now be written as:

∗
, , , (7)

The problem has now been transformed into an equation whereby the last term on the right hand
side of the integrand may be integrated by parts:

ISBN-xxx-y-zz-yyyyyyy-x 202
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

∗ | | , , , (8)

Considering the variation in due to changes in the control functions , the first variation of the
functional is obtained:
∗ | | (9)

∗
For a minimum to be reached, the variational must be equal to zero. The first two terms of
Equation (9) are set to zero because the values at the boundaries are both prescribed. From
inspection of the integral on the right hand side of the equation it follows that:

(10)

0, 1… (11)

where Equation (11) represents a set of equations where the Hamiltonian is partially

differentiated with respect to each of the control functions. Due to the fact that 0,

Equation (10) shows that the Lagrange multiplier remains constant along the span, a fact which
simplifies the process in obtaining the rotor parameters. These are known as the Euler-Lagrange
equations from Calculus of Variations.
Using the relations represented by Equation (11) and Equations (2) and (4), a set of equations
is obtained which allows the two Lagrange multipliers and each of the control functions to be
computed. These represent the necessary conditions for a minimum to be reached:

0 , 1… (12)

, 0 (13)

, ; 0 0, 1 ̅ (14)

The framework derived here is the basis on which the rotors are optimized. It remains to derive
the Blade Element and Momentum Theories which are combined in this optimization process in
order to compute the existing rotor designs of the ITR and OR and also used to obtain the MPR.

4. MOMENTUM THEORY
The incremental thrust through the disk can be obtained by deriving an expression for the
incremental mass flow rate for the hovering rotor in terms of the induced velocity. Using the
conservation of energy principle, the simple relationship between the slipstream velocity and
induced velocity can be found upon which an expression is obtained for the thrust in terms of the
induced velocity and mass flow rate. Substituting for the mass flow rate, an expression for thrust
is obtained upon which the coefficient may be found:

4 (15)
Ω Ω

ISBN-xxx-y-zz-yyyyyyy-x 203
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

where is the blade radius and Ω is the angular velocity of the rotor and all the other symbols
have their usual meaning [1]. By defining the terms Equation (15) may be simplified; in hover,
Ω

is defined as the inflow ratio given by the symbol and is the non-dimensional radial position.

This simplification gives:

4 (16)

This result contributes the first part of algebraic equality constraint represented by Equation (2).

5. BLADE ELEMENT THEORY

Blade Element Theory (BET) is the basis for the majority of all rotor analyses. It allows the
calculation of aerodynamic load distributions over the rotor by splitting each blade into 2D
sectional airfoils. It may be used to determine blade parameters such as taper, twist and airfoil
shape. Here, BET is derived using non-linear and linear aerodynamics. The 2D lift and drag on
each blade element may be defined:

1 (17)
2
1 (18)
2

These can be resolved into force component vectors which are perpendicular and tangential to
the rotor plane:

cos sin (19)

sin cos (20)

Finally, the elemental thrust and torque can be found by multiplying the force component vectors
by the number of blades :

cos sin (21)

sin cos (22)

By substituting Equations (17) and (18) into Equations (19) and (20), and using the relationship
for the thrust coefficient in Equation (15), it is possible to obtain a thrust coefficient value for the
BET. Using the commonly accepted relationship for rotor solidity and the previously defined
relationship for non-dimensional blade span-wise position, the non-linear thrust and torque
coefficients are simplified to:

cos sin (23)

2

sin cos (24)

Ω 2

By definition, the torque coefficient is equivalent to the power coefficient, ≡ . This can be
split into two components; the induced power coefficient and the profile power coefficient .
The equations for the induced power and profile power coefficients are presented below:

ISBN-xxx-y-zz-yyyyyyy-x 204
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

sin , cos (25)

2 2

The relationships in Equation (25) and the incremental thrust coefficient given by Equation (23)
are used to derive the non-linear and linear versions of the BET.

6. NON-LINEAR ANALYSIS
For hover, the climb velocity is zero so the inflow angle can be simplified using the relationships
to obtain:

tan tan tan (26)

Ωr Ω

Substituting Equation (26) into Equation (23) and also into the relationships in Equation (25), the
following non-linear derivations for Blade Element Theory are derived.

cos tan sin tan (27)

2

sin tan , cos tan (28)

2 2

7. LINEAR ANALYSIS
The small angle assumption for the inflow angle is given below:
cos 1; sin ; tan (29)

The following assumption for the inflow angle is given:

tan (30)

Equation (30) is now substituted into Equations (23) and (25). Using the small angle assumption
given by Equation (29), the following linear derivations for the thrust and profile and induced
powers are given:

(31)
2

, (32)
2 2

8. ROOT AND TIP LOSSES

The derivation of the rotor geometries is to include the root and tip loss factor which models the
trailed tip vortices produced by each blade, increasing the local induced velocity and reducing the
lifting capability at these regions. This is a function of number of blades and inflow ratio. The root
and tip loss factor increases the accuracy of the computational model but complicates the
derivation of the OR and MPR by increasing the non-linearity of the equations. It is given by:

2 (33)
cos exp

ISBN-xxx-y-zz-yyyyyyy-x 205
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

where represents a function related to number of blades , inflow angle , and the non-
dimensional blade position. The root and tip loss functions are given below:

1
, (34)

Where is the root cut-out of the rotor blade. The root losses arise from the same principle as
the tip losses except that their effects are smaller given the lower local tangential velocity.
However, airframe components such as the hub also inhibit lifting capability so it is important to
include this. The overall tip/root loss function is constructed by factoring in the relationships in
equation (35) to construct two separate functions and multiplying these together:

(35)

The tip/root loss effects are included in the analysis by modifying the algebraic constraint relation
using non-linear BET.

, cos sin 4 0 (36)

2

9. IDEAL TWIST ROTOR (ITR) AND OPTIMUM ROTOR (OR)

The well-known ITR accomplishes minimum induced power by ensuring a hyperbolic
angle of attack distribution with constant solidity to create a uniform induced velocity across the
disk. The control functions that are used to derive this rotor are the inflow ratio and angle attack
distributions to minimize the functional:

, α sin (37)
2

All the rotor parameters are found assuming a linear lift curve slope.
The OR works on the same principle as the ITR, except that in addition to minimizing the
induced power, the profile power is minimized by ensuring all blade elements operate at the
optimum angle of attack. This is achieved by a hyperbolic solidity distribution. The OR is derived
here by firstly minimizing the induced power using all three control functions:

1
, , α sin (38)
2

The inclusion of the solidity function results in there being more unknowns than equations in
order to find the rotor parameters. However, these are solved for by the second minimization
process where the profile power is minimized using only the angle of attack control function with
no constraints:
(39)
α cos
2

10. MINIMUM POWER ROTOR (MPR)

For the OR, the profile power is minimized under the constraint that the induced velocity is
constant. It is apparent that a third rotor may be derived if the total combined power is minimized
simultaneously using all three control functions. That is:
8

ISBN-xxx-y-zz-yyyyyyy-x 206
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

1
, , sin cos (40)
2

11. GENERAL PROCEDURE

The induced and profile power relationships in Equation (25) are general cases i.e. they may be
used to compute the rotors using both linear and non-linear aerodynamic analyses. All three
rotors may be derived using the linear analysis. For the non-linear analysis, the root/tip loss
function is included, and only OR and MPR are derived because the relationships for the rotor
parameters for the ITR cannot be found analytically.
The process for solving the rotor parameters using non-linear aerodynamics becomes largely
numerical and this is done using MATLAB. The basic procedure for both the linear and non-linear
analyses is the same and this is outlined in Figure 2 below:

Figure 2 - General Procedure for Numerical Analysis

For each rotor, the functional is constructed based on the control functions and the choice of
power to be minimized (induced, profile or a combination of the two). The Hamiltonian is obtained
and its partial derivatives are found as well as the two constraints to form the equations
necessary for a minimum. These are shown on the left hand side of the diagram. Steps 1 -7
outline the general procedure which is carried to obtain the final rotor derivations.

1. Solve analytically for the second Lagrange multiplier .

2. Solve analytically the algebraic constraint equation for the rotor solidity .

3. Substitute solutions from steps 1 and 2 into to obtain the relationship for maximum

aerodynamic efficiency and hence optimum angle of attack .

4. Substitute results for and steps 1 and 2 into .

5. Solve numerically to find values of inflow ratio .

6. Select the appropriate relationship of inflow ratio from step 5 and substitute into result
from step 2 to obtain a relationship for rotor solidity .
7. Substitute new value of rotor solidity into differential constraint. Solve numerically to
obtain a value for constant Lagrange multiplier and select appropriate value.
Once steps 1 – 7 have been completed, the final results for the rotor are easily obtained.

ISBN-xxx-y-zz-yyyyyyy-x 207
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

12. RESULTS
The results for the non-linear analysis incorporating the root/tip loss are presented in
Figures 3 – 5 where only the OR and the MPR are compared. The root cut-out value is 6% of the
total blade span and the aerodynamic efficiency is 20.
Figure 3 shows solidity and inflow ratio distribution. The effects of the root and tip loss
function can clearly be seen as well as the inclusion of the root cut-out. The chord variation is
directly proportional to the solidity so this is a good indication of blade planform shape. The
maximum solidity for MPR is almost double the maximum of the OR but decreases more rapidly
towards the tip, corresponding to a shorter blade chord for the outer regions.
Figure 4 shows the similarity in pitch distribution with the OR and MPR. This result is
similar to that given by the linear analysis with the exception of the obvious effects from the
root/tip loss function. Figure 5 displays the typical power distributions. The peak values of the
OR occur near the tips and are significantly higher than those of the MPR whose distributions are
generally flatter. The induced power reaches a maximum of 3 × 10-4 at around 60% blade span
before trailing off towards the tips whereas the OR follows a steeper gradient reaching its
maximum at around 90% blade span.
Table 2 shows the differences in global power coefficients for the ITR, OR and MPR
derived using the linear and non-linear analyses. It can be seen that for both sets of results, the
total power coefficient of the MPR is lower than both the ITR and OR for this particular
aerodynamic efficiency even though the global induced power coefficient is seen to be higher.
The results of the linear and non-linear analysis of the MPR are now compared. In Figure
6, apart from the region below = 0.2, the solidity variation is practically the same with only a
slight increase in peak inflow ratio. In Figure 7, the OR is compared for linear and non-linear
analyses. It may be seen how the root and tip loss dramatically affect the distribution of the inflow
ratio, with a peak value that is slightly higher than the constant value given by the linear analysis.
A commonly used method of comparing rotor performance is by obtaining the Figure of
Merit (FM) which is the ratio of actual power to ideal power of the hovering rotor. The ideal power
is that which assumes no viscous losses (given by the Momentum Theory) and therefore is
purely induced. In reality, the viscous effects contribute to both the induced and profile power and
these are represented in the actual power. In Figure 8 the FM of each rotor is measured over a
range of thrust coefficients and varying aerodynamic efficiencies of 15, 30, 60 and 100. This
method gives a non-dimensional measure of how rotors of equal disk loading perform at
generating thrust for a given power.
It is clear that the largest variations occur at the lowest values of aerodynamic efficiency
and thrust coefficient. The results show that there is a significant performance benefit to be
gained by the MPR at low Re regimes where the aerodynamic efficiency is low. It is seen that as
Re increases, the performances of the rotors become indifferent. This relates to the MPR
becoming increasingly more efficient in hover as the scale of the rotor is reduced, suggesting that
for rotary-wing MAV design, there are significant performance advantages to be had by using this
rotor. The MPR Figure of Merit can be seen to be over 1.08 times that of the OR for the lowest
obtainable aerodynamic efficiency (K = 15) at the lowest thrust coefficient ( = 0.001).

10

ISBN-xxx-y-zz-yyyyyyy-x 208
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

13. RESULTS - FIGURES AND TABLES

0.8 60
Minimum Power Rotor
Optimum Rotor 50 Minimum Power Rotor
0.6 Optimum Rotor
40
solidity ()

 (º)
0.4 30

20
0.2
10

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x x

0.065 0.01

0.06 0.008
inflow ratio ( )

0.055
0.006

dCTdx
0.05
0.004
0.045
0.002
0.04
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x x

Figure 3 ‐ Solidity and Inflow Ratio Distributions from Figure 4 ‐ Pitch and Thrust Distributions from Non‐Linear
Non‐Linear Analysis Analysis

Linear Non‐Linear
ITR OR MPR ITR OR MPR
Induced Power Coefficient ‐
2.5 2.5 2.6 2.57 2.65
( 10 )
Profile Power Coefficient ‐
1.77 1.67 1.48 1.65 1.5
( 10 )
Total Power Coefficient ‐
4.27 4.17 4.07 4.22 4.14
( 10 )
Table 1 ‐ Global Power Coefficients from Non‐Linear Analysis
-4 1
x 10
6
Linear
Minimum Power Rotor 0.8
4 Non-linear
Optimum Rotor
CP0

solidity ()

0.6
2

0 0.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
0.2
-4
x 10
6
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
4 x
CPi

2
0.08
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x 0.07
inflow ratio ( )

-3
x 10 0.06
1

0.05
0.5
CP

0.04

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.03
x 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Figure 5 ‐ Power Distributions from Non‐Linear Analysis Figure 6 ‐ Comparison of Solidity and Inflow Ratio
Distributions of the MPR
1.1
0.5 K =15
Linear K = 30
0.4 1.09 K = 60
Non-linear
K = 100
solidity ()

0.3 1.08

0.2 1.07

0.1
1.06
FM MPR / OR

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1.05
x

1.04
0.052

1.03
0.05
inflow ratio ( )

1.02
0.048
1.01
0.046
1
1 2 3 4 5 6 7 8 9 10
0.044 CT x 10
-3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Figure 8 ‐ Figure of Merit Ratio for Varying Aerodynamic

Figure 7 ‐ Comparison of Solidity and Inflow Ratio
Efficiencies
Distributions of the OR

11

ISBN-xxx-y-zz-yyyyyyy-x 209
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

14. CONCLUSIONS
The work presented here illustrates how the sub-scale rotor can be improved using Blade
Element Momentum Theory, calculus, and a rigorous yet relatively basic mathematical
framework to optimize the rotor performance at low aerodynamic efficiencies. The model proves
its worth by creating the Ideal Twist and Optimum Rotors using the standard linear assumption of
the aerodynamics, yielding the traditional and well-known results for both of the rotors. The
model then shows how a third rotor, the Minimum Power Rotor, may be produced, exhibiting the
unusual result of a non-uniform inflow velocity across the disk. By modeling root and tip losses as
well as switching to non-linear aerodynamics, more accurate models of the Optimum and
Minimum Power Rotors are produced and their results are discussed. The analysis shows how
the MPR convincingly improves on the design of the OR. It illustrates how, for decreasing
aerodynamic efficiencies, the MPR:OR Figure of Merit ratio increases and at an aerodynamic
efficiency of 15, the MPR boasts an 8% improvement in Figure of Merit. This arises from the fact
that although the induced power is comparatively higher, the lower value of the MPR profile
power overcompensates for this, resulting in a lower total power consumption at low
aerodynamic efficiencies. As the lift-to-drag ratio decreases (and thus the scale of the rotor), it
becomes apparent that the performance benefits from the MPR become more and more
significant. This has implications from the view-point of Nano Air Vehicles’ design whose
operating environments are in the ultra-low Reynolds regime where the aerodynamic efficiency
drops below 10. With a higher figure of merit, the power required to hover for a given thrust
coefficient and disk loading is reduced, relating to a higher endurance and/or higher payload for
an MAV which utilizes the MPR.

15. REFERENCES
[1] Leishman, J.G., 2nd ed., 2006. Principles of Helicopter Aerodynamics. University of
Maryland: Cambridge University Press.
[2] Bryson, A. E., Jr. and Ho, Y., 1975. Applied Optimal Control. Harvard University:
Hemisphere Publishing Corporation.
[3] Houghton, E.L., Carpenter, P.W., 5th ed., 2003. Aerodynamics for Engineering Students.
Oxford: Butterworth-Heinemann.
[4] Abbott, I.H., Doenhoff, A.E.V., 1959, Theory of Wing Sections.New York: Dover Publications.
[5] Sunada, S., Sakaguchi, A., Kawachi K., 1997. Airfoil Section Characteristics at a Low
Reynolds Number. Journal of Fluids Engineering. 119, pp.129-35.
[6] Lissaman, P.B.S., 1983. Low-Reynolds-Number Airfoils. Annual Review of Fluid Mechanics.
15,pp.233-39.
[7] Ganguli, R., 2004. Survey of Recent Developments in Rotorcraft Design Optimization.
Journal of Aircraft.41(3), pp.493-510.
[8] Bohorquez, F., Pines, D., Samuel, P.D., 2010. Small Rotor Design Optimization Using Blade
Element Momentum Theory and Hover Tests. Journal of Aircraft. 47(1), pp.268-83.
[9] Lopez-Garcia, O., Cuerva, A., Esteban, S. On the Use of Calculus of Variations to Determine
the Shape of Hovering Rotors of Minimum Power and its Application to Micro and Nano Air
Vehicles. Journal of Aerospace Engineering (Accepted for publication).
[10] Kunz, P.J., 2003. Aerodynamics and Design for Ultra-Low Reynolds Number Flight. Stanford
University.
[11] Cuerva, A., Reynolds Number Implications on the Determination of Wind Turbine Optimum
Rotors.
[12] Mueller, T.J., Sep 1999. Aerodynamic Measurements for Fixed Wing Micro Air Vehicles.
Hessert Centre for Aerospace Research, University of Notre Dame.
[13] http://www.technovelgy.com/ct/Science-Fiction-News.asp?NewsNum=2811

12

ISBN-xxx-y-zz-yyyyyyy-x 210
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Mini coaxial rocket-helicopter: aerodynamic modeling, hover

control, and implementation

E. S. Espinoza1,2 , O. Garcia1 , R. Lozano1,3 , and A. Malo1

1 Laboratoire Franco-Mexicain d’Informatique et Automatique, LAFMIA UMI 3175 CNRS-CINVESTAV Mexico

E-mail:{eespinoza,ogarcias,rlozano,alexmalo}@ctrl.cinvestav.mx
2 Universidad Politecnica de Pachuca, Hidalgo Mexico

E-mail:[email protected]
3 Laboratoire Heudiasyc, Université de Technologie de Compiègne, Compiègne, France

E-mail:[email protected]

Abstract

This paper presents the first stage of the development of a mini coaxial rocket-helicopter (MCR UAV) whose
main characteristic is that it can be catapulted and then hover at a long distance from the launching site. A
functional description of this mechanism is given and the aerodynamic model is obtained using the classical
Newton-Euler equations. The aerodynamic effects of the vehicle are obtained by studying the wind analysis
(Propeller Momentum Theory) in hover flight. An embedded system is built in order to validate the proposed
aerodynamic prototype, and a classical nonlinear control law based on backstepping procedure for the complete
system is implemented in order to test this vehicle in autonomous flight. Finally, simulation and practical results
are presented for hover flight, and the future tasks of this project are listed.

keywords: Mini coaxial rocket-helicopter, aerodynamic model, hover flight, and embedded system.

1 INTRODUCTION
In recent years, developments of mini UAVs (Unmanned Aerial Vehicles) have been growing thanks to opera-
tional and technological advances in electronics, computing science, and mechanics. Environmental monitoring,
surveillance, and homeland security are civilian missions, whereas special warfare, battle group surveillance are
common objectives for military missions.
Mini coaxial helicopters are capable of taking off and landing in reduced, and complex spaces or cluttered
environments. Due to their configuration, coaxial helicopters require mechanism such as swashplates, stabilizer
bars, and tilt-rotors in order to control the direction of rotor thrust vector. The coaxial configuration reduces
the reaction torque generated from single rotor configuration (conventional helicopters), and generates power-
ful thrust force for hover flight. In addition, coaxial helicopters are mechanically stable because the center of

ISBN-xxx-y-zz-yyyyyyy-x 211
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

gravity is below a pair of counter-rotating propellers. Many theoretical and practical developments about coax-
ial helicopters have been reported in the literature. In [12], authors present a robust control for a coaxial micro
helicopter. Design and control of an indoor coaxial helicopter are discussed in [1]. A simplified model and back-
stepping control for a coaxial helicopter can be found in the reference [4]. Nevertheless, this paper presents a mini
coaxial rocket-helicopter which possesses control surfaces (ailerons) to control the attitude flight. The complete
aerodynamic flight model is obtained using the classical Newton-Euler equations, while a classical control law
based on the backstepping procedure is used to achieve the global stability for the vehicle in hover mode.
The main contribution of this paper is to present the Mini Coaxial Rocket-Helicopter (MCR UAV) which
can be catapulted and then hover at a long distance from the launching site, see Figure 1, with the objective of
reducing the energy consumption, increasing the range, and developing surveillance missions.
This prototype is developed at LAFMIA CNRS-CINVESTAV Mexico in collaboration with the Institut Franco-
Allemand de Recherches de Saint-Louis (ISL), Laboratoire Heudiasyc (Université de Technologie de Compiègne,
France) and Centre de Recherche en Automatique de Nancy (CRAN-Nancy Université, France). The paper is or-
ganized as follows: section 2 introduces a functional description and the aerodynamic model of the vehicle. In
section 3 the control technique based on the classical backstepping methodology and the stability analysis are
presented. Simulation results of the closed-loop system and experimental results are shown in section 4 and
section 5 respectively. Finally, section 6 gives the conclusions and future works of this project.
Third task:
Hover mode
Rotor deployment

Second task:
Fin/aileron deployment

Fourth task:
Cruise mode

First task:
Launching

Camera
Observation

Five task:
Launching device
Landing in a safe area

Figure 1: Objective of the project.

2 MINI COAXIAL ROCKET-HELICOPTER

In this section, the functional description of the vehicle is described, and the complete aerodynamic model is
obtained using the classical Newton-Euler equations.

ISBN-xxx-y-zz-yyyyyyy-x 212
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2.1 Description

The mini coaxial rocket-helicopter (MCR UAV) is based on a couple of counter rotating brushless motors, and
its main characteristic of this vehicle is that it is capable of being launched through an external dispositive (e.g. a
rocket launching device, an air launcher, etc), and transforming itself into a coaxial helicopter at a long distance
from the launching site. Once the vehicle reaches the objective (a place, a building, an uninhabited area, etc.),
it performs hover flight, and can inspect the environment acquiring and transmitting information through a RF
camera to a ground station. Concerning the functional description, the MCR vehicle possesses aerodynamic
control surfaces (ailerons) which are manipulated to control the roll and pitch motion while the difference of the
velocities of the two motors regulates the yaw motion.

2.2 Mathematical equations

Consider an inertial fixed frame and a body frame fixed attached to the center of gravity of the helicopter denoted
by I={xI , yI , zI } and B={xB , yB , zB }, respectively, see Figure 2, [13].

zB
ψ
θ
B yB
φ
xB

zI
yI

I xI

Figure 2: MCR UAV.

Assume the generalized coordinates of the mini UAV as q = (x, y, z, ψ, θ, ϕ)T ∈ R6 , where ξ = (x, y, z)T ∈
R3 represents the translation coordinates relative to the inertial frame, and η = (ψ, θ, ϕ)T ∈ R3 describes the
vector of the three Euler angles with rotations around z, y, x axes. These angles ψ, θ, and ϕ are called yaw,
pitch and roll, respectively. Assume the translational velocity and the angular velocity in the body frame as
T
ν = (u, v, w)T ∈ R3 and Ω = (p, q, r) ∈ R3 , respectively.
The Newton-Euler equations of motion for a rigid object provide the dynamic model for the MCR UAV and
are given by the following expressions [13], [14]

ξ̇ = V
mV̇ = RF (1)
Ṙ = RΩ̂

I Ω̇ = −Ω × IΩ + Γ (2)

where F ∈ R3 and Γ ∈ R3 are the total force and torque acting on the vehicle, respectively. V = (ẋ, ẏ, ż)T ∈ R3
is the translational velocity in the inertial frame, m ∈ R denotes the mass of the MCR UAV, I ∈ R3×3 contains

ISBN-xxx-y-zz-yyyyyyy-x 213
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

the moments of inertia of the mini helicopter, and Ω̂ is a skew-symmetric matrix such that Ω̂a = Ω × a. Thus,
R represents the transformation matrix from the body frame to the inertial frame
 
c c sϕ sθ cψ − cϕ sψ cϕ sθ cψ + sϕ sψ
 θ ψ 
 
R =  cθ sψ sϕ sθ sψ + cϕ cψ cϕ sθ sψ − sϕ cψ 
 
−sθ sϕ cθ cϕ cθ

where the shorthand notation of sa = sin(a), ca = cos(a), and ta = tan(a) is used. For this matrix, the order
of the rotations is considered as yaw, pitch and roll (ψ, θ, ϕ) [6].
Tc Tc

zB zB

xB Y L yB
yB xB

D D
lr lr
δr δp

Lr Lp

(a) (b)

Figure 3: Schematic of the vehicle.

2.2.1 Forces

The forces acting on the vehicle include those of the propulsion system Fp , aerodynamic effects Fa and weight
Fw [8]. These forces are described as following
 
Fx
 
 
F =  Fy  = Fp + Fa + Fw
 
Fz

Propulsion forces The thrust force is generated by the two motors and is described as
 
0
 
 
Fp =  0 
 
Tc

where Tc is the thrust force of the two motors (Tc = T1 + T2 ). In this analysis, the thrust force is oriented parallel
to the axis zB of the body frame, see Figure 3.
Aerodynamic forces For the aerodynamic analysis, two rotation matrices B and W are used [14]. These
rotation matrices represent the transformation of the aerodynamic forces from the body frame to stability frame
B, and the stability frame to wind frame W , and are written as
   
cα 0 sα c sβ 0
   β 
   
B= 0 1 0  , W = −sβ cβ 0
   
−sα 0 cα 0 0 1

ISBN-xxx-y-zz-yyyyyyy-x 214
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

where α is the angle of attack and β are the sideslip angle. Thus, the aerodynamic forces in the body frame are
written as  
−L
 
T T  
Fa = B W  Y 
 
−D
where L, Y , and D are the aerodynamic forces: lift, side force, and drag, respectively.
Gravitational force The force due to the weight of the vehicle is described as
 
0
 
 
Fw = RT  0 
 
−mg

where g is the acceleration due to gravity.

2.2.2 Moments

The moments acting on the mini helicopter are due to actuators (actuator moment Γact , and gyroscopic moment
Γgyro ), and the aerodynamic effects Γa [8]. These moments are obtained as following
 
Γ
 L 
 
Γ =  ΓM  = Γact + Γgyro + Γa
 
ΓN

with    
τϕ q(Ir1 ωr1 − Ir2 ωr2 )
   
   
Γact =  τθ  , Γgyro =  p(−Ir1 ωr1 + Ir2 ωr2 ) ,
   
τψ 0
 
L̄
 
 
Γa =  M̄ 
 
N̄
where τϕ = ℓr Lr , τθ = ℓr Lp and τψ = τM1 + τM2 are the control inputs with ℓr that represents the distance
from the center of mass to the forces Lr and Lp , see Figure 3. τM1 and τM2 denote the moments produced by the
motors M1 and M2 . ωr1 and ωr2 denote the angular velocities of the rotors, Ir1 and Ir2 are the inertia moments
of the propellers. L̄, M̄ and N̄ are the aerodynamic rolling, pitching and yawing moments, respectively [3], [5].

2.2.3 Aerodynamic analysis in hover flight

Since the stability, control and dynamics of the MCR vehicle are affected by the propulsion system, the wind
produced by the propellers (Propeller Momentum Theory) is analyzed to obtain the behavior of the vehicle in
hover mode. Figure 4 shows the vehicle submerged in the propeller slipstream [9], [10], [11].
The thrust equation is described as

Tc = ṁa ∆v = ρAvs (vds − vo ) (3)

ISBN-xxx-y-zz-yyyyyyy-x 215
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Tc

vo

vi

vs = vo + vi

vds = vo + vi

Figure 4: Vehicle submerged in the propeller slipstream.

with
vs = vo + vi
vds = vo + 2vi (4)
vr = 2vi
where A represents the area of the rotor disc, ṁa is the mass flow rate, ρ denotes the air density, vo is the
freestream velocity, vi represents the induced velocity and is directed opposite to the thrust, vs is the slipstream
velocity, vds denotes the far downstream velocity, and finally, vr is the resultant velocity in the propeller slip-
stream.
Considering the hover case (α = 0, β = 0 and vo = 0), equations (3)-(4) give the induced velocity vh in
hover flight as √
Tc
vi = vh = (5)
2ρA
Finally, the aerodynamic forces and moments are written as

1 2 1 2
D = 2 ρvr SCD L̄ = 2 ρvr SbCl

Y = 1 2
2 ρvr SCY M̄ = 1 2
2 ρvr Sc̄Cm
(6)
1 2 1 2
L = 2 ρvr SCL N̄ = 2 ρvr SbCn

where S represents the fin-aileron area, c̄ is the fin-aileron chord, and b is the fin-aileron span. CD , CY and CL
are the aerodynamical non-dimensional coefficients of drag, sideforce and lift. Cl , Cm and Cn are aerodynamical
non-dimensional coefficients of the aerodynamic rolling, pitching and yawing moments.

ISBN-xxx-y-zz-yyyyyyy-x 216
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2.2.4 Equations of motion

The nonlinear model obtained by the Newton-Euler formulation is described as

Ax Ay Az
ẍ = m cθ cψ + m (sϕ sθ cψ − cϕ sψ ) + m (cϕ sθ cψ + sϕ sψ )
Ax Ay Az
ÿ = m cθ sψ + m (sϕ sθ sψ + cϕ cψ ) + m (cϕ sθ sψ − sϕ cψ )
−Ax Ay Az
z̈ = m sθ + m sϕ cθ + m cϕ cθ − g
θ̇ ψ̇ θ̇ ϕ̇sθ 1
ϕ̈ = cθ + cθ + Ixx [ΓL + qr(Iyy − Izz )]
cϕ sθ
+ cθ Izz [ΓN + pq(Ixx − Iyy )]
s s
(7)
+ cθϕIyyθ [ΓM − pr(Ixx − Izz )]
cϕ
θ̈ = −ϕ̇ψ̇cθ + Iyy [ΓM − pr(Ixx − Izz )]
ϕ s
+ Izz [−ΓN − pq(Ixx − Iyy )]
θ̇ ϕ̇ θ̇ ψ̇sθ cϕ
ψ̈ = cθ + cθ + cθ Izz [ΓN + pq(Ixx − Iyy )]
s
+ cθ Iϕyy [ΓM − pr(Ixx − Izz )]

where

Ax = −L
Ay = Y
A z = Tc − D
ΓL = τφ + q(Ir1 ωr1 − Ir2 ωr2 ) + L̄
ΓM = τθ + p(−Ir1 ωr1 + Ir2 ωr2 ) + M̄
ΓN = τψ + N̄

3 CONTROL TECHNIQUE
In this section, the control technique for the stabilization of the vehicle in hovering flight is presented using a
classical backstepping procedure [2], [7].
For simplicity, the equations of motion (7) are separated into three subsystems. One subsystem describes
the longitudinal motion (pitching, and translation in the x − z plane), the second subsystem describes the lateral
motion of the vehicle (rolling, and translation in the y − z plane), and the last subsystem describes the directional
motion (yawing, and translation in the x − y plane), [14]. Since the yaw motion is mechanically stable using
contra-rotating propellers, the gyroscopic moment Γgyro will essentially be zero.
Considering the condition ϕ = 0, and ψ = 0, the longitudinal subsystem is written as

Ax Az
ẍ = m cθ + m sθ

z̈ = − Amx sθ + Amz cθ −g (8)

ΓM
θ̈ = Iyy

where Ax = −L, Az = Tc − D and ΓM = τθ + M̄ .

Defining a change of variables
τθ M̄
uθ = + (9)
Iyy Iyy

ISBN-xxx-y-zz-yyyyyyy-x 217
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Substituting, it yields
L 1 1
ẍ = − m cθ + m Tc sθ − m Dsθ

z̈ = 1
m Tc cθ − 1
m Dcθ + L
m sθ −g (10)
θ̈ = uθ
In order to stabilize the altitude of this vehicle, a linear control law is proposed as
( )
m D L
Tc = −kz1 (z − zd ) − kz2 ż + cθ − sθ + g (11)
cθ m m

it results
L L
ẍ = −m cθ − m tθ sθ + gtθ
z̈ = −kz1 (z − zd ) − kz2 ż (12)
θ̈ = uθ

Taking a change of variables x1 = x; x2 = ẋ; x3 = θ; x4 = θ̇, the state space representation is written as follows

ẋ1 = x2
L L
ẋ2 = −m cx3 − m tx3 sx3 + gtx3
(13)
ẋ3 = θ
ẋ4 = uθ

In order to control the longitudinal subsystem (13), a nonlinear control law based on the backstepping procedure
is proposed. First, let us define the error e1 as

e1 = x1 − xd1 (14)

differentiating (14)
ė1 = ẋ1 − ẋd1 (15)

defining the following positive function

k1 2
V1 = e (16)
2 1
whose derivative is
( )
V̇1 = k1 e1 ė1 = k1 e1 ẋ1 − ẋd1 (17)
( )
and using ė1 = x2 − xd2 , it yields
( )
V̇1 = k1 e1 x2 − xd2 (18)

Now, let us define xv2 as the virtual control input, such that

xv2 = xd2 − e1 (19)

then, substituting (19) in (18), it implies

V̇1 = −k1 e21 + k1 e1 (x2 − xv2 ) (20)

Now, the error e2 is defined as

e2 = x2 − xv2 (21)

ISBN-xxx-y-zz-yyyyyyy-x 218
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

it results
V̇1 = −k1 e21 + k1 e1 e2 (22)

Now, the following positive function is proposed as

k2 2
V2 = e (23)
2 2

taking its derivative as

( { } )
L L
V̇2 = k2 e2 ė2 = k2 e2 (ẋ2 − ẋv2 ) = k2 e2 − cx3 + tx3 sx3 − gtx3 − ẋv2 (24)
m m
{L }v
Defining the virtual control input as m cx3 + m tx3 sx3 − gtx3
L

{ }v
L L k1
cx3 + tx3 sx3 − gtx3 = δ1v = −ẋv2 + e1 + e2 (25)
m m k2

and V̇2 yields ( { })

L L
V̇2 = −k2 e22 − k1 e1 e2 + k2 e2 δ1v − cx3 + tx3 sx3 − gtx3 (26)
m m
Defining the error e3 { }
L L
e3 = δ1v − cx + tx sx − gtx3 (27)
m 3 m 3 3
then, V̇2 yields
V̇2 = −k2 e22 − k1 e1 e2 + k2 e2 e3 (28)

Now, proposing a positive function V3 as

k3 2
V3 = e (29)
2 3
whose derivative is given as
[ ( ) ]
L L ( )
V̇3 = k3 e3 δ̇1v − tx3 cx3 + sx3 t2x3 − g 1 + t2x3 x4 (30)
m m
(( L ( )) )v
Let us define the virtual control input as m tx3 cx3 + m L
sx3 t2x3 − g 1 + t2x3 x4 such that
(( ) )v
L L ( ) k2
tx3 cx3 + sx3 t2x3 − g 1 + t2x3 x4 = δ2v = δ̇1v + e2 + e3 (31)
m m k3

then, V̇ yields
[ ( ) ]
L L ( )
V̇3 = −k3 e23 − k2 e2 e3 + k3 e3 δ2v − tx3 cx3 + sx3 t2x3 − g 1 + t2x3 x4 (32)
m m

finally, defining the error e4 as

( )
L L ( )
e4 = δ2v − tx3 cx3 + sx3 t2x3 − g 1 + t2x3 x4 (33)
m m

V̇3 yields
V̇3 = −k3 e23 − k2 e2 e3 + k3 e3 e4 (34)

Proposing the last positive function as

k4 2
V4 = e (35)
2 4

ISBN-xxx-y-zz-yyyyyyy-x 219
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

it results
( ( ))
V̇4 = k4 e4 ė4 = k4 e4 δ̇2v − d1 uθ + d2 x24 (36)

where

L L ( )
d1 = tx cx + sx t2 − g 1 + t2x3
m 3 3 m 3 x3
L L 2L 2L
d2 = tx3 sx3 + cx + cx t2 + sx t2 − 2gtx3 − 2gt3x3
m m 3 m 3 x3 m 3 x3

From (36), a control input uθ is proposed such that V̇4 = −k4 e24 − k3 e4 e3 , it results
( )
1 k3
uθ = δ̇2v + e3 + e4 − h2 x24 (37)
h1 k4

where

L L ( )
h1 = tx cx + sx3 t2x3 − g 1 + t2x3
m 3 3 m
L L 2L 2L
h2 = tx3 sx3 + cx + c x t2 + sx t2 − 2gtx3 − 2gt3x3
m m 3 m 3 x3 m 3 x3

now, a Lyapunov function is proposed as

V = V1 + V2 + V3 + V4 (38)

and V̇ yields
V̇ = −k1 e21 − k2 e22 − k3 e23 − k4 e24 ≤ 0 (39)

Finally, the system (8) is stable in the origin [7].

In order to control the lateral subsystem, the control methodology presented to stabilize the longitudinal
subsystem is employed.
Ay Az
ÿ = m cϕ − m sϕ
(40)
ΓL
ϕ̈ = Ixx

where Ay = Y , Az = Tc − D and ΓL = τϕ + L̄.

On the other hand, in order to stabilize the remaining directional subsystem, a linear control input is proposed
as
τψ = Izz (−kψ1 ψ − kψ2 ψ̇) − N̄ (41)

which is substituted in
τψ + N̄
ψ̈ = (42)
Izz
and it yields
ψ̈ = −kψ1 ψ − kψ2 ψ̇ (43)

Finally, the constants kψ1 , kψ2 are chosen such that the equation (43) is Hurwitz polynomial.

10

ISBN-xxx-y-zz-yyyyyyy-x 220
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

4 SIMULATION
In this section, the simulation results of the longitudinal subsystem are shown. The closed-loop system is validated
using the Matlab Simulink with the initial conditions x = 0.05 m, z = 0.1 m and θ = 0.1 rad and altitude desired
z = 1.5 m. Figure 5 illustrates that the position response is stabilized, and the desired altitude is reached. The
pitch angle and control law are shown in the Figure 6 in which the response of the pitch angle remains at zero.

0.5 2

1.5

1
Position x [m]

Position z [m]
0.5

0 0

−0.5

−1

−1.5

−0.5 −2
0 50 100 150 200 0 2 4 6 8 10
time (s) time (s)

(a) (b)

Figure 5: Position x and z.

0.2 30

0.15
20
0.1
10
Angle θ [rad]

0.05
uθ [Nm]

0 0

−0.05
−10
−0.1
−20
−0.15

−0.2 −30
0 1 2 3 4 5 0 1 2 3 4 5
time (s) time [s]

(a) (b)

Figure 6: Angle θ and control input uθ .

5 PRACTICAL RESULTS
This section presents the embedded system, and the practical results obtained in real-time during the hover flight.
The airframe of the MCR UAV is built of carbon fibre and polystyrene foam sheet (Depron). This vehicle is
powered by two brushless motors in coaxial configuration and the size of the propellers is 7×5 in, see Figure 7 (a).
The weight of the vehicle is 450 gr. The autopilot consists of a Digital Signal Controller (DSC TMS320F2808),
an inertial measurement unit (Microstrain 3DM-GX3-25 AHRS) and a communication device (XBee-PRO RF
Module), see Figure 7 (b). The microcontroller (DSC) processes the information from sensors, computes the
control law, and then sends signals to actuators. The IMU provides the angles, and angular velocities. In order to
observe the performance of the vehicle in real-time during the autonomous flight, a telemetry system is developed
using the software LabVIEW and GSM ActiveX controls which illustrate the states variables such as attitude and
position.
In this stage of the project, the aerial vehicle is stabilized in hover flight. Figure 8 shows the performance in
roll motion of the vehicle during the autonomous flight. The pitch angle and control are depicted in the Figure 9,

11

ISBN-xxx-y-zz-yyyyyyy-x 221
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(a) (b)

Figure 7: (a) Mini coaxial rocket-helicopter and (b) Autopilot.

and the Figure 10 illustrates the behavior of the yaw angle and control of this aerodynamic platform.

4 1.5

3
1

Roll control [µC signal]

2
Roll angle [deg]

0.5
1

0 0

−1
−0.5
−2
−1
−3

−4 −1.5
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
time [s] time [s]

(a) (b)

Figure 8: Roll angle and control.

3 2
Pitch control [µC signal]

2
1
Pitch angle [deg]

0 0

−1
−1
−2

−3 −2
−4
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
time [s] time [s]

(a) (b)

Figure 9: Pitch angle and control.

1 1.5

1
Yaw control [µC signal]

0.5
Yaw angle [deg]

0.5

0 0

−0.5
−0.5
−1

−1 −1.5
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
time [s] time [s]

(a) (b)

Figure 10: Yaw angle and control.

6 CONCLUSIONS
In this paper the hover flight of proposed aerodynamic prototype (called Mini Coaxial Rocket-Helicopter MCR)
has been proved. For this purpose, the complete aerodynamic flight model of the vehicle, including the wind

12

ISBN-xxx-y-zz-yyyyyyy-x 222
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

effects, is obtained using the classical Newton-Euler equations. The stability of the vehicle in hover mode has
been analyzed in order to validate the classical control technique based on the backstepping methodology. An
embedded system has been developed to test the vehicle in real-time. Finally, simulation results which highlight
the proposed nonlinear control law in closed-loop system and the experimental results obtained in real-time during
the hover flight shown an effective performance of the developed system.
The next stage of this project is to design and develop the device for launching the vehicle, to incorporate
additional instrumentation including GPS navigation, velocity sensors and vision system for observation.

REFERENCES
[1] S.Bouabdallah, R.Siegwart, and G. Caprari “Design and Control of an Indoor Coaxial Helicopter”, Inter-
national Conference on Intelligent Robots and Systems 2006, Beijing, China, 2006.

[2] P. Castillo, R. Lozano, and A. Dzul, Modelling and Control of Miniflying Machines, Springer-Verlag, Lon-
don, July 2005.

[3] M. V. Cook, Flight Dynamics Principles, Butterworth-Heinemann, second edition, USA, 2007.

[4] A. Dzul, T. Hamel, and R Lozano, “Modeling and Nonlinear Control for a Coaxial Helicopter”, IEEE
International Conference on Systems, Man and Cybernetics 2002, Hammamet, Tunisia, 2002.

[5] B. Etkin, and L. Duff Reid, Dynamics of Flight: Stability and Control, John Wiley and Sons, 3rd Ed., 1996.

[6] H. Goldstein, C.P. Poole and J.L Safko, Classical Mechanics, Adison-Wesley, USA, 2nd Ed. , 1983.

[7] H. Khalil, Nonlinear Systems, Prentice Hall, New York, 1995.

[8] R. Lozano, Unmanned Aerial Vehicles Embedded Control, John Wiley-ISTE Ltd, 2010.

[9] B.W. McCormick, Aerodynamics of V/STOL Flight, Dover Publications, 1967, USA.

[10] B.W. McCormick, Aerodynamics, Aeronautics and Flight Mechanics, John Wiley and Sons, USA, 1995.

[11] W.F. Phillips, Mechanics of flight, John Wiley and Sons, USA, 2004.

[12] D. Schafroth, C.Bermes, S.Bouabdallah, and R.Siegwart, “Modeling, system identification and robust con-
trol of a coaxial micro helicopter”, Control Engineering Practice, Elsevier, 2010.

[13] R. F. Stengel, Flight Dynamics, Princeton University Press, USA, 2004.

[14] B. L. Stevens and F. L. Lewis, Aircraft control and simulation, John Wiley and Sons, USA, 1992.

13

ISBN-xxx-y-zz-yyyyyyy-x 223
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Tools for UAV operations

P. Dias E. Marques R. Martins R. Bencatel José Pinto

and J.B. Sousa

Department of Electrical and Computer Engineering, University of Porto, Porto, Portugal

Email: pdias, bencatel, zepinto, [email protected],[email protected],[email protected]

Abstract

In this paper, we present various tools for safe UAV research missions. Most of the tools available
from off-the-shelf autopilots are not adequate to carryout research activities and have restricted
modification capability. We have developed a new command and control system (Neptus) that is
flexible in terms of planning and situational awareness, messaging system that has interoperability
capability (IMC) and an on-board software system (Dune) that interacts with various sensors and
can bridge an off-the-shelf autopilot with Neptus for operator in loop interactions. These tools have
been tested for various mission lasting from minutes to hours and also under multi-UAV operations.

keywords: UAV, command and control, communication

1 Introduction
UAVs are used in many military and civilian operations ranging from surveillance [1, 2] to agriculture
[3, 4]. Operating these UAVs requires an autopilot to control the aircraft, communication capability
for modification of plan and situational awareness, and command and control system for managing the
operations at the ground station with operator in the loop. Most of the commercial autopilots like Piccolo
[5], Micropilot [6], Kestrel [7], have proprietary softwares to handle the UAV operations. Currently, they
allow modification of waypoints, command velocity and heading using the ground station. However,
these softwares have restricted capability to modify the operations which is a necessary component in
research. Also, they do not have interoperability capability which is desired in multi-vehicle operations.
In order to have interoperability across different systems, we have developed several tools to govern
the UAV operations. First, we developed the on-board software Dune that takes care of the on-board
operations that include autopilot, system health monitoring, logging and has delay tolerant networking
capability. We developed an Inter-Messaging Protocol (IMC) that uses XML to send and receive mes-
sages [8]. The messaging system is compatible with STANAG 4586 [9]. Additionally, IMC allows low
level control commands like JAUS (Joint Architecture for Unmanned Systems)/SAE AS-4 [10] and CCL
(Compact Control Language) [11] that provide a set of commands (messages) to control the vehicle.

1
ISBN-xxx-y-zz-yyyyyyy-x 224
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

For command and control, we developed a software called Neptus [12] that has advanced capabilities
compared to the proprietary ground station softwares in terms of planning the mission, review and
analysis of mission, reporting and heterogeneous vehicle operations. In the following section, we provide
brief description of these three systems.

2 IMC Overview
The IMC protocol comprises different logical message groups for UAV operations. It defines an infras-
tructure that is modular and provides different layers for control and sensing. The use of IMC in the
context of message flow between subsystems is shown in Figure 1. The message flow corresponds to the
several control and sensing layers within IMC:

1. Mission control messages define the specification of a mission and it’s life-cycle, for the interface
between Neptus console and a mission supervisor module. A mission is a sequence or graph of
maneuvers.

2. Vehicle control messages are used to interface the vehicle from an external source, typically from
Neptus console or a mission supervisor module, for example to issue maneuver commands or other
external requests, and to monitor the vehicle state.

3. Maneuver messages are used to define maneuvers, associated commands and execution state. The
simplest maneuver types are related to waypoint tracking - encoded through a Goto message - or
loitering patterns - Loiter.

4. Guidance messages are related to the guidance commands used for maneuvering. A guidance
command generates new references for the vehicle heading, altitude, and velocity, in the form of a
Desired Guidance message.

5. Navigation messages define the interface for reporting a vehicle navigation state. The Estimated
State message defines a vehicle navigational state.

6. Sensor messages are used to report sensor readings by the respective hardware controllers. some
of the sensor messages are related to IMU, GPS, servos, motors, batteries, etc.

7. Actuator messages specify the interface with hardware actuator controllers. The actuators are
servos and propeller motor.

This layered control and sensing approach enables modular development of applications. Software com-
ponents can run in logical isolation, interfacing with other modules merely through the exchange of
IMC messages. The control infrastructure for autonomous vehicles implemented on-board, using the
DUNE framework, that enables message exchange using a message bus abstraction, and provides trans-
port mechanisms for external communications. Vehicles can be monitored and controlled externally
using Neptus consoles. Networking of vehicles and consoles, is enabled through traditional IP-based

2
ISBN-xxx-y-zz-yyyyyyy-x 225
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 1: Message flow in IMC for different systems and their interaction

communication-mechanisms, like raw UDP or TCP sockets, or using ubiquitous GSM/GPRS commu-
nications.

3 Dune
To abstract vehicle hardware and specifications, we use the DUNE on-board software. DUNE is com-
posed of a set of asynchronous tasks that communicate between each other using the IMC protocol.
Some of these tasks may be coupled to a sensor (publishing messages) or actuator (consuming mes-
sages) or may be also implementing a high-level navigation or control algorithm. All vehicles run the
same software only with an adapted configuration which states the tasks that should execute (possibly
multiple instances) and also configuration for individual tasks.

4 Command and control system

We built the Neptus command and control infrastructure in a modular way which addresses most of the
mission life cycle requirements. The life cycle of a mission is divided into planning, executing, review and
analysis and dissemination. Despite the reuse of several modules, each phase has a dedicated graphical
interface focusing on what tasks are more important in that particular phase.

4.1 Mission Planning

The planning of a mission encompasses the definition of a virtual representation of the mission site,
choosing the area for operation, and creating mission plans that the vehicle(s) will execute autonomously.
Neptus includes the Mission Planner application, which is exclusively targeted for these tasks. In the
Neptus Mission Planner, the map editor interface allows the definition of world maps in both 2D and
3D visualizations. Maps are composed of several geo-referenced elements: marks to denote points of

3
ISBN-xxx-y-zz-yyyyyyy-x 226
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 2: Neptus Operational Console

interest, geometric figures, surface images (with optional elevation map), generic 3D models, and paths
(composed of lines). Neptus mission plans are defined as a set of maneuvers that are connected through
conditional transitions, forming a finite state machine or (if one considers the continuous dynamics
of the maneuver controllers) a hybrid automaton. Mission plans can be edited directly as its graph
representation or, for usability purposes, they can be drawn directly on top of the map previously
created. As can be seen in Figure 2, to insert a new maneuver at a plan, the user may click on the
map where he wants to position the maneuver and a popup menu will appear displaying the possible
choices of maneuvers available for that particular vehicle. After a maneuver is added to the mission, it
can be dragged to other locations in the map or its parameters can also be set by double clicking the
maneuver. For increased usability, the user may also set parameters for all maneuvers at a time (when
there are shared parameter types) and translate/rotate the plan in its entirety.
Due to the numerous field-tests made with Neptus, we noticed that some of the plans carried out by
autonomous vehicles are very similar (sanity tests, regression tests, area scanning, etc). As a result, we
added the possibility of defining mission templates through a scripting language based on JavaScript.
This scripting language allows the definition of template parameters and the generation of plans. XML
(eXtensible Markup Language, VI) is the format of choice to store data in Neptus. For convenience (file
size, portability) all maps and mission plans are stored as a single Zip file.

4.2 Mission Execution

The Neptus Console application is the main interface for vehicle interaction and real-time monitoring, as
depicted in Figure 2. This application allows the establishment of connections to one or more vehicles,
display incoming data through interface plugins and control vehicles behavior by sending messages (also
through visual plugins).
In order to mitigate different communication interfaces and protocols we constructed the data feed
to Neptus in a flexible way. The Communications Manager is responsible for the I/O to and from

4
ISBN-xxx-y-zz-yyyyyyy-x 227
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

the systems. In terms of communication protocols, we set UDP to be the default data I/O, but NDDS
(Network Data Delivery System), GSM/GPRS, acoustics, TCP, and HTTP are also available with DTN
(Delay-Tolerant Networks) underway. The IMC message protocol is the main data-entry protocol used
in Neptus, but to be more flexible we define a XML Schema (XSD, VI) in which IMC can be defined.
The resulting XML is then dynamically loaded in Neptus making IMC messages known. The same can
be done for other message protocols like STANAG 4586 [9]. The complete data received is then available
in a shared data environment and then the various console components can subscribe to this data and
update themselves accordingly.
Console components are visual (or hidden) elements that display system data and can be added to
Neptus under the form of software plugins. To further improve flexibility, the console layout can be edited
allowing adding, removing, and dragging of any console components. Components have parameters that,
for instance, allow the user to select which variable to display or which vehicle to monitor. Moreover,
the consoles provide an alarm logic to which visual components can deliver events. The alarm logic
logs all events and warns the user in case any fault has been reported changing the error state of the
console. Warnings can be reported as popup messages or spoken phrases. The console layout (plugin
locations and respective parameters) can be stored in an XML file and the resulting console can then
be associated to a vehicle class.
Neptus design supports concurrent operations. Vehicles, operators, and operator consoles can be
dynamic. Operators are able to plan and supervise missions concurrently. Additional consoles can be
built and installed on the fly to display mission related data over a network.

4.3 Mission Review and Analysis

After a mission is completed, operators are interested in accessing and revising the vehicles data. For
this, there exists the Neptus Mission Review & Analysis (MRA) application (Figure 3). Again, recurring
to a plugin architecture, this software application provides various visualizations over a set of log files
produced in a single mission. By default, there are a number of predefined visualizations that are
automatically generated upon opening a mission log. However, the user may also select a number of
fields to inspect and generate new plots with a couple of mouse clicks.
Another useful visualization is the mission replay of the logged data. This comes in two flavors: the
first one a simple mode which displays the evolution of the estimated state of the vehicle in 2D or 3D;
and the other consists of taking all the logged messages and feeding them to the Neptus communications
interface. As a result, by opening the correct mission in an operational console, it is possible to follow
the evolution of the mission as the messages were being logged in the vehicle. The replay visualizations
give access to the mission timeline and it is possible to pause, advance, rewind, and speed up or slow
down the replay. MRA also allows the generation of a mission execution report with the most common
plots and statistics in the form of a PDF document. Another output of this tool is a partial automated
elevation map generation for map integration.

5
ISBN-xxx-y-zz-yyyyyyy-x 228
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 3: Neptus MRA

4.4 Disseminating Results

Dissemination plays an important role in our infrastructure because it allows external applications to
access real-time data feeds of our sensors and also to integrate external data in our system. We use
mostly web technologies (HTTP, REST, ATOM, XML/XSLT) for dissemination. The implementation
involved embedding Mortbays Jetty application server (compatible with the Servlet 2.0 API) in Neptus.
We developed various entry points for this server, both for data publication and querying. Different
data access modes (file types, queries) correspond to different URLs. For instance, there exists URLs
that display data in KML format (allowing visualization in Google Earth (see Figure 2)), and other
URLs that generate plots, HTML tables, etc.

5 Integrating Piccolo with Neptus/DUNE

The commercial autopilots and associated ground station softwares have been used over several years
for basic applications. The users are confident over these systems for many applications that require
simple waypoint and loiter maneuvers only. While carrying out the mission, it is difficult to completely
remove the autopilot system from the loop. We developed an interface DUNE with IMC and Neptus
sub-systems with the existing Piccolo autopilot system and its ground station.
A first implementation of the Piccolo controller in Neptus and DUNE was made on top of Piccolo
waypoints interface. To allow a smooth transition for both the operating team and the newly imple-
mented interface, a protocol was drafted in order to allow handover of the UAVs between PCC and
Neptus operators. This protocol is a contract between Neptus/DUNE and the PCC (Piccolo Command
Center) operator so none of the parts touch each other designated waypoints.
The 100 available waypoints are divided in 3 groups: the handover waypoints; the plan waypoints;
and all the remaining (designated as external). The handover waypoints serve, as the name indicates, to
handover the vehicle between PCC and Neptus operators. The PCC operator creates here a safe piccolo

6
ISBN-xxx-y-zz-yyyyyyy-x 229
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

plan and he is responsible for maintaining these waypoints. The Neptus operator don’t change any of
these waypoints for a safe transition of UAV command. He only sends the UAV to one of the waypoints
and takes it out of them to take full control of the vehicle. The plan waypoints are to be used by the
Neptus/DUNE for the execution of plans and must not be touch by the PCC operator. As seen earlier,
these plans are defined in a different way than in Piccolo. So DUNE translates all executing plan to a
set of waypoints and send them to Piccolo. This also makes the PCC operator aware of what to expect
from the vehicle as it executes Neptus/DUNE plans. From this period, the interaction similar to that
of operating the vehicle without piccolo and its ground station. At any time the PCC can trigger any
other waypoint. At this point the execution of the plan is aborted and the PCC is responsible for the
vehicle. Any other waypoint is considered external and of full PCC control.

6 Conclusions and future work

We presented tools that can assist UAV operations in a safe and efficient way. The IMC allows easier
integration of various standard autopilots and different communication modes. Neptus allows 2D and
3D visualization of the world and has all the components to conduct a full mission cycle. Dune allows
monitoring and execution of various tasks and is open source software. Further, we plan to integrate some
of the AI techniques into the mission planning stage for better mission design and also for situational
awareness.

REFERENCES
[1] E. Semsch, M. Jakob, D. Pavlicek, and Michal Pechoucek: Autonomous UAV surveillance in com-
plex urban environments, In Proc. of the IEEE/WIC/ACM International Joint Conference on
Web Intelligence and Intelligent Agent Technology,Washington, DC, 2009, pp. 82-85.

[2] B. Bethke, J.P. How, and J. Vian: Group health management of UAV teams with applications to
persistent surveillance,Proc. of the American Control Conference, Jun 2008, Seattle, WA, pp.3145-
3150.

[3] S.R. Herwitz, L.F. Johnson, S.E. Dunagan, R.G. Higgins, D.V. Sullivan, J. Zheng, B.M. Lobitz, J.G.
Leung, B.A. Gallmeyer, M. Aoyagi, R.E. Slye and J.A. Brass: Imaging from an unmanned aerial
vehicle: agricultural surveillance and decision support, Computers and Electronics in Agriculture,
Vol. 44, No. 1, July 2004, pp. 49-61.

[4] A. Simpson, T. Stombaugh, L. Wells and J. Jacob : Imaging techniques and applications for
UAVs in agriculture, American Society of Agricultural and Biological Engineers Annual Meeting,
St. Joseph, Michigan, 2003, #031105.

[5] www.cloudcaptech.com

7
ISBN-xxx-y-zz-yyyyyyy-x 230
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[6] www.micropilot.com

[7] procerusuav.com

[8] Martins, R.; Dias, P.S.; Marques, E.R.B.; Pinto, J.; Sousa, J.B.; Pereira, F.L.; , ”IMC: A communi-
cation protocol for networked vehicles and sensors,” OCEANS 2009 - EUROPE , vol., no., pp.1-6,
11-14 May 2009.

[9] STANAG 4586 Second Edition, Standard Interfaces of UAV Control System (UCS) for NATO UAV
Interoperability, NATO Standardization, November 2007.

[10] Open JAUS, www.openjaus.com.

[11] R. P. Stokey, L.E. Freitag, and M.D. Grund: A compact control language for AUV acoustic com-
munication,In proc. of the OCEANS Conference, Brest, FRANCE, JUN 2005, pp. 11331137.

[12] J. Pinto, P.S. Dias, G.M. Goncalves, R. Goncalves, E. Marques, J.B. Sousa, and F.L. Pereira:
Neptus a framework to support a mission life cycle, IFAC Conference on Manoeuvring and
Control of Marine Craft, Lisboa, Portugal, Sept 2006.

8
ISBN-xxx-y-zz-yyyyyyy-x 231
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Experience in Designing Autonomous Airplane

RÓBERT SCHOCHMANN, JAKUB PILKA, MICHAL KUSENDA,

ING. IVAN PESTÚN, ING. PETER VOLŠÍK, JÁN MALINA
Slovak University of Technology in Bratislava, Faculty of Electrical Engineering and Information
Technology, Ilkovičova 3, Bratislava 812 19, Slovak Republic
[email protected]
[email protected]
[email protected]
[email protected]
[email protected]
[email protected]
(website: www.animatechnika.com)

Abstract
This paper is about developement and manufacture of UAV system with further support systems. It solves
the optimal fuselage and wings profile with demanded flight options for developement of autopilot and
with adequate bur and versatility of flight (fast, and also slow flights), we solve the draft of ground station
for control and nonstop control of flying system, that means Accepting the flight informations such as roll,
pitch and yaw, relative and absolute altitude, speed, course, engine RPM, GPS position, actual
consumption of electric energy, ammount of energy in cells and so on. Also its optimalization from POV of
operator.
Later on, new features were added to the UAV functionalities such as offline or online camera systems
with additional features as gyroscopes for camera turning or more sensors to be able to get more input data
from surrounding environment. More features were added for other specific usage, such as
meteorologically specific sensors or army laser guide etc.
It describes issues with developement, such as sensor disturbance by motor vibration and applicated
solutions, but also a question of safety against the surrounding area, so as to the system itself (anticipating
of undesired behavior and minimalization of damage in case of crash).

Keywords:UAV designing, Autonomous plane, telemetry, realtime aplication, fly security

1. INTRODUCTION
This document illustrates our achievements gain so far in approximately 3 year experience. Our goal is to

-1-

ISBN-xxx-y-zz-yyyyyyy-x 232
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

design hardware and software architecture for a fully autonomous UAV capable of flying to its destination
to carry out some task (e.g. to gather visual information) and safely returning without the need for user
interference.

2. DESIGNING UAV PLANE

2.1. Main concept

Our requirements for UAV airframe were to have:

– Fuselage with interior big enough to carry all electronic and other equipment
– Plane easy to control during flight
– Stable flight characteristics
– Simple to build design
All this requirements matched concept of a tailless plane (see Fig. 1).

Figure 1 : Design of a tailless plane drawned in AutoCAD software.

Design of a tailless plane simplifies control, because it has only two control surfaces (ailerons), so the
whole flight of the airplane is controlled by two servos which also saves overall weight of the aircraft. The
fuselage itself has a round cross-section with diameter of 127mm, which gives us interior space big enough
for carrying the electronics and other necessary equipment for UAV. Other parts of the airframe to be
designed are nose cone and wings.

2.2. Designing nose cone of the airplane

Nose cone of the airplane closes front of the fuselage, so it is necessary to make it aerodynamically as
clean as possible. In designing this part of aircraft, inspiration was kept from rocketry engineering.
Fuselage of a rocket is also mainly rounded and has a nose cone, which can have different shapes.

-2-

ISBN-xxx-y-zz-yyyyyyy-x 233
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 2: Possible nose cone shapes with their drag coefficient (Cx): A: Cx=0.05, B: Cx=0.1, C: Cx=0.2,
D: Cx=0.2, E: Cx=0.34, F: Cx=0.9, G: Cx=1

A very common nose cone shape is a simple cone, which is very often chosen for its ease of manufacture,
but it is also often mis-chosen for its drag characteristics.
For our purposes, ogive shape was chosen because of it good drag characteristics in 1:3 (diameter to
lenght) ratio and easy design. In our case, the nose cone has diameter D=127mm (equals to fuselage
diameter) and length L=381mm to obtain 1:3 ratio.

Figure 3: The circle which forms the ogive with all parameters.

-3-

ISBN-xxx-y-zz-yyyyyyy-x 234
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Ogive shape is formed by a segment of a circle with radius (Ogive Radius) ρ and it is related to the length
and base radius of the nose cone as expressed by the formula:

R 2  L2
 (1)
2R

The radius y at any point x, as x varies from 0 to L is:

y   2  ( L  x) 2  R   (2)

A tangent ogive nose is often blunted by capping it with a segment of a sphere to acquire better drag
characteristics. Blunted nose is also more suitable for camera housing.

Figure 4: Spherically blunted tangent ogive.

The tangency point where the sphere meets the tangent ogive can be found from these equations:

x 0  L  (   rn ) 2  (   R) 2 (3)

rn (   R)
yt  (4)
  rn

2 2
xt  x0  rn  y t (5)

-4-

ISBN-xxx-y-zz-yyyyyyy-x 235
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Where rn is radius of the spherical nose cap, x0 is the center of the spherical nose cap and

xa  x0  rn (6)

is the apex point.

2.3. Designing wings

Airfoil
Airfoil of the wing used on the plane is a combination of RR310 and RR510 airfoil, which were designed
for tailless planes to perform maximum lift. These airfoils have also good speed range and lift/drag
characteristics [R/C Soaring Digest, Vol. 27, No.7, July 2010].

Figure 5: RR310 and RR510 airfoils.

RR510 is intended to be the root airfoil for the RR310 and has maximum lift coefficient cl=1.44 at angle of
attack α=14˚. RR310, the tip airfoil, has maximum lift coefficient cl=1.45 at angle of attack α=14˚. The
combination of this airfoils helps to keep the wings from stalling. Maximum lift / drag ratio is at cl=1.0 in
RR510 and at cl=1.1 in RR310 (angle of attack app. α=8.5˚) [R/C Soaring Digest, Vol. 27, No.7, July
2010].

-5-

ISBN-xxx-y-zz-yyyyyyy-x 236
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 6 : RR310 and RR510 airfoils.

Geometric parameters of a wing

Tailless planes and flying wings can have almost any airfoil, if sweep and twist angles are chosen
accordingly. However tailless plane with a wide operating range should have wings only with a small
amount of the twist, or none at all, to keep induced drag at reasonable levels.

Neutral Point and Stability

Stability is defined by the distance between center of gravity and neutral point, divided by the mean chord
of the wing. The center of the gravity must be located in front of the neutral point. The neutral point of an
unswept rectangular wing is approximately at the quarter of the chord line. The neutral point of the swept
wing must be calculated. Mean aerodynamic chord length of a tapered wing is defined by:

2 1    2
l  .l r (7)
3 1 

lt
 (8)
lr

Where lr is the root chord, lr is the tip chord and λ is the taper ratio.

-6-

ISBN-xxx-y-zz-yyyyyyy-x 237
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Spanwise location of the chord lµ using span b :

b l r  l
y (9)
2 l r  lt

The neutral point can be found by drawing line parallel to the fuselage center line at spanwise station y.
The chord at this station should be equal to lµ. The neutral point is approximately located at the quarter of
this chord line (see fig. 6).

Figure 7 : Geometric parameters of a tapered wing.

3. ELECTRONIC
The architecture is centred around the main 32-bit PIC (Peripheral Interface Controller) microcontroller.
Other 8-bit PICs whose purpose is to collect and process data from sensors and control servo motors are
connected to it. The reason for this architecture is distribute the A/D conversion time, whereas the main μC
can be used for other necessary calculations. The main advantage of this architecture is that the main μC
can work without having to wait for its peripheral circuits to finish a conversion from the analogue sensors.
Another advantage is the ease possibility of adding additional system components since the computing
power of the main μC is never fully used. We prepared our PCB (Printed circuit board) to be able to add
more electronic equipment if needed.
XBee RF communication modules compliant with IEEE 802.11 standards are used for wireless
communication with the model. The control station is composed of a laptop, joystick to control the aircraft
and of the XBee RF communication module.

-7-

ISBN-xxx-y-zz-yyyyyyy-x 238
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 8 : Hardware architecture

3.1. HARDWARE

The whole ,,brain” of the UAV is created by these PCB:

– Main board
– Servo board
– Sensor board
– Double Xbee board
– GPS module
The Main board, equipped by 32 bit PIC (peripheral interface controller, type 32MX775F512H) serves as a
data router (it is routing the data, which comes from other PCB boards) and directly sends them via Double
Xbee board, or process them for specific use, such as autopilot is. This board also sends commands to
peripheral boards, which are connected to it via USART (Universal Synchronous Asynchronous Receiver
Transmitter).
Servo board connected via USART is driven by one way communication from the main board. Using 8 bit
16F877A PIC, it is capable of controlling up to 5 servo motors directly and 9 servo motors isolated by
optocouple. Besides all, it can also switch 4 loads with current up to 0,5 A.
To ,,bring the environment” to the operator, Servo board sends data from attached sensors via USART.

-8-

ISBN-xxx-y-zz-yyyyyyy-x 239
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

There is also used 16F877A PIC. This board consists of one absolute pressure sensor for measuring altitude
of the UAV (using the value of absolute air pressure to calculate the altitude), one differencial pressure
sensor to calculate relative speed towards the wind it is flying in (Prandtle probe attached), two temperature
sensors for measuring temperature of electronics and the environment temperature (for calibrating sensors
calculations, frost alert and so on) and one ultrasonic sensor for landing aid (via the wide angle camera
operator can only guess the actual altitude while landing).
GPS module has its own USART interface, through which it sends data in string format. They contain
actual position (latitude) and altitude, which can be used in Ground station to display actual position on
map. The altitude calculated from pressure sensor can be inaccurate, so we compare it to that one from
GPS. This prevents the data from being misleading for further calculations.

3.2. Communication

Communication between ground station and the plane is performed via XBee RF modules at a 250 kilo
baud per sec. data rate. Communication runs on 2,4 GHz frequency. Each of the modules has its role in the
network. One acts as coordinator (ground station), the other as end-device (plane).The coordinator works in
a broadcast mode so it does not require acknowledgement messages from the end device since they are not
needed because minor data dropouts are acceptable. This further increases communication speed. Before
the network is established, the coordinator finds the most suitable channel and sets a PAN ID (Personal
Area Network) specific only to this network. The end device then scans for coordinator networks and
associates only with the one with a specific 64-bit address. This ensures no other coordinators interfere
with our communication.
Double Xbee board, as its name say, contains two Xbee modules. One is used for transmitting the data and
the other for receiving. This should speed up the communication and improve data consistency. The other
works on different channel.

4. UTILITY SOFTWARE OF THE GROUND STATION

The utility software gives the pilot all the necessary information he needs to pilot the plane. This includes a
video feed from the on-board camera, acceleration in X, Y and Z axes, roll and pitch indication is also
present. Also the joystick calibration will be available, as well as the fuel level indication and battery
charge status. Our goal is to make GUI as simple and as intuitive as possible, so that the pilot can fully
concentrate on controlling the aircraft.

-9-

ISBN-xxx-y-zz-yyyyyyy-x 240
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Requirements:
– Real time criteria
– Installation time before flight
– Power consumption (usability time)
– All in One Box Solution
– Performance
– System stability during execution time

4.1. Hardware

Figure 9 : Ground Station.

Box case
contains major hardware parts essential for remote control of UAV. The hardware was designed in
spotlight to mechanical security and power efficiency. The components were selected with
concern to maximum usability in shortest possible boot time.

Hardware components match requirements:

Motherboard
was chosen in consideration to hw security as less moving parts as possible, but with enough
calculation and bus force to support solution.

Onboard GPU

NVidia ION based chip is powerful HW for running OpenGL solution, but with low power needs.

- 10 -

ISBN-xxx-y-zz-yyyyyyy-x 241
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

SSD Hard Drive

is optimal solution in performance, power consumption and mechanical resistance.

Figure 10 : Hardware architecture of Ground Station.

All basic PC components fulfill basic requirements for hardware and are very good background for custom
hardware components and software solutions on top of hardware.

4.2. Software

To optimize speed and power consumption, the customized Bunt OS was chosen. All irrelevant services
were removed and system is optimized only to run on the Ground Station.

Performance of Ground Station application

Requirements:
– Clearly all time visible flying data
– Real-time response
– Camera view
– Human readable UI

- 11 -

ISBN-xxx-y-zz-yyyyyyy-x 242
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 11 : Screen of Ground Station.

All sensor data from plane are visualized via sets of texts and lines rendered via OpenGL technology. GUI
is made as simple as possible to keep all information clear and easily readable.
Camera View processing - As basic technology was used the OpenGL, the all data are collected via Video
Input Stream and transferred into Video Memory of the Graphic card. Then the Graphic card visualize
video in one OpenGL window. This technology is prepared for future improvements, such as Image
Processing and image analysis.
To increase robustness of Ground station, the application is designed as multithread application, where all
threads are independent and therefore plane and Ground station communication is not limited by any GUI
interaction.

5. SENSORS
5.1. GPS module

Our GPS module has an RS-232 output with 4800 baud per sec. communication speed. The data refresh
rate is 1 Hz. GPGGA is the protocol we use. The output data consists of 15 blocks, proceeded by a
synchronization symbol ‘$’, and a comma separating each data block. The data structure is:

$GPGGA,hhmmss.ss,llll.ll,a,yyyyy.yy,a,x,xx,x.x,x.x,M,x.x,M,x.x,xxxx*hh

- 12 -

ISBN-xxx-y-zz-yyyyyyy-x 243
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Table 1: Meaning of GPS datas

Rank Form Information

1 hhmmss.ss World time

2 llll.ll Latitude
3 a North/South
4 yyyyy.yy Longitude
5 a East/West
6 x GPS signal quality
7 xx indication
8 x.x Number of satellites
9 x.x Horizontal spacing
10 M Altitude (above sea level)
11 x.x Unit (meter)
12 M Altitude (geoid)
13 x.x Unit (meter)
14 xxxx Time of last refresh
15 *hh ID#
Checksum

5.2. Accelerometers and gyroscopes

To get values for pitch, roll and acceleration we use a CHAR6D module.

Figure 12 : CHR-6d module.

The CHR-6d is a 6-axis IMU, combining three accelerometer axes and three rate gyro axes in a .8" by .7"
footprint. Onboard ARM Cortex™ processor samples and filters gyro and accelerometer outputs and sends
results over a TTL (3.3V) UART.

- 13 -

ISBN-xxx-y-zz-yyyyyyy-x 244
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

● Data outputs
– 3-axis roll rates (+/- 400 deg/s)
– 3-axis acceleration (+/- 3 g)
– Pitch and roll angles
– 16-bit effective measurement resolution (after oversampling and decimation)
● Features
– Configurable digital filter (windowed Parks-McClellan FIR)
– Onboard EKF for pitch and roll angle estimation
– Automatic bias calibration
– High bias stability over temperature
– Adjustable output rates (20 Hz - 300 Hz)
– TTL (3.3V) UART interface

5.3. Ultrasound sonar

We added sonar oriented towards the ground to assist while landing, because the camera is mounted on top
of the plane making it hard to perform a visual landing since you cannot always see the ground. The range
is cca 8 meters and refresh rate is around 14 Hz.

Figure13 : Sonar radiation.

5.4. Prandtle probe

To measure flight speed of aircraft, Prandtle probes are used, which enables us to measure flow velocity of
liquids, in our case plane velocity with respect to surrounding air. It is based on measuring difference
between the total pressure and the static pressure of flow in a point of flow streamline. The velocity meter
(in central part of Prandtl probe) flow velocity is brought to zero isentropically. With velocity equal to
zero, measured pressure equals to the total pressure. On outer part of probe, where the flow streamlines are
tangent, the flow velocity is not changed and the pressure measured on that part is called the static
pressure. The difference between the total and the static pressure in one point of flow streamline is called
dynamic pressure. The dynamic pressure represents kinetic energy of the flow.

- 14 -

ISBN-xxx-y-zz-yyyyyyy-x 245
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 14 : Measurement principle.

Flow velocity can then be calculated.

For ideal gas:

( )
= (10)

For real airflow:

= . . . . − (11)

These equations work only for speeds lower than the speed of sound.
The Prandtl probe is capable of measuring speeds up to 100 m/s. This means dynamic pressure up to 6 kPa.
The most suitable pressure sensor to measure such values has proven to be MPXV4006G. This is
a piezoresistive pressure sensor with integrated temperature compensation.

Figure 15 : MPXV4006DP case and module architecture

- 15 -

ISBN-xxx-y-zz-yyyyyyy-x 246
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 16 : MPXV4006DP pressure sensor diagram

Sensor output is an analogue signal from 0.2 to 4.7 Volts. Thus, the measurement sensitivity is 766
mV/kPa. For a proper pressure measurement we needed to design appropriate filtering and A/D
conversion.

6. PRIMITIVE AUTOPILOT
This From the GPS module we get the following data:
- Latitude
- Longitude
True course in degrees is calculated using the following formula:

( ) ( )∗ ( )
= acos ( )∗ ( )
(12)

Where ‘Lat1’ and ‘Lat2’ are gathered directly from the GPS module (‘Lat1’ being the last known latitude
and ‘Lat2’ the current latitude value) and ‘d’ is the great circle distance. This is the shortest path between
two points on a sphere, in our case the globe.

- 16 -

ISBN-xxx-y-zz-yyyyyyy-x 247
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 17 : Great circle distance (‘d’)

From the gathered data we can calculate true heading. The waypoint heading is a constant telling the
autopilot where to steer the aircraft.

Figure 18 : Waypoint heading calculation

The flight direction deviation is counted as:

Error = waypointHeading – trueHeading.

if (error >= 180) error = error - 360;
if (error <= -180) error = 360 + error;
PolohaServa = stred - error;

The variable PolohaServa sets the turning of the direction flaps on the tail of the plane, to turn the whole
plane to the desired course.

- 17 -

ISBN-xxx-y-zz-yyyyyyy-x 248
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

7. PROBLEMS AND SOLUTIONS

7.1. Parachute

Since in the development of the flight control system, one has to consider potential problems that could
lead to serious malfunctions, the aeroplane should be equipped with a parachute safety system. In the case
of uncontrolled movement it should automatically start its activity and to assure sufficiently damped
landing.
Before choosing the appropriate type of parachute we need to put together some basic data. This includes:
– Max falling speed
– Weight of the plane + weight of the parachute itself
– Density of air
– Max limit for oscillation
– Deployment speed

There are two possible choices for the parachute shape. One is a more stable cross-chute and the other a
cone-shaped parachute that is more prone to oscillations but has a higher air resistance constant.

Figure 19 : Cross chute.

Air resistance constant = (0.6 – 0.85)

Deployment force constant = (1.1 – 1.2)
Oscillation = (0°± 3°)

- 18 -

ISBN-xxx-y-zz-yyyyyyy-x 249
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 20 : Our parachute.

Air resistance constant = (0.75 – 0.9)

Deployment force constant = (1.8)
Oscillation = (± 10° ± 35°)

● Equations needed
– Brake force

– G-force:
= ∗ = 44,15 (13)

– Falling speed:

∗ ∗
= ∗ ∗
= 2,12 (14)

– Surface area:
= ∗ = 5,3 (15)

– Diameter after deployment:

∗ ∗
= ∗ ∗ ∗
= 2,6 (16)

Where:
C – resistance constant
– air density
g – gravitational force constant

- 19 -

ISBN-xxx-y-zz-yyyyyyy-x 250
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The parachute test proved that our calculations were correct, at 5 m/s windspeed we measured a force of
roughly about 50 Newtons. This means it is possible to use this parachute as a backup landing system in
case of emergency, or lack of landing space.

7.2. Communication fault protection

In case of interrupt of communication, the plane could become a hazard for surroundings. To prevent such
situations, if communication is interrupted, the flaps are turned to the maximum possible position, so the
plane is forced in a corkscrew motion sending it directly to the ground instead of flying off to a populated
area. Also, to prevent damage to the plane and/or injuries to anyone standing beneath the falling plane, the
safety parachute deploys after entering communication fault mode. This ensures a softer landing.

7.3. Vibrations problems

The most serious problem solved so far is caused by vibrations generated by the engine. These are detected
by accelerometers and; in fact, they are so strong that it does not produce any relevant data. The problems
are caused by a number of factors, which include:
Combustion of fuel in the engine cylinder. Since the piston and everything connected to it have some
momentum, combustion of fuel causes a force, which forces the engine in the opposite direction of piston
movement. This is the cause of Z-axis vibrations.
Rotation axis vibrations are caused by compression and decompression of gasses in the engine cylinder.
This effect causes mild deceleration and acceleration of the propeller thus transferring force into the body
of the plane.
The CHR6d module (described in section 6.2) oversamples and decimates the ADC data on all channels to
reduce quantization noise and increase the effective ADC resolution to 16 bits. After decimation, sensor
data is processed using a configurable Parks-McClellan window FIR (Finite Impulse Response) low-pass
filter. The corner frequency of the filter is independently adjustable for each channel from 10 Hz to 140 Hz
in 10 Hz increments.
FIR (Finite Impulse Response) filters represent one of two primary types of digital filters used in Digital
Signal Processing (DSP) applications. FIR filters sometimes have the disadvantage that they require more
memory and/or calculation to achieve a given filter response characteristic. Also, certain responses are not
practical to implement with FIR filters.

- 20 -

ISBN-xxx-y-zz-yyyyyyy-x 251
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 21 : Block diagram of a simple FIR filter.

Tap - A FIR "tap" is simply a coefficient/delay pair. The number of FIR taps is an indication of:
– The amount of memory required to implement the filter
– The number of calculations required
– The amount of "filtering" the filter can do

In effect, more taps means more stop band attenuation, less ripple, narrower filters, etc.

The measurement was done for idle thrust of the engine (25 rps/ 25 Hz). The body of the plane was placed
on a fixed, rigid base. During measurement we adjusted values of the low-pass filter and tap value. These
values were set to 140Hz for the filter and 8 taps during the first part of measurements and 10Hz 64 taps
for the second part.
There is a very noticeable reduction of interference after the second setting on both the accelerometers and
gyros.

Figure 22 : Accelerometer measurement.

- 21 -

ISBN-xxx-y-zz-yyyyyyy-x 252
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 23 : Gyro measurement.

Pitch and roll angles are calculated using an extended Kalman filter, which is a comoponent of the sensor
module.
Even after filtration, there is some interference. This can be caused by imperfection of the whole system
(accumulated vibrations in the hull, irregular operation of the engine etc.).
The basic theory behind the filter is this: if the sensor is not accelerating, then the onboard accelerometers
can be used to detect gravity and (hence) estimate pitch and roll angles. In general, however, the sensor
might be moving around and vibrating, so that in the short term, the accelerometers can't be trusted. This is
where the rate gyros come in. MEMS (Microelectromechanical systems) gyro output is angular velocity. In
order to get pitch and roll angles we need to integrate this velocity. Since rate gyros are less sensitive to
acceleration, they can be used to estimate changes in pitch and roll in the short term. But angle estimates
produced by rate gyros tend to drift over time. The onboard Extended Kalman Filter is used to combine
accelerometer and rate gyro measurements in a way that removes long-term drift, and that removes the
negative effect of transient vibrations. In practice, a tradeoff must be made between trusting the rate gyros
and trusting the accelerometers. When the filter trusts rate gyros more, it is less sensitive to acceleration
and more sensitive to nonzero gyro biases. On the other hand, if the filter trusts accelerometers more, it is
more sensitive to bad acceleration, and less sensitive to nonzero gyro biases.
To get the plane’s absolute position we need to avoid long-term double integration of acceleration values
(leading to increasing relative position value errors) by using acquired GPS position data in our
calculations. By fusing this data together we get all the information about the plane’s velocity and absolute
and relative position we need.
On the CHR6d, the Extended Kalman Filter is tuned by adjusting the process variance. If the sensor were
mounted on a rotorcraft, for example, the process variance could be set very low to reduce the effect of
vibrations.

- 22 -

ISBN-xxx-y-zz-yyyyyyy-x 253
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

8. ACKNOWLEDGMENT
This work was partially supported by Project VEGA 1/0656/09: Integration and development of
nonlinear and robust control methods and their application in controlling flying vehicles and project KEGA
3/7245/09 Building virtual and remote experiments for network of online laboratories.

9. CONCLUSION
UAV development is directly tied with multiple disciples; mathematics, physics, engineering,
telecommunication, IT and many others. The project is conducted in order to achieve fully autonomous
flight equipment with no interference eith humans. In proposing solutions, we note the aircraft hazard to
people in case of errors.
We emphasize on the criterion of real-time, safety and transfer the event of failure on backup safety
systems.

REFERENCES
[1]http://www.pipeflowcalculations.com/prandtl/
[2]http://www.grc.nasa.gov/WWW/K-12/airplane/pitot.html
[3] http://www.dspguru.com/dsp/faqs/fir/basics
[4] http://en.wikipedia.org/wiki/Finite_impulse_response
[5] http://www.pololu.com/file/0J342/chr6d_datasheet.pdf
[6]http://raketove.modely.sweb.cz/popis_modelu.htm
[7]http://en.wikipedia.org/wiki/Nose_cone_design
[8]R/C Soaring Digest, Vol. 27, No.7, July 2010
[9]http://www.mh-aerotools.de/airfoils/flywing1.htm
[10]http://en.wikipedia.org/wiki/Nose_cone_design
[11]http://www.mh-aerotools.de/airfoils/flywing1.htm

- 23 -

ISBN-xxx-y-zz-yyyyyyy-x 254
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Position-based velocity control system for rotary-wing UAVs

D. Santamaría, M. García, A. Viguria and A. Ollero

Center for Advanced Aerospace Technologies (CATEC)

Parque Tecnológico y Aeronáutico de Andalucía
C/ Wilbur y Orville Wright 19 41309 - La Rinconada, Seville (Spain)
{dsantamaria, mgarcia, aviguria, aollero}@catec.aero

Abstract
This paper presents a control system for rotary-wing UAVs based on velocity references. The different
velocity control strategies and the implication of each one are discussed. The suitability of the selected
control scheme is tested first in an open loop scheme. Simulation results and flight experiments are
presented to validate the selected control methodology. Next, the final closed loop scheme is
developed and, as in the open loop case, validated first by simulation and later by experimental results
in flight experiments.

Keywords: rotary-wing UAV, position and velocity control.

1 INTRODUCTION
In the next years, unmanned helicopters are going to be used for numerous commercial, civil and
military applications. This type of platform has a VTOL (Vertical Take-off and Landing) capability
that enables UAS (Unmanned Aerial System) to operate in uneven terrains, or even to operate from
mobile platforms such as the deck of a ship. Easy to transport and deploy, these flying autonomous
vehicles are cost-effective platforms for carrying cameras and other surveillance equipment; giving
safe access to hazardous or high risk locations and environments. Some of these UAV helicopters
applications are the capability to autonomously land on a moving platform [1] or cargo deployment
and retrieval [2].

In most cases, the control system commands used to operate rotary wing UAVs are based on position
references. This is mainly given by the hovering capability (fixed point flight) of these platforms that
are not possible in fixed-wing UAVs. This control approach fits perfectly close range applications,
when the main goal is to reach the desired fixed position and it is not so important to control the
performances of the helicopter in the flight between the current and the desired position. However, in

-1-

ISBN-xxx-y-zz-yyyyyyy-x 255
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

surveillance applications that require long range navigation, it is desirable to control the flight
performances during the entire flight in order to optimize the accelerations range, cruise velocity and
maximize the aircraft endurance.

Velocity control has already been implemented for UAV helicopters with different strategies which are
designed according to specific applications ([3] and [4]). However, the velocity control scheme here
presented is based on a previously tested position controller scheme. This approach has several
advantages: the tuning of the velocity controller is easier and the helicopter is more robust to external
disturbances, such as wind, in hovering or low velocity maneuvers due to the effect of the inner
position controller.

This paper is organized as follows. The next section presents the position controller over which the
velocity controller will be designed. Then, the design framework and the flying platform are explained
in section 3. Section 4 is devoted to the proposed velocity controller approaches, including a
comparison between simulations and flying experiments results. Finally, conclusions of this work are
presented in Section 5.
The notation used in the control schemes is defined as follows: the generalized coordinates
describe the position of the helicopter center of mass (CM), describe the velocity of the
CM and indicate the orientation by the Euler angles. The main rotor force and torques are
noted by and respectively. The addition of an asterisk in the notation (*)
represents the desired value of the magnitude.

2 POSITION CONTROLLER SCHEME

We based our development on a position controller for rotary-wing UAVs that applies the inverse
kinematic and dynamic relations combined with linear PID controllers [5]. The constants of the
controllers are tuned with the pole placement technique. Other constants are tuned experimentally
introducing small disturbances in the control variables. The position controller is structured in two
nested loops: the inner and the outer loop.

The outer control loop computes the position error between the desired and current position, which is
the error associated with the three PID controllers, one for each axis. The outputs of these controllers
are accelerations, which are converted into desired tilts of the main rotor plane and required lift force
to be produced by the rotor. The conversion from accelerations to tilts and lift are based on Newton’s
Second Law. The inner control loop accepts the outputs of the outer loop, hence, it computes the error
between the desired tilts and the current orientation of the vehicle [6]. The orientation around the 3rd
axis is controlled in a separated control loop. Also, the inner loop computes the main rotor torques

-2-

ISBN-xxx-y-zz-yyyyyyy-x 256
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

from inverse kinematic and dynamic equations. Figure 1 shows a general scheme of the implemented
position controller.

Figure 1: Position controller scheme

3 DESIGN FRAMEWORK AND FLYING PLATFORM

The control algorithms were created using a model-based design tool with an automatic generation of
C code ready to run on the target. This working methodology centralizes the classical software
development process by the use of mathematical models of the system and the environment as the
backbone of the whole product creation [6]. This architecture allows an easy rapid modeling and
testing of the algorithms, and avoids many implementation problems by the use of automatic code
generation tools. Once the framework is configured, each time a modification is made in the model
the control engineer only has to update the code using a systematic procedure.

Previous to any real flight test of these algorithms, a HWIL (HardWare In the Loop) test bench was
used for a real-time testing of the algorithms and sensors [6]. The flight tests have been performed on
a modified RC helicopter that was converted into a UAV with the integration of an autopilot with
several sensors (GPS, INS and magnetometer) and a processing unit.

-3-

ISBN-xxx-y-zz-yyyyyyy-x 257
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 2: Rotary-wing UAV used to real flight tests

4 VELOCITY CONTROL APPROACHES

The presented position-based controller was use as the baseline for the development of a velocity
controller. The main goal of the velocity control is that the helicopter can fly at a certain commanded
velocity. The control is implemented on the horizontal components of the velocity (u1 and u2), so the
helicopter will move horizontally at the specify velocities. However, the altitude is controlled
independently from a specified reference in meters. Two different methods for the control of the
horizontal velocity are discussed below. Furthermore, the choice of the method that is currently
being used will be explained.

4.1 Method A: Nested velocity loop

In the method (A), a third control loop is used in cascade with the outer and the inner control loops of
the already implemented position controller. This third loop computes the error between the desired
velocity and the current velocity, resulting in position references for the position controller. These
position references are interpreted by the controller in order to produce the desired velocity.

-4-

ISBN-xxx-y-zz-yyyyyyy-x 258
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 3: Method A for velocity control. Velocity loop nested to outer and inner loop.

4.1.1 Nested velocity loop operating in open loop

First, it is advisory to implement a straightforward velocity controller operating in open loop in order
to validate the control method and to test if the dynamics of the position controller can become
quickly enough to implement the velocity control loop correctly. This control method can be treated as
the position control of the helicopter, following a virtual point moving through space. Since the loop
is not closed, the way to generate position references that can generate the desired velocity is
generating a virtual position that moves at the speed to be achieved (proportional control). Figures 4
and 5 shows simulations and flight test results of the open loop control for the two horizontal velocity
components
simulation and flight test velocity control open loop

20 u1 reference
u1 simulation
15 u1 real

10

5
u1 m/s

-5

-10

-15

-20
505 510 515 520 525 530 535 540 545 550 555
time (s)

Figure 4: Simulation and flight test results for u1 component of velocity control in open loop

-5-

ISBN-xxx-y-zz-yyyyyyy-x 259
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

simulation and flight test velocity control open loop

20 u2 reference
u2 simulation
15 u2 real

10

u2 m/s 5

-5

-10

-15

-20
495 500 505 510 515 520 525 530
time (s)

Figure 5: Simulation and flight test results for u2 component of velocity in open loop

From the previous simulation results and experimental tests, it can be observed that the helicopter
performs a right dynamic behavior. Hence, in spite of the existence of a small permanent error
between the reference and the state, this method was expected to be a good solution for the velocity
control with closed loop.

4.1.2 Nested velocity loop operating in closed loop

In order to close the loop of the controller, a velocity PID is implemented. The current position of the
helicopter is added to the reference generated by the PID controllers (one PID for each velocity
component to be controlled, u1 and u2). As in the open loop case, simulation and flight tests of closed
loop scheme are showed in Figures 6 and 7.

-6-

ISBN-xxx-y-zz-yyyyyyy-x 260
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

simulation and flight test velocity control close loop

12
u1 reference
10 u1 simulation
u1 real
8

u1 m/s 4

-2

-4

-6

-8
40 50 60 70 80 90 100
time (s)

Figure 6: Simulation and flight test results for u1 component of velocity in closed loop

simulation and flight test velocity control close loop

u2 reference
10
u2 simulation
u2 real
8

4
u2 m/s

-2

-4

-6

30 40 50 60 70 80 90 100
time (s)

Figure 7: Simulation and flight test results for u2 component of velocity in closed loop

In these figures, it can be observed that there are some oscillations in the simulation results which do
not appear in flight test data. This effect is due to the use of the same tuned values for the control
coefficients in simulation and real flights along with the fact that some dynamic effects that appears in
real flight and are not taken into account in the simulation model of the helicopter. Therefore, the
coefficients in simulation will have to be retuned in order to avoid these oscillations.

-7-

ISBN-xxx-y-zz-yyyyyyy-x 261
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

4.2 Method B: Direct application of PID controllers to velocity

The second method (B) consists of changing the structure of the position controller outer loop. In this
method, the outer loop would computes the accelerations required for the (DT)-1 block from a loop
containing three PID controllers associated to the velocity errors.

Figure 8: Option B for velocity control. Direct application of PID controllers to velocity.

The implementation of the PID controller for the velocity is similar to the one used for the position.
The general scheme of the whole system, shown in Figure 9, represents a linear control system with
the following gains: KP, KI, KD, Kq. This scheme arises from the cancellations in the inverse kinematic
and dynamic relations between the model and the controller (see [7] for more details). The
coefficients can be calculated with the pole placement technique, as in the position controller.

Figure 9: Resulting scheme for velocity control for option B

In order to proceed with the pole placement technique, it will be necessary to give an expression of
the resulting transfer function of the previous scheme:

-8-

ISBN-xxx-y-zz-yyyyyyy-x 262
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(1)

The denominator is a third degree equation, so there will be three poles. Since there are four
unknowns and three equations (one for each pole), it is needed to fix one gain in order to tune the
others:

(2)

The gain to be kept from position controller would be Kq; so the rest of gains can be calculated. As an
example, with Kq=3.3, and locating the three zeros in -1, the results are shown in Figures 10 and 11

12
u1 simulation
u1 reference
10

6
u1 (m/s)

-2
0 10 20 30 40 50 60 70 80 90
time

Figure 10: Simulation results for u1 velocity component with option B

-9-

ISBN-xxx-y-zz-yyyyyyy-x 263
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

12
u2 simulation
u2 reference
10

6
u2 (m/s)

-2
0 10 20 30 40 50 60 70 80 90
time

Figure 11: Simulation results for u2 velocity component with option B

4.3 Discussion
From the previous two methods, one of them should be selected in order to have a safe and efficient
implementation of velocity control. An analysis of both strategies has been performed to support a
suitable choice.

Method B has the major advantage of using the pole placement technique, which is similar to that
used for the position controller; it is possible to quickly get a first approximation of the coefficients
needed in the PID controllers. The main drawback of this approach is the transition between position
control mode and velocity control mode, since each controller has its own integral processes which
should be reset with a certain initial condition when the transition occurs.

The previous transition problem has been solved with the first approach (method A). In this option,
due to the implementation of the velocity loop in cascade with the controller position, the transition
between position control and velocity control modes would occur with a simple change in reference
position. Hence, in position control, the position reference is the desired position to be reached; and in
velocity control, it is the reference generated in the velocity loop. Apart from that, there is a great
advantage when tuning the coefficients of the controller, since this approach is based on a platform
previously stabilized for the position controller. Therefore, tuning the coefficients in flight tests will
result much easier with this approach. Another advantage of method A over method B is the hovering
performance (linear velocity equal to zero): while applying method A the helicopter stay still in

- 10 -

ISBN-xxx-y-zz-yyyyyyy-x 264
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

hovering in the presence of wind since the commanded position is fixed; when applying method B,
the wind can generate a small motion relative to the ground.

On the other hand, a comparison with the backstepping velocity [3] control was performed. In order to
compare the simulation results with backstepping method, a PD velocity control for the 3th axis was
added in cascade giving an altitude reference to the already implemented altitude controller. This
procedure is the same that has been used previously on the horizontal velocity controller. The
simulations start with an initial velocities condition (1, 1, 1) and the velocities references are (0, 0,
2.5).
1.2 1 2.5

1 0.8

0.8
0.6
2
0.6
V1 (m/sec)

V2 (m/sec)

V3 (m/sec)
0.4
0.4
0.2
0.2
1.5
0
0

-0.2 -0.2

-0.4 -0.4 1
0 1 2 3 4 0 1 2 3 4 0 1 2 3 4
Time (sec) Time (sec) Time (sec)

Figure 12 Simulation option A control in close loop, with 3th axis velocity control

Figure 13 Simulation of non linear backstepping method. Figure taken from work [3]
As can be seen from the comparison between Figures 12 and 13, the behavior of the developed
velocity control system is more gently, with less overshooting response and does not need to have a
toleration bandwidth of 0.5m/s of avoid singularities as the backstepping method.

5 CONCLUSIONS
Velocity control offers some advantages when comparing to position control in long range missions
allowing to better control the flight performances during the entire flight in order to optimize the
accelerations range, cruise velocity and maximize the aircraft endurance.

This paper has proposed and compared two velocity control methods. The selected method was first
tested operating in open loop in order to test its suitability. The simulation results and the subsequent

- 11 -

ISBN-xxx-y-zz-yyyyyyy-x 265
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

validation with flight experiments verified the required functionalities of the proposed controller
method. The closed loop results in simulation and in flight tests demonstrate a good performance of
the chosen control method and its suitability for velocity control.

On the other hand, the functionalities of this velocity control method can be expanded to other
applications such as tracking moving objects, control of the performance of long range flights needed
for surveillance missions and the landing on moving platforms.

ACKNOWLEDGEMENTS
This work is partially supported by the European Commission ICT Programme under project CONET
NoE (FP7-INFSO-ICT-224053), and the project PLANET (FP7 257649). The authors would like to
thank Mr. Francisco Alarcón and Mr. Antonio Jiménez for their contribution in the development of the
CB5000 helicopter, and to Mr. Allan Anderson Bell for his help during the execution of the
experiments.

REFERENCES
1. S. Saripalli and G. S. Sukhatme. Landing on a Moving Target using an Autonomous Helicopter.
s.l. : Proceedings of the International Conference on Field and Service Robotics, 2003.
2. N. R. Kuntz, and P. Y. Oh. Towards Autonomous Cargo Deployment and Retrieval by a
Unmanned Helicopter using Visual Tracking. s.l. : International Design Engineering Technical
Conferences & Mechanisms and Robotics Conference, 2008.
3. H. R. Pota, B. Ahmed and M. Garratt. Velocity Control of a UAV using Backstepping Control.
s.l. : Proceedings of the 45th IEEE Conference on Decision & Control, 2006.
4. B. Markus, K. Kondak, I. Maza and A. Ollero. Autonomous Transportation and Deployment
with AerialRobots for Search and Rescue Missions. Journal of Field Robotics, Wiley Blackwell, 2011.
5. K. Kondak, C. Deeg, G. Hommel, M. Musial, V. Renuss. Mechanical model and control of an
autonomous small size helicopter with a stiff main rotor, in the Proceedings of the International
Conference on Intelligent Robots and Systems. 2006.
6. D. Santamaría, F. Alarcón, A. Jiménez, A. Viguria, A. Ollero, M. Béjar. Model-based design,
development and validation for UAS critical software. s.l. : ICUAS (International Coference on
Unmanned Aircraft Systems), 2011.
7. K. Kondak, M. Bernard, N. Losse, G. Hommel. Elaborated modeling and control for autonomous
small size helicopters. s.l. : ISR/ROBOTIK, 2006.

- 12 -

ISBN-xxx-y-zz-yyyyyyy-x 266
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Coordinated Standoff Target Tracking of Unmanned Aerial

Vehicles Using Adaptive Sliding Mode Control

Hyondong Oh, Seungkeun Kim, Antonios Tsourdos, and Brian A. White

Autonomous Systems Group, Department of Informatics and Systems Engineering, Cranfield University,
Defence Academy of the UK, United Kingdom, [email protected]

Abstract

This paper proposes a coordinated standoff target tracking guidance for unmanned aerial vehicles
using a sliding model control based on a tangent vector field. Particularly, this study exploits
additional adaptive terms in an existing vector field approach using sliding mode control to reduce
the effect of unmodeled dynamics and disturbance, as well as a decentralised angular separation
control between UAVs using either velocity change or orbit radius change by different information
architectures. Numerical simulations are performed to verify the feasibility and benefits of the
proposed approach under a realistic civilian ground vehicle tracking scenario using three cooperating
UAVs having unknown parameters in the system.

keywords: standoff tracking, vector field guidance, sliding mode control, coordination, UAV

1 Introduction
In the recent decades, the benefits of using cooperative unmanned aerial vehicles (UAVs) have been
actively investigated for both military and civil applications because it can have higher survivability in
dangerous circumstances due to inherent redundancy and then contribute to maintain or succeed the
given mission. Especially searching and subsequent tracking of moving ground target interested is one of
the important capabilities of UAVs required to increase an overall knowledge to surrounding environment
and to take proactive measures for preventing safe-critical events from happening. In performing such
missions, UAVs should be able to keep a certain distance from the moving target known as a standoff
distance with prescribed inter-vehicle angular spacing in order to track it without being noticed and
at the same time to acquire the accurate target information. In a domain of standoff target tracking,
Frew et. al. [1] proposed to apply Lyapunov vector fields for standoff coordination of multiple UAVs.
They invented a decoupled control structure in which speed and rate of heading change are separately
controlled for standoff distance and phase angle keeping, respectively. Summers et. al. [2] extended this
phase-keeping idea to multiple UAVs using information architectures in vehicle formations. Moreover,

1
ISBN-xxx-y-zz-yyyyyyy-x 267
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

they addressed a variable airspeed controller to maintain the circular orbit despite unknown wind and
unknown constant velocity target motion using adaptive estimation. Similarly, Kingston et. al. [3]
used vector field approach, however, they introduced a sliding mode control and orbit radius change
without velocity change for phase-keeping of multiple UAVs. Wise and Rysdyk [4] well surveyed and
compared the different methodologies for standoff tracking: Helmsman behaviour, Lyapunov vector
field, controlled collective motion, and model predictive control. Recently, Kim et. al. [5] applied a
receding horizon model-based model predictive control for standoff tracking.
This paper proposes a coordinated standoff tracking guidance using a adaptive sliding model control
(SMC) based on the tangent vector field. Motivated by [3], this study exploits additional adaptive terms
in the existing vector field approach using the sliding mode control to reduce the effect of unmodeled
dynamics and disturbance. As to the coordinated tracking using multiple UAVs, in general, the angular
separation keeping has been performed by a secondary guidance which adjusts the velocity of the vehicle
as explained above, independent of the standoff distance keeping. However, considering small allowable
velocity bounds of the UAV, since the convergence to the variable velocity might be slow or hard to
achieve, maintaining UAVs at their constant fuel efficient velocity can be more desirable from a mission
duration point of view. Thus, this study introduces a decentralised angular separation control based on
the tangent vector field guidance (TVFG) for multiple UAVs using either velocity change or orbit radius
change by different information architectures following [2, 3]. Numerical simulations for a standoff target
tracking are performed to verify the feasibility and benefits of the proposed approach under a realistic
civilian ground vehicle tracking scenario using three cooperating UAVs having unknown parameters in
the system. The remainder of this paper is organised as follows. Several vector field approaches are
first reviewed in Section 2, and followed by TVFG using the adaptive SMC. In Section 4, this paper
introduces the coordinated TVFG for angular separation keeping. At the last section, the performance
and the properties of the proposed algorithm are verified via numerical simulations.

2 Vector Field Characteristic

This section introduces UAV dynamic model, and briefly reviews several vector field approaches.

2.1 System dynamics

This study assumes that each UAV has a well-designed low level flight controller composed of SAS
(Stability Augmentation System) and CAS (Controllability Augmentation System) for heading, velocity,
and altitude hold functions. In order to deliver guidance inputs to this low level controller, let us consider
a three dimensional UAV kinematic model as:

ẋ = uV cosψ + Wx

ẏ = uV sinψ + Wy

ψ̇ = uψ (1)

2
ISBN-xxx-y-zz-yyyyyyy-x 268
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

where [x, y] and ψ are the inertial position and heading of the UAV, respectively and [Wx , Wy ] is the
background wind velocity. [uV , uψ ] are the commanded speed and turning rate.

2.2 Lyapunov vector field

Lyapunov vector field was initially proposed by [6] and further developed by [1] as:

V (x, y) = (r2 − rd2 )2 (2)

p p
where r = δx2 + δy 2 = (x − xt )2 + (y − yt )2 is the distance of the UAV from the ground vehicle.
Herein (xt , yt ) is the position of the ground vehicle which can be estimated from the tracking filter, and
rd is a desired standoff distance from the UAV to the ground vehicle. Differentiating Eq. (2) gives:

V̇ (x, y) = ∇V [ẋ, ẏ]T (3)

The Lyapunov vector field uses the following desired velocity [ẋd , ẏd ]T :
   
δx 2
ẋd −V d (r − rd2 ) + δy(2rrd )
 =  kl  (4)
r(r2 + rd2 ) δy 2 2
ẏd kl (r − rd ) − δx(2rrd )

where kl is positive constant, and Vd is a desired UAV speed. Substituting Eq. (4) into Eq. (3) yields
4Vd r
V̇ (x, y) = − V (5)
kl (r2 + rd2 )
As seen in the above equation, kl newly introduced in this study can be used to adjust the converging
speed to the standoff circle. The desired heading can be decided using the desired two dimensional
velocity components in Eq. (4) as:
ẏd
ψd = tan−1 . (6)
ẋd

2.3 Supercritical Hopf bifurcation

The second vector field is supercritical Hopf bifurcation which was initially proposed in [7] based on
the theory of [8]. The Supercritical Hopf bifurcation is known to mathematically produce the spiral
trajectory which converge to a limit cycle no matter where the starting position is inside and outside of
the limit circle. Desired vector field based on the system equation of the Supercritical Hopf bifurcation
is given as:
   
ẋd δx
(r2 − δx2 − δy 2 ) − δy
ks rd2 d
 =  (7)
δy
ẏd (r2 − δx2 − δy 2 + δx
ks rd2 d

where ks is a positive constant. Substituting Eq. (7) into Eq. (3) yields
4r
V̇ (x, y) = − V (8)
ks rd2
Similarly as the Lypunov vector case, ks can be used to adjust the converging speed to the standoff
circle.

3
ISBN-xxx-y-zz-yyyyyyy-x 269
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2.4 Tangent vector field

Tangent vector field [3] is generated by a desired heading angle as shown in Fig. 1, and given as:

Figure 1: Geometry of tangent vector field

ψd = ψp + tan−1 (kt d) (9)

where d = r − rd is the distance from the standoff orbit and ψp = tan−1 (y/x) + π/2 is tangent to the
standoff orbit along the ray connecting the UAV and the target position. For the standoff orbit following,
p
vehicle dynamics is expressed in polar coordinates by letting r = x2 + y 2 and θ = tan−1 (y/x) and
differentiating these as given:
Vd
ṙ = −Vd sin ψ̃, θ̇ = cos ψ̃ (10)
r
where ψ̃ = ψ − ψp . By using the Lyapunov function W = 12 ψ̃ 2 and Eq. (10) gives:

˙ kt Vd ψ̃ sin ψ̃
Ẇ = ψ̃ ψ̃ = − <0 (11)
1 + (kt d)2

This shows that the heading angle converges to the angle ψp tangent to the orbit and d converges
zero (that is, r → rd ) asymptotically by LaSalle’s invariance principle. Figure 2 shows the example

(a) Lyapunov vector field (b) Supercritical Hopf bifurcation field (c) Tangent vector field

Figure 2: Example UAV trajectories for a stationary ground target using vector field approach

of UAV trajectories for a stationary ground target using different vector field approaches. Although

4
ISBN-xxx-y-zz-yyyyyyy-x 270
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

aforementioned vector fields have different characteristics, they can generate a similar trajectory as red
dashed line in Fig. 2 by adjusting the field gain such as kl , ks and kt . It is worth noting that the field
gain should be determined not to exceed the turning rate constraint of the vehicle, ψ̇max , along with
control gain. This paper uses tangent vector field approach since it allows to apply the adaptive sliding
mode control concept robust to uncertainty of the system and disturbance as well as to use the orbit
radius for angular separation control between UAVs as will be explained later.

3 Tangent Vector Field Guidance with Adaptive SMC

This section deals with the tangent vector field guidance (TVFG) using the sliding mode control concept.
First of all, the UAV is assumed to have a first-order heading dynamics by the autopilot as given:

ψ̇ = α(ψc − ψ) + ν (12)

where ψc is the commanded heading angle, α is a known positive constant that characterizes the speed of
response of heading-hold autopilot loop, and ν represents unmodeled dynamics or faults of the autopilot
loop. Under the assumption that α is accurately known and ν is small enough to ignore, to obtain
the desired heading ψd , the guidance command ψc is selected as the sum of proportional feedback and
feedforward term as:
1
ψc = ψ + (ψ̇d − kψ (ψ − ψd )) (13)
α
where kψ represents the proportional gain. However, in general, α is difficult to determine experimentally
and inevitably contains error in its estimated value as well as ν can be large value due to the unmodeled
dynamics, disturbance or fault. Since these factors could result in less precise target tracking, this
study uses the adaptive sliding mode control which estimates and compensates the effect of unknown
parameters. First of all, let us define the sliding surface as:
Z t
S = e + kI edτ (14)
0

where e = ψ − ψd is the state error, and kI is an integral gain. Differentiating the sliding surface with
respect to time as:
Ṡ = ψ̇ − ψ˙d + kI e = α̂(ψc − ψ) + ν̂ − ψ˙d + kI e (15)

where α̂ and ν̂ are the estimation of α and ν, respectively. Let Lyapunov function candidate W1 = 21 S 2
and take the derivative to obtain:
 
Ẇ1 = S Ṡ = S α̂(ψc − ψ) + ν̂ − ψ˙d + kI e (16)

Then, the control command can be made as:

1 1 kt
ψc = ψ + (−ν̂ + Vd cosψ̃ − Vd sinψ̃ − kI e − c1 S − c2 sign(S)) (17)
α̂ r 1 + (kt d)2
and c1 and c2 are a positive constant. Then,

Ẇ1 ≤ −c1 S 2 − c2 |S| (18)

5
ISBN-xxx-y-zz-yyyyyyy-x 271
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

from which it can be concluded that S goes to zero in finite time. Substituting obtained control command
into Eq. (16) gives:

Ẇ1 = S Ṡ = S(αα̂−1 (−ν̂ + ψ˙d − kI e − c1 S − c2 sign(S)) + ν − ψ˙d + kI e)

' S(−ν̃ − c1 S − c2 sign(S) + α̃ψs ) (19)

where ν̃ = ν − ν̂, α̃ = α − α̂ and ψs = ψc − ψ. To obtain the adaptation rule for the parameter error,
let us consider another Lyapunov function candidate W2 as:

1 2 1 −1 2 1 −1 2
W2 = S + γν ν̃ + γα α̃ (20)
2 2 2

where γν and γα are a positive constant. Differentiating Eq. (20) gives:

Ẇ2 = S(−ν̃ − c1 S − c2 sign(S) + α̃ψs ) + γν−1 ν̃ ν̃˙ + γα−1 α̃α̃˙

˙ + α̃(ψs S − γ −1 α̂)
= −c1 S 2 − c2 |S| + ν̃(S − γν−1 ν̂) ˙ (21)
α

where the relation ν̃˙ = −ν̂˙ and α̃˙ = −α̃˙ are used under the assumption that ν and α are constant. Then,
the adaptation law for ν̂ and α̂ can be obtained as:

ν̂˙ = γν S, α̂˙ = γα ψs S (22)

Then,
Ẇ2 = −c1 S 2 − c2 |S| ≤ 0 (23)

from which it can be concluded that the error state e tends to zero in finite time.
Moreover, to take a target velocity into account, let us consider the behaviour of a point orbiting a
constant speed target at fixed radius rd , then, the position of the point can be expressed as:

xtp = rd cosθ + Tx t, ytp = rd sinθ + Ty t (24)

where Tx and Ty are the velocity of the target. Maintaining the commanded vehicle speed Vd of the
UAV leads the following expression:

Vd2 = (ẋtp )2 + (ẏtp )2 = (−rd θ̇sinθ + Tx )2 + (rd θ̇cosθ + Ty )2 (25)

Substituting θ̇ obtained by solving Eq. (25) into following equation gives the final path heading angle.
 
ẏtp
ψp = tan−1 (26)
ẋtp

4 Decentralised TVFG for Angular Separation

In performing coordinated target tracking mission, UAVs should keep prescribed inter-vehicle angular
separation to maximise sensor coverage or enhance the accuracy of the target information, as well as
a standoff distance. Moreover, in terms of estimation performance, policies that maintain the optimal

6
ISBN-xxx-y-zz-yyyyyyy-x 272
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

distance while allowing the angular separation to vary are more robust over the entire set of possible
sensor characteristics than approaches that maintain optimal angular separation at the expense of
distance [9]. For this, in general, the angular separation keeping has been performed by a secondary
velocity control. However, considering small allowable velocity bounds of the UAV, maintaining UAVs
at their constant fuel efficient velocity can be more desirable from a mission duration point of view at the
expense of performance in the standoff distance. Thus, this study introduces the decentralised angular
separation control of multiple UAVs using either velocity change or orbit radius change by different
information architectures.

4.1 Minimally persistent leader-follower information architecture

In this architecture, one of the UAVs (leader) maintains the orbit around target with nominal airspeed
and orbit radius, and the remaining vehicles (followers) maintain the orbit around target and through
adjustments in airspeed or orbit radius, maintain a prescribed angular spacing with the neighboring
vehicle ahead of it. This architecture requires a minimum possible number of communication links to
achieve the circular orbit and angular spacing. For orbit radius change case, among total n UAVs, let
one UAV have the constant orbit radius rd , and the remaining n − 1 UAVs have a variable orbit radius
by changing heading command of tangent vector field as:

ψid = ψip + tan−1 (kt (di − ko δθim )) (27)

where ko is a control gain weighting the convergence to a desired orbit, and δθim is i-th angular spacing
error given by:
δθim = θi+1 − θi − θd (28)

where θd is the desired angular separation between UAVs. Note that d = r − rd is a distance from the
desired orbit to the UAV position in a normal TVFG as Eq. (9), and now it is adjusted by additional
term (ko δθim ) according to the angular spacing error, which results in the temporary change of orbit
radius. Similarly, for the velocity control case, let one UAV have the nominal desired velocity input Vd ,
and the remaining n − 1 UAVs have a variable velocity input according to the angular spacing error as:

uV i = Vd + ∆Vmax tanh(kV δθim ) (29)

where kV is a control gain, and δθim is the same as Eq. (28), and ∆Vmax > 0 is a design parameter to
be met for a speed constraint of the UAV. When the angular separation is different from the desired
value, velocity input of each UAV is adjusted accordingly.

4.2 Nonminimally persistent information architecture

In this architecture, the airspeed or orbit radius is adjusted such that each vehicle moves toward the
midpoint of its two nearest neighbors on the standoff orbit. It has the advantage that the vehicles do not
need to know the number of engaging vehicles so that the control structure can be fully decentralised.

7
ISBN-xxx-y-zz-yyyyyyy-x 273
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

For this, let i-th UAV have a variable orbit radius or velocity using the same equations as Eq. (27) and
Eq. (29) but with different angular spacing error δθin , given by:
1
δθin = (θi−1 + θi+1 ) − θi (30)
2

5 Numerical Simulations
To verify the feasibility and benefits of the proposed approach in this paper, numerical simulations are
performed by using the ground vehicle tracking scenario as shown in Fig. 3(a) and three cooperating
UAVs having unknown parameters, the time constant of heading-hold loop α and disturbance ν. The
vehicle trajectory data is obtained using a S-Paramics traffic model, and the EKF (Extend Kalman
Filter) and the state-vector fusion were used for the localisation of the target as shown in Fig. 3 (Details
can be found in [5]). The parameter values used is shown in Table 1. Firstly, the standoff tracking
without any coordination between UAVs (i.e. no angular separation keeping) is performed using both
of the normal SMC and adaptive SMC in TVFG to investigate the effect of adaptive terms. As can be
seen in Table 2, the standoff tracking performance of the adaptive SMC is much better than that of the
normal SMC due to the estimation of unknown parameters. Figure 4-5 display the relative trajectories of
UAVs with respect to the ground target using the adaptive SMC and the angular separation keeping for
different information architectures under the assumption of the perfect communication between UAVs.
Nonminimally persistent architecture shows better performance in terms of angular separation as shown
in Table 2 since all three UAVs adjust their velocity or orbit radius compared to minimally persistent case
adjusting parameters of only two UAVs. Moreover, although velocity change shows better performance
than that of orbit radius change, in case that frequent velocity change is undesirable or unattainable,
the angular separation can be achieved without velocity control within some boundary by adjusting
orbit radius appropriately at the expense of performance in standoff distance.

(a) (b) Trajectory (c) Velocity

Figure 3: The scenario description in the civilian traffic and state-vector fusion based on EKF

6 Conclusions
This paper proposed the coordinated standoff target tracking guidance using the sliding model control
based on tangent vector field. Particularly, this study used additional adaptive terms in the existing

8
ISBN-xxx-y-zz-yyyyyyy-x 274
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Table 1: Parameter values used in simulation

Parameters Value

Desired values rd , θd , Vd 500 m, 23 π, 40 m/s

Maximum speed increment ∆Vmax 10 m/s
Control gain [kV , ko , kI , kt , c1 , c2 ] [30, 3, 0.02, 0.02, 0.3, 0.6]
Maximum heading rate ψ̇max 0.2 rad/s
heading-hold time constant [α, α̂0 ] [1.5, 3.0]
Disturbance [ν, ν̂0 ] [(0.1,-0.1,0.15), 0.0]

Table 2: Mean error on standoff distance and angular separation

No coordination Orbit radius change Velocity change
Mean error
SMC ASMC Minimal Nonminimal Minimal Nonminimal

er = r − rd (m) 53.51 25.17 100.57 111.71 26.94 25.95

eθ = δθ (deg) - 25.60 17.09 16.42 5.20

(a) Orbit radius change (b) Velocity change

Figure 4: Trajectories of UAVs relative to the target using adaptive TVFG (minimally persistent)

(a) Orbit radius change (b) Velocity change

Figure 5: Trajectories of UAVs relative to the target using adaptive TVFG (nonminimally persistent)

9
ISBN-xxx-y-zz-yyyyyyy-x 275
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

SMC approach, as well as the decentralised angular separation control based on TVFG using either
velocity or orbit radius change by different information architectures. Numerical simulation for the
realistic ground vehicle tracking scenario shows benefits of using adaptive terms in case that there are
uncertainties in the system. Moreover, nonminimally persistent architecture shows better performance in
terms of the angular separation. Lastly, although velocity change case showed better performance than
that of orbit radius change, the angular separation was achieved within some boundary without velocity
control by adjusting orbit radius only, which could be desirable for the mission duration standpoint. As
a future work, the effect of imperfect or inconsistent information exchange between UAVs depending on
the number of communication links will be investigated by adopting a realistic communication process.

REFERENCES
[1] E.W. Frew, D.A. Lawrence, and S. Morris. Coordinated standoff tracking of moving targets using
lyapunov guidance vector fields. Journal of Guidance, Control, and Dynamics, 31(2):290–306, 2008.

[2] T.H. Summers, M.R. Akella, and M.J. Mears. Coordinated standoff tracking of moving targets:
Control laws and information architectures. Journal of Guidance, Control, and Dynamics, 32(1):56–
69, 2009.

[3] D. Kingston and R. Beard. UAV spaly state configuration for moving targets in wind. Lecture Notes
in Control and Information, 369:109–128, 2007.

[4] R.A. Wise and R.T. Rysdyk. UAV coordination for autonomous target tracking. AIAA Guidance,
Navigation and Control Conference, Keystone, Colorado, USA, 2006.

[5] S. Kim, H. Oh, and A. Tsourdos. Nonlinear model predictive coordinated standoff tracking of moving
ground vehicle. to be appeared in AIAA Guidance, Navigation, and Control Conference, Potland,
Oregon, USA, 2011.

[6] D.A. Lawrence. Lyapunov vector fields for UAV flock coordination. 2nd AIAA Unmanned Unlimited
conference, workshop, and exhibit, Reston, VA, USA, 2003.

[7] M. Quigley, M.A. Goodrich, S. Griffiths, A. Eldredge, and R.W. Beard. Target acquisition, local-
isation, and surveillance using a fixed-wing mini-UAV and gimbaled camera. IEEE international
conference on robotics and automation, Barcelona, Spain, 2005.

[8] J.E. Marsden and M. McCracken. The Hopf Bifurcation and its applications, Vol 19 of Applied
Mathematical Sciences. Springer-Verlag, 1976.

[9] E.W. Frew. Sensitivity of cooperative target geolocalization to orbit coordination. Journal of Guid-
ance, Control, and Dynamics, 31(4):1028–1040, 2008.

10
ISBN-xxx-y-zz-yyyyyyy-x 276
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Nonlinear Optimal Control of Holomonic Indoor Airship

Yin Yang, Inna Sharf

Department of Mechanical Engineering

McGill University, Montreal, Canada
[email protected]

Abstract

Two full-state feedback controllers have been proposed to fulfill the requirements for fly-
ing an indoor holonomic airship based on its dynamics model, namely, hovering control and
continuous reference tracking. In the hovering control design, the airship is assumed to be a
quasi-stationary plant. An infinite horizon linear quadratic regulator (LQR) operating in a gain
scheduling manner is employed. In addition, the LQR controller can also be applied for simple
set-point maneuvers, for example a step change in the position of the airship center. Experi-
mental results are presented for this maneuver which show that the LQR controller has a better
performance in terms of pose regulation and energy cost as compared to the PD controller.
Then, a continuous tracker is designed to reject all disturbances along an arbitrary reference
trajectory. The controller is implemented and tested on the airship for tracking a rectangular
trajectory in the lab. With little computation cost, the proposed controller achieves signifi-
cant improvement in terms of overshoot over the PD controller and 2-3 times smaller attitude
tracking error than that obtained with the PD controller.

keywords: indoor airship, hovering control, reference tracking, unmanned aerial vehicle

1 INTRODUCTION

Over the last few decades, indoor airships (blimps) have attracted interest in various contexts. Most
of the airships used are simply miniature versions of traditional airships: ellipsoidal in shape with

ISBN-xxx-y-zz-yyyyyyy-x 277
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

thrusters at the rear and fins for stabilization. The Aerospace Machatronics Laboratory (AML) at
McGill University has been home to a novel indoor airship, shown in Fig. 1. The design of the 1.8m
diameter, spherical airship permits holonomic 6-degree-of-freedom (DOF) operation. The airship
consists of a bladder bag circumscribed by three rigid hoops with all onboard equipment fixed to
the hoops, including the six ducted fans symmetrically arranged to provide the 6-DOF control.
At the present, full-state estimation for the airship is accomplished by using the motion capture
system from Vicon Motion Systems Inc., and an on-board 3DM-GX1 inertial measurement unit
(IMU). Originally developed to emulate a tumbling satellite, the airship is being adapted for use
as a surveillance platform and it also serves as a safe test-bed for developing control algorithms for
aerial vehicles. The airship employs a ground control station running a 100 Hz feedback loop with
Quanser Quarc real-time platform, and control signals are transmitted via RF to on-board thruster
driving circuits. A PD controller was implemented in the early stages of the airship development.
However, more complicated navigation tasks require more robustness and precision from the control
system.

Z’
B

C Y’
E
A

X’
D

(a) Airship hovering in AML (b) Airship frame and propeller arrangement, with
thruster direction denoted by arrows

Figure 1: Indoor airship developed in AML

Model-based control has been successfully used in many airship and unmanned aerial vehicle (UAV)

ISBN-xxx-y-zz-yyyyyyy-x 278
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

applications. Unlike model-free approaches, such as reinforcement learning control and neural
network control [1], it takes advantage of the prior knowledge of the system, and the algorithm is
suitable for real-time implementation. PID controllers are commonly used in UAV control [2][3],
but nonlinear design methods, for example, feedback linearization [4] and sliding mode control [5]
have also been attempted.

For our holomonic airship, modern optimal control is selected as the foundation of the airship
control for a couple of reasons. First of all, it offers standard and easy-to-implement algorithms for
selecting the inner loop feedback gains automatically. Hence, closed-loop stability and performance
are guaranteed in theory. In contrast to the classical controller design, in optimal control, all
the feedback loops are closed simultaneously by solving standard matrix design equations. This is
extremely advantageous for controlling a system with relatively large dimensions and coupled states.
Secondly, it offers enough robustness in terms of handling disturbance and modelling uncertainty.
Since the airship itself is a time-variant system due to helium leakage and battery drain, a model-
sensitive strategy is not suitable for our application. At last, the desired strategy needs to be
implemented in real-time. Approaches that require large computational resources, such as the
nonlinear model predictive control [6], have therefore not been considered in this work.

This paper is organized as follows. The simplified airship dynamics model is presented in Section
2. In Section 3, two control strategies of the holomonic airship are laid out after reviewing the
maneuvering requirements for indoor navigation tasks. Experimental results demonstrating the
advantages of our approaches over the PD controller are described in Section 4.

2 DYNAMICS MODEL OF HOLONOMIC AIRSHIP

To model our airship, we assume it to be neutrally buoyant and perfectly balanced, i.e., the center of
mass and the center of buoyancy are coincident at the geometric center of the sphere. Therefore no
buoyancy nor gravity forces need to be included in the dynamics equations. Also, any unpredictable
disturbances, for example, air drafts, have been ignored and the system is assumed to be time-
invariant over the flight period, i.e., quantities, such as helium leakage and battery drain, that
may change over time are assumed to be constant. The system dynamics can be formulated in the

ISBN-xxx-y-zz-yyyyyyy-x 279
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

airship frame as:

X
(ma + madd )(v̇ + ω × v) = fT i + fD + fL (1)
X
Ia ω̇ + ω × (Ia ω) = ri × fT i + MD (2)

where v is the translational velocity of the airship and ω is the angular velocity; Ia is the inertia
matrix of the airship; ri denotes the location of the thrusters relative to the airship center; fT i
are the thruster inputs; fD is the translational drag on perfect sphere; fL is the aerodynamic lift
(Magus force) and MD is the rotational drag. In Eq. (1), we also correct the airship mass with the
added mass calculated as:
1 1
madd = ρa Va = ma (3)
2 2
where ρa is the density of air, and Va and ma are the volume and mass of the airship respectively.

Extensive simulations have shown that the aerodynamics terms in Eqs. (1-2) calculated according
to empirical formulas ([7], [8] and [9]) are negligible for our indoor airship since the maximum
velocities attempted within the confines of our lab are: ωmax = 1 rad/s for rotational velocity and
vmax = 0.5 m/s for translational velocity. Defining the system state as:
h iT
x = qT pT ωT vT (4)

where q = [q0 q1 q2 q3 ]T is the quaternion of the airship (in the global frame); p is the position
of the airship center in the global frame. The dynamics equation combined with translational and
rotational kinematics can be written in the first-order form as:
   
1
q̇ 2 Qω Rω
   
   
 ṗ   Rv 
 =  (5)
   P 
 ω̇   I−1 ( ri × fT i − ω × (Ia ω)) 
   a 
2 P
v̇ 3ma fT i − ω × v

where R is the rotation matrix from airship to the global frame, and Qω transforms angular velocity
components to the derivative of the quaternion [10].

For controller design, Jacobian lineraization can be performed based on the nonlinear dynamics Eq.
(5) and equilibrium point x0 defined on:

ω 0 = 0, v0 = 0, u0 = 0 ∀q0 , p0 (6)

ISBN-xxx-y-zz-yyyyyyy-x 280
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Defining perturbation vectors as:

δx(t) , x(t) − x0 (t) (7)

δu(t) , u(t) − u0 (t) = u (8)

with Jacobian linearization technique, we obtain the linearized system:

δ ẋ(t) = A0 (t)δx(t) + B0 (t)u(t) (9)

where A0 B0 are Jacobian matrices computed at x0 .

3 CONTROL STRATEGY

To apply the optimal control theory, we first subdivide the airship control problem into three design
tasks. The simplest one is the hovering task, where we want to keep the airship stationary at a
given pose in the work space, while rejecting perturbations including air flow disturbances, control
errors caused by modelling uncertainties and linearization, as well as the airship’s imbalance, which
is unavoidable on the real system. It is pointed out that the hovering pose may not be unique, and
the airship must be capable of shifting from one pose to another as long as this process is quasi-
stationary or quasi-time-invariant. Therefore, in designing the hovering controller, the airship can
be treated as a time-invariant plant, operating at or near the equilibrium family derived in the
Jacobian linearization. Therefore, the infinite horizon formulation of the optimal controller design
can be employed.

The second control objective for our airship is set-point control: with the knowledge of the current
state x, it is desired to find the optimal control input to bring the airship to the desired state
xd . This type of maneuver is especially useful in navigation. Since the motion of the airship is
generally slow, it is possible to extend the quasi-stationary assumption of the hovering control to
the set-point tasks and use the aforementioned controller.

Trajectory control is another yet very important problem to tackle. A reference trajectory r(t) is
prescribed along with a time constraint t ∈ [t0 , T ]. The goal is to follow the reference input with
minimal tracking error and control effort. Fortunately, by using the Jacobian linearization along
the reference trajectory, we are able to apply the quadratic performance index as the objective

ISBN-xxx-y-zz-yyyyyyy-x 281
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

function for the tracking problem. The classical optimal control theory can be applied after we
solve the inverse dynamics problem and the Riccati differential equation.

The design and the implementation of the hovering controller and the trajectory tracking con-
troller are discussed in detail in the following subsections. Experimental results demonstrating
their performance as compared to that of the PD controller are also presented.

3.1 Hovering Control

As indicated before, the basic idea for the hovering control is to achieve perturbation rejection
around the equilibrium point of a quasi-time-invariant system. Applying the quadratic regulator
method to the linearized system Eq. (9), we define the infinite horizon performance index (PI) as:
Z ∞
 T 
J= δx (t)QL δx(t) + δuT (t)RL δu(t) dt (10)
t0

where RL = RTL > 0 is the control weighting matrix; QL = QTL ≥ 0 is the state weighting matrix.
The control law is determined by:

u(t) = −R−1 T
L B0 Kδx(t) (11)

where K is the solution of the algebraic Riccati equation (ARE):

KA0 + AT0 K + QL − KB0 R−1 T

L B0 K = 0 (12)

Given the computational capabilities of the modern computers, combined with the tools available
in MATLAB Simulink/Quanser QuaRC, it is feasible to solve ARE on-line. Thus, for the case when
control for different hovering poses is required, it becomes feasible to recompute the gain matrix
on-line at some specified rate and to dynamically “schedule”the feedback gains. This technique is a
trade-off between high computation cost (lowest when using the precomputed gain) and adaptability
to the change of the state (best when updating the gain at the speed of the control loop). Since the
system nonlinearity mainly arises from the orientation transformation and A0 , B0 in Eq. (12) are
dependent on the attitude, the scheduling variable is chosen to be the quaternion of the airship.

ISBN-xxx-y-zz-yyyyyyy-x 282
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3.2 Reference Tracking

The tracking algorithm chosen for our airship is a continuous version of linear quadratic regulator
in finite time(CLQR). Recalling the perturbation representation of the system dynamics (Eq. (9)),
in the reference tracking problem, we redefine the PI as:
Z T
 T 
J(δx(t0 ), δu(·), t0 ) = δxT (T )FL δx(T ) + δx (t)QL δx(t) + δuT (t)RL δu(t) dt (13)
t0

where FL is the final state weighting matrix, which is zeroed since the tracking error should converge
at the end of a task. The perturbation vectors in Eqs. (7-8) can be reused, with the modification
that u0 6= 0 is calculated by solving the inverse dynamics equation of the system, and the Riccati
equation now becomes a matrix differential equation, which can be solved by using the Runge-Kutta
integration backwards (off-line):

−K̇(t) = QL (t) − K(t)B0 (t)R−1 T T

L B0 (t)K(t) + K(t)A0 (t) + A0 (t)K(t) (14)

The control inputs of the tracking controller are given by:

u(t) = u0 (t) − G(t)x(t) = u0 (t) − R−1 T

L B0 (t)K(t)x(t) (15)

4 EXPERIMENTAL RESULTS

To compare the performance of the designed controllers and the PD controller, we first define the
performance index J in terms of φ, p and W .
 
1
φ(t) , cos−1 (trace (∆R(t)) − 1) (16)
2

p(t) , kp(t) − pr (t)k (17)

Z TX 6
W , u2i (t) (18)
0 i=1

In the above, φ denotes the error angle computed for the rotation ∆R = RR−1
r , where R is the

actual orientation matrix and Rr is the reference orientation matrix; p is the distance between the
actual airship center and the reference center; W is the squared norm of actuator inputs.

ISBN-xxx-y-zz-yyyyyyy-x 283
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

After implementing and testing the proposed controllers in simulation, the two controllers are tested
in real flights of the airship. Fig. 2 shows results from a typical maneuver in translational control
for set-point changes of 1 m in −Y direction. The slight misalignment in the initial pose of the
two controllers (LQR and PD) is caused by the imbalance. The results clearly demonstrate that
the LQR control results in much smaller overshoot and shorter settling time in p while maintaining
a lower rotational error. In terms of energy cost, LQR also shows advantages over PD due to the
smaller control efforts.

1
PD
p (m)

0.5
LQR

0
0 5 10 15 20 25 30 35 40

PD
φ (rad)

0.2

0.1 LQR

0
0 5 10 15 20 25 30 35 40
1
||u||2 (N2)

PD
0.5

LQR
0
0 5 10 15 20 25 30 35 40
Time (s)

Figure 2: Performance comparison between LQR and PD control: translational step response

A significant improvement of the proposed control designs over the PD controller is found to be in the
reference tracking task. We demonstrate this with an experiment with the airship is commanded
to follow a rectangular trajectory in the horizontal (XY) plane with no rotation. A constant
acceleration process is designed to smoothly drive the airship from stationary to the maximum
speed vm = 0.2 m/s and then to slow it down along each edge of the rectangle. The results of
continuous tracking this rectangular trajectory three times with the PD and CLQR controllers are
shown in Fig. 3a. The CLQR result is very close to the reference except for somewhat larger
errors along the right vertical edge of the rectangle where the ventilation system of the lab causes
disturbances. On the other hand, the result of PD controller shows a much higher overshoot and
deviation from the reference. Meanwhile, the attitude error of the PD control fluctuates during the
flight as shown in Fig. 3b. The average error of attitude tracking is 2-3 times higher than the result
of the CLQR control.

ISBN-xxx-y-zz-yyyyyyy-x 284
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

1 0.35
CLQR CLQR
PD PD
0.5 Reference 0.3

0
0.25

−0.5
0.2

φ (rad)
Y (m)

−1
0.15
−1.5

0.1
−2

−2.5 0.05

−3 0
−3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0 50 100 150 200 250
X (m) Time (s)

(a) Trajectory on the XY plane (b) Attitude error

Figure 3: Experimental comparison of the trajectory tracking results of the CLQR controller and
the PD controller

5 CONCLUSION

Two optimal controllers have been described in this paper to accomplish different tasks of flying
the holonomic indoor airship, namely hovering control, quasi-static set-point control and continuous
reference tracking. The final envisioned solution for the airship is a hybrid scheme which switches
among the controllers depending on the pilot command or the desired high-level task specification.
The proposed controllers have been validated in experiments. A translational maneuver and a more
complicated reference tracking tasks are demonstrated in this paper. The results show substantial
advantages of the designed controllers over the PD controller in terms of the speed of response,
tracking error and actuator effort. Our future work will focus on testing the capability of the
designed controllers and optimal state estimation for the feedback system.

REFERENCES
[1] J. Zufferey, A. Guanella, A. Beyeler, and D. Floreano, “Flying over the reality gap: From
simulated to real indoor airships,” Autonomous Robots, vol. 21, no. 3, pp. 243–254, 2006.

ISBN-xxx-y-zz-yyyyyyy-x 285
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[2] H. Kawamura, H. Kadota, M. Yamamoto, T. Takaya, and A. Ohuchi, “Motion design for
indoor blimp robot with PID controller,” Journal of Robotics and Mechatronics, vol. 17, no. 5,
pp. 500–508, 2005.
[3] T. Suzukia, “A memory-based pid controller for indoor airship robot,” Intelligent autonomous
systems 9: IAS-9, vol. 9, pp. 341–348, 2006.
[4] B. Bijnens, Q. Chu, G. Voorsluijs, and J. Mulder, “Adaptive feedback linearization flight
control for a helicopter uav,” in 2005 AIAA Guidance, Navigation, and Control Conference
and Exhibit; San Francisco, CA, pp. 1–10, American Institute of Aeronautics and Astronautics,
1801 Alexander Bell Drive, Suite 500, Reston, VA, 20191-4344, USA, 2005.
[5] T. Kim, K. Stol, and V. Kecman, “Control of 3 DOF quadrotor model,” Robot Motion and
Control 2007, pp. 29–38, 2007.
[6] F. Allgöwer and A. Zheng, Nonlinear model predictive control. Birkhäuser, 2000.
[7] B. Munson, D. Young, and T. Okiishi, Fundamentals of fluid mechanics. John Wiley & Sons,
1998.
[8] J. Davies, “The aerodynamics of golf balls,” Journal of Applied Physics, vol. 20, no. 9, pp. 821–
828, 2009.
[9] R. Clift, J. Grace, and M. Weber, Bubbles, drops, and particles, vol. 380. Academic press New
York, 1978.
[10] S. Kim, J. Crassidis, Y. Cheng, A. Fosbury, and J. Junkins, “Kalman filtering for relative
spacecraft attitude and position estimation,” Journal of Guidance Control and Dynamics,
vol. 30, no. 1, pp. 133–143, 2007.
[11] J. Pearson, “Approximation methods in optimal control,” International Journal of Electronics,
vol. 13, no. 5, pp. 453–469, 1962.
[12] J. Cloutier, C. D’Souza, and C. Mracek, “Nonlinear regulation and nonlinear H-infinity control
via the state-dependent Riccati equation technique. I- Theory,” in International Conference
on Nonlinear Problems in Aviation and Aerospace, Daytona Beach, FL, pp. 117–130, 1996.
[13] C. Mracek and J. Cloutier, “Control designs for the nonlinear benchmark problem via the state
dependent Riccati equation method,” International Journal of Robust and Nonlinear Control,
vol. 8, no. 4–5, pp. 401–433, 1998.
[14] T. Çimen, “State-dependent Riccati equation (SDRE) control: a survey,” in Proc. of the 17th
IFAC World Congress, Seoul, Korea, pp. 3761–3775, 2008.

10

ISBN-xxx-y-zz-yyyyyyy-x 286
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Attitude Control of a Quad-rotor based on Speed Sensing

in Brushless DC Motors

I. Gonzalez, S. Salazar H. Romero R. Lozano and J. Torres

UMI-LAFMIA 3175 CNRS, CINVESTAV

Av. Instituto Politécnico Nacional 2508. 07360 México D.F., México
[email protected], [email protected], [email protected]
[email protected], [email protected]

Abstract

Recently, the research in the area of autonomous miniature flying robots is growing thanks to
the development of new configurations and prototypes of Unmanned Aerial Vehicles (UAV). In
this paper, we introduce the problem of attitude and altitude control of a mini-quadrotor UAV
in hovering-mode to indoor and outdoor applications. Therefore, we have implemented a control
strategy based on speed sensing in each brushless motors. As result, the miniature aerial vehicle
has a good and robust stabilization on the horizontal plane. This paper also describes a control
strategy to stabilize the quad-rotor using a simple linear control algorithm based on PD controller
as well as the speed sensing scheme. Finally, the experimental results of speed sensing control for
stabilizing the quad-rotor at hover are presented.

keywords: Embedded flight system, Real-Time Stabilization, Motor Feedback Control, Quadrotor, UAVs

1 INTRODUCTION
Nowadays, one of the most appealing topics among the research control community is the application of
modern control theory to Unmanned Aerial Vehicles (UAV). The operation and development of UAVs
has increased during the last few years, due to the wide range of applications such each aerial inspection
in constrained locations, work in hazardous environments, forest fire detection, aerial photography and
so on. For these reasons UAVs aroused a great interest in the research and academic circles in recent
years, using it as a platform for many applications and for pure academic research [7], [8], [12]. Attitude
and altitude control in a mini-UAV involves research in various areas such as digital filtering, estimation
of angular position, data fusion of sensors, and so on. Flexibility provided by these aircraft to be compact
is a great advantage in small locations where you want to arrive. However, these characteristics limit the
load that can carry these vehicles due to the small size that characterizes them. It meanwhile, the aerial
vehicle considered in developing this work is classified as a mini-flying robot which payload is limited to

ISBN-xxx-y-zz-yyyyyyy-x 287
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

less than 1 kg. Therefore implementation of a light and small Inertial Measurement Unit together with
the altitude sensor is required in order to have an embedded UAV platform. Advantages of implementing
an embedded system (i.e, IMU-Inertial Measurement Unit with a Microcontroller) are several including
low power consumption and low weight. The performance of the speed sensors implemented in the
brushless DC motors is shown in a practical application. Now, the paper is organized as follows: section
2 presents the dynamic model of the Quad-rotor. The linear control PD is described in section 3. The
architecture of the prototype are described in section 4. The speed sensor system on brushless DC
motors is presented in section 5. Then, the experimental results of the attitude and altitude control
using speed sensing are described in section 6. And finally, we present some conclusions about our work,
in the section 7.

2 DYNAMICAL MODEL
As you known, Quad-rotors are mechanically simpler because it does not require a complicated mechan-
ical structure like conventional helicopters and potentially it is easier to be controlled. A Quad-Rotor
Flying robot is controlled by varying the angular speed of each motor. The front and rear motors rotate
counterclockwise (CCW), while the other two motors rotate clockwise (CW). Gyroscopic effects and
aerodynamic torques tend to cancel in trimmed flight. The main thrust is the sum of the thrusts of
each motor, the pitch torque is a function of the difference f1 − f3 , the roll torque is produced by the
difference f2 − f4 , and the yaw torque is the sum of τM1 + τM2 + τM3 + τM4 (see Figure 1), where τMi
is the reaction torque of motor i due to shaft acceleration and the blades drag. Let ΓI = {iI , jI , kI } be
the inertial frame, ΓB = {iB , jB , kB } denote a set of coordinates fixed to the rigid aircraft as is shown in
Fig. 1. Let q = (x, y, z, ϕ, θ, ψ)T ∈ R6 = (ξ, η)T be the generalized coordinates vector which describe
the position and orientation of the flying machine, so the model could be separated in two coordinate
subsystems: translational and rotational. They are defined respectively by

• ξ = (x, y, z)T ∈ R3 : denotes the position of the aerial vehicles center of mass relative to the
inertial frame ΓI .

• η = (ϕ, θ, ψ)T ∈ R3 : describe the orientation of the aerial vehicle and (ϕ,θ,ψ) are the three Euler
angles: roll, pitch and yaw, respectively.

The dynamic model is obtained using the Euler-Lagrange approach. We can decompose the equations
into translational and rotational displacement. As first step, we obtain the Lagrangian of the aerial
vehicle, which is given by the following equation

L(q, q̇) = (Ttrasl + Trot ) − U (1)

• Ttrasl : It is the translational kinetic energy of the quad-rotor.

1 ˙T ˙
Ttrasl = mξ ξ (2)
2

ISBN-xxx-y-zz-yyyyyyy-x 288
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 1: Inertial and Body frame of the four-rotor aircraft.

• Trot : It is the rotational kinetic energy of the quad-rotor.

1 T
Trot = η̇ Jη̇ (3)
2

• U : It is the potential energy of the quad-rotor.

U = mgz (4)

In the above equations we have that m denotes the mass of the vehicle, J represents the inertia
matrix which is symmetric and defined positive, g denotes acceleration due to the gravity and z is the
altitude desired of the vehicle. Following the approach, the Lagrangian is given by
1 ˙T ˙ 1 T
L(q, q̇) = mξ ξ + η̇ Jη̇ − mgz (5)
2 2
The model for the full four-rotors aircraft dynamics is obtained from the Euler-Lagrange equations
with external generalized force  
d ∂L ∂L FI
− =   (6)
dt ∂ q̇ ∂q τ
where FI defines the translational force applied to the aerial robot due to the control inputs and relative
to the frame ΓI and τ is the generalized moments vector. Doing the current calculations, the complete
dynamic model of Quad-rotor is

FI = mξ¨ + mg
( T ) (7)
τη = Jη̈ + J̇η̇ − 1 ∂
2 ∂η η̇ Jη̇

Now, we have a vector FB can be expressed in the inertial frame FI as follows

FI = RB→I F B (8)

ISBN-xxx-y-zz-yyyyyyy-x 289
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

where RB→I is the rotation matrix. Finally, the dynamic model of our miniature aerial vehicle is given
by      
mẍ u sen θ 0
     
     
 mÿ  =  −u sen ϕ cos θ + 0  (9)
     
mz̈ u cos ϕ cos θ −mg
   
ϕ̈ τ̃
   ϕ 
   
 θ̈  =  τ̃θ  (10)
   
ψ̈ τ̃ψ

3 CONTROL STRATEGY
In this section we present a control law for the attitude stabilization and control altitude of the four-
rotor aircraft at hover. The goal of the control law is to bring the quad-rotor to rest at the zero attitude
(aligned with the inertial frame). Furthermore, we are able to propose a simple linear control PD
strategy. The PD controller is relatively simple and each one of the control inputs can operate in either
manual or automatic mode. In addition, one of the main advantages of this algorithm is that the gains
are easy tuning and implementing on a microcontroller. Finally, the collective input u is essentially used
to reach a desired altitude value.

3.1 Altitude control

To altitude control, we consider the dynamic in the z-axis from (9),

1
z̈ = (u cos ϕ cos θ − mg) (11)
m
using the following control input
mcz + mg
u= (12)
cos ϕ cos θ
where the variable cz is a PD controller, which is defined by

cz = −kDz ż − kPz (z − zd ) (13)

with kDz ,kPz > 0 and zd the altitude desired. From (12) we can see that control u is necessary bounded
in: − π2 < ϕ < π
2 and − π2 < θ < π
2. Then, using (12) into (11) we get

z̈ = cz (14)

3.2 Attitude control

Attitude control is the most important part of control system, it stabilizes the orientation of the heli-
copter at a desired value. Therefore, the main aim of the attitude control is to stabilize the vehicle in
the desired orientation ηd = (ϕ, θ, ψ)T = (0, 0, 0)T . We first consider the non-linear model

M (η)η̈ + C(η, η̇)η̇ = τη (15)

ISBN-xxx-y-zz-yyyyyyy-x 290
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

with η = (ϕ, θ, ψ)T . Now, we propose the following Lyapunov candidate function
∑3
1 T
V (η, η̇) = η̇ M (η)η̇ + Kai ln(cosh(Kpi ηi )) (16)
2 i=1

where, ηi represent the i-th element of the vector η, Kai and Kpi represent the i-th diagonal element of
positive definite matrices Ka and Kp ∈ R3x3 , respectively. The Lyapunov candidate function V (η, η̇) ≥ 0
since M (η) > 0 and ln (cosh (·)) is a positive definite increasing function. Differentiating (16) with
respect to time, we obtain
∑3
1 T
V̇ (η, η̇) = η̇ Ṁ (η)η̇ + η̇ T M η̈ + η˙i Kai Kpi tanh(Kpi ηi ) (17)
2 i=1

then (17) may be rewritten as

1 T
V̇ (η, η̇) = η̇ Ṁ (η)η̇ + η̇ T M η̈ + η˙i T Ka Kp tanh(Kp η) (18)
2
with
T
tanh = (tanh (kp1 ϕ) , tanh (kp2 θ) , tanh (kp3 ψ))

Considering the attitude control as a regulation problem (η = ηd ) and replacing (15) into (18) we get
· ( ·
) ·T (1 · ( · )) · ·T
V η, η = η M (η) − C η, η η + η (τη + ka kp tanh (kp η)) (19)
2
using the property proposed in [3], we have
( )
1
uT Ṁ (η) − C(η, η̇) u = 0 ∀ u ∈ R3 (20)
2
leads to
· ( ·
) ·T ( )
V η, η = η τη + ksp tanh (kp η) (21)
k sp
where, it was considered ka = kp . Now, we introduce the following saturated PD control:
( ·)
τη = −ksp tanh (kp η) − ksv tanh kv η (22)

where Ksp and Ksv ∈ R3×3 are positive diagonal matrix. Then, substituting (22) in (21) leads to
· ( ·
) ·T
( ·)
V η, η = −η ksv tanh kv η (23)

Finally, it follows that V̇ < 0. Therefore, the origin is stable and the solutions η(t) and η̇(t) are
bounded.

4 PLATFORM ARCHITECTURE
In this section we present the prototype built of the Quad-rotor. The overall system structure is shown
in Fig. 2. It basically consists of four-rotors arranged in a Microkopter’s aluminium frame, Measuring
stage (Sensors), Control stage (Embedded system: DSP1 ), Tx/Rx stage (Radio-control ) and Power stage
(Li-Po batteries). Futhermore, this experimental platform serves as didactic prototype in the field of
miniature aerial vehicles research because we can test different control laws to verify the efficiency of
these controls on flying.
1 Digital Signal Processing

ISBN-xxx-y-zz-yyyyyyy-x 291
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 2: The control stages implemented on Quad-rotor.

4.1 Hardware structure

First of all, the structure of the Quad-Rotor flying robot is modified from a popular quad-rotor RC toy,
the Mikrokopter [14]. The main frame (crosshead ) of the Quad-Rotor is constructed of aluminum tubes.
They are connected together by a center plastic square hub which is available as a spare part for the
Mikrokopter. The motors and rotors used for the Quad-rotor are available at any RC hobby shop as a
spare part for the RC aircraft. The motors chosen are the Robbe Roxxy BL-Outrunner 2824-34 because
they have low vibration and noise while the rotors used is EPP1045 counter rotating propellers, one
propeller rotates clockwise CW and the other propeller rotates counter clockwise CCW (1 left hand &
1 right hand rotation). The length is 10 inches and the pitch is 4.5 inches per revolution. The following
figure sums up the entire structure described above (see Figure 3).

Figure 3: The Quad-rotor frame set (Mikrokopter) and miniature prototype.

4.2 Measuring stage

This stage contains a homemade Inertial Measurement Unit (IMU) based on Sparkfun’s Analog Combo
Board Razor-6DOF Ultra-Thin (see Figure 4). This board removes the high-pass filters which were
creating some issues with the output, that includes three surface mount devices: the ST’s LPR530AL

ISBN-xxx-y-zz-yyyyyyy-x 292
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(pitch and roll) and LY530ALH (yaw) gyros, as well as the popular ADXL335 triple-axis accelerometer,
to give you six degrees of measurement on a single, flat board. The signals from IMU are sampled
( )
using a 12-bit analog-to-digital converter. The IMU measures three angular rates ϕ̇, θ̇, ψ̇ and two

Figure 4: IMU-homemade (left) and Compass (right) to measure and control the orientation of the
Quad-rotor.

angular positions (ϕ, θ) after to measure the yaw angle (ψ) we use a compass (CMPS03) (see Fig. 4).
The CMPS03 can produce a unique number to represent the direction the flying robot is facing. It uses
the Philips KMZ51 magnetic field sensor, which is sensitive enough to detect the Earths magnetic field
(see Figure 4). The output from two of them mounted at right angles to each other is used to compute
the direction of the horizontal component of the Earths magnetic field [15]. We can also measure the
position and velocity of Quad-rotor directly from the GPS receiver that has a sampling rate of up to 5
Hz.

4.3 Control stage

A mini-core TMS320F2808-DSP (32-Bit Digital Signal Controller with Flash) is employed at this stage
to control the attitude and altitude of our Quad-rotor. It have the following main features: operates
at 100 Mhz (10-ns Cycle Time), with 256K bits serial I2C EEPROM memory, low-power (1.8-V Core,
3.3-V I/O), 12 PWM channels (16-bit resolution), 16 ADC channels (12-bit resolution), 2 serial ports, 4
input-capture channels, 3 timers (32-bit resolution) and 1 I 2 C port. It is programmed using the C2000
Code Composer from Texas Instruments. The DSP runs the control algorithm in real time to stabilize
the Quad-rotor managing the information provided by the inertial measurement unit (IMU) and the
GPS sensors and sends the control signals to the four motors. The following block diagram sums up the
entire process of the control stage (see Figure 5).

4.4 Tx/Rx stage

We use a commercial 7-Channel 2.4GHz radio control system (RC) from FUTABA, whose main signal
is captured and decoded in such a way that it allows the user can introduce an external input to have
either manual and/or automatic operation. The radio can be used for trimming small offset deviations.
This allows that position feedback is triggered by an external RC signal from radio.

ISBN-xxx-y-zz-yyyyyyy-x 293
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 5: Embedded control system based on TMS320F2808-DSP.

4.5 Power stage

This stage is responsible for supplying all the stages that made up the helicopter. We use a stable +5V
and +3.3V supplies for the digital electronic devices of our flying robot. They have a LM7805 and
LF33CV regulators are connected to the battery through a 1A diode. The purpose of the diode is to
help protect the regulator from voltage drops that can be caused by the Brushless DC motors when
they draw large amounts of current. Since driving the speed motors requires a high current demand, we
decided to use lithium-polymer batteries (Li-Po). These kind of batteries have a wide range of power
management and an good advantage is that they are reasonably small and not too heavy. The features
of the Li-Po battery that we use are: 1350mAh, 11.1V (3-Cell pack). Quad-rotor have a flight time of
approximately 5 to 10 minutes with this type of Li-Po battery.

5 Speed sensing implementation

The speed sensing hardware system to perform an autonomous flight by controlling the orientation of
each Brushless DC motors is shown in Fig. 6. It consist of a Speed sensor block, a PD controller block,

Figure 6: Block diagram of the Speed control using PD method.

an I 2 C control block, a Power stage block and the Brushless DC motor block. The actual motor speed
value is captured through CNY70 sensor. The Frequency to Voltage Converter (FVC) takes the output

ISBN-xxx-y-zz-yyyyyyy-x 294
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

signal (pulse train) from CNY70 and converts this signal to an equivalent value of voltage with respect
to the input frequency. Then motor speed converted into voltage is compared to the desired speed, and
a correction signal is generated, if necessary.

e(k) = r(k) − y(k) (24)

where e(k) actual error value, r(k) is the set speed value:=accelerometer and gyro offset and y(k) is the
actual motor speed value. Errors are used in PD control calculation afterwards DSP is programmed to
measure the motor’s speed through FVC and decide the direction in which it is drifting. A proportional-
derivative (PD) routine produces one output signal control (25).

( )
kd ∗ [e (k) − e (k − 1)]
u (k) = kp ∗ e (k) + (25)
Ts

where u (k) is the input control to the brushless DC motors and Ts is DSP’s sampling period. The
gains of both kp and kd are chosen to provide a quick response with little instability. The motor’s
load, inertia and torque performance are the factors that determine the PD gain constants. Finally,
PD controller output generates I 2 C signals to the motors drivers. Fig. 7 shows a printed circuit board
(PCB) of Brushless DC motors speed sensors, consisting of a Frequency-to-Voltage converter, op-amps
and low-pass filters (RC) and the speed sensor fixed close to brushless DC motor to measure its speed.

Figure 7: Speed sensor (reflex ) PCB implemented in the Quad-rotor.

6 EXPERIMENTAL RESULTS
In this section we present the experimental results that validate the performance of the Quad-rotor
during autonomous hover flight. The control gains were adjusted to obtain a fast system’s response
but avoiding mechanical oscillations as much as possible. Now, we present some experimental results
obtained by applying the algorithms proposed control.

ISBN-xxx-y-zz-yyyyyyy-x 295
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 8: a), b), c) show the behavior of the quad-rotor in hover in the face of disturbances in roll, pitch
and yaw (attitude) respectively and d) show the behavior of the quad-rotor onto the vertical (altitude).

The Fig. 8 show the behavior of the quad-rotor in hover in the face of disturbances and it also show
the behavior of the quad-rotor onto the vertical, equally in the face of perturbations.

7 CONCLUSIONS
In this paper, we have presented the hardware implementation of a speed sensor system for measur-
ing and controlling the angular velocity on Brushless DC Motors of a Quad-rotor, by combining an
embedded-PD controller and an DSP on-board. In addition, it has successful been flown by a human
operator (see Figure 9). The dynamic model of the miniature flying robot (Quad-rotor) was obtained
using an Euler-lagrange approach. The linear control algorithm embedded-PD (Proportional-Derivative)
is used to stabilize the attitude and altitude dynamic system around the origin where the input control
satisfies some given disturbances, locally. The brushless DC motors or better known as: BLDC are a
good option in the construction and development of a Quad-rotor because this kind of motors are clean,

10

ISBN-xxx-y-zz-yyyyyyy-x 296
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

fast and efficient, and creates no sparks, making it a safer choice in explosive environments. The linear
current/torque relationship results in smoother acceleration. In speed sensing control, it has a response
very fast to handle the attitude desired of the Quad-rotor. In addition, this motor type has a longer
life than typical Brushed DC motors. Finally, the experimental results show that speed control on
brushless DC motors works satisfactorily to stabilize the Quad-rotor on attitude for indoor and outdoor
applications.

Figure 9: Photo of the Quad-rotor at hover flight.

In order to show the Quad-rotor performance, a video can be found in the following link:

http://www.youtube.com/watch?v=y870AsGkTS4

REFERENCES
[1] I. Fantoni, and R. Lozano, Non-linear control for underactuated mechanical systems, Communica-
tions and Control Engineering Series, Springer-Verlag. London (2001).

[2] Etkin, B., Duff, R., Dynamics of flight, John Wiley and Sons, Inc. New York (1959).

[3] R. Kelly, and V. Santibanez, Control de Movimiento de Robots Manipuladores, Prentice Hall,
Madrid (2003).

[4] P. Castillo, R. Lozano and A. Dzul, Modelling and Control of Mini-Flying Machines, Springer-
Verlag. London (2005).

[5] H. Goldstein, C.P. Poole and J.L. Safko, Classical Mechanics, Addison-Wesley Publishing Company,
Inc., Massachusetts (1983).

[6] Fantoni, I. and Lozano, R., European Journal of Control, 7, 328-348 (2001).

[7] J. Escareo, S. Salazar-Cruz and R. Lozano, Embedded control of a four-rotor UAV, in Proc. Amer-
ican Control Conference, Minnesota, pp.3936-3941 (2006).

11

ISBN-xxx-y-zz-yyyyyyy-x 297
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[8] Samir Bouabdallah and Roland Siegwart, Full Control of a Quadrotor, in IEEE/RSJ International
Conference on Intelligent Robots and Systems, San Diego, CA, pp.153-158 (2007).

[9] S. Bouabdallah et al., Design and control of an indoor micro quadrotor, in Proc. (IEEE) Interna-
cional Conference on Robotics and Automation (ICRA), New Orleans, (2004).

[10] P. Castillo, A. Dzul et R. Lozano, Real-time stabilization and tracking of a four rotor mini rotorcraft,
in IEEE Transactions on Control Systems Technology, Vol. 12, No. 4, pp.510-516 (2004).

[11] Chang-Sun Yoo, and Iee-Ki Ahn, Low cost GPS/INS sensor fusion system for UAV navigation, in
Proc. Digital Avionics Systems Conference, Vol. 2, pp.8.A.1-1 - 8.A.1-9, (2003).

[12] S. Salazar-Cruz, J. Escareo, D. Lara and R. Lozano, Embedded control system of a four-rotor UAV,
in International Journal of Adaptive Control and Signal Processing, volume 21, pp.189-204 (2007).

[13] Kong Wai Weng and Abidin, M.S.B., Design and Control of a Quad-Rotor Flying Robot For
Aerial Surveillance, in 4th Student Conference on Research and Development (SCOReD), Malaysia,
pp.173-177 (2006).

[14] MikroKopter
http : //www.mikrokopter.de/ucwiki/en

[15] CMPS03 - Compass Module

http : //www.robot − electronics.co.uk/htm/cmps3tech.htm

12

ISBN-xxx-y-zz-yyyyyyy-x 298
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

An INS / GPS Integrated Navigation Filter for a UAV Based on

a Mobile Phone

B. Fons-Albert J. Vila-Carbó Enrique Hernández-Orallo J. Tur-Massanet

Universitat Politècnica de València

[email protected], [email protected]

Abstract

This paper shows the feasibility of implementing and execute with real-time performance a
complex INS/GNS integrated filter for attitude and navigation estimation on a small mobile device
that can be equipped on a UAV. The proposed integration is based on a Kalman filter and improves
the performance of both INS and GPS systems: the attitude estimation of the INS is made more
accurate during turns and sudden accelerations while the frequency of the GPS estimations can be
increased by making predictions with the INS. The paper shows that a good accuracy in attitude
estimation and navigation can be obtained using the low cost sensors of mobile devices.

keywords: Inertial Navigation Systems (INS), Global navigation, GPS, Kalman filtering, mobile devices.

1 Introduction
The proliferation of small devices equipped with inertial sensors, GPS, cams and various connectivity
facilities (3G, wifi) has made highly attractive to use these devices in UAV control and navigation.
The UAV (Unmanned Aerial Vehicle) project that we are involved at the UPV uses an Apple iPhone
for the control of two UAV prototypes developed for research purposes: a quadrotor (fig. 1) and a delta
wing. The iPhone terms of service prohibits the device being used to control remote vehicles and being
used for emergency or life-saving purposes (iPhone SDK, Section 3.3.7). However, this is not the only
nor the most serious safety limitation or regulations that UAVs will have to face in the future if they
are widely used. Despite this inconvenient, we have found it very useful for rapid prototyping since
it has a very powerful and compact hardware and also a highly available and excellent cross Software
Development Kit (SDK).
This paper only concentrates on the attitude and navigation filters that have been designed for
our UAV prototypes. Several INS/GPS integration methods have been proposed in the bibliography,
from loosely-coupled to tightly-coupled implementations: see [2] [3] [4] [6], among others. This paper
introduces improvements on loosely-coupled schemes based on low-cost MEMs devices.

1
ISBN-xxx-y-zz-yyyyyyy-x 299
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2 System description

This section presents a loosely-coupled INS/GPS

integration method for attitude, position, and
speed computation. The implementation is based
on the iPhone 4, although it can be easily ported
to any embedded system, since no iPhone specific
facility for INS or GPS filtering has been used; only
the raw access to the sensors. Furthermore, some
iPhone limitations, like the maximum frequency of
sensor reading (50 Hz) or the sensor accuracy, could
be easily overcome on other platforms.
Figure 1: Quadrotor prototype of the UPV As a matter of fact, we have ported it to an
LPC2468 (ARM7 based) microcontroller.

Figure 2 shows the system structure where the INS is composed by the three axis gyroscope /
accelerometer / compass that equip the iPhone They are low cost MEMs technology sensors. The
iPhone GPS is also a low cost unit that provides location, speed and course. The software design to
compute attitude and position has two main blocks: the Attitude Kalman Filter, which provides the
Euler angles, and the Navigation Kalman Filter, which provides position and speed. Getting a high
accuracy in attitude estimation has shown to be key for making the navigation filter feasible. This has
motivated a lot of efforts for improving attitude estimation.
The implementation of this filters on the iPhone uses the graphical interface of fig. 2. It has three
main views: the artificial horizon, the map view (not shown) and the control panel, which allows to
make real-time transmissions of data sensors and filter results to a host computer using the wifi.

2.1 Attitude filter

Attitude estimation is the result of filtering two independent measures: the gyroscopes on one hand,
and accelerometers and magnetometers on the other. The attitude filter is an Extended Kalman Filter
filter with seven states: the four quaternion components plus the three gyros biases. The quaternion
has been chosen as the attitude representation because it highly improves the accuracy of floating point
calculations and avoids singularities of trigonometric functions.
Computing the attitude using the gyros basically consists of integrating the angular speeds Ωb that
they measure. This is performed using the quaternion propagation eq. [1]: q̇ = 21 q×Ωb . The integration
accuracy relies on determining the gyros bias [4], since its integration can lead to non-bounded errors
in attitude. For this reason, the gyros biases have been chosen as a part of the filter state.
Computing the attitude using the accelerometers and magnetometers is based on finding out the

2
ISBN-xxx-y-zz-yyyyyyy-x 300
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 2: Block diagram of the system and iPhone Graphical User Interface

angles of the gravity vector ~g and the geomagnetic field vector m

~ with the Body Fixed Frame (BFF) axis
system. Based on this, the attitude angles (φ, θ, ψ) can be calculated using well known formulas [1] [2]:

φf = atan2 (−gy , −gz ) mE = my cos φ + mz sin φ

 q 
θf = atan2 gx , gy2 + gz2 mN = mx cos θ + my sin φ sin θ + mz cos φ sin θ
ψ = arctan(mE , mN )

The problem of attitude computation using the accelerometers is that they do not only measure the
gravity vector, but the total acceleration. That includes so called effective forces, like the ones due to
tangential or centripetal accelerations. These forces make the above attitude computation erroneous.
Kalman filtering of the attitude estimation of the gyros on one hand, and the accelerometers /
magnetometers on the other, produces a sensor fusion that improves the estimation. However, in our
experience the attitude estimation still results inaccurate, especially during turns and accelerations.
Solving this requires to compensate the accelerometers readings with the effective forces, so the gravity
vector can be correctly evaluated. We evaluate the effective forces using the GPS.
Tangential accelerations can be obtained from the speed difference of two consecutive measures of the
GPS along a time interval. This acceleration is in the direction of motion, and needs to be transformed
through the rotation matrix to the BFF. Keep in mind that if the longitudinal axis of the device remains
aligned with the direction of motion or forming a given angle, then the rotation between the reference
systems is straightforward; otherwise it should be rotated an additional angle, which is the angle between
the direction of motion and the longitudinal axis.
Estimating centripetal accelerations involves more measures. They can be calculated through the
following expression: ac = Ωb × v b , where Ωb is the angular speed vector obtained from the gyros and

3
ISBN-xxx-y-zz-yyyyyyy-x 301
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

v b is the speed vector of the BFF obtained from the GPS. Based on the speed, course and attitude
readings, the components of v b can be obtained. This components have to be first rotated from the
NED (North East Down) to the BFF axis and then, they can be used to calculate the cross product of
the previous equation.
The former computed tangential and centripetal accelerations are subtracted from the readings of
the accelerometers for the correct estimation of the gravity vector.

2.2 Navigation filter

The navigation filter computes the speed and position concurrently with the attitude filter. It was
originally a six states filter composed by the three components of the speed and the three of the position.
It was later extended to twelve, in order to include an error model for compensating the accelerometers
bias and gain factors. The prediction of these magnitudes basically involves the integration of the
accelerations due to effective forces. The Kalman filter measurements are the GPS readings. The
navigation equations of the system model for the filter prediction relate the derivatives of speed and
position to the filter state variables and the inputs [1] [2]:
2
vN vD −vE tan L
L̇= Rv0N+h v̇N =fN + R0 +h
˙
l= vE
v̇E =fE + vE
(R0 +h) cos L R0 +h (vD + vN tan L)
2 2
vN +vE
ḣ=−vD v̇D =fD − R0 +h +g

where L, l and h are the latitude, longitude and height, vN , vE and vD are the North-East-Down
(NED) components of the speed, R0 is the radius of the Earth and the terms fN , fE and fD are the
NED components of the specific forces. They are calculated using the accelerometers readings and
transforming them to the NED system through the rotation matrix computed in the attitude block.

3 Results
This section shows the results of the attitude and navigation filters shown in section 2.
Effective forces : the first important result is the feasibility of accurately calculating effective forces
(tangential and centripetal acc.) based on the GPS. This is used to compensate the accelerometers
readings in order to compute the gravity vector. It improves quite a lot the attitude determination
in turns and sudden accelerations. Figure 3 (1) shows the tangential acceleration measured using the
GPS and the measures of the motion oriented accelerometer. It can be seen that there is a vertical
displacement between both signals which is due to the gravity component in that direction. Some delay
can also be appreciated, since GPS calculations are delayed by one iteration. Figure 3 (1) shows similar
results for the centripetal acceleration of the y axis of the BFF.
Attitude determination: The calculation of the roll (φ) and pitch (ψ) angles with the proposed
attitude filter is really accurate, once the compensation of the effective forces is performed. The most
challenging and difficult issue is the calculation of the yaw angle ψ. Figure 4 (1) shows the comparison for

4
ISBN-xxx-y-zz-yyyyyyy-x 302
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Tangential acceleration
Centripetal acceleration
0.4
Accelerometer
0.3 Accelerometer
GPS
GPS
0.3

0.2

0.2

0.1
acceleration (g)

acceleration (g)
0.1

0
0

−0.1 −0.1

−0.2 −0.2

−0.3
40 50 60 70 80 90 100 110 0 50 100 150 200 250 300
time (s) time (s)

Figure 3: (1): Comparison of tangential acc. measured using the GPS and the motion oriented ac-
celerometer. (2): Comparison of centripetal acc. measured using the GPS and the y (BFF) axis accel.

this angle using the the GPS and the attitude filter. Except for the initial moments, when the vehicle
is stopped, and the GPS readings are incorrect, both estimations match almost perfectly. Another
remarkable result of the attitude filter is its ability to calculate an stationary value for the bias of the
gyros without any initial calibration (figure 4 (2)).
−3 Bias gyroscopes
Course x 10
10
200
GPS
Kalman filter
150 8

100 6

bx
50
4 by
course (deg)

bias

0
bz
2

−50

0
−100

−2
−150

−4
−200 0 50 100 150 200 250 300 350 400 450
0 50 100 150 200 250 300 350 400 450
t (s) t (s)

Figure 4: (1): Course measured using the GPS and the Attitude Kalman filter. (2): Bias of the gyros.

Position determination The proposed navigation filter allows to make the position calculation at
a higher frequency (50 Hz) than the GPS (1 Hz) and still estimate it with reasonable accuracy during
short periods of GPS unavailability. Figure 5 (1) shows the path used for one of the experiments.
The most innacurate measure with the GPS is always the altitude. Figure 5 (2) shows a comparison
between the GPS reading and the improved estimation using the filter. Experiments show that the
GPS readings are better at higher speeds and when no turns are performed. Despite that, the IMU
significantly improves the GPS readings in turns and sudden accelerations or decelerations. On the
other hand the filter also allows to decompose the GPS speed and get its values in each navigation axis
(NED). Figure 5 (2) shows an example of the N component which is very good, except for the initial
moment when the gyros bias is still not compensated and the attitude calculation is not reliable.

5
ISBN-xxx-y-zz-yyyyyyy-x 303
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

North speed
Path 4
Latitude
39.346 Kalman filter
Kalman filter GPS
2
39.344 39.3425 GPS

39.342 39.342 0

39.34
39.3415 −2

north speed (m/s)

latitude (deg)
39.338
39.341
latitude (deg)

−4
39.336
39.3405
39.334 −6
39.34
39.332
−8
39.3395
39.33

39.339 −10
39.328

39.3385 −12
39.326 100 110 120 130 140 150 160 170 180 190 0 50 100 150 200 250 300 350 400 450
−0.492 −0.49 −0.488 −0.486 −0.484 −0.482 −0.48 −0.478 −0.476 −0.474
longitude (deg) t (s) t (s)

Figure 5: (1): Path of the experiment. (2): Height using the Kalman filter and GPS (3): North
component of the speed using the Kalman filter and GPS.

4 Conclusions
The paper has shown the feasibility of implementing a complex INS/GNS integrated attitude and
navigation filter on a small device that can be equipped on a UAV and getting a reasonable accuracy.
One of the most interesting conclusions is that the in order to be able to improve the GPS results
with the INS, it is required a highly accurate attitude determination, specially during turns and sudden
accelerations. This improvement in attitude only can be achieved by estimating the effective forces with
some other sensor which, in this case is the GPS itself. Another important problem that was solved was
the model errors for the bias compensation of the gyros (attitude filter) and accelerometers (navigation
filter). In both cases they were included as part of the filter state and it proved to be effective in
its calculation. Finally, the accuracy of the attitude determination can be considered excellent. The
navigation accuracy is also high, although in the altitude determination is somewhat lower.

References
[1] Mohinder, S. et al. Global Positioning Systems, Inertial Navigation, and Integration, Wiley.

[2] Rönnbäck, S. Development of a INS/GPS navigation loop for an UAV, Lulea Univ. of Technology, Feb.2000.

[3] Kingston, D.B. and Beard, R.W. Real-Time Attitude and Position Estimation for Small UAVs Using Low-
Cost Sensors, Department of Electrical and Computer Engineering, Brigham Young University, Provo,
Utah.

[4] El Hadri A. and Benallegue A. Attitude estimation with gyros-bias compensation using low-cost sensors,
Joint 48th IEEE Conference on Decision and Control and 28th Chinese Control Conference Shanghai, P.R.
China, December 16-18, 2009

[5] Jung, B. and Tsiotras, P. Inertial Attitude and Position Reference System Development for a Small UAV,
Georgia Institute of Technology, Atlanta, GA, 30332-0150

[6] Ferrer Mnguez , G. Integracin Kalman de sensores inerciales INS con GPS en un UAV, RSLab TSC UPC
Abril, 2009

6
ISBN-xxx-y-zz-yyyyyyy-x 304
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

A test bed for rapid flight testing of UAV control algorithms

Tiago Oliveira ∗ , Gonçalo Cruz † , Eduardo R. B. Marques ‡

§
and Pedro Encarnação
∗
Academia da Força Aérea Portuguesa, Portugal, [email protected]
†
Academia da Força Aérea Portuguesa, Portugal, [email protected]
‡
Faculdade de Engenharia da Universidade do Porto, Portugal, [email protected]
§
Faculdade de Engenharia da Universidade Católica Portuguesa, Portugal, [email protected]

Abstract

This paper presents a test bed for rapid flight testing of control algorithms for unmanned aerial
vehicles (UAVs), framed in the Portuguese Research and Development Program on Unmanned Aerial
Vehicles (PITVANT). The test bed can be used to validate UAV control algorithms at the kinematic
level, being the aircraft dynamics controlled by commercially available autopilots. It supports all
stages from numerical simulations to flight testing (with control on the ground) using the same
framework. It allows testing multi-vehicle operations and the integration of on board systems with
data from and to those systems transmitted through a single communications data link. A common
console integrates all information and controls every aspect of multi-vehicle missions. The paper
describes the architecture of the test bed and its deployment, in particular its operational framework,
software and hardware components and user interaction.

keywords: test bed, UAV, control algorithms, hardware-in-the-loop simulation, software architecture

1 Introduction
The Portuguese Research and Development Program on Unmanned Air Vehicles (PITVANT), supported
by the Portuguese Ministry of National Defense, has been carried out since 2009 by the Portuguese Air
Force Academy (AFA) and the Faculty of Engineering of the University of Porto (FEUP). The main
objectives of this seven-year project are the exploration of small platforms and the development of new
technologies and new concepts of operation, with an emphasis on the exploration of cooperative control
systems for teams of autonomous aerial, marine and land vehicles. The envisioned applications of the
systems to be developed are forest and coastal patrolling, military aerial surveillance, search and rescue
and support/tracking of land or marine vehicles.
Currently, the AFA owns a variety of platforms with different wingspans (from 2 m to 7 m) and max-
imum takeoff weights (from 2Kg to 100Kg) built from scratch, allowing for the testing of technological
systems on board and different control algorithms.

1
ISBN-xxx-y-zz-yyyyyyy-x 305
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Given the early stage of the project, testing of the control algorithms has been carried out on
small platforms. Namely, the ANTEX-M X02 platforms (Figure 1) have been widely used due to their
convenient size/payload ratio Table 1).

Maximum takeoff weight 10 Kg

Wingspan 2.415 m
Payload 4 Kg
Maximum Speed 100 Km/h
Autonomy 1.5 h

Figure 1: ANTEX-M X02. Table 1: ANTEX-M X02 - Main features.

The main goal of the work reported here was to develop a software and hardware architecture that
allows rapid hardware-in-the-loop simulations and subsequent flight tests of different high level multi-
vehicle control systems.
The standard approach with UAV control is to assume that the vehicle has an off-the-shelf inner loop
controller that accepts references at kinematic level (angular rates and linear velocities) and generates
the UAV control signals necessary to follow those references in the presence of model uncertainty and
external disturbances, like wind. See for example [1, 2, 3, 4, 5, 6, 7]. Outer loop control laws are thus
derived using a kinematic model of the vehicle and provide the references to the inner control loop.
This approach permits the development of control algorithms based on simpler models. The use of an
off-the-shelf autopilot also adds to the operational safety of the aircraft since it is always possible to
switch to autopilot control mode or even to remote control mode, if a potentially dangerous situation
is detected. The downside is that it usually complicates formal proofs of convergence and performance
analysis of the designed control algorithms, due to the inner-outer loop structure. Following the same
strategy, all the PITVANT platforms are equipped with a Piccolo II controller [8]. Within this R&D
project, the UAV control algorithms are usually developed and tested in simulation using Matlab [9].
Thus, one of the main features of the here reported test bed is that it provides a mean to rapid testing
of kinematic control laws running under Matlab, allowing for the validation and implementation of the
high level controllers in different controlled conditions before final implementation. Firstly, it provides
hardware-in-the-loop simulations, where the vehicle dynamics is simulated by software provided by
Piccolo. Secondly, it allows for control algorithm tests in real flight situations with the control loop
closed by a Matlab routine running on a PC on the ground. Thirdly, it is possible to close the control
loop with an embedded computational system (e.g. PC-104). The latter flight tests only differ from the
final application by the localization of the computational system (on ground instead of on on board the
aircraft). In fact, these flight tests are more demanding since with the on board software it is possible
to achieve faster sensor data and control updates. Finally, the presented architecture allows the fast
migration of the embedded system to the aircraft.
Test beds for UAV testing have been reported in the literature. They differ from each other essentially

2
ISBN-xxx-y-zz-yyyyyyy-x 306
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

in the system architecture utilized. In [7], both high level planning strategies and low-level servo
commands are computed on the same hardware. Other research groups use a separate off-the-shelf
hardware to compute low-level control commands, such as aileron and elevator servo commands [1, 5].
For collaborative control applications, the high-level planning may occur on a ground station computer
[2], or onboard the aircraft [6]. The hardware and the software used for the flight tests may also be
commercial [5] or proprietary [3]. In these architectures, the telemetry from the on board systems
is usually transmitted to the ground using a dedicated wireless data link. For command and control
purposes, a second telemetry link operates between the autopilot and the ground control station.
In the here proposed test bed, the telemetry from all the on board systems and their control com-
mands can be performed through a single communications data link and a common command and
control console, shared by all the vehicles of the network. This feature enables testing missions with
multiple platforms saving hardware. The proposed test bed was successfully utilized to test several
different control algorithms designed for the PITVANT project, namely algorithms for target tracking
[10], obstacle avoidance [11] and thermals navigation [12].
The paper describes the hardware and software architecture of the PITVANT test bed and is orga-
nized as follows. Section 2 presents the basic control architecture used for the autonomous flights using
autopilot tools and the proposed test bed software and hardware architectures. Section 3 describes the
entire process of control algorithms validation and implementation, from numerical simulations to flight
tests, including hardware-in-the-loop validation. Finally, section 4 presents the main conclusions and
future work.

2 Test bed software and hardware architecture

2.1 Operation with a Piccolo autopilot

All PITVANT platforms are equipped with a Piccolo II controller that deals with the inner control loop
of the UAVs. It relies on a mathematical model parameterized by the aircraft geometric data and has
a built-in wind estimator. Several model and controller parameters can be set by the user [8]. The
ANTEX-M X02 model parameters were fine tuned in more than 30 hours of autonomous flights.
A Piccolo Ground Station and a Piccolo Command Center (PCC) can be used for basic UAV oper-
ation (see Figure 2). The Piccolo Command Center (PCC) is a graphical user-interface for operators
developed by Cloudcap that shows flight information and allows the operator to send waypoints, velocity,
bank or altitude references to the aircraft (Figure 3).
Telemetry data is transmitted to the ground station via a 2.4 GHz Piccolo link. If an on board
system is added (e.g. a video camera), another data link must be used to relay the video signal to a
different monitor on the ground. Although simple and easy to operate, this architecture does not allow
the control of the on board systems that are not compatible with the Cloudcap’s software. Additionally,
the PCC does not support the inclusion of command and control windows devoted to the operation

3
ISBN-xxx-y-zz-yyyyyyy-x 307
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

with the control algorithms implemented. Next section describes the here proposed architecture that
overcomes the limitations of this basic setup.

Figure 2: Operation with a Piccolo autopilot. Figure 3: Piccolo Command Center.

2.2 Proposed test bed

All development and test stages within the PITVANT project share the software architecture depicted
in Figure 4. It has a layered structure, comprising network control modules, DUNE software tasks, and
physical or simulated hardware.

Figure 5: Neptus command and control

console.

Figure 4: Software architecture.

The physical hardware comprises Piccolo autopilots and ground stations. These may also be simu-
lated using software tools provided by Cloudcap.
DUNE is an embedded software platform used in several autonomous vehicles developed by the
Underwater Systems and Technology Laboratory (LSTS) at FEUP (please refer to [13] for an application
to autonomous underwater vehicles). DUNE provides programmers with a C++ abstraction of real-
time tasks, and runs on several heterogeneous operating systems (e.g., Linux or Windows) and hardware
platforms (e.g., Intel x86 or ARM). In this context DUNE is used to implement software tasks that:

• handle the physical/simulation connection to an autopilot or ground station;

4
ISBN-xxx-y-zz-yyyyyyy-x 308
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

• provide a network gateway for that connection to Cloudcap software;

• decode and encode, respectively, telemetry and control packets [14], from Piccolo and other on
board systems (e.g. gimbal camera), from and to IMC packets [15];

• locally control on board systems and aircraft systems that can run independently from the ground
station.

IMC is a message-oriented interoperability protocol [15] that is used in all LSTS software deployments
of autonomous vehicles. When a new system is included on board (e.g. gimbal) the DUNE software is
used to convert its communication protocol to IMC, thus allowing its control from the ground, without
relying on the commercial software that comes with the equipment. Therefore, the telemetry from all
on board sensors and their control commands can be performed through a single communications link,
instead of using a particular control software and a dedicated communication link for each system.
Moreover, in the proposed architecture, IMC is shared by all vehicles, thus allowing the exchange of
information between heterogeneous systems (e.g., aerial and marine vehicles) and the use of a single
console for mission control.
Network modules may comprise instances of Matlab, Neptus [16], and GUI software from Cloudcap.
Cloudcap GUI applications can be used to operate UAVs equipped with the Piccolo autopilot, and
interface with DUNE’s gateway functionality for that purpose. Neptus is a command and control
infrastructure for autonomous vehicle missions, covering diverse aspects such as mission planning, online
vehicle control and monitoring, or data analysis (Figure 5) [16]. The Neptus console gives the operation
team the freedom to create a terminal suited to the needs of the mission and allows for the inclusion
of new command and control windows associated with the control algorithms implemented. Thus, the
proposed architecture enables the simultaneous use of Matlab, Cloudcap and Neptus consoles to control
the missions. This is useful since different stages of algorithm validation require different consoles. While
Matlab is in the loop, dedicated Matlab consoles can be used to monitor the algorithm performance, and
PCC provides the safety of an alternative way to monitor and control the aircraft. When the algorithms
migrate from Matlab to C++, everything can be monitored and controlled from Neptus, using again
the PCC as a redundant system.
Figure 6 presents the hardware architecture of the test bed. PCC, Matlab and/or Neptus can receive
telemetry data from the aircraft, and send commands to the aircraft using the DUNE gateway. When
the PCC is used to control the mission, commands to the aircraft are sent through the Piccolo Ground
Station. That is also the case when MATLAB is in the loop. An IMC data link is used when the mission
control is done through Neptus.

5
ISBN-xxx-y-zz-yyyyyyy-x 309
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 6: Hardware architecture.

3 Implementation and validation process of control algorithms

Validation and implementation of control algorithms usually goes through the following steps: numerical
simulations, hardware-in-the-loop simulations, flight testing with the control algorithm implemented on
a computer on ground (standard PC and then a dedicated computational system), and finally imple-
mentation of the control algorithm on an on board embedded computational system. The here proposed
test bed supports all stages from numerical simulations to flight testing (with control on the ground)
using the same Matlab code.

3.1 Numerical simulations

The first phase of implementation and testing of the control algorithms is their numerical validation,
considering only the kinematics of the system (Figure 7). At this stage, the development team has the
freedom to choose the architecture and the syntax of the code being written, including a Matlab console
to monitor simulation data. In fact, different applications require different parameters and data, thus
different consoles to match each application needs are usually designed.
These consoles are often created utilizing GUIDE [9] to rapidly create a suitable Graphical User
Interface (GUI).

Figure 7: Numerical simulations structure.

Despite the freedom involved in this first stage, the code implementation should be done keeping in
mind its future hardware-in-the-loop and real flight tests. Thus, the control algorithm and its console
should be implemented independently of the vehicle kinematics.

6
ISBN-xxx-y-zz-yyyyyyy-x 310
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3.2 Hardware-in-the-loop simulations

Hardware-in-the-loop (HIL) simulations refer to simulations were the Piccolo autopilot and the Piccolo
Ground Station are included in the loop, replacing the model simulation used for numerical simulations
(Figure 8). A communication block to allow the exchange of data between the controller and the
Piccolo autopilot must be included. In the here proposed test bed, the UAV dynamics is simulated by
the Simulator software provided by Cloud Cap (Figure 9). It receives the commands from the Piccolo
autopilot and provides the UAV state information via a CAN bus interface. The communication between
the Piccolo and the algorithm implemented on Matlab is mediated by the Ground Station and DUNE.
The Ground Station is connected to the Piccolo autopilot via an UHF link, receiving the UAV state
and transmitting the control references, and to the PC that runs the algorithm through a serial cable
(Figure 9).

Figure 8: Hardware simulations and flight tests basic structure.

Hardware-in-the-loop simulations are instrumental in identifying interface problems between different

subsystems and in validating the control algorithm now including the dynamics inner loop closed by the
autopilot. The proposed setup still relies on a Matlab implementation of the control, thus facilitating
control algorithm modifications. Additionally, Matlab consoles can still be utilized to monitor the control
system’s performance.

Figure 9: Control system architecture used in the HIL simulations. Adapted from Piccolo Setup Guide
[8].

7
ISBN-xxx-y-zz-yyyyyyy-x 311
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 10 presents a hardware-in-the-loop simulation of an obstacle avoidance algorithm [11]. The

usage of MATLAB at this stage enables an easy design of intuitive interfaces (left-hand side of Figure 10)
to tune the algorithm before testing it on a real flight. Theses interfaces provide a better understanding
of the system’s behaviour and can be used during the flight tests. From the Matlab routine point of view,
there is no difference between the hardware-in-the-loop and the real flights operations. After tuning the
algorithm, the user can evaluate its performance by using Matlab’s set of tools, as shown in right-hand
side of Figure 10.

Figure 10: Trajectory obtained in a obstacle avoidance test, using hardware-in-the-loop simulations.

This test bed was also used to perform hardware-in-the-loop simulations of target tracking [10]
multi-UAV missions, and thermal estimation algorithms [12].

3.3 Flight tests

For the flight tests, the Piccolo autopilot is on board the aircraft, along with a PC-104 and several
other systems, including video cameras and other sensors. The autopilot sends the telemetry data to
the ground station through a 2.4 GHz Piccolo protocol link. The data received can be used to close
the UAV control loop using a standard PC. This is useful for the first flight tests where the control
algorithms are still implemented in Matlab and run on a computer on ground.
Using DUNE software [16] and an ethernet connection, the computer running Matlab is enabled to
receive telemetry data, compute the new control references and send them to the aircraft via Piccolo
Ground Station (please refer to Figure 6). Using the Neptus console [16] is possible to monitor the
telemetry and video data.

8
ISBN-xxx-y-zz-yyyyyyy-x 312
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 12: Aircraft’s trajectory following

Figure 11: Fligth test operation frame for target tracking. a target on the ground.

The above described hardware architecture was successfully utilized to test a wide range of high-level
controllers within the PITVANT project, namely a ground target tracking algorithm, using a GPS in
the loop to provide the target state (Figures 11 and 12) [10], and a thermal estimation algorithm[12].
The following validation and implementation stages require the implementation of the control al-
gorithms in an embedded computational system. This can be done within the DUNE framework [13],
following specific rules and design methods still under development within the PITVANT project.

Figure 13: System’s final hardware architecture.

Figure 13 presents the envisioned system’s final architecture after the migration of the control algo-
rithms from Matlab to DUNE running at the on board computational system. Neptus console will be
used to send commands to the UAV and to on board systems. It will allow for the design of different
windows to control and monitor each system. Through the IMC data link, different vehicles will be able
to exchange information and thus high-level planning for cooperative missions may occur on board the
vehicles or on a ground station.

9
ISBN-xxx-y-zz-yyyyyyy-x 313
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

4 Conclusions
The test bed for rapid flight testing of UAV control algorithms developed under the PITVANT research
project allows for validation and implementation of high level control algorithms, from hardware-in-
the-loop simulations to real flight tests with multiple vehicles. It provides a hardware and software
architecture to progressively test new control algorithms in controlled situations before the final im-
plementation on board the aircraft. Future work will include the definition of detailed operational
procedures for flight testing, the establishment of software rules and design methods to implement the
algorithms within the DUNE software to be run on an on board dedicated computational system, and
the development of software tools to include its consoles within NEPTUS.

REFERENCES
[1] I. Kaminer, O. Yakimenko, V. Dobrokhodov, and K. Jones, “Rapid flight test prototyping system
and the fleet of UAVs and MAVs at the Naval Postgraduate School,” in Proceedings of the 3rd
AIAA “Unmanned Unlimited” Technical Conference, Workshop and Exhibit, 2004, pp. 20–23.

[2] J. How, E. King, and Y. Kuwata, “Flight demonstrations of cooperative control for UAV teams,” in
Proceedings of 3rd AIAA Unmanned Unlimited Technical Conference, Workshop and Exhibit, 2004.

[3] M. Quigley, M. Goodrich, S. Griffiths, A. Eldredge, and R. Beard, “Target acquisition, localization,
and surveillance using a fixed-wing mini-UAV and gimbaled camera,” in Proceedings of the 2005
IEEE International Conference on Robotics and Automation, ICRA, 2005.

[4] T. Mclain and R. Beard, “Unmanned air vehicle testbed for cooperative control experiments,” in
Proceedings of the 2004 American Control Conference, 2004, pp. 5327–5331.

[5] L. Ma, V. Stepanyan, C. Cao, I. Faruque, C. Woolsey, and N. Hovakimyan, “Flight test bed for
visual tracking of small UAVs,” in Proceedings of the 2006 AIAA Guidance, Navigation, and Control
Conference and Exhibit, 2006.

[6] J. Tisdale, A. Ryan, M. Zennaro, X. Xiao, D. Caveney, S. Rathinam, J. Hedrick, and R. Sengupta,
“The software architecture of the Berkeley UAV Platform,” in Proceedings of the 2006 IEEE In-
ternational Conference on Control Applications, 2006, pp. 1420 –1425.

[7] J. Jang and C. Tomlin, “Design and implementation of a low cost, hierarchical and modular avionics
architecture for the DragonFly UAVs,” in Proceedings of the 2002 AIAA Guidance, Navigation, and
Control Conference, 2002.

[8] B. Vaglienti, R. Hoag, M. Niculescu, J. Becker, and D. Miley, Piccolo System User’s Guide, Cloud
Cap Technology, 2009.

[9] MATLAB, version 7.10.0 (R2010a). Natick, Massachusetts: The MathWorks Inc., 2010.

10
ISBN-xxx-y-zz-yyyyyyy-x 314
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[10] T. Oliveira and P. Encarnação, “Ground target tracking for unmanned aerial vehicles,” in Proceed-
ings of the 2010 AIAA Guidance, Navigation, and Control Conference, 2010.

[11] G. Cruz and P. Encarnação, “Obstacle avoidance for unmanned aerial vehicles,” in Journal of
Intelligent and Robotic Systems, in press, 2011.

[12] R. Bencatel, “Thermal localization,” in Proceedings of the 2010 AIS Conference, 2010.

[13] A. Tinka, S. Diemer, L. Madureira, E. Marques, J. Sousa, R. Martins, J. Pinto, J. Silva, P. Saint-
Pierre, and A. M. Bayen, “Viability-based computation of spatially constrained minimum time
trajectories for an autonomous underwater vehicle: implementation and experiments,” in Proceed-
ings of the 2009 American Control Conference. IEEE Computer Society, 2009, pp. 3603–3610.

[14] B. Vaglienti, “Communications for the Piccolo Avionics, version 2.1.2,” Cloud Cap Technology,
2011.

[15] R. Martins, P. Dias, E. Marques, J. Pinto, J. Sousa, and F. Pereira, “IMC: A communication
protocol for networked vehicles and sensors,” in Proceedings of the 2009 IEEE Oceans Europe’09,
2009.

[16] P. Dias, G. Goncalves, R. Gomes, J. Sousa, J. Pinto, and F. Pereira, “Mission planning and
specification in the Neptus framework,” in Proceedings of the 2006 IEEE International Conference
on Robotics and Automation, (ICRA), 2006, pp. 3220–3225.

11
ISBN-xxx-y-zz-yyyyyyy-x 315
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Adaptive Control of Airplane Longitudinal Flight Dynamics

Francisco Gavilan† , Rafael Vazquez† , and Jose Angel Acosta‡

† ‡
Departamento de Ingenierı́a Aeroespacial, Departamento de Ingenierı́a de Sistemas y Automática
Universidad de Sevilla, Camino de los Descubrimientos s.n., 41092 Sevilla, Spain.
(e-mails: fgavilan,[email protected],[email protected])

Abstract

In this paper, an adaptive backstepping control law is designed for the longitudinal dynamics
of an Unmanned Air Vehicle (UAV). The controller makes the system follow references in the
aerodynamic velocity and flight path angle, using elevator deflections and thrust as actuators. The
proposed strategy allows to design an explicit output-feedback controller without any knowledge of
the aerodynamic coefficients; only well-known qualitative physical properties from aerodynamics are
used. Simulations are performed in a complete nonlinear 6 DOF UAV model, where control signal
saturations and limitations in the actuators rates are present.

keywords: flight control, UAVs, Lyapunov stability, nonlinear control, adaptive control.

1 Introduction
Recent years have seen a considerable growth in the development of unmanned air vehicles (UAVs).
These kind of aircrafts have demonstrated to be able to carry out a wide variety of missions with less
cost than traditional manned airplanes. Thus, interest in development of flight control systems for such
airplanes has increased considerably.
Traditionally, flight controllers have been designed based on a linearized aircraft model. However,
when the flight condition is changed, the model is no longer valid and the controller performance can
be compromised. To overcome this difficulty, gain scheduling methods have been applied in the past [6].
However, these methods have the need of computing different controllers for different operating points
and estimating aircraft stability derivatives for a wide range of flight conditions, which can be a very
cumbersome task.
In addition, since there are a lot of different airplanes, not always good aerodynamic models are
available. In this cases analytic computation of controller gains can be imprecise, thus tuning must be
done in real flight situations, a procedure which is more expensive, time-consuming and risky.
Another way to handle the real nonlinear behaviour of the aircraft is using nonlinear control tech-
niques. For instance, feedback linearization [7] or backstepping methods [4] can be applied to nonlinear
systems, leading to controllers which can guarantee the stability for a wide range of operating points.

ISBN-xxx-y-zz-yyyyyyy-x 316
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

However they are based on a good knowledge of the aerodynamic model, unavailable in most practical
situations.
These kind of systems with model uncertainties can be handled with adaptive strategies such as
adaptive backstepping [5] control. This technique can be used in nonlinear systems with a cascade
structure, and is able to guarantee the global stability even in the presence of parametric uncertainties
in the model.
Several examples of adaptive backstepping can be found in the literature. For instance, in [2]
an adaptive backstepping flight controller for a high-performance UAV is developed by using a linear
aerodynamic model and adaptation laws to estimate online the stability derivatives in the model. A
similar approach is described in [8], where a constrained adaptive backstepping controller is designed for
the F-16/MATV simulation model, representing its aerodynamics with neural networks whose weights
are estimated through adaptation laws.
The objective of this paper is to develop an output-state feedback law (based on our previous
full-state feedback design presented in [3]) for aircraft longitudinal dynamics that works for all the
normal operating regimes of the aircraft, and needs minimal information of the aerodynamic model.
The control objective is to seek references in the aerodynamic velocity and flight path angle, using as
actuators the elevator deflections and the thrust level. An adaptive backstepping strategy is proposed,
which exploits the structure of the system and well-known qualitative properties from aerodynamics. The
nonlinear longitudinal aircraft model is used, and, since only some specific properties of the aerodynamic
coefficients are known, an adaptation law is designed for their online estimation. The controller for
velocity and flight path angle are designed separately. The resulting control laws are explicit and
simpler than those produced by the previously cited works, and do not require much computational
power on board.
The paper is structured as follows: First, in Section 2 the aircraft model used in this work is presented.
The controller design is detailed in Section 3, which begins with the velocity controller (3.1) and follows
with the flight path angle controller (3.2). Simulations are shown in Section 4, and Section 5 closes the
paper with some concluding remarks.

2 Aircraft longitudinal flight model

The equations of motion of the aircraft longitudinal flight dynamics can be formulated as [9]:

1
V̇a = (−D + FT cos α − mg sin γ) , (1)
m
1
γ̇ = (L + FT sin α − mg cos γ) , (2)
mVa
θ̇ = q, (3)
M (δe )
q̇ = , (4)
Iy

where Va is the aerodynamic velocity; q is the pitch angular velocity; γ, θ and α are the flight path,
pitch and attack angles respectively (notice that α(θ, γ) = θ−γ); FT is the thrust force, δe is the elevator
deflection; m and Iy are the mass and the inertia; and L, D and M (δe ) are the aerodynamics forces lift,

ISBN-xxx-y-zz-yyyyyyy-x 317
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

drag and pitching moment, respectively.

The aerodynamic forces and moments are computed through their non-dimensional coefficients as
follows
1 2 1 1
L= ρV SCL , D = ρVa2 SCD , M = ρVa2 Sc̄Cm (δe ), (5)
2 a 2 2
where ρ is the air density, S is the reference wing surface, c̄ is the mean chord and CL , CD and Cm (δe )
are the lift, drag and pitching moment coefficients, respectively. Moreover, we consider the following
models for the drag and moment coefficients:

CD = CD0 + k1 CL + k2 CL2 , (6)

Cm (δe ) = Cm0 + Cmα α + Cmq q + Cmδe δe , (7)

where CD0 , k1 , k2 , Cm0 , Cmα , Cmq and Cmδe are aircraft aerodynamic coefficients. In this work, all
coefficients are considered to be unknown parameters, except for the well-known fact that Cmδe < 0.
This is an improvement over [3] where knowledge of Cmδe was assumed in the design.
Regarding the lift coefficient model, only the following assumption is considered.

Assumption 1 The lift coefficient CL is only a function of α. The reference axis xB is chosen so that
CL (0) = 0, i.e. xB is parallel to the aircraft zero-lift line. Then, the property x · CL (x) ≥ 0 is satisfied
for all x ∈ R.

This assumption is satisfied by all conventional airplanes in the non-stalled regime1 . We underscore
that this is the only assumption about CL , which otherwise is considered completely unknown.

3 Controller design
The control objective is to design feedback laws for (FT , δe ) which make the system seek for known
references in velocity and flight path angle (Vr , γref ). In [3], as a first attempt to solve this control
problem, we designed a full-state control law accomplishing this objective. However, while the state
vector (Va , γ, θ, q) is physically measurable in an aircraft, the reference value θref needed to reach the
flight condition (Vr , γref ) is unknown and so, the state error θ − θref used in [3] is not measurable. It
is shown below that θref = α0 + γref , with α0 being the trim angle of attack, which is not known since
it depends on the desired flight condition (Vr , γref ) and on the lift coefficient function CL . Thus, our
previously derived full-state control law has to be suitably modified to obtain an output feedback law
not requiring the knowledge of θref .
The controller is designed considering first the velocity dynamics, given by Equation (1), and then the
pitch dynamics given by (2)–(4). Thus, two different controllers are designed: the aerodynamic velocity
is controlled using only the thrust (FT ) and the flight path angle (pitch dynamics) is controlled with
the elevator angle (δe ). The compound controller makes the closed-loop system follows the references
as desired.
Even though the novelties with respect to [3] are only on the flight-path-angle control design (Sec-
tion 3.2), for the sake of completeness, we reproduce next in Section 3.1 the design of the velocity control
law; we refer the reader to [3] for the proof of stability.
1 See for instance [1], where an extensive compendium of lift curves with this property can be found.

ISBN-xxx-y-zz-yyyyyyy-x 318
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3.1 Control of aerodynamic velocity

The velocity dynamics are governed by the equation (1), where after substituting the moment model D
from (5) reads  
1 1
V̇a = − ρVa2 SCD + FT cos α − mg sin γ , (8)
m 2
where α and γ are measurable quantities and the engine thrust FT is the control input. In addition,
based on (6), we use the following drag model:

CD = CD0 + k1 α + k2 α2 , (9)

where CD0 , k1 and k2 are unknown parameters. Denote Vr to the reference velocity and define the error
zV := Va − Vr . Thus, the evolution of the error from (8) becomes

cos α
żV = −β1 zV2 + Vr2 + 2zV Vr ϕV (α)T · θV + FT − g sin γ − V̇r , (10)
m
where we have defined
 T ρS
ϕV (α) := 1 α α2 , θV := [CD0 k1 k2 ]T, β1 := , (11)
2m
where θV ∈ R3 is the unknown parameters vector, the ϕV ∈ R3 is defined using (9) as CD = ϕV (α)T · θ V > 0
and it holds that β1 > 0.
The velocity feedback and adaptation law is as follows:
m  
FT = g sin γ + V̇r + β1 (zV2 + Vr2 )ϕV (α)T · θ̂ V − κV1 zV , (12)
cos α
˙

θ̂ V = −β1 zV3 + zV Vr2 ΓV ϕV (α), (13)

This control and adaptation law is shown in [3] to guarantee global boundedness of zV and θ̂ V and
convergence of zV to zero.

3.2 Control of the flight path angle

In this section a new adaptive output-feedback controller for the flight path angle is proposed, improving
the adaptive full-state design given in [3]. There are two novelties; on the one hand, the value α0 is
considered unknown, which leads to a more involved output-feedback design. On the other hand, the
parameter Cmδe is here considered unknown but negative. Both α0 and Cmδe were considered known
in [3], paving the way for the design provided here.
The pitch dynamics are governed by equations (2)–(4) after substituting (5) and (12). We next state
some assumptions regarding the pitch dynamics.

Assumption 2 The following assumptions are made:

- It is assumed that cos γ ≈ cos γref , as proposed in [4].

- γ̇ref is assumed to be zero.

- Aircraft engine can not produce negative thrust. Thus FT is always nonnegative.

ISBN-xxx-y-zz-yyyyyyy-x 319
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Under Assumption 2 the equation (2) becomes

γ̇ = f (α) = f (θ − γ), (14)

where the scalar function f is defined as

 
1 1 2
f (α) := ρVa SCL (α) + FT sin α − mg cos γref .
mVa 2

Note that f depends implicitly on γref .

Property 1 Let α0 be the trim angle of attack, which is the value of α that makes f (α) zero (for a given
γref ), i.e., f (α0 ) = 0. Then, under the Assumption 1, the function f (α) satisfies (α − α0 )f (α) > 0.
Since f (α) is unknown, α0 is not computable. In addition, in what follows α0 is assumed to be constant.

Let us first shift the equilibrium to zero defining the following vector of error coordinates z ∈ R3 as
h iT h iT
z= z1 z2 z3 := γ − γref θ − γref − α0 q (15)

Note that since α0 is unknown it follows that z2 is not a measurable quantity by itself. The equations
(2)–(4) together with (14) in the new set of coordinates read

ż1 = η(z2 − z1 ), (16)

ż2 = z3 , (17)
 
ż3 = β2 Cm0 + Cmα (z2 − z1 + α0 ) + Cmq z3 + Cmδe δe , (18)

ρVa2 Sc̄
where we have defined β2 := 2Iy and

η(x) := f (x + α0 ). (19)

Notice that Property 1 makes the scalar function η(x) to satisfy x · η(x) ≥ 0.
The control problem. To make the origin of (16)–(18) asymptotically stable, that with (15) becomes
(γ, θ, q) = (γref , θref , 0), through the input δe and under the following conditions:

C1. γref is a given reference.

C2. α0 is unknown and so, θref := α0 + γref too.

C3. Cmδe from (7) is unknown but assumed to be negative.

C4. The measurable output vector y ∈ R3 is defined as

h iT h iT
y := γ − γref α q ≡ z1 z2 − z1 + α0 z3 . (20)

Remark 1 A backstepping control law was designed in [4] for the cascade structure (16)–(18), but using
in (4) the aerodynamic moment model M as the control input. Then, a control allocation scheme was
used to estimate the elevator deflections δe ; additionally, Property 1 was invoked assuming knowledge of
α0 . In this work M (δe ) is a function of the physical control input δe , given by (7) with all aerodynamic
coefficients unknown.

ISBN-xxx-y-zz-yyyyyyy-x 320
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

In what follows, we stabilize each step of the cascade explicitly using the backstepping approach.
Step 1. First, equation (16) is stabilized using z2 as a virtual control. Defining the Lyapunov function
as
1 2
z , W1 =
2 1
the derivative reads Ẇ1 = z1 η(z2 − z1 ); we select the control z2 = u1 (z1 ) = −κγ1 z1 , where notice that
z1 = y1 and thus measurable. Thus,

Ẇ1 |z2 =u1 (z1 ) = z1 η(−(1 + κγ1 )z1 ),

and hence Ẇ1 |z2 =u1 (z1 ) is negative definite for κγ1 > −1.
Step 2. Defining now the error variable z̃2 := z2 − u1 (z1 ), the equations (16)–(17) can be rewritten as

ż1 = η(z̃2 − (1 + κγ1 )z1 ), (21)

z̃˙2 = z3 + κγ1 η(z̃2 − (1 + κγ1 )z1 ), (22)

The Lyapunov function for (21)–(22) is

Z z̃2 −(1+κγ1 )z1
W2 = c1 W1 + η(s)ds,
0

which is a positive definite function by the Property 1. Calculating Ẇ2 we get

Ẇ2 = c1 z1 η + (−η + z3 )η = −η 2 + (c1 z1 + z3 )η,

where we have omitted the argument of η for clarity, minding that η = η(z̃2 − (1 + κγ1 )z1 ). Selecting
the virtual control as z3 = u2 (z1 ) = −c1 z1 , (23) becomes Ẇ2 = −η 2 and then negative semidefinite,
using again only the available output y1 . Invoking LaSalle’s Theorem, z1 and z̃2 tend to the largest
invariant set inside the set {(z1 , z̃2 ) ∈ R2 : η = 0} which in turn implies from Property 1 and (19) that
z̃2 − (1 + κγ1 ) z1 = 0. The residual dynamics on this set become

ż1 = 0, (23)
z̃˙2 = −c1 z1 , (24)

and then (23) implies that z1 a constant. The derivative of the set is z̃˙2 − (1 + κγ1 ) ż1 = 0, so it follows
that z̃˙2 = 0, which together with (24) implies that the largest invariant set is the origin z1 = z̃2 = 0.
Moreover, since W2 is radially unbounded the origin is globally asymptotically stable. Notice that this
implies γ → γref , θ → θref even though θref is unknown, or equivalently α → α0 with α0 unknown.
Step 3. In this last step, we extend the backstepping design to generate the elevator deflections laws.
We employ an adaptive scheme to estimate online the aerodynamic moment coefficients. As commented
before a further improvement upon [3] comes from dropping the hypothesis that Cmδe is known. Only
the physical fact that Cmδe < 0 is used.
The system, for this last step of the design, is composed by the subsystem (21)-(22) and the equation
(18) in the new error coordinate defined as z̃3 := z3 − u2 (z1 ), and then becomes

ż1 = η, (25)
z̃˙2 = z̃3 − c1 z1 + κγ1 η, (26)


z̃˙3 = β2 Cmδe ϕTγ · θγ + δe − β2 κγ3 z̃3 + c1 η, (27)

ISBN-xxx-y-zz-yyyyyyy-x 321
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

where we have defined a scaled vector from (7) of unknown aerodynamic coefficients θγ ∈ R4 as
 T
Cm0 Cmα Cmq 1
θ γ := , (28)
Cmδe Cmδe Cmδe Cmδe

and a vector of measurable quantities ϕγ (y) ∈ R4 as

h iT h iT
ϕγ (y) := 1 α z3 κγ3 z̃3 = 1 y2 y3 κγ3 (y3 + c1 y1 ) , (29)

and recall that δe is the elevator deflection, which is the real control input of the aircraft.

Remark 2 The controller proposed in this step is designed to stabilize the system (25)–(27) with an
adaptation law to estimate the parameters (28) and using the only available outputs given by (20).

Let us define the compound Lyapunov function as

c3 2 |Cmδe | T −1
W3 = W2 + z̃ + θ̃γ Γγ θ̃γ , (30)
2 3 2
where c3 > 0, Γγ = ΓTγ > 0 is the adaptation gain matrix, θ̂γ is the estimate of θγ and θ̃ γ := θγ − θ̂γ
is the estimation error vector. Then, the Lyapunov function derivative becomes


Ẇ3 = −η 2 + z̃3 η + c3 z̃3 β2 Cmδe ϕTγ · θγ + δe + c3 z̃3 c1 η − c3 β2 κγ3 z̃32 + |Cmδe |θ̃γ Γγ−1 θ̃˙ γ .
T
(31)

Defining the control and the adaptation laws as

δe := −ϕTγ · θ̂γ , (32)

˙
θ̂ = −θ̃˙ γ := −c3 β2 z̃3 Γγ ϕγ , (33)
γ

1
and selecting c3 = c1 then (31) yields
   
β2 κ γ 3 2 2 β2 κ γ 3
Ẇ3 = −η 2 + 2z̃3 η − z̃3 ≤ − 1 − η2 − − 2λ z̃32 ,
c1 λ c1
where we used the Young’s inequality. Thus, pick λ = 4 and κγ3 > 8c1 /β2 and we get
1 1
Ẇ3 ≤ − η 2 − z̃32 ,
2 2
which is a negative semidefinite function, as before. Invoking again LaSalle’s Theorem, it is straightfor-
ward to see that the largest invariant set inside the set {(z1 , z̃2 , z̃3 ) ∈ R3 : η = 0, z̃3 = 0} is the origin
z1 = z̃2 = z̃3 = 0, because z1 = z̃2 = 0 implies z2 = 0 by the same arguments as before, and additionally
z1 = 0 and z̃3 = 0 implies directly z3 = 0.
The resulting feedback and adaption laws are summarized next:

δe = −ϕγ (y)T · θ̂γ , (34)

˙ β2
θ̂γ = − (q + c1 (γ − γref )) Γγ ϕγ (y), (35)
c1

with c1 > 0 and κγ3 > 8c1 /β2 , Γγ = ΓTγ > 0 and
h iT
ϕγ (y) = 1 α q κγ3 (q + c1 (γ − γref )) . (36)

These feedback and adaption laws guarantee that the equilibrium (γ, θ, q) ≡ (γref , θref , 0) is globally
asymptotically stable.

ISBN-xxx-y-zz-yyyyyyy-x 322
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

4 Simulation Results
The simulation testbench chosen is a small electrical UAV, with a mass m = 14 kg and a wing surface
S = 1 m2 .
Instead of using only the longitudinal dynamics (which is the model used throughout this paper), the
simulation model is built based on the standard nonlinear 6 DOF aircraft equations of motion in body
axes. In addition, a nonlinear aerodynamic model has been used, obtaining the aerodynamic forces and
moments coefficients using Datcom methods.
For a more realistic simulation, saturations in the control signals and in their rates are also considered.
Thus, the elevator deflection and the thrust force must satisfy the limits δe ∈ [−30o , 30o ] and FT ∈
[0 N, 50 N] respectively, whereas the maximum elevator rate is δ̇e,max = 30 deg/s and the maximum
thrust rate is ḞT,max = 40 N/s.
The tuning parameters for the velocity controller are κV1 = 10, ΓV = 0.001I 3 (where I 3 is the
identity matrix of dimension 3). For the flight path angle controller, the parameters are
 
1 0 0 0
 
 
 0 1 0 0 
κγ3 = 3; c1 = 1; Γγ = 0.5  
.

 0 0 1 0 
 
0 0 0 20
The initial estimate of the unknown parameters is
h iT h iT
θ̂V = 0.05 0.05 0.05 , θ̂γ = 0.5 1 15 −0.3 .

In Fig. 1 is shown the time evolution of the aerodynamic velocity and the flight path angle. The
dashed green lines represent the outputs signals using the controller formulated in this paper, whereas
the solid blue lines are the references. It can be seen that the airplane can follow the references in
velocity and flight path angle, but there is some overshooting in the airspeed response, that can be
excessive if the reference step is high (for instance in the last step).
Looking now at Fig. 2, where the control signals are shown, it can be seen that the controller tends
to be aggressive when a reference step is commanded. Thus, the real control signal applied is not the
same than the computed one, due to the limitations in the control signal values and rates.
Particularly, in the airspeed subsystem, it can be seen that when the reference step is high, the thrust
saturates during long times. This leads to a windup effect in the adaptation law that causes excessive
overshooting in the output (notice that when a control signal saturates, the parameters adaptation is
inaccurate, since errors introduced by control saturations are not modelled).
To overcome this windup effect, the velocity controller has been modified. Thus, when the thrust
computed exceeds its limits, the adaptation law is stopped, and it is restarted again when the control
law demands feasible thrust forces.
The airplane response with this modified velocity controller can be seen in the red solid lines of Fig. 1.
Comparing with the original velocity controller, there is a noticeable improvement in the response, which
is faster and with negligible overshooting.
Finally, in Fig. 3 it can be seen the time evolution of the estimated parameters, which converge
towards certain constant values. It is remarkable that the final values of the parameter CD0 = 0.026 is
close to the real value of the aircraft coefficient, which is CD0 ,real = 0.023.

ISBN-xxx-y-zz-yyyyyyy-x 323
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

35
Reference
Anti−windup off
30
Anti−windup on
V [m/s]

25
a

20

15
0 10 20 30 40 50 60 70 80 90 100
t [s]

15

10
γ [deg]

−5
0 10 20 30 40 50 60 70 80 90 100
t [s]

Figure 1: Time evolution of the aerodynamic velocity and the flight path angle. Two simulations are
shown, switching on and of the anti-windup correction.

50

40

30
F [N]
T

20

10

0
0 10 20 30 40 50 60 70 80 90 100

Commanded Applied

30
20
10
δ [deg]

0
e

−10
−20
−30
0 10 20 30 40 50 60 70 80 90 100
t [s]

Figure 2: Time evolution of the control signals. The dashed line represents the control signal com-
puted by the control law, while the solid line shows the real control signal applied, taking into account
saturations and limitations in the actuators rates.

5 Conclusion
In this work, we presented the design of an adaptive and output-feedback controller for the longitudinal
flight dynamics of an UAV that is able to make the aircraft follow references in velocity and flight path

ISBN-xxx-y-zz-yyyyyyy-x 324
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

CD0 k1 k2
0.055 20

0.05 10

0.045
0

0.04
−10
0.035
−20 Cm0/Cmδ
0.03 e

Cmα/Cmδ
−30 e
0.025
Cmq/Cmδ
e

0.02 −40 1/C

mδ
e

0.015 −50
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
t [s] t [s]

Figure 3: Time evolution of the estimated parameters.

angle. The design is built upon our previously developed full-state control law [3], it is explicit, easy
to implement, only employs measurable quantities, and does not require to estimate any aerodynamic
coefficient. Simulations have been performed in a realistic 6 DOF model, showing that the controller
can make the system follow the references, even in the presence of actuator saturations and limitations
in the actuators rates. As a next step the control laws will be implemented on board the aircraft to
perform experiments and further validate the results.

REFERENCES
[1] Ira H. Abot and Albert E. Von Doenhoff. Theory of wings and sections. Dover, 1959.

[2] J. Farrell, M. Sharma, and M. Polycarpou. Backstepping-based flight control with adaptive function
approximation. J. Guid. Contr. Dynam., 28(6):1089–1102, 2005.

[3] Francisco Gavilan, Jose Angel Acosta, and Rafael Vazquez. Control of the longitudinal flight dy-
namics of an uav using adaptive backstepping. In IFAC World Congress, 2011.

[4] Ola Härkegård. Backstepping and Control Allocation with Applications to Flight Control. PhD thesis,
Linköping Universtity, 2003.

[5] Miroslav Krstić, Ioannis Kanellakopoulos, and Petar Kokotović. Nonlinear and Adaptive Control
Design. John Wiley, 1995.

[6] Robert A. Nichols, Robert T. Reichert, and Wilson J. Rugh. Gain scheduling for H∞ controllers: A
flight control example. IEEE Trans. Contr. Syst. Tech., 1(2):69–78, 1993.

[7] Yoshimasa Ochi and Kimio Kanai. Design of restructurable flight control systems using feedback
linearization. J. Guid. Contr. Dynam., 14(5):903–911, 1991.

[8] L. Sonneveldt, Q.P. Chu, and J.A. Mulder. Nonlinear flight control design using constrained adaptive
backstepping. J. Guid. Contr. Dynam., 30(2):322–336, 2007.

[9] B. L. Stevens and F. L. Lewis. Aircraft Control and Simulation. John Wiley, Second edition, 2003.

10

ISBN-xxx-y-zz-yyyyyyy-x 325
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

AIRCRAFT STRUCTURAL HEALTH MONITORING USE CASE

Joaquín Pérez Asier Rabanal Ander Gorostiza
I+D Proyect Manager I+D Proyect Manager I+D Director
SISTEPLANT S.L. SISTEPLANT S.L. SISTEPLANT S.L.

ABSTRACT
High technology and system integration are growing up and inevitably availability and safety
conditions worsen. The failure to meet requitements is due to systematic faults, accidental faults
and aging effects. The evidency proves that devices fail, and then is mandatory to introduce a
verification methodology both in the design stage and during lifecycle. Is in this latter context
where the devices works in its environment, and where the verification system acts as an
observer. In this field the SIMAP (Advanced Predictive Intelligent System) platform provides an
analysis methodology oriented to the failure mode, where vital signs of a system, detection
models, prognosis and decision support models are related. As an example a use case is
presented, and an aircraft flight envelope is identified to provide a failure detection schema
design.

1 INTRODUCTION
Technologic revolution in data acquisition systems, sensors and processing units have
encouraged the evolution of corrective and preventive maintenance towards new strategies,
that consider the health status in real time. This type of maintenance is referred as condition
based maintenance (CBM) [1][2][3] and its main feature is the ability to monitor, detect and
predict failures and consequently establish optimal maintenance actions [4].
Condition based maintenance is directly related to real time system verification, where system
response is checked against system specification. This approach is relevant both in
maintenance and in safety, because the device is observed during operation and then it
becomes a logic redundancy scheme [5], so the probability of failure is reduced [6] and safety is
improved. Altough the advantages of contidion based maintenance are prominent it requires a
high investment in resources, so its only justified in high value-added systems because of safety
requirements, maintenance cost or availability conditions [7][8]. Undoubtedly the aerospace
industry is pioner in condition based maintenance, and is specially suitable for UAVs
(Unmanned Aerial Vehicles) where operation observability is an aim.
Real time system verification is a continuously developing area and several detection and
isolation techniques can be found in the bibliografy. Among them it is a fact that model based
analysis [9][10][11] is the way forward, specially data oriented models because of the reduction
of experts contributions and improvents in system identification techniques. Thereby system
identification [14][15] plays a crucial role in any verifiction platform, and is particularly interesting
when the identification pattern concers several abstraction levels. In this context the SIMAP
solution offers a verification methodology based on system identification, where the main
information unit is given by the failure mode [5][24]. Its methodology allows to transform the
specification of an electromechanic design into a formal and executable specification, called
Failure Detection Schema (FDS). The initial information comes from the requirements document
and from data collected during simulation (operation), and after several refinement stages a
system identification model is obtained. Not only is an executable specification but also a
knowledge source to be shared by the system experts. The FDS comprises detection models at
different levels of abstraction and contains explicit and implicit knowledge covered by experts
and data analysis. FDS describes system dinamics at physical level by differential equations,
while at higher level of abstraction relation among events an transactions is caputred by
temporal logic languages. To verify the proposed methodology an use case is presented.
Aircraft structural health monitoring is analysed in order to detect structural and sensor faults.
Several faults has been injected and successfully detected.

ISBN-xxx-y-zz-yyyyyyy-x 326
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2 FAILURE MODE IDENTIFICATION

The Failue Detection Schema (Figure 1) is the main concept in the SIMAP solution because it
states how to detect failures, and conditions prognosis and decision support (this area is
beyond the scope of this article). Then, failure mode identification establishes a relationship
between the vital signs and events of a system and the corresponding failure mode excitation.
In this way the FDS is the transfer function of the failure mode.
The implicit knowledge in the FDS covers many and different concepts, and the best
introduction point is the detection channel. The detection channel groups all the techniques and
models implied in the activity of the failure mode, and is executed in parallel with the observed
phenomenon. In other words it follows the failure mode dynamics. For example if the failure
mode refers to the overheating of a device, a detection channel working at 1 Hz is enough.
Also several channels can be deployed to observe the same signal if different features appear
along the frequency spectrum. Inside the detection channel, models manage the interaction
between events, transactions, physical quantities, probabilities, etc.
Vital signs [U , Y ] of the system under verification are the input to the detection channel and the
output is determined by the failure mode [FM ] excitation. A detection channel example is
depicted in Figure 1 where several models could be identified:
 The system dynamics evaluator is excited by the vital signs of the device, and at the
output a metric of distance d  y  yˆ between the device response and the estimated
response is provided [14][15]. The dynamics evaluator evals the differential equations of
the systems to measure a quantitative magnitude of the drift.
 The probability evaluator performs a statistical analysis of signals. When the input is
single statiscal process control [18] is the reference, but multivariable analysis [19]
when is multidimensional: under normal conditions the input is described by a
probability density function (pdf) which may be invariant until a fault occurs.
 The last identification unit is the failure mode evaluator. Its output is the failure mode
excitacion degree, a dimensionless magnitude betwenn 0 and 1. The level of
abstraction of this module is high and the unit of information is the event. Behavioural
expressions are formulated by FTL-Cfree [5], a fuzzy temporal logic derived from LTL
[12].

Figure 1 Failure Detection Schema

3 USE CASE
In this section an use case is presented to guide the application of the methodology around
SIMAP [5]: the flight dynamics of an aircraft is identified to perform an structural health
monitoring (SHM) example. The source of information comes from several data files
corresponding to five hours of flight, with 26 signals sampled at 40Hz (Figure 2 depicts the block
diagram of the system under test). The transfer function betwen flight envelope and actuator
signals is described by the “Aircraft Dynamics Phase I” block, and the main signal, SHM(t), is

ISBN-xxx-y-zz-yyyyyyy-x 327
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

goberned by the dynamics captured in the “Aircraft Dynamics Phase II” block. Among others the
envelope signals are the roll rate, pitch rate, yaw rate, lateral acceleration, longitudinal
acceleration, normal acceleration, barometric altitude, equivalent air speed, angle of attack,
sideslip angle, vertical velocity, basic engine data, etc. The actuators control several engine
parameters and primary control surfaces, and finally, the SHM signal captures the global
structural health state. The goal of the use case is to demostrate the ability to detect additive
and multiplicative faults perturbating the structural health dynamics, so the block diagram can
be simplied. Note that the identification model will be part of a real time detection schema and
an accurate analytical model of the system it is not required, as shown in the following sections.
Then, if the transfer function between SHM(t) and actuators is considered negligible, a new
Multiple Input (16 envelope signals) Single Output block diagram depicted in Figure 3 is
proposed.

Figure 2 System under verification block diagram

Figure 3 Simplified Muliple Input Single Output (MISO) Aircraft system

Provided that f Env (t ) , f SHM (t ) are the additive perturbations and f Airc (t ) the multiplicative
faults to be injected in a simulated enviroment, the failure detection schema depicted in Figure 4
is proposed. The schema comprises a single detection channel working at 40Hz, where two
main blocks perform a widespread residual generation an residual evaluation schema [9][10]:
 The first module generates the residual d  SHM (t )  SˆHM (t ) to measure the drift between
the real signal SHM and the estimated signal.
 The second module evaluates the residual to decide if a fault occurs. A safety assertion
states that “the residual must be alway less than an upper bound D ”.
Both modules are data oriented, and then, after the design of the initial structure a training
phase must be computed. Three quartes of the data (550000 samples) are devoted to the
training phase and the rest (180000 samples) to the validation phase.

Figure 4 Detection channel

ISBN-xxx-y-zz-yyyyyyy-x 328
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

A preliminary signal processing analysis suggests the classification of the flight envelope into
two groups: the first group is the Low Frequency Envelope (LFE) and comprises 4 signals
(barometric altitude, equivalent air speed, vertical velocity and all-up weight). The second group
is attributed to the High Frequecy Envelope (HFE) and comprises 12 signals (longitudinal
acceleration, lateral acceleration, normal acceleration, roll rate, pitch rate, yaw rateangle of
attack, sideslip angle, etc). Figure 5 and Figure 6 are examples of LFE and HFE time charts.
Note that all signals are dimensionless (normalized to zero mean and unit variance) in order to
apply correctly the model parameter estimation methods [14][15]. The signal processing
analysis include a time domain study based on autocorrelation and a frequency domain study
based on power spectral density and FFT (Fast Fourier Transform). Figure 7 depicts the
autocorrelation of LFE and HFE signals, wich confirms the difference in the energy distribution.

Figure 5 Barometric Altitude

Figure 6 Longitudinal acceleration signal

Figure 7 LFE (top) vs HFE(bottom) autocorrelation

4 FLIGHT ENVELOPE IDENTIFICATION

The signal classification made is vital to address the residual generation d  SHM (t )  SˆHM (t ) ,
because the group LFE sets the working point of the flight envelope and the HFE describes the
dynamics around the working point. To ensure the integrity of the global model all working
points are described by the same identification model type. In this case an ARX (Autoregressive
Exogenous) model is selected due to its simplicity and and fast training algorithm. However
other types could be selected like the ARMAX, Box Jenkins, state space, even non linear
models like the MLP neural network [16][17]. The behaviour of the composite model is similar to
the LOLIMOT approach presented in [20].

ISBN-xxx-y-zz-yyyyyyy-x 329
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 8 Identification model based on subespaces

When a sample excites the model (see Figure 8), the degree of membership to each subespace
is evaluated first and then the global response is computed as a weighted output from the local
models by the following expression.
M
SˆHM (t )    i ( LFE (t )) ARX i ( Env (t ), SHM (t ))
i 1

Where M is the number of subespaces (working points) identified, ARX i ( Env (t ), SHM (t )) is the
response of the i local model (en este caso de tipo ARX) and  i ( LFE (t )) the normalized
degree of membership:
i LFE (t ) 
 i LFE (t )   M

  LFE (t )
j 1
j

i LFE (t )  is the metric of distance (exactly the euclidean distance) between the LFE (t ) and
the centroid of the subespace i .
In order to design the composite model, the subespace locations an the number of subespaces
must be specified. This is a data mining work, and specifically a non supervised process [13].
The Kohonen feature map [16] is a widespread option in data clustering methods if spatial
locality is the classification target, so in those regions where the density of samples is greater
more subspaces will be located. Applying the non supervised clustering algorithm to the training
dataset (more than 550000 samples), the working points on Figure 9 are calculated. It is
remarkable that the figure is two-dimensional and the original dataset is five dimensional, so a
dimension reduction has been applied. Principal Component Analysis [21] is a mathematical
procedure that uses an orthogonal transformation to accomplish cardinality reduction. After
computing the eigenvalues and eigenvectors of the dataset, the two eigenvectors with more
information (variance) are selected as the new coordinate system. The dynamic of each
working point is governed by a local ARX model. Prior to configure the ARX models the training
samples are classified in the subespace datasets depicted in Figure 10. It is worth to mention
that samples in local subespaces must be consecutive in time to describe a dynamic process,
so jumps between subespaces must be discarded. Also the local datasets must be in
compliance with the persistence of excitation condition requited in linear systems [15], which
states that input data must present more energy points than the order of the model.

ISBN-xxx-y-zz-yyyyyyy-x 330
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 9 Flight envelope (bi-dimensional reduction) and localized working points

Figure 10 Training dataset classification in local subespaces

There is no analytical rule to resolve the number of subespaces, so an iterative procedure is

proposed. A number is prefixed at the begining and the model is validated against the real
response of the system. In this context the graph depicted in Figure 11 is helpful. The z-axis
represents the magnitude of error, and the xy-axis the first and second principal component of
the low frequency envelope. It can be seen that identification error is focused on a delimited
region, probably due to non-linearity conditions. To reduce the identification error more working
points could be placed, but this solution is not always possible because of the persistence of
excitation requirement. Another alternative is to improve the local models, because the ARX is
linear and presents a mischaracterization of noise.

Figure 11 Identification surface error against the first and second principal components of the
low frequency flight envelope

ISBN-xxx-y-zz-yyyyyyy-x 331
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The size of the validation dataset is 180000 samples, and the validation error can also be
quantified analytically by the following expression.

 N

  abs(SHM (i)  SˆHM (i)) 
Fit  100  1  i 1N 
 


i 1
abs(SHM (i)  SHM ) 

If the Fit coefficient is zero the identification is as poor as the mean value of the signal. If it is
less than zero is worse than the mean value and if is equal to 100 is perfect. The SHM (t )
signal and the estimate SˆHM (t ) are depicted in Figure 12. The prediction model on the left
computes the value SHM (t ) from input signals and past values ( SHM (t  1) , SHM (t  2) , ...),
but the simulation model is not aware of the real response. The prediction model estimation is
almost perfect ( Fit  97% ), and not so much the simulation model Fit  70% as expected

Figure 12 SHM (t ) vs SˆHM (t ) prediction model (left) and simulation model

5 FAILURE MODE EVALUATION

The residual generated inside the detection channel (see Figure 4) is the input to the failure
mode evaluation module, which sates if a failure occurs. The residual is evaluated by the
following safety property [12], formulated in the FTL-Cfree language:

10 d  D 
The safety property states that “after a transitory of 10 seconds, the residual d must be always
less than an upper bound D ”. The upper control limit conditions the failure mode sensitivity,
because the smaller D the higher is the probability to detect failures but also the higher is the
probability to signal false positives. This tradeoff can be approached from a statistical point of
view, particularly applying hypothesis testing [18]. If the model identyfies correctly the system
dynamics the residual will follow a normal distribution so it is inmediate to calculate the quantile
that satisfies P ( d  D )   . If the distribution is non normal there are several approaches: first
of all the central limit theorem [22] states that the sample mean of a moving window average
follows always a normal distribution so the solution is the same. It is easy to see that with small
values of the window n  4 the distribution tends to normality. Other alternative comes from the
Chevychev theorem which states that any distribution satisfies the inequality
1 . Given that 1 is equal to the significance level ,
P ( d    k )   u and  are the
k2 k2
sample mean and standar deviation, it is inmediate to calculate the upper control limit. It is also
possible to fit the empirical distribution to a kernel smoothing function [23].
Figure 13 (left) depicts the histogram of the residual signal d compuetd as the difference
between the measured signal SHM(t) and the model estimation. The solid line superimposed

ISBN-xxx-y-zz-yyyyyyy-x 332
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

corresponds to the normal distribution fitting. Considering the residual d is normally distributed
and settiging the probability  to 1E-9 the upper control limit D (quantil distribution) is fixed to
0.16. The right chart depicts the residual values bellow the upper control limit D , so the system
is failure free.

Figure 13 Residual histogram (left) and control chart (right)

6 FAILURE INYECTION
Once the detection channel is designed and validated, it proceeds to the failure inyection phase
according to the schema in Figure 4. We demonstrate the platform’s ability to detect sensor
breaks (total sensor breakdown), additive noise and multiplicative faults.
The most common scenario occurs when additive noise is injected by the expression
x(t )  x(t )  e(t )  K   x to any of the signall in the dataset, where e(t ) is white noise and  x the
standard deviation. Figure 14 depicts the response of the control chart when the additive noise
( K  0.3 ) is injected to the SHM signal and then, in another experiment, to two of the input
signals (equivalent air speed and longitudinal acceleration). As seen in both simulations the
residual signal exceeds the control limit ( D  0.16 ), so the dectection channel is able to the
detect the additive perturbance.

Figure 14 Control chart with additive noise in SHM signal (left) and additive noise in two input
signals(right)

Another typical scenario occurs when the sensor breaks. The effect on the sensor breakdown is
depicted in Figure 15, where the model residual exceeds the upper control limit, and then the
failure is also detected.

ISBN-xxx-y-zz-yyyyyyy-x 333
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 15 SHM break (left) y two input sensor break (right)

Finally the performance of the failure detection channel is evaluated when a multiplicative fault
is inyected to the system. In that case it is unfeasible to cause a failure in the aircraft and we
lack the analytical equations of the transfer function, so we propose a simple fault by the
following expression: SHM (t )  SHM (t )  SHM (t )  ( Input7  Input11 ) . The residual exceeds the
upper control limit as depicted in Figure 16.

Figure 16 Multiplicative fault, control chart

7 CONCLUSIONS
We have presented the system verification methodology around SIMAP, which establishes a
direct relation between the vital signs of a devide and its health status. The solution is failure
mode oriented and its effects are identified by a failure detection schema, useful during the
process operation in real time, but also later in prognosis and decission support [5][24].
Several failure detection modules has been presented at the physical and the functional level
and the ability of the system has been demonstrated by a simulated use case. The example
covers flight envelope identification in order to detect structural and sensor faults. Several faults
has been injected and successfully detected.

8 BIBLIOGRAPHY
[1] Williams J.H., Davies A., Drake P.R. Condition-based maintenance and machine
diagnostics. CHAPMAN & HALL, 1994
[2] Hadden G.D., Bergstrom P., Samad T. Application Challenges: System Health Management
for Complex Systems. Lecture Notes In Computer Science, 2000
[3] Wang H., Pham H. Reliability and Optimal Maintenance. Springer-Verlag, 2006

ISBN-xxx-y-zz-yyyyyyy-x 334
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[4] Crespo A., Moreu P. Ingeniería de Mantenimiento, Técnicas y Métodos de Aplicación a la

Fase Operativa de los Equipos. AENOR, 2004
[5] Perez J., Gorostiza A., Borda J. SIMAP, Sistema Inteligente de Mantenimiento Avanzado
Predictivo. XII Congreso de Confiabilidad de la AEC, España, 2010
[6] Koren I., Krishna C.M. Fault Tolerant Systems. Morgan Kaufmann, 2007
[7] Dhillon B.S. Engineering Maintenance, a Modern Approach. CRC Press, 2002
[8] Matzelevich, William W. Real-Time Condition Based Maintenance for High Value Systems.
Society for Machinery Failure Prevention Technology, 2001
[9] Isermann R. Fault-Diagnosis Systems, An Introduction from Fault Detection to Fault
Tolerance. Springer, 2006
[10] Simani S., Fantuzzi C., Patton R.J. Model Based Fault Diagnosis in dynamic system using
identification techniques. Springer-Verlag, 2002
[11] Witczak M. Modelling And Estimation Strategies For Fault Diagnosis Of Non-linear
Systems: From Analytical To Soft Computing Approaches. Springer-Verlag, 2007
[12] Baier C., Katoen J.P. Principles of Model Checking. The MIT Pres, 2008
[13] Hernández J., Ramírez M.J., Ferri C. Introducción a la Minería de Datos. Prentice Hall,
2005
[14] Zhu I. Multivariable System Identification for Process Control. Elsevier, 2001
[15] Ljung L. System Identification, Theory for the User. Prentice Hall, 1999
[16] Isasi P., Galvan I.M. Redes Neuronales Artificiales, un Enfoque Práctico. Prentice Hall,
2004
[17] Gurney K. An Introduction to Neural Networks. Routledge, 1997
[18] Thompson J.R., Koronacki J. Statistical Process Control, The Deming Paradigm and
Beyond. Chapman & Hall/CRC, 2002
[19] Mason R.L., Young J.C. Multivariate Statistical Process Control with Industrial Applications.
ASA SIAM, 2002
[20] Nelles, o. Nonlinear System Identification. Springer, 2001
[21] Raikov T, Marcoulides G.A. An Introduction to Applied Multivariate Analysis. Routledge,
2008
[22] Martín-Pliego F., Ruiz-Maya L. Fundamentos de Probabilidad. Thomson, 2005
[23] Chou Y.M., Mason R.L., Young, J.C. The Control Chart For Individual Observations from a
Multivariate Non-Normal Distribution. Comm. Statist. 2001
[24] www.sisteplant.com

ISBN-xxx-y-zz-yyyyyyy-x 335
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Autonomic communications for joint maritime surveillance missions

enabled by UAVs

José Manuel Sanchez*, Vicente Baena+, Darío Pérez+, Patricio López+

*
IntegraSys; Esquilo 1, 28232, Las Rozas, Madrid; [email protected]
+
University of Seville; Camino de los Descubrimientos,41092, Seville; [email protected];
[email protected]; [email protected];

Abstract
We present a cooperative satellite communications (SATCOM) network management
solution for a complex border surveillance scenario supporting sensor, voice and video
services for UAV and fixed ground platforms. Such a system has been developed within the
INTEGRA project under the Spanish CENIT-INGENIO 2010 Programme, which is run by a
consortium of 15 main enterprises of the security sector. Our approach is the design of a
SATCOM-friendly management framework inspired on policy-based autonomic
management principles, able to automatically translate operational mission policies into
system/network/device level actions at different hierarchical vertical levels, as well as to
perform horizontal policy transformation to achieve interoperability among autonomous
organizations taking part in the mission. The framework is based on abstract architecture,
information and service models which, by means of model-based techniques, are instantiated
and integrated either for production in real-life platforms or for testing & training in a
simulation platform.

Keywords: satellite, autonomic, virtualization, UAV, maritime

1. INTRODUCTION
International institutions have a growing demand of satellite communications over wide areas for a
number of applications such as: emergency and rescue, border surveillance, external security actions or
disaster management. Among other topics, the INTEGRA project investigates new schemes to support
the provisioning of satellite networking capacity in a federated scheme allowing for a convenient
arrangement and execution of collaborative institutional missions supported by both government and
commercial network resources. Interoperability is a key requirement to achieve seamless operation of
application and services running over federated network segments under different administrative
authorities, but it can become an extremely challenging issue if no sufficient degree of flexibility is
provided to properly handle the potential complexity level of the global system. Some novel
technologies such as autonomic networking and network virtualization offer clear advantages to

-1-

ISBN-xxx-y-zz-yyyyyyy-x 336
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

integrated large, heterogeneous systems in a reasonably and manageable fashion. Throughout the paper
we will describe how these paradigms are applied in the INTEGRA project, always with an eye on the
key need of efficiently implementation, as required by high capacity demands and the unfriendly
scenario represented by the SATCOM channel and aerial platforms with high maneuverability.

1.1. Federation of network resources

A federated communications (COM) system is understood in the INTEGRA project as a joint capacity
built up from a set of COM resources ─e.g. spectrum frequencies, communication satellites, ground
stations, routing devices etc─ under responsibility of different autonomous organizations which agree
on the common use of such a capacity under certain conditions, which is achieved through
interoperability mechanisms. Potential contributors are obviously institutional organizations ─e.g.
homeland security agencies─, but also commercial network operators which are gaining importance as
capacity providers of more and more bandwidth-hungry security missions1.
Figure 1 below depicts the interconnection of different COM management systems in a federated
environment through a harmonized management interface; which does not preclude the possibility for
every contributor to maintain its internal (non interoperable) system to manage its own communications
or any other resource of electronic nature such as applications or databases. Organizations labeled A and
B in the figure contribute with part of their COM resources to the joint capability at the disposal of all
of the remaining entities, while the organization C contributes with the whole of its resources. The three
organizations together build up a federated management system, assuming one of them a global
coordination role to organize and plan the set of federated resources according to a established set of
mission goals. This central role is basically of operational nature; since from a technical viewpoint,
systems must have a high uncoupling and distribution degree so as to enable the agility and dynamisms
required in a wide area surveillance scenario.

1
http://defensesystems.com/articles/2011/02/28/cover-story-commercial-satellites-evolve.aspx

-2-

ISBN-xxx-y-zz-yyyyyyy-x 337
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 1 : Federation of COM management systems in INTEGRA

2. Autonomic networking and management
The today’s Internet, widely present in the business and society worlds, has its origin in the defense
world; and, at the same time, current networks used in security and defense applications are constantly
moving towards Internet-based technologies developed in the civil area. The Internet was founded on a
fundamental premise: one network service is used universally to interconnect intelligent end systems. It
focuses on the ability of dynamic selection at the network layer of a convenient route for data packets,
from origin to final destination, with no committed traffic limits nor delivery assurance. Allocating most
of the complexity, if needed, to the end points, has allowed the Internet to reach an impressive scale in
terms of connected devices. However, this has come to a price. Nowadays, we are facing with the
relative inflexibility of the Internet to provide highly demanded features both in consumer and
professional fields: i) Operations Support Services ─e.g. service installation, service discovery, service
provisioning or network management─; ii) mobility of devices and networks; iii) facilities for quality of
service (QoS), security and reliability; or iv) practical addition of new service features.
The so called Autonomic Networking is a novel concept derived from the more general Autonomic
Computing idea, first presented by IBM in 2001 (http://www.research.ibm.com/autonomic). The
essence of this new vision is taken from the human nervous system which is able to maintain the body
in the appropriate equilibrium by interpreting thousands of parameters and accordingly adapt (handle)
body functions in a very complex manner. The Autonomic networking suggests a transition from
non-sensitive Internet services to sensitive, smart network services, managing resources through the
application of autonomous principles. To achieve this vision it proposes an overlay of virtual resources
that supports mobility, security, QoS and reliability overcoming inflexibility barriers. Under this overlay,
multiple physical networks are expected to coexist on the basis of a shared and uniform management.
Autonomic management concepts are very well in line with unmanned vehicle operations which require
a certain degree of autonomy and automation of all of the vehicle functions, including COM.

2.1. Autonomic management concept in INTEGRA

The autonomic management system architecture adopted in the INTEGRA project is shown in Figure 2,
which depicts how different system entities are distributed over different organizations in a networked
scenario with: wide are broadband (DVB-RCS [1]), wide area narrowband (Iridium [2]) and local area
network bearers. The autonomic system is organized around four abstraction planes which reflect
observations, limitations, specifications and rules used to conveniently respond to network and end-user
events through management functionalities embedded in network hosts, devices and servers:
 The management plane observes a group of COM resources managed by an autonomous entity
through a set of common policies and available context information. Two different key system
entities are situated in this plane: the IAM (INTEGRA Autonomous Management) performs the

-3-

ISBN-xxx-y-zz-yyyyyyy-x 338
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Policy Decision Point (PDP) function in a typical policy-based management system; while the
IDM (INTEGRA Device Management) acts as the Policy Enforcement Point (PEP), residing
commonly in every node containing managed resources, e.g. satellite modems, satellite ground
stations, network routers or service platforms.
 The service plane aims to automate the deployment and execution of new network services;
dealing either with COM resources ─management services under technical operators
competence, such as a new admission control algorithm or a new mobility protocol─ or with
end-user functions, i.e. application-related services such as requesting resources for urgent
transmission of a large dataset. The core entity at this plane is the so called ISM (INTEGRA
Service Management), which is associated both to managing entities and to managed resources
able to host new services.
 The coordination plane hosts the IDC (INTEGRA Domain Coordinator) entity, which
supervises global resource sharing among the different contributing organizations, acting as
needed, either as a supra-manager of individual organizations or as a mediator between them.
 The context plane has the essential role of acquiring and delivering network-related
information at different complexity levels, from basic factual data to more elaborated
information and knowledge. The ICM (INTEGRA Context Management) system entity is
strategically placed at system nodes to monitor their status and help to generate knowledge and
experience so as to allow network self-diagnostic, self-healing and self-optimization at
different time-scales.
All the planes and entities located therein build up together a software-based management layer which
runs on top of the physical (hardware based) network and service devices commonly deployed in the
overall networking system supporting security missions. This system typically presents a number of
distributed nodes across the so-called strategic domain (headquarters, permanent service centers) and
the so-called tactical domain (air, land and maritime patrols; portable control centers etc). The
software-based management layer constitutes a harmonized core tool for interconnection and service
provisioning of all of the involved nodes from different autonomous entities, which a priori may
constitute a highly heterogeneous set.

The described system architecture of the INTEGRA management system has been translated into a
Service Oriented Architecture (SOA) functional architecture supporting simultaneously distributed
services of different nature, e.g. having request-reply or publish-subscribe interaction patterns; or
having stateless or stateful operation. The functional architecture is being implemented today under
three key premises:
 Management services with interfaces defined in XML, which gives the highest flexibility to
support different data models related to managed resources
 Core services supporting registration, discovery and use of management services

-4-

ISBN-xxx-y-zz-yyyyyyy-x 339
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

 A common inter-communication tool (see Figure 1) supporting the interaction between

management services. This function is provided by a COM-oriented middleware which is able
to run optimally though narrowband links such as Iridium.

Figure 2 : INTEGRA autonomic management scenario

(R)NCC: (Regional) Network Control Center; GW: (satellite) Gateway; EUT: End User terminal; BG: Border Gateway;

3. Virtual Networking in INTEGRA

The networking approach in INTEGRA consists of a virtual networking overlay on top of a physical
network substrate. It relies on network virtualization 2 technology which allows the simultaneous
operation of multiple logical networks (also known as overlays) on a single physical platform, and
provides a solid foundation for the flexibility demanded by the autonomic management principle. The
overlay is deployed and managed in real time according to current applications demands and to defined
operational policies, e.g. security, resource load or mobility policies. Network virtualization is not a
new technique, it can be seen as an evolution of the well known Virtual Private Networks (VPN) which
is attracting a lot of attention today thanks to the success of complementary virtualization initiatives
such as Cloud Computing3 and to the research on the so-called Future Internet4

2
http://en.wikipedia.org/wiki/network_virtualization
3
http://en.wikipedia.org/wiki/Cloud_computing

-5-

ISBN-xxx-y-zz-yyyyyyy-x 340
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The network virtualization technique adopted primarily in INTEGRA is based on system virtualization
─virtualization of OS of terminals, access points and routers─ while alternative advanced technologies
allowing features such as non-IP forwarding or data-centric forwarding etc have being researched at
study level, including security implications.
INTEGRA interoperability specifications ─which all of the organizations have to adhere to in the
federated environment─ have been defined at the virtual networking level. They include provisions for
virtual network instantiation/de-instantiation, link provisioning, QoS handling, mobility management
and security management. Such a set of network services are complemented by corresponding
management functions with high automation degree, so that they not only make it possible the
interconnection of different networks and autonomous domains, but also decrease dramatically the
complexity and time dedicated to operate the services.
In addition to constitute the key tool to implement interoperability, the software-based virtual
networking choice offers some other valuable features in the INTEGRA scenario such as: dynamic
creation of Communities of Interests (CoI) according to different surveillance tasks within the same
mission; strong isolation of traffic per security requirements; convenient accommodation of ad-hoc
network protocols; or dynamic inclusion of performance enhancement modules adapted to hostile
conditions. Nevertheless, such benefits are subject to the efficient implementation of the virtualization
mechanisms at substrate level. Furthermore, the virtualization of SATCOM resources within an
organization, makes it possible to have a finer allocation for resources offered for a particular joint
mission and resources reserved for other utilization in the organization.

4. Substrate networks
Two different satellite networks have been selected to provide the network substrate in INTEGRA,
counting also on the need to interoperate with external networks, such as fixed networks in the strategic
domain and WLAN networks in the tactical domain. The first satellite network is based on DVB-RCS,
a broadband GEO SATCOM system standardized by ETSI; which, offering high performance and
openness has enjoyed recent years a significant use in institutional markets and is also being used by
defense agencies. Nowadays a new version of the standard (named DVB-RCS2) is pending approval
from ETSI. The second one is based on a foreseen evolution of the LEO Iridium system, which actually
offers a rather narrowband channel ─between 600 and 2400 bps─ but with a global coverage in the
L-band, which means a much easier and robust reception compared to the broadband data-link which is
of a highly directional nature.

4.1. Iridium-based COM services

4
http://www.future-internet.eu/

-6-

ISBN-xxx-y-zz-yyyyyyy-x 341
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The primary use of the Iridium-like data-link in INTEGRA is the inter-connection of management
entities. Thus, it is used as an out-of-band channel to handle the establishment, maintenance and release
of broadband connections; as well as to realize generic management functions such as services
instantiation or platform monitoring. Furthermore, there is another key role played by this LEO based
data-link: thanks to enhanced robustness and low-latency characteristics, it is proposed for the remote
control of the UAV platform beyond line-of-sight (BLOS), both in coordinated teleoperation and
supervisory control modes. In order to make this possible with enough safety guarantees, several
measures have been taken, namely: the shift of error correction functions from link level to application
level; the optimization of control data volume; the efficient integration of the Iridium data-link with the
system software; and the use of adequate mission planning tools to predict and mitigate potential
data-link outperformance. In this way, appropriate UAV control loops can be designed accommodating
expected time delays (in the order of 600 milliseconds), data rates (in the order of 2400 bps) and
reliability levels (close to 97%).

4.2. DVB-RCS-based COM services

The DVB-RCS system architecture adopted in INTEGRA includes features particularly relevant for the
maritime surveillance scenario. Both star and mesh connection modes are supported so that flexible
distribution of data can be realized between and within the tactical and strategic domains. In addition,
both Ku band and Ka band are also supported, since a requirement for INTEGRA scenario is being able
to use as many satellites/networks as possible in the area of interest.
The INTEGRA DVB-RCS system uses the DVB-S2 waveform in the forward link, while two different
waveforms and access modes have been defined for the return link: the primary Multi-Frequency Time
Division Multiple Access (MF-TDMA) scheme specified in the DVB-RCS standard and a Single
Channel Per Carrier (SCPC) mode. In turns, the SCPC mode splits into a traditional sub-mode based on
explicit network allocations, and a more advanced sub-mode with no network coordination based on
radio sensing from the terminal side and pre-established transmission policies. The SCPC mode is
based on the DVB-S2 specification and is particularly targeted to UAV platform terminals. The use of
small antennas necessary for typical UAVs employed in maritime surveillance due to size and weight
constraints is incompatible with the off-axis emission limits defined for Ku Band by ETSI/FCC
regulations if the TDMA DVB-RCS waveform is used. Therefore UAVs will necessarily use the SCPC
mode in Ku band; while both TDMA and SCPC modes can be used in Ka band. Ground terminals can
use both modes in whatever band.
The broadband DVB-RCS network carries most of the payload traffic and supports end-user virtual
connections over the established virtual networks. A detailed survey has been accomplished on the
adequate support of the INTEGRA virtual network procedures at the DVB-RCS physical, link and
network layers; defining the needed data, control and management mechanisms to isolate bandwidth,
traffic, forwarding tables and topology between multiple overlays deployed over the substrate network.

-7-

ISBN-xxx-y-zz-yyyyyyy-x 342
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Particular attention has been paid to minimize traffic overhead and avoid extra delays due to the
virtualization mechanisms, so as to ensure optimal transport over the intrinsic bandwidth-constrained
and long-delay nature of satellite links. In the remaining sections of this clause we will describe how the
DVB-RCS data-link has been defined and characterized targeting a mobile aeronautical scenario.

4.3. DVB-RCS-based data-link characterization

The default DVB-RCS return link configuration uses a MF-TDMA transmission scheme, together with
a double binary turbo code, see Figure 3. It is also possible to configure the transmission chain with a
concatenated convolutional plus Reed Solomon code. This second choice enables the system to provide
high bandwidth efficiency for multiple users. A key issue to achieve high efficiency is a well-designed
demand-assignment algorithm which will use several capacity mechanisms to fulfill the QoS of
different classes of applications, e.g. real-time voice/video streaming or best-effort file transfer. The
INTEGRA management system includes services so that the end-user (or a system operator) can
dynamically define the QoS for every used application.
The DVB-RCS terminal onboard the UAV has a small satellite tracking antenna (typically 45 dBW
EIRP) which, depending on remaining link budget parameters (particularly satellite G/T) is expected to
provide up to 1.2 Mbps peak capacity in the return link and several Mbps in the downlink.

4.4. Aeronautical broadband channel model

A Rician (Line of Sight) channel has been chosen as reference for the DVB-RCS based data-link in Ku
and Ka bands, since it represents a good approach for the aeronautical scenario in general [3] and for
our maritime UAV case in particular, which overflies the sea a high percentage of the time. The Rice K
factor ─rate between main path power and the sum of secondary paths power─ strongly depends on the
antenna elevation, as it can be seen in Figure 4.

Figure 3 : DVB-RCS transmitter Figure 4 : Geometry of the Aircraft-Satellite channel

Several maximum Doppler spread frequencies have been defined, corresponding to the estimated speed
in the expected two main flight stages for the aircraft: i) UAV flying going from one interest area (or
landing/takeoff place) to another (150 Km/h) and ii) UAV hovering over an interest area (80 Km/h).
In order to take into account the whole set of potential impairments in the channel, a thorough link

-8-

ISBN-xxx-y-zz-yyyyyyy-x 343
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

budget has been added to the model, including effects from propagation (multipath and signal blockage
due to aircraft structure); weather (rain, scintillation, depolarization), interference/jamming ─linked to
regulatory aspects─; as well as detailed RF system features (antenna patterns and tracking
performances in particular). This detailed channel model is used at two stages: first to characterize the
physical layer performance by means of a bit-by-bit Matlab simulation, and then during system level
simulations or emulations.

5. Simulation and emulation for assessment

An ad-hoc SATCOM test platform has been developed in INTEGRA with both simulation and
emulation capabilities to verify and validate the designed COM management solution. Based on the
adaptation and extension of open source tools, the platform core elements are: i) a network-wide
simulation and emulation engine able to handle both real and statistical traffic; ii) a digital globe for
overall scenario configuration and analysis results (see Figure 5); iii) high-fidelity
satellite-terminal-platform models, including antenna patterns, mobility, 3D aircraft models and full RF
chain performance etc); and iv) basic human behavior models. Furthermore, the platform has been
properly designed to include highly realistic physical layer performance models of the DVB-RCS and
Iridium data-links. Such a models are given in the form of Packet Error rate (PER) versus C/N (carrier
to noise ratio) curves under given channel conditions (terminal speed, elevation angle, multipath etc.)
which have been obtained through extensive bit-level simulations. The platform is able to realistically
generate the temporal behavior of the COM channel (i.e. the C/N temporal evolution) and select the
proper physical-layer curve to apply the corresponding packet loss rates. Figure 6 and Figure 7 show
respectively an example of physical-layer performance curve and a diagram of the function carrying out
physical-layer packet loss processing.

Figure 5 : Digital globe interfaces in the system simulation platform

The implemented SOA-based COM management system depicted in Figure 2 is being integrated at

-9-

ISBN-xxx-y-zz-yyyyyyy-x 344
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

present with the SATCOM test platform for emulation testing. The management system is based on
abstract architecture, information and service models which by means of model-based techniques are
instantiated in concrete data models and protocols. The same model transformation techniques are used
to map the management models into the data structures of the test platform.

PER (Eb/No)
PHY simulations
AWGN + HPA

Uniform [0:1]
Mapping Random number
LMS model generator
function

C/N

C/N > Threshold

Correct
Packet loss
transmission

Figure 6: Simulation results for QPSK, Ricean Figure 7: Physical-channel error performance
channel, Doppler 1100 Hz and coding rate ½ implementation

6. Conclusions
A powerful network and service management framework using different satellite data-links has been
designed in the INTEGRA border surveillance project. An autonomic system approach has been
followed, so that all resources at system, subsystem and component level can be jointly conducted to
achieve overall mission objectives and individual performance can be kept high. Particular attention has
been given to the support of UAV COM, designing data-links suitable for an aeronautical satellite radio
channel in high frequency bands. The designed solutions are being tested today in a simulation and
emulation platform developed ad-hoc, allowing for in depth verification and validation, as well as for
personnel training and operational planning of networks and missions. The designed framework
proposes a flexible and automated solution to complex spectrum management in future security
missions while attaining state-of-the-art Bits per Hertz performance.

REFERENCES
[1] ETSI EN 301 790 V1.5.1
[2] www.iridium.com
[3] Neul, A., Hagenauer, J., Papke, W., Dolainsky, F., & Edbauer, F. (1987). Propagation Measurements
For The Aeronautical Satellite Channel. Proceedings of the 37th IEEE Vehicular Technology
Conference , 90-97.

- 10 -

ISBN-xxx-y-zz-yyyyyyy-x 345
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

BUILDING A UAS BASED ON PBL CONCEPT

RODRÍGUEZ-SEVILLANO, A.A.*, GANDÍA-AGÜERA, F.*, BARCALA-

MONTEJANO, M.A.*
* Professors at the UPM (Universidad Politécnica de Madrid), Madrid, Spain.

Abstract
During last decade, UAS platforms have been used as a tool bench for innovation in
aeronautical designs: innovation in propulsion, in structures, in flight control, in
aerodynamic design, … Moreover, many investigation centers, enterprises and universities
set the design and building of an UAS as a goal (a showcase) to show their capacities to the
aeronautical community. The enormous variety of design options, in any area of applied
engineering to UAS, leads to a wide dissemination of the research results; moreover, the
associated costs of the project are quite moderate.
Here we present the work carried out by a team of professors and undergraduate students -
at Universidad Politécnica de Madrid- to build and design an UAS. The paper will provide
a brief outline of the whole evolution, from the beginning to the first flight data.

Keywords: UAS design; undergraduate students; RP&M; PBL concept

1. INTRODUCTION
It is not infrequent for Academic Research Centres, especially Engineering Schools to offer
students collaborate in developing projects for the design and building of prototypes [1]. In several
occasions, it is based on student competition, to stimulate interest in UAS technology among
undergraduate students [2].
In this respect, our proposal does not differ much from the rule. We have not set out to establish a
new mark or conquer a pre-established level, either. We consider that the challenge of placing the
responsibility for the whole project in the hands of the students, who were carrying out their End of
Degree Project, would be valuable on its own. Because it would imply nothing less than being
responsible for the conceptual and preliminary design, the manufacturing, the design of the flight
control, the assembly and finishing of a Unmanned Aerial Vehicle (UAS) [3], [4].
It will be described the principal stages of the project, from conceptual design to the first flight. We
will show some building details and processes of the UAS. In that sense, RP&M (rapid prototyping
and manufacturing) has played an important role in the project. The concept “rapid prototyping”
(quick prototype generation) has become usual in conceptual design processes and prototype design
applied to engineering and other parts of the applied sciences [5], [6].
A group of professors from the EUIT Aeronáutica who had undertaken the design and building of
Unmanned Aerial System (UAS), soon realized that this engineering project could be approached as a
Project Based Learning activity to be developed by the students as a part of their compulsory Final
Project.
We expected to evaluate whether the adoption of this Project Based Learning approach [7], [8]
would be useful to actuate on those learning areas which would have to be enhanced in the new
curriculum, namely personal and professional skills, team work, communications and those activities
enabling the integration of conceptual knowledge and technical skills to the designing and building of
engineering products.
Thus the objective of PBL activity described in this paper was twofold: designing and building a
real and industry competing UAS and testing the adoption of a PBL as an influential part of our new

-1-

ISBN-xxx-y-zz-yyyyyyy-x 346
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

curriculum without changing it to a pure PBL.

2. UAS CONFIGURATIONS
Nowadays UAS present a very wide variety of types and characteristics. The diferencies are not
only in their geometry or configuration but also in performance. There are UAVs of over 14 tons of
weight and able to carry more than 1300 kg payload while there are others which can be carried in
one’s hand and manually launched. There is even a wide range of small dimensions micro-UAVs
which can fly indoors.
Firstly, because of this variety in the designs and applications of present UAS, it is quite complex to
establish a unique classification. Of all possible classifications we have chosen the one shown in [9],
[10].
The range of configurations available for UAS´ airframes is as wide as that for their designs o even
wider. It should be taken into account that UAS’ airframes are generally much smaller than manned
aircraft and that potential operators are more willing to accept unconventional solutions.
Runway performances (take off/landing procedures) are very convenient criteria for UAS
classification [11].
The configurations are the following:
• Horizontal Take-off and Landing, (HTOL).
• Vertical Take-off and Landing, (VTOL).
• Hybrid configurations.

3. DEFINITION OF REQUIREMENTS
The aim of the project is to design an UAS for civil aerial observation. Once its mission has been
defined, a wide research analysis of similar airplanes is carried out since, in the initial stage of the
project, it is quite common to resort to similar airplanes for estimating some important data of our
design. Next, the preliminary sizing is undertaken, taking as a basis weight and wing loading
estimations, the latter parameter being essential because it will allow us to study one of the most
important elements of the plane: the wing.
The objective of the first working team was to establish an initial airplane layout, taking into
account that many modifications will add up as the development of the project progresses.
By means of carrying out a comparison with similar planes, on a first approach, parameters were
estimated: maximum takeoff weight, payload, range, altitude, and cruising speed.
In this case, we decided an (HTOL) configuration. The control and stability features is based on a
conventional tailplane; a Tail-aft on Boom.UAS take-off. This design makes possible the power plant
installation aft the wing, with a pusher propeller. The forward part of the plane is free for the payload.

Figure 1. Ploting Draft. Conceptual design

-2-

ISBN-xxx-y-zz-yyyyyyy-x 347
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

4. PRELIMINARY SIZING
The internet was used as source of information for the study of similar UAS. First, a group of charts
of several airplanes sharing common characteristics (thrust –weight, payload-weight) will be
presented.

Figure 2. Weight to thrust ratio

PAYLOAD (Kg) PAYLOAD vs MTOW

120

100

80

60

40

20

0
0 100 200 300 400 500 600
MTOW (Kg)

Figure 3. Payload vs weight

By means of carrying out a comparison with similar planes, on a first approach, we estimated:
maximum takeoff weight, payload, range, altitude, and cruising speed.

-3-

ISBN-xxx-y-zz-yyyyyyy-x 348
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The profile of the planed flight mission will be:

Figure 4. Flight mission profile

By means of carrying out a comparison with similar planes, on a first approach, we estimated:
• Maximum takeoff weight: 30 kg.
• Payload: consisting of a video camera (or webcam), batteries and an air-to-ground radio
transmitter. Estimated weight: 8 kg.
• Range: an estimated value of 425 to 450 km with maximum payload and a 10% fuel reserve.
• Altitude: approximately 800 m.
• Cruising speed: around 30 to 34 m/s.
We concluded that the preliminary values for weights will be: WPL = 8 kg, MTOW = 30 kg, EW =
13 kg, FW = 9 kg.
In the pre-design stage [13], [14], [15] we will point out two main parameters: weight estimation
and wing loading estimation.
Before calculating the wing area, and, therefore, being able to establish our prototype’s wingspan,
we needed to calculate the wing loading. This value is conditioned to the corresponding phase of flight
we are going through; as a consequence, we had to determine the wing loading for each of the possible
situations. We calculated the wing loading variation, for each of the conditions which have been
studied. These were considered as final numerical values with respect to estimation of the wing area.
In the same way it could be defined the region of the wing area which verifies all different conditions
and the final estimated wing loading numerical value in that region.
Once all the preliminary sizing data had been determined, the following stage of the project was
deciding the final aircraft’s layout.
Making a decision on the type of fuselage to be used was one of the first objectives of this stage.
The main characteristics (requirements) that the fuselage structure should meet are the following:
• An aerodynamic design.
• Light weighted (an essential element for our design).
• Not difficult to manufacture and assemble.
• Easy to maintained.
• Maximum cargo capacity (open planed structure).

-4-

ISBN-xxx-y-zz-yyyyyyy-x 349
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 5. Preliminary design of the UAS

5. AERODYNAMIC AND PERFORMANCE ANALYSIS

The following stage in the project, after determining the preliminary configuration, was calculating
all aerodynamic features and the aircraft’s performance so that they could later be checked and
recalculated.
To complete all these steps, a sound analysis of the most important elements of the aircraft had to
be done, from the geometric and the aerodynamic points of view.
Once the required power and the power available have been defined, we can embark on the analysis
of the basic performance.
We concentrated on testing aerodynamic forces using our wind tunnel. The next figure presents the
model used in wind tunnel testing of the whole aircraft.

Figure 6. Wind tunnel tests model

After the study of the performance, we proceeded to check dynamic and static stability, together
with the aircraft control characteristics.

6. STRUCTURAL ANALYSIS
Once the study of the airplane from the aerodynamic point of view was completed, we had to carry
out a basic structural study. Before constructing the final prototype and as an essential part of the
whole project, we carried out a series of test at our workshop. We also checked the most critical
structural elements by means of static load tests.
The following photos show one stage of the static loading test. In this stage, we load the primary
longeron of the wing until its collapses.

-5-

ISBN-xxx-y-zz-yyyyyyy-x 350
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 7. Loading tests until crash

7. BUILDING THE AIRPLANE

After we concluded the stage corresponding to the preliminary essential calculations we moved
forward to the detailed designed and manufacturing stages. A global designed was developed and we
had to define materials, joining elements and fasteners, the most important of which are shown here.
We also include some comparisons between initial design of the most important elements of the
aircraft and the final manufactured ones.

Figure 8. Comparison between design and final mounting. Fin

Some covers and aircraft fairings are designed –based on RP&M- like surfaces whose geometry is
quite difficult to build by other methods, besides the obvious advantage of speed and versatility of
assembly.
Within the framework of the UAS design project, it was addressed the design of a rapid prototype
generation process for design and building some important elements of the UAS.
As an example of this, one of the elements manufactured using this technique is shown below: the
UAS´ tail-boom fairing. Figure shows the area resulting of joining the forward and aft parts of the
fairing.

-6-

ISBN-xxx-y-zz-yyyyyyy-x 351
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 9. Preliminary design and final prototype manufacturing process

Figure 10. Final result

8. GUIDANCE AND CONTROL

The purpose of the first stage of the Project is to be able to carry out several test flights in order to
check that all major systems and components meet the requirements they were initially assigned.
In order to confirm the checks carried out, we must employ guidance and control system which
meets the objectives previously defined and which, at the same time, is simple, reliable and economic.
The system selected is a radio controlled device similar to that of any remote radio-controlled
model airplane but with some differences. So as to guarantee a better protection, should any aircraft
component fail, we will use professional equipment with enough channels to handle more aircraft
components (4 being the minimum required for any model) or to have a twin-drive guidance system
(by means of duplicating the boosting mechanisms –servomechanisms, “servos”).
At later stages of the project we intend to provide the vehicle with an upgraded and autonomous
guidance system with a much longer range. The guidance system envisaged for the final stage of the
project is an automatic control system (autopilot).

-7-

ISBN-xxx-y-zz-yyyyyyy-x 352
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 11 shows the initial UAS model of the project, with the indication of the selected elements
which a will be acted by each servo.

Figure 11. Virtual model showing the 7 servos configuration

9. FIRST FLIGHTS
The purpose of the first phase of flights is to check if the most basic and important issues fulfill
their original tasks. These basic issues are:
• To check that the aerodynamic airfoil of the wings and fuselage are suitable. Taking into
account the possible payload and additional fuel that might be needed, as well as withstand
the structural load in the later phases.
• To check that the engine power is sufficient enough to be able to reach the desired velocity,
with the aircraft weight as in the initial phases as it is considered to be in the later stages.
• To check the video system works properly, with an optimal transmission.
• To check the guidance and control systems function correctly, that the servos are adequate in
strength and numbers, with a suitable receiver. Also, to check it is installed in a way as to not
present the pilot with problems when piloting the aircraft from the ground, with the selected
radio transmitter.
The first flights were carried out in June 2011, with very satisfactory results. We can see several
photos of it.

Figure 12. Maiden flight. June 2011

-8-

ISBN-xxx-y-zz-yyyyyyy-x 353
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

10. ACQUISITION OF THE GENERIC COMPETENCES BY PBL

When we first thought of designing and building a UAV at the EUIT Aeronáutica, in the last years
of the previous century, our aim was basically the achievement of various engineering objectives, such
as: to acquire the technological knowledge in this field, to propose some modest innovations for
processes and procedures, to apply our knowledge to a wholly aeronautical project, to develop our
own design of the cell and to create the software for managing its flight.
We conceived the Project to be developed entirely by final year students so that each of their End of
Degree Projectworks would constitute a part of the general Project. From this it logically followed
that, once the initial stages had started, it was clear that the project offered another research
perspective in addition to the one originally proposed. This was related to the development of the
skills and competences that the students needed to acquire if they were to carry out the Project
successfully. The following were considered the most relevant ones:
• Ability for teamwork.
• Manual skills for building and assembling.
• Ability for coordinating and leadership.
• Capacity for communicating ideas and proposals.
• Capacity for the interpretation of drafts and schematics.
• Quite an important knowledge of other disciplines such as electrical plants and electronics.
• Competence in the process for selecting the materials in accordance to their requirements.
• Establishing the technical requirements for a system.

11. CONCLUSSIONS
It has already been stated that the initial aim of our Project was the design and building of a UAS,
as a purely engineering task. In that sense, the results of the successful first flight are quite
satisfactory.
However, the evolution of the students’ involvement in the project along these years has led us to a
point where, while being still important, the initial aim is not more important than the development of
the learning environment, which is essential for the acquisition of generic competences.
The End of Degree Projectwork has traditionally been considered as a very valuable teaching
instrument. It facilitates the development of an integrative process where students have to apply all the
knowledge they have acquired along their course. The development of our project has shown that it
can also be used for the acquisition and assessment of generic competences, which otherwise would be
difficult to acquire and evaluate in an engineering course.
The students, being fully aware of how important these competences are for the development of
their careers, value very positively the opportunity they have been given to develop them.
We also wanted to know whether this kind of PBL activity would be useful to enhance personal and
professional skills. Although no quantitative assessment has been carried out yet, we are certain that in
our case the capacity to communicate with others, working in teams and integrating conceptual
knowledge into real products have largely and positively increase by the students involved in the
design and building of the Eyefly 1.
As opposed to this, some of the greatest difficulties which both students and teachers have had to
face have been in the area of team coordination, ability for building and assembling, lack of solid
financial support and the constant doubt of reaching success.
Nothing would have been done without the enthusiasm and the dedication of all the students who
have taken part in the project.

-9-

ISBN-xxx-y-zz-yyyyyyy-x 354
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

REFERENCES
[1] Chartier, B.J.; Gibson, B.A.. Project-Based Learning: A Search and Rescue UAV-Perceptions of
an Undergraduate Engineering Design Team: A Preliminary Study. Proceedings of the 18th
Annual Conference of the Australasian Association for Engineering Education, AaeE Conference,
Melbourne, 2007.
[2] Brown, D., Collins, C., McClure, D., Meriden, A.A., Profitt, P., Smith, M., Yadack, V.
“University of Kentucky Aerial Robotics Team: 2006 AUVSI Student UAV Competition
Design”. University of Kentucky, Lexington.
[3] Gandía-Agüera, F., Rodríguez-Sevillano, A. A., Barcala-Montejano, M. A., & Pérez-Álvarez, J.
(2008). The design and building of an UAV: an actual engineering project for student cooperative
work. International Technology, Education and Development Conference (INTED). Valencia, 3-5
March.
[4] Barcala-Montejano, M. A., Gandía-Agüera, F., Rodríguez-Sevillano, Á. A., Crespo-Moreno, J.,
Pérez-Álvarez, J., Gómez-Pérez, J. P., et al. (2010). The application of rapid prototyping in the
design of an UAV. 27th International Congress of the Aeronautical Sciences. Nice (France).
[5] Heyes, A. L., & Smith, D. A. (2004). Rapid Technique for Wind-Tunnel Model Manufacture.
Journal of Aircraft , 41 (2), 413-415.
[6] deWeck O.L., Kim I. Y., Graff C., Nadir W., Bell A.. Engineering design and rapid prototyping:
A rewarding CAD/CAE/CAM and CDIO experience for undergraduates. 1st Annual CDIO
Conference. Queen’s University. Kingston, Ontario, Canada. June 7 to 8, 2005.
[7] Synteta, P. (2003). Project-Based-e-Learning in Higher Education: The model and the method, the
practice and the portal. Studies in Communication, New Media in Education, 263-269.
[8] Thomas, J.W. (2001). A Review of Research of PBL. Accessed at
http://www.autodesk.com/foundation. The Autodesk Foundation (San Rafael, California).
[9] Visión Estratégica Española. Sistemas Aéreos No Tripulados (UAS). Plataforma Aeroespacial
Española. (SPANISH).
[10] Unmanned Systems Roadmap 2007-2032. Integrated Office of the Secretary of Defense.
Department of Defense.
[11] Austin, R. Unmanned Aircraft Systems. UAVS Design, Development and Deployement.Chichester,
Reino Unido，John Wiley & Sons Ltd.，2010.
[12] Preliminary design of a low speed, long endurance remote piloted vehicles (RPV) for civil
applications. Martínez-Val, R., Hernández, C. Aircraft Design 2 (1999) 167-182. Elsevier Science
Ltd.
[13] Raymer, D. P. (1992). Aircraft Design: A Conceptual Approach. Washington, DC: American
Institute of Aeronautics and Astronautics, Inc.
[14] Roskam J. Airplane design. Part I: preliminary sizing of airplanes. Ottawa (KA): Roskam
Aviation Engineering, 1985.
[15] Torenbeek E. Synthesis of subsonic airplane design. Delft: Delft University Press, 1982.

-10-

ISBN-xxx-y-zz-yyyyyyy-x 355
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ISBN-xxx-y-zz-yyyyyyy-x 356
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ISBN-xxx-y-zz-yyyyyyy-x 357
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ISBN-xxx-y-zz-yyyyyyy-x 358
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ISBN-xxx-y-zz-yyyyyyy-x 359
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ISBN-xxx-y-zz-yyyyyyy-x 360
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ISBN-xxx-y-zz-yyyyyyy-x 361
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ISBN-xxx-y-zz-yyyyyyy-x 362
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ISBN-xxx-y-zz-yyyyyyy-x 363
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ISBN-xxx-y-zz-yyyyyyy-x 364
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ISBN-xxx-y-zz-yyyyyyy-x 365
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Coordination of Multiple UAS for Tracking under Uncertainty

Jesús Capitán1 , Luis Merino2 and Anı́bal Ollero1,3

1
University of Seville, Camino de los Descubrimientos s/n, 41092, Seville, Spain
2
Pablo de Olavide University, Crta. Utrera, km. 1, 41013, Seville, Spain
3
Center for Advanced Aerospace Technology, C. Wilbur y Orville Wright 17-19-21, 41309, La Rinconada, Spain

Abstract

This paper presents how Partially Observable Markov Decision Processes (POMDPs) can be used
for controlling fleets of UAS under uncertainty1 . POMDPs provide a sound mathematical framework
to deal with planning actions when tasks outcomes and perception are uncertain, although their
computational complexity have precluded their use for multi-robot applications. However, in this
work a scalable approach based on POMDPs is presented. Instead of solving a complex model for
the whole team, distributed models in which the UAS have no knowledge about others’ actions are
considered. Then, a decentralized data fusion algorithm is used to share information from all the
sensors and obtain an implicit coordination between the UAS. Moreover, the proposed approach is
applied to a tracking application with a fleet of UAS, and some simulations are presented to show
its advantages and how the mentioned coordination arises between the members of the team.

keywords: coordination, multiple UAS, decentralized data fusion

1 INTRODUCTION
Many applications with UAS, such as tracking, surveillance or environmental monitoring, require to deal
with uncertainties. For instance, in UAS surveillance, where one UAS or a team of UAS should monitor
an area and track potential targets of interest, imperfect sensors, incomplete models and other sources of
uncertainty require taking into account the uncertainty on the target’s position when planning the best
actions [5]. In addition, if the application involves several UAS with different sensors, the cooperation
between vehicles is also desirable.
Techniques for decision-theoretic planning under uncertainty are becoming more popular in the field
of cooperative robotics. For instance, in [15, 17] different probabilistic approaches for mapping and
tracking with robotic networks are presented. [16, 2] also propose decentralized Bayesian approaches for
multiple-target optimal search with UAS. Usually, these approaches are greedy, in the sense that only
the next action is considered, and no planning is involved. Additionally, [4, 8] make a distinction between
cooperative and coordinated information-theoretic solutions, and propose the latter to control fleets of
1 This work has been partially supported by the EU-funded project EC-SAFEMOBIL (FP7, ICT-288082)

1
ISBN-xxx-y-zz-yyyyyyy-x 366
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 1: Multiple UAS can cooperate in tasks like surveillance and tracking. Left: two autonomous
helicopters in a surveillance mission outdoors. Right: three quadrotors following targets in an indoor
testbed. Robust mission execution requires to take into account uncertainties about the target’s position
and motion model.

UAS. In a coordinated approach, the members of the team have no knowledge about the others’ models
or control actions, but they exchange some information that may influence implicitly other members’
subsequent decisions. Hence, sharing fused perception information or the impact of others’ control
actions over a certain objective function, and acting locally, a coordinated behavior can be obtained.
Within a Bayesian framework, Partially Observable Markov Decision Processes (POMDPs) provide
an elegant way to model the interaction of a team of UAS with their environment. Based on prior
knowledge of the sensors and actuators models and the environment dynamics, policies which indicate
the UAS how to act can be computed. These policies can be extracted by optimizing iteratively a
certain value function over the belief space. The main problem is that the information space (the set
of potential actions and observations), and hence the computational complexity of POMDPs increases
dramatically with the number of UAS in the team.
There exist approximate methods to obtain POMDP policies that can deal with large belief spaces.
In this sense, point-based algorithms [9, 14, 12, 13] represent the value function by a set of vectors over
the belief space (α-vectors) and restrict the optimization procedure to a bounded set of feasible beliefs.
Particularly, [14] propose a point-based solver, called Perseus, where no computation is needed for all
the belief points at every iteration, hence improving performance. In [10], an extension of Perseus is
presented. It is called Symbolic Perseus and uses Algebraic Decision Diagrams (ADDs) [6] in order to
optimize the operations in the original Perseus for factored POMDPs. Besides, [7] propose SARSOP,
which maintains a tree-shaped set of reachable beliefs that is expanded at every iteration using the best
policy so far.

2
ISBN-xxx-y-zz-yyyyyyy-x 367
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

However, the above methods still find hard restrictions when applied to multiple UAS, since the
space of actions and observations grows exponentially together with the belief space. This leads to
very complex POMDPs whose policies are highly difficult to compute but for the simplest problems.
Hence, this paper describes a different method that can overcome this issue for certain applications.
In particular, instead of solving a large POMDP considering all the UAS, a decentralized approach
where each UAS acts without any knowledge about the others’ actions is proposed. Then, an implicit
coordination is achieved (according to the idea in [4]) by sharing a fused belief of the state among the
team members. Moreover, sensors properties and desired objectives could be changed from one UAS to
another in order to combine different behaviors within the team.
The application of decision-making under uncertainty that is addressed in this paper is target tracking
with a team of UAS. Bearing-only sensors (cameras) on board the UAS are considered to detect the
target. In this case, a single sensor does not provide enough information about the distance to the
target, what makes tougher a proper tracking. Moreover, the motion of the target is not predictable.
Thus, a team of UAS can act together in order to achieve a more efficient performance.
The paper is structured as follows: Section 2 gives some theoretical background about POMDPs;
Section 3 presents a Decentralized Data Fusion approach in order to estimate a shared belief; Sec-
tion 4 details the coordinated tracking approach for multiple UAS; and Sections 5 and 6 explain the
experimental results and conclusions, respectively.

2 POMDP MODEL
Formally, a POMDP is defined by the tuple hS, A, Z, T, O, R, h, γi. The meaning of every component is
the following:

• State space: All the information available of the environment is encoded in the state, which is not
fully observable. The system’s state at time step t is defined by st ∈ S, where S is the finite set
of all the possible states.

• Action space: Every UAS can take an action each time step. A is the finite set of possible actions,
whereas at ∈ A is the action taken at a certain time step.

• Observation space: Given a time step t, after executing an action, a UAS can make a measurement
or observation zt ∈ Z, where Z is the finite set of all the possible observations.

• Transition function: After executing an action, the state can vary probabilistically. This prob-
ability density function is modeled by the transition function T : S × A × S −→ [0, 1], where
T (s0 , a, s) = p(st = s0 |at−1 = a, st−1 = s). It represents the probability of ending in the state s0
if the UAS performs the action a in the state s.

• Observation function: The observations gather information from the current state and are related
probabilistically to this state. This probability density function is modeled by the observation

3
ISBN-xxx-y-zz-yyyyyyy-x 368
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

function O : Z × A × S −→ [0, 1], where O(z, a, s0 ) = p(zt = z|at−1 = a, st = s0 ). It gives the

probability of observing z if the action a is performed and the resulting state is s0 .

• Reward function: A POMDP selects the best actions so that a reward function is optimized.
Thus, the behavior of the system is determined by this function. R : S × A −→ < is the reward
function, where R(s, a) is the reward obtained by executing action a in state s. Here, the reward
is assumed to be bounded. Since it allows to model quite complex goals, the reward function is a
very powerful tool, and hence, its right design is crucial.

• Horizon and discount factor : Then, the goal of a UAS is to maximize the expected reward earned
over some time frame. The horizon h defines this time frame by specifying the number of time
hP i
h t
steps the UAS must plan for. Thus, the objective is to maximize the sum E t=0 γ rt , where rt

is the reward at time t, E[ ] is the mathematical expectation, and γ ∈ [0, 1) is a discount factor,
which ensures that the sum is finite when h → ∞.

Given that it is not directly observable, the actual state cannot be known by the UAS. Instead, the
information about the environment must be encoded in a belief distribution, which indicates a probably
density function over the state space. Moreover, under the Markov assumption, this belief can be
updated (Bayesian filter) recursively using the previous belief bt−1 (s), the most recent action a and the
most recent observation z:
X
bt (s0 ) = ηO(z, a, s0 ) T (s0 , a, s)bt−1 (s) (1)
s∈S

where η acts as a normalizing constant such that bt remains a probability distribution.

Once the UAS has computed the belief state at a certain time, the best action is determined by a
policy π(b), which maps all the possible beliefs to their optimal actions. The policy is computed in such
a way that it maximizes the value function:
" h
#
X
π t
V (b) = E γ r(bt , π(bt ))|b0 = b (2)
t=0
P
where r(bt , π(bt )) = s∈S R(s, π(bt ))bt (s).

3 DECENTRALIZED DATA FUSION

In the multi-UAS case, maintaining a belief over the state space according to (1) is not trivial. Given
a team with N UAS, a centralized node with access to all the information would update the belief as
follows:
X
0 J 0
bcen J
t (s ) = ηO(z , a , s ) T (s0 , aJ , s)bcen
t−1 (s) (3)
s∈S

where aJ = ha1 , · · · , aN i is the joint action and z J = hz 1 , · · · , z N i the joint measurement. However,
if the belief estimation is decentralized and each UAS i uses only its local information (action ai and

4
ISBN-xxx-y-zz-yyyyyyy-x 369
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

observation z i ), some communication must be allowed among the UAS so that they can recover this
centralized belief locally [11].
The conditional independence assumption of the measurements (given the state at that instant s0 )
is typical in Bayesian data fusion. This is reasonable when the measurements obtained by each agent
do not depend on the state of the other agents. Therefore, assuming this independence (O(z J , aJ , s0 ) =
Q i i 0
i O(z , a , s )) and assuming that actions are known when predicting, it is possible to combine locally

received beliefs from other UAS with the one from UAS i, bit (s0 ), to recover the centralized belief:
Y bj (s0 )
0 i 0 t
bcen
t (s ) ∝ bt (s ) (4)
j6=i
bij 0
t (s )

Equation 4 fuses the belief in UAS i with the one received from j by multiplying them. The common
information previously exchanged by the UAS bij 0
t (s ) must be removed not to count it twice. This

common information can be maintained by a separate filter called channel filter. Further details about
the channel filter and the equation above can be seen in [1].
It has been shown [3] that if the state is dynamic, it is possible to obtain locally the same belief as in
a centralized node with access to all the information available, by including delayed states in the belief.
This represents a common belief signal for all the team members. If only the current state is considered,
some information is lost with respect to an ideal centralized fusion unless the UAS communicate every
time they gather new information [1].

4 COORDINATED TRACKING WITH MULTIPLE UAS

An application for tracking a target by means of multiple UAS is considered here. In this problem there
is a moving target and a team of N hovering-capable UAS (like an helicopter or a quadcopter) which
are the pursuers (see Fig. 1). Each of these UAS carries a camera in its nose which determines whether
the target is visible or not (see Fig. 2, right). The UAS can communicate among themselves by wireless
links. Then, the objective is to find the target in the environment and localize it as well as possible.
The state is composed of the position of the target (tl ) and the position of the pursuers (p1l , · · · , pN
l ).

Since bearing-only sensors are used, the headings of every UAS (p1h , · · · , pN
h ) are also considered and

included in the state. Moreover, the space is discretized into a cell grid, and a map of the scenario is
assumed to be known. Thus, an occupancy grid can be obtained indicating which cells are attainable
and which are non-flying zones. Then, the locations of the target and every pursuer are specified by
cells, and just non-occupied cells are possible values. There are also four possible headings for every
robot: north, west, south or east.
Every sensor provides a boolean measurement: detected or non-detected. These sensors proceed as
it follows, if the target is out of its field of view, the sensor produces a non-detected measurement.
However, when the target is within its field of view, it can be detected with a probability pD .
In addition, each UAS can choose at each time step between four possible actions: hover, turn right,
turn left or go forward. hover explains itself; when turning, the robot changes its heading 90o degrees;

5
ISBN-xxx-y-zz-yyyyyyy-x 370
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 2: Simulated environment (left). The black cells are non-flying zones, whereas the yellow ones
represent the field of view of the camera on board the UAS (right). If the target is in one of the cells
with crosses, a high reward is obtained.

and when going forward, it moves to the cell ahead.

Finally, the key point is how to design a reward function so that the target is localized and tracked
by the team. All the UAS are supposed to provide additional information about the target, so their
individual reward functions {R1 (s, a), R2 (s, a), · · · , RN (s, a)} encode a high reward when the target is
within their fields of view and no reward otherwise. For all the members of the team, a cost of 1 is
associated with the actions.
Solving the above POMDP for the whole team is a highly non-scalable approach regarding the
number of UAS. Hence, a different scheme is proposed where each UAS i solves its own POMDP without
considering the other UAS. That POMDP has state si = (pil , pih , tl ), and reward Ri (s, a). Then, once
the policies have been calculated, the coordination is achieved during the execution phase by sharing a
fused belief state bcen (s) that considers information from all the robots. This belief can be computed
in a decentralized manner with the method in Section 3, what makes the approach scalable for multiple
UAS.
With this approach based on POMDPs, coordination arises implicitly due to the fused belief, and
there is no need to solve the original POMDP, which was far more complex. Thus, the approach can scale
with the number of UAS in the team, since this does not influence the complexity of the independent
models to solve. The drawback is that the policies are sub-optimal, as the UAS do not reason about the
others’ actions, but through proper design of the rewards and the shared information that is provided
by the decentralized data fusion, a coordinated behavior useful for many applications can be obtained.

5 EXPERIMENTS
In order to test the proposed methods for target tracking, several simulations are presented here. A
simulated environment was created in Matlab and discretized into the 12x10 grid shown in Fig. 2. For
these simulations, a team with four homogeneous UAS is considered. All of them carry bearing-only
sensors with the same field of view (see Fig. 2) and detection probability (pD = 0.9). The role of the

6
ISBN-xxx-y-zz-yyyyyyy-x 371
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(a) (b)

(c) (d)

Figure 3: Sequence of an experiment using totally independent UAS (steps 8, 16, 26 and 34). The
surface represents the belief of the red UAS on the target position. The solid lines are the trajectories
of the UAS, representing the dots their initial positions. The UAS fly at different heights to avoid
collisions.

UAS is to get as close as possible to the target to reduce its uncertainty. Thus, the UAS get a high
reward (+100) when the target is in one of the closest cells of their field of view, otherwise the reward
is zero. Fig. 2 shows an example of the field of view of a UAS. Besides, the cells that would get high
rewards are marked with crosses.
In these experiments, no uncertainty is considered for the UAS’ control. Thus, every time a UAS takes
one of the actions described in the previous section, it is supposed to be executed perfectly. Nevertheless,
there is some uncertainty associated with the observations and the target model. Actually, the target
is assumed to move randomly. Therefore, the transition function for its position tl indicates that, from
one time step to the next, the target can move to any of its 8-connected cells with the same probability
(only non-obstacle cells are considered in order to calculate that probability).
Two main experiments are presented here. In the first one, each UAS calculates its policy indepen-
dently and run it without considering the other. In the second experiment, the approach proposed in
this paper is tested. Even though the policies are calculated separately, the UAS share a belief during
the execution. In both experiments, the initial position of the UAS as well as the target trajectory were
equal. Nonetheless, in order to include some uncertainty in its behavior, at every time step, the target
could (with equal probability) either stay in the same cell or follow the prefixed path. Moreover, all the

7
ISBN-xxx-y-zz-yyyyyyy-x 372
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(a) (b)

(c) (d)

Figure 4: Sequence of an experiment using the coordinated approach (steps 5, 10, 15 and 22). The
surface represents the fused belief for the red UAS.

policies were calculated with the algorithm SARSOP.

Results of the first experiment are shown in Fig. 3. Since each UAS looks for the target on its own,
they encounter some difficulties due to the lack of range measurements. In fact, the belief of the UAS
is quite flat during the whole experiment. Of course, when one of them detects the target the others do
not notice it and keep searching around, so the UAS trajectories are not coordinated.
The results of the second experiment are summarized in Fig. 4. In this case, the localization of the
target is improved and beliefs with high peaks are achieved. This is because all the UAS know about
the target position once it is detected (fused belief), so they coordinate their trajectories toward the
target to take more measurements.
Both approaches, considering communication between the robots (a shared belief) and no commu-
nication, were also compared running 500 simulations of 20 time steps each one. In all the runs, the
four UAS started at the same positions as above and the target followed the same path (also with the
P
randomness explained before). The average expected reward and the average entropy (− i p(st =
si ) log(p(st = si ))) were computed for all the simulations. The results are shown in Table 1. It can be
seen how the implicit coordination achieved through decentralized data fusion leads to higher rewards
and lower entropy, i.e. the proposed approach reduces the uncertainty in the target position.

8
ISBN-xxx-y-zz-yyyyyyy-x 373
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Coordinated Non-coordinated
Reward 976.91 ± 13.81 540.14 ± 8.6
Entropy 1.85 ± 0.01 3.35 ± 0.01

Table 1: Average reward and entropy for 500 simulations with coordination and without it. Standard
deviations are also shown.

6 CONCLUSIONS
This paper has presented preliminary simulations that show how POMDPs can be used for multi-UAS
planning under uncertainty. Even though the intractability of some models is still a drawback for multi-
UAS teams, the paper has shown how solving the POMDPs in a decentralized manner and sharing
information can lead to implicit coordination within the team, without having to solve the full POMDP
model. In this sense, results with non-coordinated and coordinated UAS were compared and a better
performance was achieved with the latter.
Since each UAS does not reason about the others’ actions, no intentional cooperation is considered.
However, the proposed approach is rather scalable and the whole scheme (including the belief estimation)
is totally decentralized.
Next steps include the implementation of the proposed method in a fleet of UAS; particularly, in
the indoor CATEC’s testbed with up to 10 quadrotors (see Fig. 1, right). Furthermore, future works
will address techniques with more explicit cooperation between the UAS. And, although one particular
application has been presented, due to their generality and adaptability the same techniques can be
applied to other scenarios, like environmental monitoring, as well as to the cooperation between static
sensors and UAS for cooperative navigation when the UAS lack of accurate sensors for localization.

REFERENCES
[1] F. Bourgault and H.F. Durrant-Whyte. Communication in general decentralized filters and the
coordinated search strategy. In Proc. of The 7th Int. Conf. on Information Fusion, 2004.

[2] F. Bourgault, T. Furukawa, and H.F. Durrant-Whyte. Decentralized bayesian negotiation for co-
operative search. In Proceedings of the IEEE/RSJ International Conference on Intelligent Robots
and Systems, pages 2681–2686, 2004.

[3] J. Capitan, L. Merino, F. Caballero, and A. Ollero. Delayed-State Information Filter for Coop-
erative Decentralized Tracking. In Proceedings of the International Conference on Robotics and
Automation, 2009.

9
ISBN-xxx-y-zz-yyyyyyy-x 374
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[4] Ben Grocholsky, Alexei Makarenko, Tobias Kaupp, and Hugh F. Durrant-Whyte. Lecture notes
in Computer Science, volume 2634, chapter Scalable Control of Decentralised Sensor Platforms.
Springer, 2003.

[5] Ruijie He, Abraham Bachrach, and Nicholas Roy. Efficient Planning under Uncertainty for a Target-
Tracking Micro-Aerial Vehicle. In IEEE International Conference on Robotics and Automation,
2010.

[6] J. Hoey, R. St-Aubin, A. Hu, and C. Boutilier. SPUDD: Stochastic planning using decision dia-
grams. In Proceedings of the Fifteenth Conference on Uncertainty in Artificial Intelligence, 1999.

[7] H. Kurniawati, D. Hsu, and W.S. Lee. SARSOP: Efficient point-based POMDP planning by ap-
proximating optimally reachable belief spaces. In Proc. Robotics: Science and Systems, 2008.

[8] George Mathews, Hugh Durrant-Whyte, and Mikhail Prokopenko. Advances in Applied Self-
Organizing Systems, chapter Decentralized Decision Making for Multiagent Systems. Springer-
Verlag, 2008.

[9] J. Pineau, G. Gordon, and S. Thrun. Point-based value iteration: An anytime algorithm for
POMDPs. In Proceedings of the Eighteenth International Joint Conference on Artificial Intelligence,
pages 1025–1032, 2003.

[10] Pascal Poupart. Exploiting Structure to Efficiently Solve Large Scale Partially Observable Markov
Decision Processes. PhD thesis, University of Toronto, 2005.

[11] D. V. Pynadath and M. Tambe. The communicative multiagent team decision problem: Analyzing
teamwork theories and models. Journal of Artificial Intelligence Research, 16:389–423, 2002.

[12] T. Smith and R. Simmons. Heuristic search value iteration for POMDPs. In Proceedings of the
20th Conference on Uncertainty in Artificial Intelligence, pages 520–527, 2004.

[13] T. Smith and R. Simmons. Point-based POMDP algorithms: improved analysis and implementa-
tion. In Proceedings of the 21th Conference on Uncertainty in Artificial Intelligence, pages 542–547,
2005.

[14] Matthijs T. J. Spaan and Nikos Vlassis. Perseus: Randomized point-based value iteration for
POMDPs. Journal of Artificial Intelligence Research, 24:195–220, 2005.

[15] A.W. Stroupe and T. Balch. Value-based action selection for observation with robot teams using
probabilistic techniques. Robotics and Autonomous Systems, 50:85–97, February 2005.

[16] E.-M. Wong, F. Bourgault, and T. Furukawa. Multi-vehicle bayesian search for multiple lost targets.
In Proceedings of the International Conference on Robotics and Automation, pages 3169–3174, 2005.

[17] Feng Zhao, Jaewon Shin, and J. Reich. Information-driven dynamic sensor collaboration. Signal
Processing Magazine, IEEE, 19(2):61–72, March 2002.

10
ISBN-xxx-y-zz-yyyyyyy-x 375
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Electromechanical Actuation Systems for Unmanned Systems with Zero

Environmental Impact
VERONICA GONZALEZ, AITOR ISTURIZ, JAVIER VIÑALS
Sener Ingenieria y Sistemas SA, Av. Zugazarte 56 – 48930 (Getxo), SPAIN
[email protected]; [email protected]; [email protected]

Introduction
SENER participates in the CENIT SINTONIA project in the actuation and control surfaces area, in
two parallel activities: electromechanical actuators (EMA), and adaptive control surfaces (winglets),
with the capacity to modify their shape or morphing capacity.

For EMAs, SENER is developing two smart actuator types for typical kinematics in unmanned
systems, with special care in space and weight critical requirements. The systems are linear and rotary
actuators for rudder or winglet-type surfaces. EMA actuators will have health monitoring systems
integrated for predictive maintenance. The intent is to finally compare the two actuation system types
and their services to develop the optimum unmanned system actuation. The design will be as modular
as possible in order to reach different operational ranges for different types of unmanned systems,
trying to establish common EMAs for different applications inside the UAS / UAVs. The
standardization issues are also included in the design in order to reduce and optimize spare and
maintenance tasks.

Keywords: UAS, UAV, electromechanical actuators, health monitoring

A state of the technique

The number of Unmanned Aerial Vehicles (UAVs) being developed and flown is growing in the
recent years. UAVs are used in a wide variety of applications like scientific missions, coast watch,
news broadcasting, fire fighting and the most common, defense applications including Intelligence,
Surveillance, Target Acquisition and Reconnaissance missions (ISTAR).

This continuous development demands from to the aerospace industry better performances, higher
efficiency lower operational costs and improved reliability of its UAVs. These system level
requirements are translated consequently to the subsystems suppliers (structures and materials,
communication equipments, power plants, actuation systems…). This growing market opens to
suppliers’ new business opportunities.

-1-

ISBN-xxx-y-zz-yyyyyyy-x 376
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Conventional hydraulic actuators are used but require a high amount of pipes for the actuator fluid.
This implies a high vulnerability and maintenance level. The new technology referred to EMA
actuators does not show the same vulnerability as a conventional hydraulic system, regarding the
flammable oil and maintenance of the tube systems. In addition to the higher efficiency, lower cost
and weight offered by the EMA solutions there is also a possibility to re-generate energy from the
motion of the actuators back to the power system.

The motion of the control surfaces of EMA systems shows a dynamic nature. On top of the dynamic
behavior these movements are often applied under very short periods of time, which require quite
high loading output of the actuator. A simplified motion profile of an Unmanned Airborne Vehicle,
UAV actuator is given in the following table.

Table 1 - An example of an UAV mission profile [Reference 1]

The most UAVs actuation systems are based on electromechanical technology (EMA). It is used for
primary and secondary flight control as for other functions like deck grid locks or emergency
parachute flight. The major benefits of this technologies compared to the technologies used in manned
aircrafts like central hydraulics or Electro Hydraulic Actuation (EHA) are a better dynamical behavior,
configuration flexibility, lower maintenance costs and weight/volume reduction.

Today, this technology has to increase its maturity to be integrated on manned civil aircrafts basically
because of reliability issues but seems to be enough to fit the lower reliability requirements of UAVs.
The operation of EMAs in primary and secondary UAV flight control represents an excellent data
base in terms of operational faults, reliability statistics or failure modes.

There is a trend for small & smart actuators in UAV military applications. For example, in 1996
DARPA initiated the Micro Air Vehicle (MAV) Program initiative, seeking to develop emerging
technologies for military surveillance and reconnaissance applications [Reference 2]. MAVs are micro

-2-

ISBN-xxx-y-zz-yyyyyyy-x 377
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

or nano Unmanned Aerial Vehicles (UAVs) requiring tiny actuators & motors. Another type of
example is given by the complex optic instruments, such as Infra Red (IR) cameras, embedded on
military vehicles or aircrafts. Such cameras also request small& smart actuators working in severe
environment. [Reference 3 and 4] are focused on the research in the field of smart & small actuators
for Space & Military fields, updating with a complementary approach: To illustrate the trend, the
actuators and mechanisms, the development and qualification to meet technical requirements but
logically found also applications in military and micro aerial vehicles fields, for various
micro-mechatronic functions.

EMA challenges
As explained in the introduction of this paper, the wide variety of UAVs applications demands
different EMAs specifications. Each new development requires several engineering hours (mechanical,
electrical, software, quality…) and material costs (prototypes, test benches, …) distributed in several
months that impacts as non recurring cost the final price of actuation systems. Several times, this
enormous effort can not be justified for small series productions or development duration does not
result admissible.

A simplification and a standardization of the preliminary design phase carries out a shortening of the
complete development schedule and a reduction of the non recurring costs. This methodology permits
more efficient more rapid response to the new specifications. To make possible this improvement, all
the components involved in actuations systems shall be previously designed taking in account two
characteristics, the scalability and the modularity.

SENER and SINTONIA CENIT PROJECT

SENER participates in the SINTONIA project in the actuation and control surfaces area, carrying out
two parallel projects: one dedicated to electromechanical actuators (EMA) and the other regarding
control surfaces (winglets), adaptive and with the capacity to modify their shape or morphing
capacity.

For electromechanical actuators, SENER is studying two smart actuator types for typical kinematics
in unmanned systems, where available space and weight are very critical requirements. The systems to
be developed are linear and rotary actuators for rudder or winglet-type surfaces. EMA actuators will
have monitoring systems integrated for predictive maintenance. The intent is to finally compare the
two actuation system types and their services to develop the optimum unmanned system. The design
will be as modular as possible in order to reach different operational ranges for different types of
unmanned systems.

-3-

ISBN-xxx-y-zz-yyyyyyy-x 378
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 1 UAV rotary actuators, developed by SENER

The main topics developed here are focusing on the following features:
 Looking for modularity (modules common for different actuators and applications)
 Looking for scalability (family of actuators suitable to be scalable for different loads,
stroke…)
 Looking for reducing development time and NRC (non recurrent costs)

One of the topics covered in the project developed by SENER is the Health Monitoring approach for
EMAs in UAV and UAS applications. UAVs are good candidates for model-based approach and have
specific requirements for fully automated embedded diagnostics and prognostics. Health management
is a design-in feature for UAVs and faults are handled through automated contingency management
methods.

The Health monitoring systems will allow developing “smart” EMAs that reduce the environmental
impact, by means:
- Increase reliability
- Reduce maintenance tasks and operations
- Reduce spares and repairs (that means less machine operations)
- Reduce EMA life cycle cost
- EMA against EHA (or hydraulic) will permit reduction in the fluid systems, pumps, tubes … and the
corresponding contamination possibilities

The Health Monitoring will be applied to the mechanical and electronic systems and with direct
impact in the total system behavior.

-4-

ISBN-xxx-y-zz-yyyyyyy-x 379
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Results
Project is on going and expected results are:
- UAV EMA validation in SENER test facilities
- UAV EMA modularity optimization
- UAV EMA commonalities in design features: similar mechanical and electronic architecture that
implies reduction in the non recurring development costs
- UAV EMA similar for different application inside the UAV platform: to use the same design (this is
the intention) for different functionalities (to move an aileron, for landing gear …) that makes easy the
UAV control and power laws

CONCLUSION
Big effort is inside the Health Monitoring system as a tool that allows a predictive and intelligent
maintenance protocol to be implemented in EMA systems.
Some of the expected innovations are:
• Definition of EMA Health Monitoring System Strategy
• Establish EMA Health Management
• Develop sensor and signal processing technologies
• Integrate wireless data transmission technology
• Selection of Health Assessment technologies
• Definition of the predictive maintenance strategy
• Innovative and strong EMA architecture with reliable and safe health monitoring
• Define feasible solutions for integration of sensors within a EMA, without reduction in performance
or reliability of "intelligent" EMAs
• Identify solutions with associated cost-benefits during whole life cycle

References
Ref.1: Analysis of Electro-Mechanical Actuator Systems in More Electric Aircraft Applications,
Mohsen Torabzadeh-Tari, DEPARTMENT OF ELECTRICAL ENGINEERING ROYAL INSTITUTE
OF TECHNOLOGY SWEDEN, ISBN 91-7178-099-8; May 2005.
Ref. 2: http://www.defense-update.com/features/du-2-04/mav-darpa.htm
Ref. 3: P. Jänker, New Actuators for Aircraft and Space Applications, Proc. Actuator 2006, Ed Messe
Bremen, 2006, pp 324-330
Ref.4: P. Jänker(1), F. Claeyssen(2), B. Grohmann(1), M. Christmann(1), T. Lorkowski(1), R.
LeLetty(2), O. Sosniki(2), A. Pages(2) (1) EADS, MUNICH, GERMANY;(2) CEDRAT
TECHNOLOGIES, MEYLAN, FRANCE, New Actuators for Aircraft and Space Applications

-5-

ISBN-xxx-y-zz-yyyyyyy-x 380
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

OpenFOAM as Free CFD Tool for Education, Research and Design in

UAS

DIEGO DOMÍNGUEZ FERNÁNDEZ, DEIBI LÓPEZ RODRÍGUEZ,

JESÚS GONZALO DE GRADO
Área de Ingeniería Aeroespacial. Escuela de Ingenierías Industrial, Informática y Aeronáutica.
Campus de Vegazana. Universidad de León. 24071 León (Spain)
[email protected] [email protected] [email protected]

Abstract
The aim of this paper is to show an open source and free license software for educational
and research purposes which are directly applied to UAV´s design process. Moreover, the
analysis of several common cases of Fluid Mechanics Theory can be better understood via
CFD, which allows students to enhance the theoretical concepts and become familiar with
numerical techniques, a powerful tool in today´s industrial activity.

Keywords: CFD teaching, UAS design, Design tools

1. INTRODUCTION
Solving fluid dynamics problems in UAS design by means of computational analysis is commonly
known as Computational Fluid Dynamics (CFD). This name covers techniques which allow to solve
numerically the equations that govern fluids, usually called Navier-Stokes equations: continuity,
momentum and energy equation. Although these techniques started to be used in the mid 60s, greatest
advances took place at the end of the 20th century thanks to the impressive development of computers,
which allows to resolve fluid governing equations quickly and accurately [1]. Computational fluid
dynamics is a powerful tool to solve engineering problems involving fluid flow and heat transfer, as in
UAS design and teaching.

2. WHAT IS CFD?
The continuity, momentum and energy equations are known as conservation principles and can be
expressed in terms of mathematical functions usually called Navier-Stokes equations. Together with
the perfect gas equation of state, they can be used to calculate the fluid’s velocity (V), temperature (T),
pressure (p) and density (ρ). These equations are [2]:

 Continuity equation:

-1-

ISBN-xxx-y-zz-yyyyyyy-x 381
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(1)

 Momentum equation:

(2)

where are the mass dependant forces and the viscous stress tensor.

 Energy equation:

(3)

where is the internal fluid energy per mass unit, is the conductive heat flux, is the

radiation heat flux, is the Rayleigh dissipation function and, finally, is the heat addition by

any other mechanism.

To solve these equations, each problem imposes its own boundary and initial conditions which allow
the problem to be solved numerically.
It is almost impossible to find an analytical solution for the Navier-Stokes equations, and it is only
possible in some educational cases when large amount of simplifications are made, allowing a
reduction in the dimensions number or removing terms which are negligible [3]. To obtain numerical
solutions, space and equations must be discretized; in this way, the solution is only obtained at certain
points of the control volume, which are defined by a mesh. Algorithms implemented in CFD software
iteratively resolve the equations applied over the mesh.

3. CFD, A POWERFUL EDUCATIONAL TOOL IN UAS TRAINING

Nowadays, more and more companies are introducing CFD tools in their work methodology and
designing procedures. However, the success of its utilization depends not only on the code, but also,
and most common, on users’ skills. CFD is still largely underutilized in the undergraduate curriculum,
usually due to both time constraints during degree studies and a lack of knowledge of the CFD tools
available and the possible uses of CFD for enhancing teaching [4].
CFD exercises are analogous to physical experiments. Students collect data after each run, plot and
identify trends in the data, and learn something about both CFD and UAS design in the process.
Furthermore, it has some advantages over physical experiment: CFD is cheaper, portable and allows
to conduct essays and obtaining measurements which are very difficult to acquire with experiments.

-2-

ISBN-xxx-y-zz-yyyyyyy-x 382
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

4. OPENFOAM, BASELINE OPTION FOR EDUCATION AND

RESEARCH
OpenFOAM is an open source software package developed by OpenCFD Ltd. It can actually solve
any continuum mechanics problem. It has complete equipment behind the software: supporting teams,
developers and consulters willing to solve any problem. However, this service has to be paid.
One of the most advantageous things of OpenFOAM is that it is completely free, what explains its
increasing competitiveness. This fact is especially important for students, universities or research
institutions with constrained budget, where it can be used as an educational package.

Table 1: One‐year license of different CFD software [5]

Company Code Price

Ansys, Inc. Fluent 12 k$
CFX N/A
Flow Science, Inc. FLOW-3D 4 k$
CHAM Ltd. Phoenix 8.74 k$
OpenCFD Ltd. OpenFOAM Free

Nevertheless, some researchers think that the main advantage of OpenFOAM is its open source
code. It is composed by lots of solvers and models, all open, which make it one of the most powerful
tools in its field. This enables anybody with programming knowledge to open its source, watch it,
change it, improve it and get a new solver recompiling the new code.
.
5. SOME EXAMPLES OF CFD TUTORIALS
CFD can be considered as a numerical experiment which can elucidate physical aspects of a flow field
in a manner not achievable in a real laboratory experiment. For instance, it is possible to switch
between laminar and turbulent flow for a certain high Reynolds problem. This is an easy way to
understand turbulence effects and allows to study the physical differences between them in detail [3].
Besides, more practical exercises can be conducted in order to teach how to employ CFD as a
design tool. A great point is that CFD provides a means to calculate the detailed flow field around a
complete airplane configuration, including the pressure distribution over the three-dimensional
surface. This is really useful, since it is impossible to obtain the whole aircraft aerodynamic
coefficients just with analytic methods. A good example of its potential could be the difference
between finite and 2D wings.

-3-

ISBN-xxx-y-zz-yyyyyyy-x 383
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

As it is known, the net

imbalance of the pressure
distribution over the top and
bottom surface creates the lift.
However, as a consequence of this
imbalance, the flow near the wing
tips tends to curl around the tips.
As a by-product of this curl, the
produced lift diminishes and the
induced drag increases [6].
The pressure near the tip in the

Figure 1: Contours of pressure distribution over the wing top surface is increased due to the

arrival of the flow from the lower

surface of the wing, so the lift
produced in this section is quite
lower and its distribution along the
span of a wing is one of the central
problems of finite-wing theory.
More complex studies can be
conducted, taking into account that
nowadays, many modern UAS’s
have very unusual aerodynamic

Figure 2: Contours of pressure distribution normal to the wing configurations (Figure 3) so global
coefficients, and other parameters
like pressure centre, will be extremely difficult to calculate without numerical or experimental
methods.

-4-

ISBN-xxx-y-zz-yyyyyyy-x 384
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 3: Some UAV’s with unusual configuration

5.1. Proposed exercise for students

Applying CFD techniques to UAS design is not trivial and it could be difficult to get appropriate
results without previous training in CFD codes. Due to this reason, this kind of exercises can be
conducted just in the context of a more general CFD course, like those that have taken place at the
Universidad de León, where some exercises related with UAS design have been realized by
aeronautical students. The most successful one is explained here.
A finite 3D wing is provided by the teacher and, as it can be seen in Figure 4, a remarkable trailing
vortex is created at wing tip. These wing-tip vortices downstream of the wing induce a small
downward component of air velocity in the neighborhood of the wing itself. As a result, the aircraft
performance is worse [6]. Students are encouraged to improve the wing adding some kind of winglet
or wingtip. In this way, competition between them is stimulated and their interest in the problem
increases.
Students are free to choose a valid computational domain, turbulence model, solution algorithm
and boundary conditions, applying the knowledge acquired during the course. Boundary conditions
for this case are the common ones for an external aerodynamics problem: velocity and turbulent
magnitudes (k and ε for the k-ε turbulence model or ν for the Spalart-Allmaras) are known at the inlet
and, as the flow reached a fully developed state at the outlet, gradients of all variables (except
pressure) at the outlet are zero in the flow direction. Boundary conditions for the pressure should be
established as zero gradient at the inlet and fixed value at the outlet. Also, wall functions must be used
at the walls.
Table 2: Boundary Conditions in OpenFOAM for external aerodynamics

Velocity Pressure Turbulence Variables

Inlet fixedValue zeroGradient fixedValue
Outlet zeroGradient fixedValue zeroGradient
Walls fixedValue zeroGradient wallFunctions

-5-

ISBN-xxx-y-zz-yyyyyyy-x 385
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

A possible solution to the problem is also provided, in order to illustrate its goal. The proposed
solution in Figure 5 is not the optimal one, it is just a suggestion, but the simulation conducted shows
an improvement of about 6.7% in the aerodynamic efficiency. As it is shown in Figure 4 and Figure 5,
the rotation magnitude of the flow is quite smaller in the proposed solution than in the initial wing.

Figure 4: Rotation of velocity value for a 2418 finite wing profile without winglet at 60 m/s

Figure 5: Rotation of velocity value for a 2418 finite wing profile with winglet at 60 m/s

Although these results must be reviewed with an empirical experiment, there are a lot of references
in literature about winglets benefits for aircraft performance. For instance, in [7] several wingtips are
tested with CFD and they achieve a 5.4% of aerodynamics efficiency improvement, similar to ours as

-6-

ISBN-xxx-y-zz-yyyyyyy-x 386
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

they employed a different wing configuration and airfoil profile.

Approximately, the maximum improvement accomplished in modified commercial aircraft is
about 4 – 5%, Boeing 737-NG in [8], less than half of the upper limit of 9 percent suggested by wind
tunnel tests in the ACEE program. Anyway, the improvement is high enough to be carefully analyzed
in order to fix the better solution. Must be noticed that not all the winglets for UAS are adequate
because of the large weight needed for the installation and the long-term effects on the structural
integrity of the wing due to stress imparted by winglet forces [8].
6. CONCLUSIONS
Free CFD software like OpenFOAM is an excellent tool in order to accomplish college practices. The
examples have demonstrated its viability to enhance the student knowledge of fundamental concepts
in fluid mechanics. The application of different cases, allows the student to compare CFD solutions
and results of non-numerical techniques, such as analytical solutions of math models and
experimental data.
Furthermore, students not only obtain a basic understanding of the principles of CFD, but also gain
knowledge of the tools available to solve complex, real-world problems.
Because of these reasons, increasing undergraduate students’ CFD skills is a mandatory action, as
it has been shown that companies and industry are demanding them for their UAS projects and other
issues.

REFERENCES
[1] Date, a. Wanam, Introduction to Computational Fluid Dynamics, New York, Cambridge
University Press (2009)
[2] J. Meseguer, A. Sanz, J. M. Perales and S. Pinado Aerodinámica Civil: Cargas de viento en las
edificaciones, Madrid, McGraw-Hill (2001)
[3] J. D. Anderson, Computational Fluid Dynamics, New York, McGraw-Hill (1995)
[4] J. S. Curtis, Enhancing Undergraduate Understanding of Trasport Phenomena via Application of
CFD, in AIChE Annual Meeting, University of Florida (2007)
[5] Kopyt, Pawel and Wojciech Gwarek, A Comparison of Commercial CFD Software Capable of
Coupling to External Electromagnetic Software for Modeling of Microwave Heating Process, in 6th
IMMG Seminar “Computer Modeling & Microwave Power Engineering”, Texas, pp 33-39 (2004)
[6] J. D. Anderson, Fundamental of Aerodynamics, New York, McGraw-Hill (2001)
[7] Wan, Tung, K.-W. Lien, and H.-C. Chou, Aerodynamic Efficiency Study of Modern Spiroid
Winglets, in 25th Congress of International Council of the Aeronautical Sciences, Hamburg, pp 1-7
(2006)
[8] National Research Council, Air Force Studies Board, Assessment of wingtip modifications to
increase the fuel efficiency of Air Force aircraft, Washington D.C., National Academies Press (2007)

-7-

ISBN-xxx-y-zz-yyyyyyy-x 387
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

-8-

ISBN-xxx-y-zz-yyyyyyy-x 388
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Real-time localization using a Wireless Sensor Network:

Application to an Unmanned Aerial Vehicle

J.-L. Rullán-Lara, G. Sanahuja, S. Salazar, and R. Lozano

Laboratoire HEUDIASYC UMR CNRS 6599, Université de Technologie de Compiègne, France.

{jrullanl,gsanahuja,rlozano}@hds.utc.fr
Laboratoire Franco-Mexicain d’Informatique et Automatique
LAFMIA UMI 3175 CNRS-CINVESTAV México.
[email protected]

Abstract

A real-time comparative analysis of four localization methods applied to an Unmanned Aerial

Vehicle (UAV) is presented in this paper. Range measurements from a Wireless sensor networks
(WSN) based on Chirp Spread Spectrum (CSS) communication protocol are used like observed
process. Trilateration and least squares approaches performances are compared to Kalman filter
estimation. The localization methods are implemented in real-time in an embedded UAV system.
The performance of each algorithm are compared for hover flight test. From experimental results,
Kalman filter outperformed the estimation from least square based methods.

keywords: Time Differential of Arrival, Kalman Filter, Localization, Unmanned Aerial Vehicle, Wireless
Sensor Networks.

1 INTRODUCTION
Localization is an important research subject that has been attracted a lot of interest during the recent
years. Two important aspects are essential in the localization problem: the kind of sensor and the
mathematical methods employed.
Wireless sensor network (WSN) is present as a favorable choice to solve the indoor location challenge.
These sensors are capable of measuring some radio signal metrics such as RSS (received signal strenght),
TOA (time of arrival), and TDoA (time difference of arrival). In addition, WSN can be used in both
indoors and outdoors applications. More precisely, Chirp Spread Spectrum (CSS)-based communication
protocol and chipset technology have been dominated the indoor localization market [1]. The major
advantage of the CSS technology is to provide robust performance for low rate wireless personal areal
network even in the presence of path loss, multi-path, and reflection [2]. Symmetric double side two way
ranging (SDS-TWR) is the key of CSS [3], [1]. SDS-TWR consists in repeating the packet exchange

ISBN-xxx-y-zz-yyyyyyy-x 389
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

twice, inverting the role of the two devices in the second exchange. By evaluating the propagation
time using the sum of the two measurements, the impact of clocks offset is highly reduced. SDS-
TWR technology demonstrates a superiority over other localization methods because removes the time
synchronization between devices, which is a quite demanding requirement and leading to an improved
ranging accuracy.
Noise in radio signals poses a problem for all location methods. Several localization algorithms
have been developed based on different approaches, the most popular are the least squares (LS) and
Kalman filter (KF), see [4], [5], [6]. Based on range measurements, some works have addressed to target
positioning using LS. In [7], the authors prove by simulations a weighted least squares (CWLS) solution
based on range measurements. In [8] and [9], closed-form solutions are derived with the Spherical
Interpolation method [10], [11].
Kalman filter is a well-known and widely-used method to deal with noisy data and sensor fusion. In
addition, internal states of a process can be estimated. Kalman filter consider two main assumptions
[12]: target motion and observation models are linear and, theirs errors and initial estimated probability
distribution are Gaussian. Nevertheless, the first assumption is not always true, however variants such
as Extended Kalman Filter (EKF) [13] or Unscented Kalman Filter (UKF) [14] can be employed.
A lot of applications of location using KF can be found in the literature. In [15] and [16], the authors
propose the computer equipment tracking and animals location is made in [17]. Using CSS ranging, a
static wireless node location is made in [18] with some practical improvements. Localization of pallets
and forklift truck in warehouse is proposed in [19] and [20] using SDS-TWR ranging.
The problem addressed in this paper is the determination of the 2D coordinates of an UAV based on
SDS-TWR ranging [21]. Two least squares methods and Kalman filter are tested in real time on a small
UAV. The performances of each algorithm were tested on the location of the aircraft in hover flight.
The paper is organized as follows. In Section 2 we state the localization problem. The main aspects
of positioning algorithms are reviewed in Section 3. Experimental setup and results are described in
Section 4. And the concluding remarks are finally given in Section 5.

2 Problem statement
The localization problem can be stated mathematically for 2D as follows: A set of N wireless nodes,
called base stations (BS) or anchors, are placed in the xy plane at known positions xi = (xi , yi )T ,
i = 1, 2, . . . , N , see Fig. 1-left. Within the workspace delimited by anchors, a wireless node (mobile
target (MT)), is at unknown position p = (x, y)T . The relationship between the measurable variables
and the unknown position can in general be written as follows [4]:

r = h(x) + e (1)

where h(·) is a function which implicity contains the positions of the base stations, and e is the error
∑
affecting the measurement and is assumed to follow a zero-mean Gaussian distribution e ∼ N (0, ).

ISBN-xxx-y-zz-yyyyyyy-x 390
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 1: Left: Location problem geometry. Right: Experimental set up, quad-rotor UAV.

Based on TOA or TDoA measurements, usually called range measurements, the function h(·) is
nonlinear and is given by the Euclidean distances between each base station and the mobile target
√
di = ∥p − xi ∥ = (x − xi )2 + (y − yi )2 , i = 1, 2, ..., N. (2)

The goal of the localization algorithms is to find out the closest position to the actual position of the
target by using the measurements (2). Trilateration is a method which consist in to find out the point
of intersection of N − 1 circles whose centers are at points xi given the range (2). Without noise in the
measurements, the circles will intersect at exactly one point. However, in real case the noise is always
present and the circles may intersect in an area or not intersect at all. In addition, if the number of
anchors increase, the resulting set of equations to solve is over-determined.
Several different approaches have been proposed for such over-determined system which are usually
based on maximum likelihood (ML) and least-squares approach (LS). Due to nonlinearities, the ML
cost function has not convex topology as in the LS case. In addition, to reach the estimates, numerical
methods are needed and must to be well initialized. These disadvantages of ML becomes the LS approach
in one of the workhorses for location problems and many improvements can be found in the literature.
In [8] and [9], improvements to the least squares were made based on the Lagrangian multipliers
optimization technique. In [9], quadratic-correction (QCLS) version is developed for TOA, TDOA, and
RSS measurements whilst in [8], an linear-corrector(LCLS) factor is added to standard least square solu-
tion. The LCLS was applied in real time to localize a static sound source. Other study for static target
was presented in [7] with a constrained weighted least squares (CWLS) using TDOA measurements.
In the linear-gaussian case, the Kalman filter has a low computational complexity because only the
two first moments of the probability distribution need to be stored [12], [22]. In addition, the Extended
Kalman filter (EKF) can be employed when nonlinear relationship are present. In real-time applications
like unmanned aerial, ground and underwater vehicles computation time is an important factor to reach
task. In this sense, Kalman filter is attractive since make estimation recursively and is easy to implement.

ISBN-xxx-y-zz-yyyyyyy-x 391
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The common inconvenient to get the actual position of the target is imposed by the noise in radio
signals and NLOS conditions. In addition, the UAV is an unstable system with very fast dynamic by
which, even in hover flight, the condition of static target can not be considered at all as in [18] and [20].
Then, the location should be sufficient to at least locate the UAV in an acceptable neighborhood of the
real position.

3 Positioning algorithms
In this section we present fourth positioning algorithms, in all cases, we consider four base stations and
one target mobile.

3.1 Trilateration

Using the range measurements the unknown position can be found. By squaring (2) and substracting
d1 from di for i = 2, 3, 4, yields
     
x − x1 y2 − y1 x (x22 + y22 ) − (x21 + y12 ) − (d22 − d21 )
 2    1 
     2 
 x3 − x1 y3 − y1    =  (x + y22 ) − (x21 + y12 ) − (d23 − d21 ) 
    2 3 
(3)
x4 − x1 y4 − y1 y (x24 + y22 ) − (x21 + y12 ) − (d24 − d21 )
| {z } | {z } | {z }
M p T

Equation (3) is over-determined since there exists more equations that unknown variables. Using the
pseudo inverse method, the tag position yields

p = (MT M)−1 MT T (4)

3.2 Least squares approach

This approach makes no probabilistic assumptions about range measurements and usually produces a
closed-form estimate that is desirable in real-time applications. In order to get an algebraic solution,
it is necessary to minimized an error function which is zero in the noiseless case. For the localization
problem, hyperbolic and spherical least squares error functions are well known and they are based on
the range difference measurements

gij = dj − di , i, j ∈ {1, 2, 3, 4}, i ̸= j. (5)

In a two-dimensional space, gi1 defines a hyperbole where all points lying on it are potential target
positions and all have the same range difference to the two base stations i and 1. The hyperbolic least
squares error function is defined as
eh (p) = ĝ − g, (6)

Due to highly nonlinear relationship, minimizing (6) leads to a mathematically intractable solution as
N gets large and is very sensitive to noise. On the other hand, spherical error function is obtained by

ISBN-xxx-y-zz-yyyyyyy-x 392
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

reorganizing the nonlinear equations in (5) into a set of linear equations by introducing an intermediate
variable, which is a function of the source position. This error function is described in subsection 3.2.2.
The iterative Taylor series method is also reviewed. In this case, the error function is defined as the
difference between the linearized equations and the new approximated position.

3.2.1 Taylor series

The range differences are linearized by expanding them in Taylor series around a point (initially an
estimate of the actual position) and keeping only the terms below second order. The set of linearized
equations is solved to produce a new approximate position and the process continues until a pre-specified
criterion is satisfied. Thus, from (5) we define the function qi (pls ) as follows

qi (pls ) = di − d1 + ωi,1 , i = 2, 3, 4 (7)

where ωi,1 is the corresponding range difference error with covariance matrice Rϵ of appropriate dimen-
sion. Considering an arbitrary starting point of the target pls,0 = (x0 , y0 )T , then the equation (7) can
be expanded in Taylor series and retaining the first two terms. This yileds
∂qi (p0 ) ∂qi (p0 )
(x − x0 ) + (y − y0 ) ≈ di − d1 − qi (p0 ) + ωi,1 (8)
∂x ∂y
In matrix form, it yields
       
x−x2 x−x1 y−y2 y−y1
d2
− d1 d2
− d1
x − x0 (d2 − d1 ) − q2 (p0 ) ω2,1
       
 x−x3
− x−x1 y−y3
− y−y1    =  (d3 − d1 ) − q3 (p0 )  +  ω3,1 
 d3 d1 d3 d1       
x−x4 x−x1 y−y4 y−y1 (9)
d4
− d1 d4
− d1
y − y0 (d4 − d1 ) − q4 (p0 ) ω4,1
| {z } | {z } | {z } | {z }
G δ b eε

The least square error function is then expressed from above equation as ets (δ) = Gδ − b and the
weighted least-square estimator is given by

δ = (GT R−1 G)−1 GT R−1

ϵ b. (10)

Then, a recursive solution can be as p̂ls,k = p̂ls,k−1 + δk . However, the estimation do not take any
additional assumption about the distribution of measurements errors and only depends on the solution
of (10). In addition, if the initial guess pls,0 is not well stated the convergence is not always assured.

3.2.2 Quadratic constrained least squares (QCLS)

Based on spherical interpolation, some improvements have been done to LS approach in [9] which a
constrained weighted solution is developed. First of all, the range differences in (5) can be rewritten,
after some algebraic manipulations, in the matrix form
     
x − x1 y2 − y1 g12 x − x1 (x − x1 )2 + (y2 − y1 )2 − g12
2
 2    1  2 
     
 x3 − x1 y3 − y1 g13   y − y1  =  (x − x1 )2 + (y3 − y1 )2 − g13
2 
    2 3 
(11)
x4 − x1 y4 − y1 g14 R1 (x4 − x1 )2 + (y4 − y1 )2 − g14
2
| {z } | {z } | {z }
A pqcls m

ISBN-xxx-y-zz-yyyyyyy-x 393
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Using the intermediate variable R1 = ∥x − x1 ∥, the system of equations in (11) is linear in pqcls . In
presence of noise, the Spherical Interpolation technique determine the target position by simply solving
the above equation via standard least squares. Thus, the least squares error function is then given by
esp (pqcls ) = Apqcls − m. In [9], an improvement to standard least squares is proposed by solving the
LS cost function
p̂qcls = esp (p̌qcls )T Y−1 esp (p̌qcls ) (12)

subject to the quadratic constraint

p̌Tqcls Σp̌qcls = 0. (13)

where Σ = diag{1, 1, −1} and p̌ is an optimization variable vector. The symmetric weighting matrix
Y is in function of the estimate of R̂1 . The problem stated in (12)-(13) is solved by using the technique
of Lagrange multipliers and the solution for (12) yields

p̂qcls = (AT Y−1 A + λΣ)−1 AT Y−1 m. (14)

Further details of the whole development of this algorithm can be find in [9] and [11].

3.3 Extended Kalman filter

The Kalman filter is an algorithm for efficient state estimation minimizing the covariance error [12].
The basic filter is well-established, and it assumes that the transition and the observation models are
linear. Due to nonlinearities in range measurements (see eq. (2)), the Extended Kalman Filter must be
applied. When the input is zero, the evolution of the state in time is given by a stochastic difference
equations
xk+1 = fk (xk , wk ), yk = hk (xk , vk ) (15)

where xk is the state vector at time k, yk is the range vector process, and wk and vk are respectively noise
process and measurement vectors. Noises are assumed to be independent with gaussian distribution,
that is, wk ∼ N (0, Q) and vk ∼ N (0, R). The initial state x0 is assumed to be normally distributed
with mean x0 and covariance matrix P0 . The process described in (15) is linearized around the new
estimates x̂k at time step k. Thus, the state and range functions are linearized Fk = ∇x fk (xk )|xk =xˆk
and Hk = ∇x hk (xk )|xk =x̂k−1 . Matrices Fk and Hk are the state transition matrix and the observation
matrix, respectively. Thus, the system becomes

xk+1 = Fk xk + wk , yk = Hk xk + vk . (16)

The time and measurement update for the EKF can be verified in [22].
In (15), h(·) is the same as in (2), however, the function f (·) depends on the kinematic state move-
ment. If we want to obtain the position and velocity of an UAV, then a continuous white noise acceler-
T
ation model can be considered. Thus, xk is defined as xk = (p, ṗ) and, consequently, the discrete-time

ISBN-xxx-y-zz-yyyyyyy-x 394
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

state transfer matrix, Fk and the observation matrix Hk yields the 4 × 4 matrices
   
xk −x1 yk −y1
1 0 ∆ 0 0 0
   d1 d1 
   xk −x2 yk −y2 
 0 1 0 ∆   d 0 0 
Fk = 
 , Hk = 

2 d2
 xk −x3 yk −y3

 (17)
 0 0 1 0   d 0 0 
   3 d3 
xk −x4 yk −y4
0 0 0 1 d4 d4 0 0

where ∆ is sampling period. Notice that all elements of above equation have singularities if di = 0,
i = 1, 2, 3, 4.

4 Experimental results

4.1 Set up

The flying robot is the classical quad-rotor which is mechanically simpler than a classical helicopter
since it does not have a swashplate and have constant pitch blades, see Fig. 1-right. It has two counter
rotating pairs of propeller arranged in a square. In the middle of the square, the onboard computer, the
inertial measurement unit (IMU), the location sensor board, and the batteries are mounted.
The thrust is provided by four Robbe Roxxy BL-Motor 2827-35 brushless motors and four Mikrokopter
v1.2 speed controllers. A MicroStrain 3DM-GX3 IMU is used to measure the roll (ϕ), pitch (θ), and
yaw (ψ) attitude angles. The altitude is measured using an ultrasonic sensor (SPF03) and a PS3 camera
is used to compute optic flux.
The onboard hardware used is the IGEPv2 board based on the OMAP3530 from Texas Instruments.
The attitude dynamic η̈ = τ , η : ϕ, θ, ψ, is stabilized onboard using the control input based on the
integrators in cascade with bounded input as in [23]

τ = −σ1 (k1 η̇) − σ2 (k2 η) (18)

The NanoLOC development kit from Nanotron Technologies has been used to measure the distances
between the base stations and target. Wireless communications as well as the ranging methodology SDS-
TWR are integrated in a single chip, the nanoLOC TRX Transceiver [1]. The NanoLOC development
kit contains five sensor boards with sleeve dipole omnidirectional antennas. One target board, TAG
board, is mounted in the UAV and is connected by serial communication link to onboard hardware. A
frame of 68 bytes is decoded to find out the four actual range measurements d1k , d2k , d3k , and d4k .

4.2 Hover test

The experiment is carried out considering the UAV in hover flight over the point xref = (400cm, 300cm)
for a while of one minute with sampling period ∆ = 0.08 seconds. The base stations are placed at points
BS1 = (0, 0), BS2 = (800cm, 100cm), BS3 = (800cm, 700cm) and BS4 = (0, 600cm). The vehicle is totaly
stabilized in orientation and is launched from the point xref and kept flying around a neighborhood of

ISBN-xxx-y-zz-yyyyyyy-x 395
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

xref . During hovering, small translational movements are induced by the pilot to keep the UAV close
to the starting point. After about a minute, the UAV is landed on the same starting point.
The initial estimates for all methods were obtained from (4) and the noise covariance matrices are
2
initialized as follows: Rε = σts × I, R0 = diag{830, 650, 250, 905}cm2 , Y0 = σqcls
2
× I. The values in R0
2 2
were identified by experiments and the values for σts and σqcls are chosen as mean noise covariance of all
experiments. After that, matrix Y is updated according to [9]. Matrix Rk in Kalman filter is updated
by using the concept of covariance matching [24], i.e., to compare the behavior of actual innovation
vector yk − ŷk against its theoretical value. The actual residual covariance can be approximated by its
sample covariance, through averaging inside a moving window of size m. The value of m was fixed after
some tests.
Figure 2-left shows the estimated position when algorithms are applied in real-time. Trilateration
estimate is the worst of all, not seeing any estimates (”◦” points) in the workspace where the UAV
is flown. In this figure we observe that the estimation from Kalman filter converges quickly to the
desired value. Moreover, note that p̂ls and p̂qcls have sudden changes in magnitude almost all the time
while the estimate from Kalman filter is more smoother (see fig. 3-left). These values are outside the
workspace delimited by the base stations and should be mitigated if the position estimation will be use
in the position control loop of the UAV, however, x̂ could be used immediately in hover. This smooth
behavior in the estimation is an advantage when new points on the workspace are achieved by the UAV.
Thus, the estimate might be merged with other inertial sensors, for a more precise localization. The fig.
2-right shows the error obtained from the estimates.

Table 1: Performance of the algorithms

Method mean value (cm) standard deviations (cm) MAPE (cm)

Trilateration − − 1694.4
Taylor µx = 825.26, σx = 413.22 µy = −161.20, σy = 400.83 729.99
QCLS µx = 387.27, σx = 214.42 µy = 368.46, σy = 377.01 286.39
Kalman µx = 411.06, σx = 43.68 µy = 383.87, σy = 86.367 119.43

Estimated position error are listed in Table 1. Note that the estimated mean values from QCLS and
Kalman filter are near to xref however, Taylor series are far from the desired values. In addition, the
Kalman filter has a smaller standard deviation showing a more accuracy. This is because the Kalman
filter takes into account the past and reduces state error with respect to time. Table 1 also shows
the mean absolute positioning error (MAPE). Kalman filter improves the performance of least squares
methods with an average position error of 119.43cm. These results are quite similar to those reported
in [25] and [26] where the same wireless sensor network is used.
Since oscillations are presented in the estimates from the Taylor series and QCLS, these data were
smoothed to improve the results. Position estimates were smoothed using the target motion model used

ISBN-xxx-y-zz-yyyyyyy-x 396
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Position estimation
800 3000
BS3 Trilateration
700 Taylor
QCLS
BS4 2500
Kalman
600

500 2000

error (cm)
400
y (cm)

1500
300 Trilateration
Taylor
200 QCLS 1000
Kalman
100
BS2 500
0
BS1
−100 0
−100 0 100 200 300 400 500 600 700 800 900 0 10 20 30 40 50 60
x (cm) t (sec)

Figure 2: Comparison of results: Estimated position in xy (left), error (right).

Position estimation
2000 1500
Desired Taylor
1500 Trilateration QCLS
1000
Taylor Kalman
x (cm)

x (cm)
1000 QCLS
500
Kalman
500
0
0
−500 −500
0 10 20 30 40 50 60 0 10 20 30 40 50 60
t (sec) t (sec)

1500 1500

1000 1000
y (cm)

y (cm)

500 500

0 0

−500 −500
0 10 20 30 40 50 60 0 10 20 30 40 50 60
t (sec) t (sec)

Figure 3: Estimates p̂ (left). Estimates p̂ after smothing (right).

by the Kalman filter. From figure 3-right we can see how the data from QCLS are very similar to those
of Kalman filter in both coordinates after softer. This means that the QCLS method and the smoother
can be implemented in real time and could be used in the position control loop as mentioned above.
Smoother data in figure 3-right reported a MAPE de 138.53cm and means values of µx = 391.46cm and
µx = 410.93cm and standards deviations of σx = 70.171cm and σy = 80.668cm.

5 Conclusions
Four localization methods are compared in real-time. A Wireless Sensor Network, based on the chirp
spread spectrum (CSS) communication protocol is used to measure distances between node. Trilatera-
tion, Taylor Series, Quadratic-Correction Least Squares and Extended Kalman Filter are implemented
in closed-loop in real-time experiments for hover flight of an Unmanned Aerial Vehicle.
Trilateration method reported the worse estimation. This is an expected result because it is based in

ISBN-xxx-y-zz-yyyyyyy-x 397
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

absolute error and hence not accurate. According to our study, least squares based methods have very
similar behavior but QCLS reports a more accurate performance that Taylor series. However, Kalman
filter estimates were being more closer to the actual UAV position than the other methods.
The measurement covariance matrix for Kalman filter was updated by using the concept of covariance
matching. This allowed us to obtain a smoother estimation and to reach the position of the target even
when the initial position was far enough. Moreover, and according to our study, to implement in real-
time the least squares methods should be improvement with a smoother to avoid sudden magnitud
changes in estimation.
Current research work is focused on including the proposed localization algorithm based on Kalman
filter into the stabilization control loop of the UAV and realize path tracking.

6 Acknowledgement
The first author thanks the Universidad Autónoma del Carmen (UNACAR) and the Secretarı́a de
Educación Pública (SEP) for their financial support PROMEP/103.5/07/2057.

REFERENCES
[1] Nanotron Technologies GmbH, ”Real Time Localization Systems (RTLS)”, White paper, 2006

[2] L. De Nardis, M.-G. Di Benedetto, Overview of the IEEE 802.15.4/4a standards for low data rate
Wireless Personal Data Networks, 4th Workshop on Positioning, Navigation and Communication,
Hannover, Germany, 2007.

[3] H.-S. Ahn, H. Hur, W.-S. Choi, One-way Ranging Technique for CSS-based Indoor Localization,
IEEE International Conference on Industrial Informatics, Daejeon, Korea, 2008.

[4] D.J. Torrieri, Statistical Theory of Passive Location Systems, IEEE Transaction on Aerospace and
Electronic Systems, Vol. AES-20 Issue 2, pp. 183-198, March 1984.

[5] G. Mao, B. Fidan, B. D. O. Anderson, Wireless sensor network localization techniques, Computer
Networks, Vol. 51, Issue 10, pp. 2529-2553, 11 July 2007.

[6] F. Seco, A. Jiménez, C. Prieto, J. Roa, and K. Koutsou, A survey of Mathematical Methods for
Indoor Localization, IEEE 6th International Symposium on Intelligent Signal Processing, 26-28
August, 2009, Budapest, Hungry.

[7] K. Dogancaya, , A. Hashemi-Sakhtsari, Target tracking by time difference of arrival using recursive
smoothing, Signal Processing 85 (2005) 667–679.

[8] Y. Huang, J. Benesty, G. W. Elko, and R. M. Mersereati, Real-time Passive Source Localization:
A Practical Linear-Correction Least-Squares Approach, IEEE Transactions on Speech and Audio
Processing, vol. 9, no. 8, pp 943-956, 2001.

10

ISBN-xxx-y-zz-yyyyyyy-x 398
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[9] K. W. Cheung, H. C. So, W.-K. Ma, Y. T. Chan, A Constrained Least Squares Approach to
Mobile Positioning: Algorithms and Optimality, EURASIP Journal on Applied Signal Processing,
Vol. 2006, Article ID 20858, pp. 1-23, 2006.

[10] J. O. Smith and J. S. Abel, Closed-form least-squares location estimation from range-difference
measurements, IEEE Trans. Acoust. Speech, Signal Processing, vol. ASSP-35, pp 1661-1669, Dec.
1987

[11] Y.T.Chan and K.C.Ho, A simple and efficient estimator for hyperbolic location, IEEE Transaction
on Signal processing, August, 1994.

[12] R.E. Kalman, ”A new approach to linear filtering prediction problems”, ASME Journal of Basic
Engineering, Vol. 82, pp. 34-45, 1960.

[13] V. M. Becerra, P. D. Roberts, G. W. Griffiths, Appliyingthe extended Kalman filter to systems

described by nonlinear differential-algebraic equations, Control Engineering Practice, Volume 9,
Issue 3, March 2001, pp 267-281.

[14] S. J. Julier, J. K. Ulhmann, A new extension of the Kalman filter to nonlinear systems, in Pro-
ceedings of the American Control Conference, 1995, pp. 1628-163.

[15] J. Yim, C. Park, J. Joo, S. Jeong, Extended Kalman Filter to wireless LAN based indoor psoitioning,
Decision Support Systems, Elsevier, Vol. 45, pp. 960-971, 2008.

[16] J. Yim, S. Jeong, K. Gwon, J. Joo, Improvement of Kalman filters for WLAN based indoor tracking,
Expert Systems with Applications, Elsevier, Vol. 37, Issue 1, pp. 426-433, 2010.

[17] F. A Togersen, F. Skjoth, L. Munksgaard, S. Hojsgaard, Wireless indoor tracking network based
on Kalman filters with an applications to monitoring dairy cows, Computer and Electronics in
Agriculture, Elsevier, Vol. 72, pp. 119-126, 2010.

[18] H. Cho, C. W. Lee, S. J. Ban S. W. Kim, An enhanced positioning scheme for chirp spread spectrum
ranging, Expert Systems with Applications, Volume 37, Issue 8, August 2010, pp 5728-5735.

[19] C. Röhrig and S. Spieker, Localization of Pallets in Warehouses using Wireless Sensor Networks,
16th Mediterranean Conference on Control and Automation, 2008, pp 1833-1838.

[20] C. Röhrig and S. Spieker, Tracking of Transport Vehicles for Warehouse Management using a Wire-
less Sensor Network, 2008 IEEE/RSJ International Conference on Intelligent Robots and Systems,
Nice, France, Sept, 22-26, 2008.

[21] Nanotron Technologies GmbH, ”nanoLOC Development Kit User Guide”, Technical Report NA-
06-0230-0402-2.0, 2008

[22] Y. Bar-Shalom, X.-Rong Li, T. Kirubarajan, Estimation with Applications to Tracking and Navi-
gation, John Wiley Sons, Inc. New York, 2001.

11

ISBN-xxx-y-zz-yyyyyyy-x 399
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[23] G. Sanahuja, P. Castillo, A. Sanchez, Stabilization of n integrators in cascade with bounded input
with experimental application to a VTOL laboratory system, International Journal of Robust and
Nonlinear Control, Vol. 20, Issue 10, pp 1129-1139, 10 July 2010, DOI: 10.1002/rnc.1494.

[24] R. K. Mehra, On the identificqtion of variances and adaptive kalman filtering, IEEE Trans. Autom.
Control, Volume AC-15, No. 2, pp 175-184, April, 1970.

[25] C. Röhrig and M. Müller, Indoor Location Tracking in Non-line-of-Sight Environments Using a
IEEE 802.15.4a Wireless Network, IEEE/RSJ International Conference on Intelligent Robots and
Systems, October, 2009, pp 552-557.

[26] C. Röhrig, M. Muller, Localization of Sensor Nodes in a Wireless Sensor Network Using the
nanoLOC TRX Transceiver, IEEE 69th Vehicular Technology Conference Spring, 26-29 April 2009
Barcelona, Spain.

12

ISBN-xxx-y-zz-yyyyyyy-x 400
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Real Time Stabilization of a four-rotor UAV using rotor speed

feedback

I. Gonzalez, S. Salazar H. Romero R. Lozano and J. Torres

UMI-LAFMIA 3175 CNRS, CINVESTAV

Av. Instituto Politécnico Nacional 2508. 07360 México D.F., México
[email protected], [email protected], [email protected]
[email protected], [email protected]

Abstract

Recently, the research in the area of autonomous miniature flying robots is growing thanks to
the development of new configurations and prototypes of Unmanned Aerial Vehicles (UAV). In
this paper, we introduce the problem of attitude and altitude control of a mini-quadrotor UAV
in hovering-mode to indoor and outdoor applications. Therefore, we have implemented a control
strategy based on speed sensing in each brushless motors. As result, the miniature aerial vehicle
has a robust stabilization on the horizontal plane. This paper also describes a Lyapunov candidate
function which allows to have stable attitude control of the Quad-rotor . Finally, the experimental
results of speed sensing control for stabilizing the quad-rotor at hover are presented.

keywords: Embedded flight system, Real-Time Stabilization, Motor Feedback Control, Quadrotor, UAVs

1 INTRODUCTION
Nowadays, one of the most appealing topics among the research control community is the application of
modern control theory to Unmanned Aerial Vehicles (UAV). The operation and development of UAVs
has increased during the last few years, due to the wide range of applications such each aerial inspection
in constrained locations, work in hazardous environments, forest fire detection, aerial photography and so
on. For these reasons UAVs aroused a great interest in the research and academic circles in recent years,
using it as a platform for many applications and for pure academic research [7], [8], [13]. Attitude and
altitude control in a mini-UAV involves research in various areas such as digital filtering, estimation of
angular position, data fusion of sensors, and so on. In [14] the authors proposed a method for improving
attitude control of the UAV regulating the RPM of motors based on sensorless control motor. This
approach is based on an electronic circuit which measures the voltage applied in each phase of motors.
In the same view, another approach is presented in [15] where the authors propose a Sensorless Motor
Control technique using current measurement. This paper presents a different technique which involves

ISBN-xxx-y-zz-yyyyyyy-x 401
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 1: Photo of the Quad-rotor at hover flight.

direct measurement of the speed of each DC Brushless motors using a REFLEX sensor and a FCV
(Frequency to Voltage Converter). This circuit gives a voltage that corresponds to the current RPM of
the motor. The advantage of measuring the speed of the motors with this technique is that it is possible
to obtain a more accurate reading of the RPM. The prototype built to test the speed sensing is shown
in Fig. 1. The performance of the speed sensors implemented in the brushless DC motors is shown in
a practical application. Now, the paper is organized as follows: section 2 presents the dynamic model
of the Quad-rotor. The linear control PD is described in section 3. The architecture of the prototype
are described in section 4. The speed sensor system on brushless DC motors is presented in section 5.
Then, the experimental results of the attitude and altitude control using speed sensing are described in
section 6. And finally, we present some conclusions about our work, in the section 7.

2 DYNAMICAL MODEL
As you known, Quad-rotors are mechanically simpler because it does not require a complicated mechan-
ical structure like conventional helicopters and potentially it is easier to be controlled. A Quad-Rotor
Flying robot is controlled by varying the angular speed of each motor. The front and rear motors rotate
counterclockwise (CCW), while the other two motors rotate clockwise (CW). Gyroscopic effects and
aerodynamic torques tend to cancel in trimmed flight. The main thrust is the sum of the thrusts of
each motor, the pitch torque is a function of the difference f1 − f3 , the roll torque is produced by the
difference f2 − f4 , and the yaw torque is the sum of τM1 + τM2 + τM3 + τM4 (see Fig. 2), where τMi is the
reaction torque of motor i due to shaft acceleration and the blades drag. Let ΓI = {iI , jI , kI } be the
inertial frame, ΓB = {iB , jB , kB } denote a set of coordinates fixed to the rigid aircraft as is shown in
Fig. 2. Let q = (x, y, z, ϕ, θ, ψ)T ∈ R6 = (ξ, η)T be the generalized coordinates vector which describe
the position and orientation of the flying machine, so the model could be separated in two coordinate
subsystems: translational and rotational. They are defined respectively by

ISBN-xxx-y-zz-yyyyyyy-x 402
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 2: Inertial and Body frame of the four-rotor aircraft.

• ξ = (x, y, z)T ∈ R3 : denotes the position of the aerial vehicles center of mass relative to the
inertial frame ΓI .

• η = (ϕ, θ, ψ)T ∈ R3 : describe the orientation of the aerial vehicle and (ϕ,θ,ψ) are the three Euler
angles: roll, pitch and yaw, respectively.

The dynamic model is obtained using the Euler-Lagrange approach. We can decompose the equations
into translational and rotational displacement. As first step, we obtain the Lagrangian of the aerial
vehicle, which is given by the following equation

L(q, q̇) = (Ttrasl + Trot ) − U (1)

• Ttrasl : It is the translational kinetic energy of the quad-rotor.

1 ˙T ˙
Ttrasl = mξ ξ (2)
2

• Trot : It is the rotational kinetic energy of the quad-rotor.

1 T
Trot = η̇ Jη̇ (3)
2

• U : It is the potential energy of the quad-rotor.

U = mgz (4)

In the above equations we have that m denotes the mass of the vehicle, J represents the inertia
matrix which is symmetric and defined positive, g denotes acceleration due to the gravity and z is the
altitude desired of the vehicle. Following the approach, the Lagrangian is given by
1 ˙T ˙ 1 T
L(q, q̇) = mξ ξ + η̇ Jη̇ − mgz (5)
2 2

ISBN-xxx-y-zz-yyyyyyy-x 403
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The model for the full four-rotors aircraft dynamics is obtained from the Euler-Lagrange equations
with external generalized force  
d ∂L ∂L FI
− =   (6)
dt ∂ q̇ ∂q τ

where FI defines the translational force applied to the aerial robot due to the control inputs and relative
to the frame ΓI and τ is the generalized moments vector. Doing the current calculations, the complete
dynamic model of Quad-rotor is

FI = mξ¨ + mg
( T ) (7)
τη = Jη̈ + J̇η̇ − 1 ∂
2 ∂η η̇ Jη̇

The dynamic model of the Quad-rotor can be rewritten including the Coriolis terms as,

mξ¨ = FI + mg
(8)
Jη̈ = τη + C(η, η̇)η̇

Now, doing a change of coordinates, we propose new variables to the control input τη :
 
τ̃ϕ
 
 
τ̃η =  τ̃θ  = J−1 (τη − C (η, η̇) η̇) (9)
 
τ̃ψ

The force vector on the body frame, denoted as FB = (0, 0, u)T where u is the sum of mechanical thrust
forces, fi for (i=1,2,3,4), (see Fig. 2). It can be expressed in the inertial frame as

FI = RB→I F B (10)

where RB→I is the rotation matrix which is defined by

 
cθ cψ sψ cθ −sθ
 
B→I  
R =  cψ sθ sϕ − sψ cϕ sψ sθ sϕ + cψ cϕ cθ sϕ  (11)
 
cψ sθ sϕ + sψ cϕ sψ sθ sϕ − cψ cϕ cθ cϕ

Finally, replacing (9) and (10) into (8) we have that the dynamic model of our miniature aerial vehicle
is given by      
mẍ u sen θ 0
     
     
 mÿ  =  −u sen ϕ cos θ + 0  (12)
     
mz̈ u cos ϕ cos θ −mg
   
ϕ̈ τ̃
   ϕ 
   
 θ̈  =  τ̃θ  (13)
   
ψ̈ τ̃ψ

ISBN-xxx-y-zz-yyyyyyy-x 404
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3 CONTROL STRATEGY
In this section we present a control law for the attitude stabilization and control altitude of the four-
rotor aircraft at hover. The goal of the control law is to bring the quad-rotor to rest at the zero attitude
(aligned with the inertial frame). Furthermore, we are able to propose a simple linear control PD
strategy. The PD controller is relatively simple and each one of the control inputs can operate in either
manual or automatic mode. In addition, one of the main advantages of this algorithm is that the gains
are easy tuning and implementing on a microcontroller. Finally, the collective input u is essentially used
to reach a desired altitude value.

3.1 Altitude control

To altitude control, we consider the dynamic in the z-axis from (12),

1
z̈ = (u cos ϕ cos θ − mg) (14)
m
using the following control input
mcz + mg
u= (15)
cos ϕ cos θ
where the variable cz is a PD controller, which is defined by

cz = −kDz ż − kPz (z − zd ) (16)

with kDz ,kPz > 0 and zd the altitude desired. From (15) we can see that control u is necessary bounded
in: − π2 < ϕ < π
2 and − π2 < θ < π
2. Then, using (15) into (14) we get

z̈ = cz (17)

3.2 Attitude control

Attitude control is the most important part of control system, it stabilizes the orientation of the heli-
copter at a desired value. Therefore, the main aim of the attitude control is to stabilize the vehicle in
the desired orientation ηd = (ϕ, θ, ψ)T = (0, 0, 0)T . We first consider the non-linear model

M (η)η̈ + C(η, η̇)η̇ = τ̃η (18)

with η = (ϕ, θ, ψ)T . Now, we propose the following Lyapunov candidate function
∑3
1 T
V (η, η̇) = η̇ M (η)η̇ + Kai ln(cosh(Kpi ηi )) (19)
2 i=1

where, ηi represent the i-th element of the vector η, Kai and Kpi represent the i-th diagonal element of
positive definite matrices Ka and Kp ∈ R3x3 , respectively. The Lyapunov candidate function V (η, η̇) ≥ 0
since M (η) > 0 and ln (cosh (·)) is a positive definite increasing function. Differentiating (19) with
respect to time, we obtain
∑3
1 T
V̇ (η, η̇) = η̇ Ṁ (η)η̇ + η̇ T M η̈ + η˙i Kai Kpi tanh(Kpi ηi ) (20)
2 i=1

ISBN-xxx-y-zz-yyyyyyy-x 405
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

then (20) may be rewritten as

1 T
V̇ (η, η̇) = η̇ Ṁ (η)η̇ + η̇ T M η̈ + η˙i T Ka Kp tanh(Kp η) (21)
2

with
T
tanh = (tanh (kp1 ϕ) , tanh (kp2 θ) , tanh (kp3 ψ))

Considering the attitude control as a regulation problem (η = ηd ) and replacing (18) into (21) we get
· ( ·
) ·T (1 · ( · )) · ·T
V η, η = η M (η) − C η, η η + η (τη + ka kp tanh (kp η)) (22)
2

using the property proposed in [3], we have

( )
1
uT Ṁ (η) − C(η, η̇) u = 0 ∀ u ∈ R3 (23)
2

leads to
· ( ·
) ·T ( )
V η, η = η τη + ksp tanh (kp η) (24)
k sp
where, it was considered ka = kp . Now, we introduce the following saturated PD control:
( ·)
τ̃η = −ksp tanh (kp η) − ksv tanh kv η (25)

where Ksp and Ksv ∈ R3×3 are positive diagonal matrix. Then, substituting (25) in (24) leads to
· ( ·
) ·T
( ·)
V η, η = −η ksv tanh kv η (26)

Finally, it follows that V̇ < 0. Therefore, the origin is stable and the solutions η(t) and η̇(t) are
bounded.

4 PLATFORM ARCHITECTURE
In this section we present the prototype architecture of the Quad-rotor. The overall system structure is
shown in Fig. 3.

Figure 3: The control stages implemented on Quad-rotor.

ISBN-xxx-y-zz-yyyyyyy-x 406
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

It basically consists of four-rotors arranged in a Microkopter’s aluminium frame, Measuring stage

(Sensors), Control stage (Embedded system: DSP1 ), Tx/Rx stage (Radio-control ) and Power stage
(Li-Po batteries). Futhermore, this experimental platform serves as didactic prototype in the field of
miniature aerial vehicles research because we can test different control laws to verify the efficiency of these
controls on flying. The structure of the Quad-Rotor flying robot is modified from a popular quad-rotor
RC toy, the Mikrokopter [16]. A homemade Inertial Measurement Unit (IMU) based on Sparkfun’s
Analog Combo Board Razor-6DOF Ultra-Thin is used to measure three angular instantaneous rates
( )
ϕ̇, θ̇, ψ̇ and two angular positions (ϕ, θ) and we use a compass CMPS03 to measure the yaw angle (ψ).
Afterwards a mini-core TMS320F2808-DSP (32-Bit Digital Signal Controller with Flash) is employed at
this stage to control the attitude and altitude of our Quad-rotor. The DSP runs the control algorithm
in real time to stabilize the Quad-rotor managing the information provided by the inertial measurement
unit (IMU) and the GPS sensors and sends the control signals to the four motors.

5 Speed sensing implementation

The speed sensing hardware system to perform an autonomous flight by controlling the orientation of
each Brushless DC motors is shown in Fig. 4. It consist of a Speed sensor block, a PD controller block,

Figure 4: Block diagram of the Speed control using PD method.

an I 2 C control block, a Power stage block and the Brushless DC motor block. The actual motor speed
value is captured through CNY70 sensor. The Frequency to Voltage Converter (FVC) takes the output
signal (pulse train) from CNY70 and converts this signal to an equivalent value of voltage with respect
to the input frequency. Then speed motor is converted a voltage signal which is then compared with
the desired speed, and a correction signal is generated, if necessary.

e(k) = r(k) − y(k) (27)

where e(k) is the actual error value, r(k) is the reference speed value which is given by the accelerometer
and the gyro offset while y(k) is the actual motor speed value. The errors are used by the PD control
algorithm guarantees a control signal calculation, then a DSP algorithm delivers a measurement of the
1 Digital Signal Processing

ISBN-xxx-y-zz-yyyyyyy-x 407
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

motor’s speed through FVC and decide the direction in which it is drifting. A proportional-derivative
(PD) routine produces one output signal control (28),
( )
kd ∗ [e (k) − e (k − 1)]
u (k) = kp ∗ e (k) + (28)
Ts

where u (k) is the input control to the brushless DC motors and Ts is DSP’s sampling period. Notice
that, the gains kp and kd are chosen to provide a quick response with little instability. The motor’s
load, inertia and torque are key the factors that determine the PD gain constants. Finally, PD controller
generates I 2 C output signals to the motors drivers. Fig. 5 shows a Printed Circuit Board (PCB) of
Brushless DC motors speed sensors, consisting of a Frequency-to-Voltage converter, op-amps and low-
pass filters (RC) and the speed sensor fixed close to brushless DC motor to measure its speed.

Figure 5: Speed sensor (reflex ) PCB implemented in the Quad-rotor.

6 EXPERIMENTAL RESULTS
In this section we present the experimental results that validate the performance of the Quad-rotor
during autonomous hover flight operation. The control gains were adjusted to obtain a fast system’s
response but avoiding mechanical oscillations as much as possible. Now, we present some experimental
results obtained by applying the proposed control algorithms. The Fig. 6 shows the behavior of the
quad-rotor in hover flight in presence of external disturbances. As we can see, the attitude control of the
Quad-rotor is satisfactory in the sense that the attitude is kept and stable. Figures 6a and 6b shows that
in hover-mode the speed sensing (RPM) in each DC motor brushless provides robust stability against
disturbances. Fig. 1 shows the four-rotor UAV aircraft hovering autonomously.

7 CONCLUSIONS
In this paper, we have presented a new hardware implementation of a speed sensor system for mea-
suring and controlling the angular velocity on Brushless DC Motors of a Quad-rotor, by combining
an embedded-PD controller and an DSP on-board. The dynamic model of the miniature flying robot
(Quad-rotor) was obtained using an Euler-lagrange approach. The linear control algorithm embedded-
PD (Proportional-Derivative) is used to stabilize the attitude and altitude dynamic system around the

ISBN-xxx-y-zz-yyyyyyy-x 408
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 6: a), b), c) show the behavior of the quad-rotor in hover in the face of disturbances on roll, pitch
and yaw (attitude) respectively and d) show the behavior of the quad-rotor onto the vertical (altitude).

origin where the input control is able to reject some external disturbances, locally. The brushless DC mo-
tors or better known as: BLDC are a good option in the construction and development of a Quad-rotor
because this kind of motors are clean, fast and efficient, and creates no sparks, representing safer choice
in explosive environments. The linear current/torque relationship results in smoother acceleration. In
speed sensing control, it has a response very fast to handle the attitude desired of the Quad-rotor. In
addition, this motor type has a longer life than typical Brushed DC motors. Finally, the experimental
results show that speed control on brushless DC motors works satisfactorily to stabilize the Quad-rotor
on attitude for indoor and outdoor applications. In order to show the Quad-rotor performance, a video
can be found in the following link:

http://www.youtube.com/watch?v=y870AsGkTS4

REFERENCES
[1] I. Fantoni, and R. Lozano, Non-linear control for underactuated mechanical systems, Communica-
tions and Control Engineering Series, Springer-Verlag. London (2001).

[2] Etkin, B., Duff, R., Dynamics of flight, John Wiley and Sons, Inc. New York (1959).

ISBN-xxx-y-zz-yyyyyyy-x 409
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

[3] R. Kelly, and V. Santibanez, Control de Movimiento de Robots Manipuladores, Prentice Hall,
Madrid (2003).

[4] P. Castillo, R. Lozano and A. Dzul, Modelling and Control of Mini-Flying Machines, Springer-
Verlag. London (2005).

[5] H. Goldstein, C.P. Poole and J.L. Safko, Classical Mechanics, Addison-Wesley Publishing Company,
Inc., Massachusetts (1983).

[6] Fantoni, I. and Lozano, R., European Journal of Control, 7, 328-348 (2001).

[7] J. Escareno, S. Salazar-Cruz and R. Lozano, Embedded control of a four-rotor UAV, in Proc.
American Control Conference, Minnesota, pp.3936-3941 (2006).

[8] Samir Bouabdallah and Roland Siegwart, Full Control of a Quadrotor, in IEEE/RSJ International
Conference on Intelligent Robots and Systems, San Diego, CA, pp.153-158 (2007).

[9] S. Bouabdallah et al., Design and control of an indoor micro quadrotor, in Proc. (IEEE) Interna-
cional Conference on Robotics and Automation (ICRA), New Orleans, (2004).

[10] P. Castillo, A. Dzul et R. Lozano, Real-time stabilization and tracking of a four rotor mini rotorcraft,
in IEEE Transactions on Control Systems Technology, Vol. 12, No. 4, pp.510-516 (2004).

[11] Chang-Sun Yoo, and Iee-Ki Ahn, Low cost GPS/INS sensor fusion system for UAV navigation, in
Proc. Digital Avionics Systems Conference, Vol. 2, pp.8.A.1-1 - 8.A.1-9, (2003).

[12] Kong Wai Weng and Abidin, M.S.B., Design and Control of a Quad-Rotor Flying Robot For
Aerial Surveillance, in 4th Student Conference on Research and Development (SCOReD), Malaysia,
pp.173-177 (2006).

[13] S. Salazar-Cruz, J. Escareno, D. Lara and R. Lozano, Embedded control system of a four-rotor
UAV, in International Journal of Adaptive Control and Signal Processing, volume 21, pp.189-204
(2007).

[14] X. Liu, X. Zhao, A. Sanchez and R. Lozano, Design and Implementation of an Embedded Control
System for Small Unmanned Aerial Vehicles: Application to a Four-rotor Mini Rotorcraft, in 4th
IEEE Conference on Industrial Electronics and Applications, pp.414-419 (2009).

[15] A. Sanchez, L. R. Garca Carrillo, E. Rondon, R. Lozano and O. Garcia, Hovering Flight Improve-
ment of a Quad-rotor Mini UAV Using Brushless DC Motors, in Journal of Intelligent and Robotic
Systems, volume 61, pp.85-101 (2010).

[16] MikroKopter
http : //www.mikrokopter.de/ucwiki/en

10

ISBN-xxx-y-zz-yyyyyyy-x 410
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Small UAS for Montagu´s Harrier´s (Circus pygargus) nests monitoring.

MARA MULERO PAZMANY, JUAN JOSÉ NEGRO BALMASEDA

Doñana Biological Station. Spanish Council for Scientific Research (CSIC).

Avda. Américo Vespucio s/n. 41092, Seville, Spain. www.ebd.csic.es

http://aeromab-english.blogspot.com, [email protected]

Abstract

This presentation explains the rationale, objectives and results of Montagu´s Harrier’s nests surveillance
experiments conducted in the Aeromab Project using small UAS (Unmanned Aerial Systems) in Seville
countryside (Southwest of Spain). It presents the situation of the species, whose nests are threatened by
harvesters and the observational techniques that are being used nowadays to monitor the animals. We
describe how a small UAS was used to get aerial photographs that were georeferenced to locate the
nests. We demonstrated that it is possible to use SUAS to complement the nests monitoring, saving a lot
of personnel effort and costs and potentially with less impact on the animals than conventional manned
aircrafts. This also would avoid the necessity of access by foot to the nests and that minimizes the
associated potential risk of predation.

Keywords: small UAS (Unmanned Aerial Systems), Montagu´s Harrier (Circus pygargus), civil UAS applications,
wildlife monitoring, bird nests monitoring.

1. INTRODUCTION

The Montagu´s Harrier (Circus pygargus) is a medium-sized diurnal

raptor. This is a migratory species that uses the Iberian Peninsula to breed,
selecting cereal crops for this purpose. In March or April they breed in
colonies, and build the nest formed by plant stems, resulting in a platform
that leaves several tunnels that serve as shelter for the chicks. On this
platform, they lay the eggs that the female incubates for around 30 days.

In August to September they return to its winter quarters in Africa. [1]

Figure 1: Montagu´s Harrier, adult male.
From the data obtained by SEO Birdlife in 2006, the total breeding pairs
of Montagu´s Harrier in Spain is estimated in the range of 6.093-7.389 pairs (CI 5095-8888) and its population in
Andalusia is about 1.300-1.500 pairs. [2]

Since 1.950 the species has declined dramatically in many countries and in the Iberian Peninsula in the last 20 years
there has been a major regression. The main cause of mortality in Spain is the crushing of eggs and chicks by
harvesters. The fact that the species build their nests on the ground within crops, combined with the advancement in the
collection of crops due to the mechanization of agriculture, have the consequence that these tasks are nowadays
conducted when the chickens have not yet flown. Although there are no exact data to quantify these losses, they have
been estimated in some areas up to 90%. [3]

1 Red UAS 2011. Doñana Biological Station. CSIC

ISBN-xxx-y-zz-yyyyyyy-x 411
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Predation is one of the parameters that affect the reproductive success of the
species. It has been observed that this factor is responsible for nearly 30% of the
nests failed during the years 1.998-2.003 (mainly by foxes, dogs and wild boar).
[4]

The Montagu´s Harrier is catalogued as "vulnerable" in the National Catalogue of

Endangered Species and Red Data Book of Birds of Spain, 2003. It is legally
protected by the Birds Directive 79/409/EEC, Annexes I, II, the Berne
Convention: Annex II, the Bonn Convention: Annex II and the CITES
Convention: C1. [2]

In recent years, several campaigns have been carried out for tracking and locating
nests, and some environmental education has been conducted in several areas of
Figure 2: Harvesters and
Spain. Some regional governments have signed agreements with private property Montagu´s Harrier´s chicks on a
to respect the nesting areas or getting insurance to cover the losses associated to nest.

delay harvesting of crops or partial purchase agreements in the winter sowing.

The location of nests of Montagu´s Harrier is nowadays done through the joint efforts of several technicians who
monitor the areas where traditionally there have been nests. A technician watches the plot and looks for the adults
accumulating material for the nest or for the female perching on the floor to lay down. He gives indication to another
technician who covers the plot on foot in order to locate the nest. This effort represents a high cost of personnel working
hours.

There are indications that access to the plot by technicians in order to locate the nest and monitor eggs and chicks
increases the rate of predation, as it leaves an odorous trail and signs that may attract predators and shows them the
route to the nest. Each new visit that is made to the nest increases the trail detection by predator.

The detection of the nest by aerial photography could be a solution to this problem, as it potentially could be used to
locate and track the nest without leaving traces that could attract predators.

The potential advantages of using small UAS versus conventional aircraft are several: first: economic and logistical
investment is much lower in the case of the UAS, easily transportable by car and not dependent on an airfield to take
off. Moreover, the potential impact on wildlife of the UAS looks smaller than the produced by a conventional manned
plane.

2. OBJECTIVES

The objectives of this study are to evaluate the detectability of nests, eggs and chicks by aerial cameras embarked on
SUAS and to develop a safe methodology for searching and monitoring nests of Montagu´s Harrier by SUAS.

2 Red UAS 2011. Doñana Biological Station. CSIC

ISBN-xxx-y-zz-yyyyyyy-x 412
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

3. METHODOLOGY

General SUAS Requirements for field biology:

A typical mission in field biology is usually quite urgent and not very flexible (animal’s behavior and its locations are
not easily predictable) and normally has to take place in natural areas. These areas usually have mud, water, scrub or
trees, and it’s very common that the best available place to take off and land is a trail or just a patch of grass. During
field campaigns there is not access to labs or specific installations, so if necessary, the repairs must be solved
immediately.

All these conditions determine that the UAS requirements for field biology are: easily transportable: light and small,
durable, modular, easy launchable and recoverable, operable with minimal training, able to collect georeferenced
imagery, easy to maintain, repair and substitute parts and silent (electric powered). [5]

The technical characteristics of the UAS we used are resumed on the following table.

Table 1: UAS Characteristics

Autonomy 30-90 minutes

Range Data Link: 35 Km

Video Link: 18 Km

Altitude 50-200m

MTOW Max 4,5 Kg

Flight Control GPS+Inertial

System and Navigation

Telemetry Diversity in frequency

OSD

XBee 868 Figure 3: SUAS and GCS.

Ground Control Station Portable Netbook with

and GDT Google Earth

Payload 0,5 Kg

Type of payload Cameras, sensors

-1. Simulated nests experiment:

In April 2011 we conducted an experiment in the airfield of Bollullos de la Mitacion (Seville), which has some
cultivated areas around it.

We simulated the appearance of:

-A nest: using three hen eggs placed together.

3 Red UAS 2011. Doñana Biological Station. CSIC

ISBN-xxx-y-zz-yyyyyyy-x 413
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

-An adult Montagu´s Harrier: using a brown leather object of the same size.

The UAS performed several flights at various heights between 200 and 30m AGL over the area where the targets were
placed. The plane was equipped with a digital camera Panasonic Lumix LX3 10 Mpixels resolution with a focal length
of 24mm (35mm camera equivalent). Positional parameters including barometric altitude were recorded with an
Eagletree data logger. The images and data were downloaded and analyzed using specific software (Arc GIS for
georeferencing of images and EagleTree Software for the tracks of the aircraft).

2. Real nests experiment:

The study area of the real experiments was located in the municipality of Gerena, (Seville). It is an agricultural area of
dry cereal, mostly wheat and corn.

We dedicated two days in spring 2011 to locate nests by technicians, observing with binoculars from a car, recording
their position by GPS, and registering the distance and angle to the places where breeding activity was detected with a
rangefinder and a compass. This data were processed to determine the possible nests´ areas using Arc GIS software.

In a first review of the images obtained, it was not possible to distinguish clearly what was a real nest and what was a
hole in the crops. There were many silhouettes that were labeled as possible nests. So in order to get a clear real nest
reference to compare, we made a second experiment in the breeding area. One of the technicians accessed a real nest on
foot and we took photographs of it from the UAS. Then, with that reference, we analized the images of the area again,
discarded the dubious nests and located the real ones.

3. RESULTS

1. Simulated nests experiment:

The optimum height of the plane for taking pictures of the targets was 70 meters. At this altitude it was possible to
discern the objects, and zooming on the picture we were able to identify the simulated nest and the eggs. Flying at lower
altitude, obviously the quality of the pictures is better, but the risk for the system in case of any malfunctioning is
higher, as is the potential impact on the animals. We decided a 70m altitude, which is the minimum for acceptable
results and the maximum for avoiding risk and impact.

4 Red UAS 2011. Doñana Biological Station. CSIC

ISBN-xxx-y-zz-yyyyyyy-x 414
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 4: Simulated nest and Montagu´s Harrier. Top: simulated adult. Bottom:
eggs.

2. Real nests experiment:

Fifteen possible nests were located in Gerena by direct observation of breeding activity. We flew over eleven of them
with the SUAS a week later in order to cover the whole area with the aerial photographs obtained.

After the comparison of all the possible nests labeled with the real nest, we applied a filter with the criteria obtained and
seven targets were identified as real nests. One of them was photographed with an adult female on it, probably
incubating the eggs. No reactions of the animals in the area were detected during the flights.

Figure 5: Left: Montagu´s Harrier´s nest located and observer. Right: nest located on a mosaic of pictures
of the area.

5 Red UAS 2011. Doñana Biological Station. CSIC

ISBN-xxx-y-zz-yyyyyyy-x 415
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 6: Left: 8 Montagu´s Harrier´s nests located. Right: map of the tracks made by the UAS in the
area.

Figure 7: Left: Montagu´s Harrier´s female. Right: Montagu´s Harrier´s nest.

4. CONCLUSSIONS

It is possible to locate a nest of Montagu´s Harrier using aerial photography on an unmanned aircraft flying at an
altitude of 70m AGL.

The animals in the working area did not show any escape, alarm or attack reactions to the plane, which indicates that
the impact of the drones on the fauna is very low. The results suggest that the use of this technology could be very
interesting in areas of difficult access or to avoid possible predation associated with nest entrance by foot.

This task would be more difficult later on when the crop is dry (yellow grass) because there would be less color contrast
between the nests and the environment, so other tests will be necessary in those conditions.

We conclude that SUAS can be a very useful tool in field biology. SUAS offer the attractive possibility to obtain high
quality data in quasi-real time, and sometimes substitute or improve the traditional methods for wildlife monitoring.

6 Red UAS 2011. Doñana Biological Station. CSIC

ISBN-xxx-y-zz-yyyyyyy-x 416
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

UAS manufacturers and developers should take into account that field biologists demand low cost equipment, as the
budgets in these kinds of projects is usually low compared with other fields. Also, should consider that due to the
urgency and unpredictability of their typical missions, they generally prefer to buy the equipments and be independent
than to rent the services. We highly recommend a multidisciplinary approach, and to consider the opinion of the end
users on development projects.

Acknowledgements:

This study was funded by the Autonomous Andalusian Government, within the Aeromab (Aerospace Technologies
applied to Biodiversity Conservation) project, whose aim is to assess and develop methodologies for the study of the
environment through the use of aerospace technology.

The authors wish to acknowledge the work of Miguel Angel Aguilar and Esteban Guerrero, who piloted and maintained
the UAS. We thank Grupo Ecologista Gerena Verde for their cooperation, especially Julio Roldán with his knowledge
of the species and the area, which was essential on this project.

5. REFERENCES

[1] B. Arroyo and J. García. Aguilucho cenizo (Circus pygargus). In, A. Madroño, C. González y J.C. Atienza (Eds.).
Libro Rojo de las Aves de España. Direción General para la Biodiversidad-SEO/Birdlife. Madrid. (2004)

[2] B. Arroyo. and J. García. El aguilucho cenizo y el aguilucho pálido en España. Población en 2006 y método de
censo. SEO/BirdLife. Madrid. (2007)

[3] B. Arroyo, J. García and V. Bretagnolle. Conservation of the Montagu´s harrier (Circus pygargus) in agricultural
areas. Animal Conservation, 5, pp 283-290. (2002)

[4] http://www20.gencat.cat. Accessed on October 10th 2011.

[5] G. P Jones, L.G. Pearlstine, and H.F. Percival. An assessment of small unmanned aerial vehicles for wildlife
research. Wildlife Society Bulletin 34:750–758. (2006)

7 Red UAS 2011. Doñana Biological Station. CSIC

ISBN-xxx-y-zz-yyyyyyy-x 417
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Tail Motion Model Identification for Control Design of an Unmanned

Helicopter

BERNARDO MORCEGO, ELEAZAR SALDÍVAR and FATIHA NEJJARI

Automatic Control Department, Universitat Politècnica de Catalunya,
TR11, Rambla Sant Nebridi 10, 08222 Terrassa, Spain.
e-mail: [email protected]

Abstract
This paper explains the methodology developed to design the yaw control system (heading
control system) of the α-SAC UAV. The problem of modeling and controlling the tail motion
of this UAV along a desired trajectory is considered. First, the response data of the system are
collected during special flight test and a linear time invariant model is extracted by
identification techniques. Then, the control system is designed and implemented using a PID
feedback/feedforward control method. The technique is tested in simulation and validated in
the autonomous flight of the small scale helicopter.

Keywords: UAV, identification, yaw control.

1. INTRODUCTION
The Advanced Control Systems research group at Universitat Politècnica de Catalunya decided to
develop a platform suitable for control algorithms test and design. The selected platform was a fleet of
small scale unmanned autonomous helicopters (the acronym UAV is used for convenience). The long
term research objective of the group is to design a fully autonomous and reliable helicopter autopilot,
able to perform well with different UAV models and in different operating conditions.
Helicopters are mechanically complex and their dynamic behavior is very unstable, apart from being
nonlinear and multivariable, with strong couplings between variables. Those are the main reasons why
they require a lot of work by the pilot during flight and the main motivation for the design of a reliable
autopilot. The viability of the small scale helicopter as a multipurpose research vehicle has driven great
interest these days [1]-[6].
This paper explains the hardware and software system designed and mounted on the commercial R/C
models the group has. This system converts the models into real UAVs. The paper also describes the
control architecture developed and shows results of the methodology carried out to design the yaw
control system (lateral control system). The small scale helicopter used for the results given here is a

-1-

ISBN-xxx-y-zz-yyyyyyy-x 418
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Thunder Tiger Raptor 30 model, named hereafter α-SAC. Fig. 1 (left) shows a picture of the α-SAC
carrying all the instrumentation during one of the experiments.
Human-machine Command computer
interface control

Pilot
Helicopter System Xbee Radio
transmitter

On ground computer WIFI

Figure 1: α-SAC helicopter in autonomous flight (left) and block diagram (right)
The paper is organized as follows: in section 2, the α-SAC hardware description is given; section 3 is
devoted to the software design; section 4 describes the overall low-level autopilot control structure; the
heading identification methodology is presented in section 5; the control design of the yaw dynamics
and some results are given in section 6; finally, in Section 7 the main conclusions are summarized.

2. HIGH LEVEL SYSTEM DESCRIPTION

The UAV is configured around three subsystems: the helicopter, the command computer and the
on-ground computer, as shown in fig. 1. The helicopter carries instrumentation, an on-board computer
and a specialized microcontroller (µC). All the hardware devices are off the shelf commercial devices
which were assembled, connected and programmed by the members of the research group.
The command computer is used to operate the helicopter manually. It is used as a safety device in
case there is a system malfunction. The human pilot controls the helicopter with a radio control attached
to a special interface in the command computer. The computer sends the pilot’s commands to both the
on-board computer and the µC via radio using a MaxStream XBeePRO modem.
The helicopter system receives the pilot’s commands from the command computer with a 990.001
XBee module and distributes them to the on-board computer and the µC through serial/USB
connections. Several sensors are also attached to the on-board computer: an Inertial Measurement Unit
(IMU) that measures, among other variables, the orientation of the helicopter (pitch, roll and yaw); a
barometer, used to obtain the altitude and a camera, that can be used to have an onboard perspective.
The µC is used to drive the helicopter’s servos. The information transmitted to the on-board
computer by the sensors and/or the command computer is processed and transmitted via a wired serial
connection to the specialized microcontroller. Nevertheless, when the human pilot needs to take control
of the system, he turns a safety switch of the radio controller on and then the µC ignores everything but
the pilot’s commands.

-2-

ISBN-xxx-y-zz-yyyyyyy-x 419
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The on-ground computer connects to the on-board computer via standard Wi-Fi. The on-ground
computer is used to monitor and control the UAV system.

3. SOFTWARE DESCRIPTION
The UAV system depends on the correct operation of several subsystems. Each of them is programmed
for a specific task as described in the previous section. The most delicate part is the on-board computer
since it is in charge of receiving data from sensors, receiving commands from the pilot (command
computer), interacting with the supervising user, calculating control actions and finally issuing servo
commands to the µC.

Figure 2: Diagram of the α-SAC software system. Dotted boxes represent applications, each of them
running in a different computer; grayed boxes represent physical devices; and white boxes represent
threads.
The on-board computer is a Linux machine with the real time kernel patch. Its processor is an Intel
Atom Z530 at 1.6GHz with 1GB of board memory and the total weight of the computer is about 300gr.
The program running in the on-board computer is organized as a multithreaded process. The
communication between threads is accomplished through message queues. This way it is possible to
have different cycle times for each thread and different types of communication needs solved in a very
robust and homogeneous way.
Sensor reading threads (labelled IMUsensor and AltSensor in fig. 2) are in charge of regularly
collecting data from the inertial unit and the pressure sensor, respectively. They process the readings if
necessary; for example, the altitude needs to be calculated from a filtered reading of the atmospheric
pressure. Finally, they transmit the data through their message queues, which are of size one, allowing
this way to send only the latest information from the sensors. Thread IControl behaves similarly but
collecting data frames from the command computer instead of sensor readings.
The Logger thread buffers all the data the DataController thread sends. Its cycle time is the largest in
the application and so is its priority (large numbers mean less priority). When it has collected enough

-3-

ISBN-xxx-y-zz-yyyyyyy-x 420
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

data it sends the data through a standard Wi-Fi connection using the UDP protocol.
The Supervisor thread is in charge of controlling the application operation mode. It is an
asynchronous thread since it only reacts to commands sent from the on-ground computer application.
Typical actions that this thread carries out are: changing the control mode from automatic to manual,
changing the controller from one type another one, setting new values for setpoints, etc.
Finally, the Control thread implements the control algorithms that drive one or more variables of the
UAV autonomously. All the values of the helicopter servos are calculated each cycle, or obtained
directly from the human pilot. Then they are sent to the Servos thread, which delivers them to the µC.
The software system introduces unavoidable delays to the control loop. When the on-board computer
is controlling directly the servos of the helicopter, there is a maximum delay of 30ms between sensor
data acquisition and servo movement. The sampling time is 20ms so the delay introduced by the control
system is between one and two samples. When the human pilot is controlling the helicopter the delay
can rise up to approximately 55ms, including all the communications and computing delays. This delay
is short enough to allow an experienced pilot to control the helicopter.

4. CONTROL STRUCTURE
The control architecture of an autonomous helicopter is a controversial topic, according to the variety of
options found in the literature. The reason for that is that helicopter control involves several variables
(attitude, position, velocity) which can be dealt with in various ways.
Hierarchical control is often employed to separate attitude control from velocity or position control,
as in [9]. In this case attitude control is in the lowest level loop.
Even attitude can be treated in various ways. In [10] and [11] an independent control loop for pitch,
roll and yaw is proposed. On the opposite side [12] uses a MIMO controller with four control actions
and 8 measures.
The control of a real helicopter is achieved through four flight controls operated by the pilot: the
collective lever, the cyclic joystick, the lateral control pedals and the throttle. The first two are used to
control the main rotor, the third one is for the tail rotor system and the last one is usually connected to a
governor that depends on the amount of pitch in the main blades.
Although the effects of each control handle are not strictly related to only one variable in the system,
it is true that there is a dominant relation between each control and the key variables of the system as
follows: the cyclic stick works in two dimensions affecting roll and pitch angles; the collective lever is
used to control the altitude of the craft; finally, the lateral control pedals change the yaw angle or
heading of the helicopter.
The control structure used here for attitude control, shown in fig. 3, follows the structure of a real
helicopter. There is a control loop for each actuator.

-4-

ISBN-xxx-y-zz-yyyyyyy-x 421
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Pitch Lat. cyclic

Pitch/Roll Control
Roll
Long. cyclic
Yaw
Yaw Control
Antitorque

Z position Altitude Control Collective

Figure 3: Control structure diagram. It shows the relations between the control variables that are taken
into account.

5. IDENTIFICATION OF α-SAC HEADING DYNAMICS

To ensure the quality of data and obtain a model which represents the operation point dynamics reliably,
flight experiments have to be carefully designed. The system is unstable and it cannot receive arbitrary
input signals in open loop. Nevertheless, with the help of a skilled pilot, it is possible to place the
helicopter in hover position and inject predesigned input sequences for a period of time during which
the pilot does not operate on the commands. After that, or interrupting the experiment if the pilot feels it
is necessary, the pilot recovers the control over the helicopter and it is safely returned to the ground.
This way, all the experiments are virtually open loop experiments. Typical duration of an experiment is
about 20s, which yields 1000 samples.
The data sets used as inputs are the following: a 10s MLBS (Maximum Length Binary Signal)
spanning frequencies from 0.1 Hz to 25 Hz; a 5s MLBS spanning frequencies from 0.2 Hz to 25 Hz; a
20s multisine signal spanning frequencies from 0.5 Hz to 5 Hz; and a step signal. The amplitudes of the
signals were tuned manually to make them as large as possible within the range of behaviors from
which the pilot could recover the control of the helicopter. For the pedals input, u2, binary and multisine
signals were used. For the collective input, u1, pulses inserted by hand were used. The amplitude was
also tuned to be the largest. Fig. 4 shows sample fragments of the signals captured during the
experiments.
System identification, based on the prediction error method, is applied to α-SAC UAV as a MISO
system. The ident Matlab toolbox is used and the reader is addressed to [7] for a very comprehensive
review of the methods used along this section. The auto-regressive moving average exogenous
(ARMAX) models and the state space (SS) models are used to describe the dynamics of this system.
The SS and ARMAX models incorporate a polynomial that models additional information that is not
a part of the model and is treated as noise. When the experiments are well designed and carried out, the
data obtained has enough useful information of the models and this term and its effects are small.
The noise term of the ARMAX model obtained has a large influence in the output, which means that

-5-

ISBN-xxx-y-zz-yyyyyyy-x 422
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

the resulting model does not describe well the helicopter.

exp101112-164548 exp101217-165044.csv
150

100 100

50 u1
0
u2
y

-100 0

-200 -50

-100
-300

-150
-400

-200
20 25 30 35 40 0 10 20 30 40 50 60
Time Time

Figure 4: Fragments of the signals captured during identification experiments. Red line, y, is the yaw
angle; green line, u1, is the lateral control pedal signal; blue line, u2, is the collective lever signal.

The selected model for the pedals-yaw relation is

A( q ) y (t ) = B (q )u (t ) + C (q )e(t )
A( q ) = 1 − 1.5( ±0.11)q −1 + 0.2( ±0.2)q − 2 + 0.3(±0.09)q −3
B1 (q ) = −0.009(±0.04) q − 5 + 0.008( ±0.04)q − 6 (1)
−8
B 2 (q ) = 0.02(±0.002) q
C ( q) = 1 − 0.03( ±0.1)q −1 + 0.25( ±0.04)q − 2
with a loss function of 0.643393 and an FPE of 0.647958.
For the collective-yaw relation we have:
A( q ) y (t ) = B ( q )u (t ) + C (q )e(t )
A( q ) = 1 − 2.34( ±0.035)q −1 + 1.7(±0.06) q − 2 + 0.41( ±0.03) q −3
B1 ( q ) = −0.00009(±0.0009)q −5 + 0.0009(±0.0009) q − 6 (2)

B 2 ( q) = 0.0006(±0.0002)q −8
C (q ) = 1 − 1.3( ±0.02)q −1 + 0.66(±0.02)q − 2
with a loss function of 0.0248903 and an FPE of 0.0250202.
The other method of identification was done considering a state space pre-fixed structure based on
the physics model of the helicopter’s tail dynamics and on [8]. The identification algorithm used was
PEM. The advantage of using this type of identification is that additional information of the system
known beforehand can introduce in the model.
The initial structure is

-6-

ISBN-xxx-y-zz-yyyyyyy-x 423
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

 x1 (k + 1)   1 Ts   x1 ( k )   0 0  u1 (k ) 
 x (k + 1) = a  + b22  u 2 (k )
 2   21 a 22   x 2 (k ) b21 (3)
 x1 ( k ) 
y = [1 0]  
 x 2 (k )
where x1 represents the angle, x2 the angular velocity, u1 the collective input, u2 the pedals input and y
the yaw angle. The seed parameters are taken from [9]. a21=-.3, a22=.9 b21=.1 and b22=.2.
In general these values show that the collective (u1) has a smaller influence in the yaw output than the
pedals (u2) and that the system has and integrator effect. This linear model is assumed to be good
enough when the output and inputs are small, the helicopter is hovering and the angular velocities are
small.
From the experiments of the pedals-yaw it is obtained:
 1 0.02   0 0 
A= , B=  (4)
− 0.00376 0.88796 0.04716 1.009
with a loss function of 0.639837 and an FPE of 0.642947.
For the collective-yaw relation it is obtained:
 1 0.02   0 0 
A= , B=  (5)
− 0.021203 0.89778 − 0.082775 0.56623
with a loss function 0.0254125 and FPE 0.0255121.
The two model structures are evaluated in table I. The parameters compared are the 5-step prediction fit,
FPE, loss function, the noise-output/u-output coefficient and the model order; the first parameter should
be as high as possible, and for the rest of the parameter the lower the number is, the better.
Table 1: ARMAX and SS Model Comparison
ARMAX SS
Collective Pedals Collective Pedals
Fit id-data 95.57% 97.18% 96.02% 97.25%
Fit val-data 96.39% 91.43% 95.97% 92.32%
Noise-Y/u-Y 53 26.8 37 20.3
FPE 0.02502 0.6479 0.02551 0.6429
Loss 0.02489 0.6433 0.02541 0.6398
Model Order 3 3 2 2

Based on this comparison, that shows the slightly better results of the SS model compared to the
ARMAX one, the SS model will be used along the rest of this work.
Fig. 5 shows the time responses of the 5-step predicted output of the chosen model and the measured
output.

-7-

ISBN-xxx-y-zz-yyyyyyy-x 424
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

y1. (5-step pred) y1. (5-step pred)

50 45

40

0 35

30

-50 25
y1

y1
20

-100
15

10
-150 DATTR; measured
5 DATTR; measured
pedals; fit: 92.32%
colective; fit: 95.97%
0
2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20 22

Figure 5: Measured and model 5-step ahead predicted output for the pedals-yaw relation (left) and for
the collective-yaw relation (right). Validation data set is used.

The two transfer functions obtained from the SS model are:

− 0.001656 0.02018
G1 ( z ) = , G2 ( z) = 2 (6)
z 2 − 1.898 z + 0.8982 z − 1.888 z + 0.888

6. CONTROL DESIGN
A PID feedback controller is used for the pedals-yaw relation, and a feedforward controller for the
collective-yaw relation. The control diagram is shown in fig. 6. Disturbance rejection is managed by the
feedback loop but it also helps correcting the disturbances caused by changes on the cyclic commands.
Disturbances correspond mainly to air gusts. The changes in collective are compensated for with the
feedforward controller.

collective input feedforward

PID

yaw
reference + yaw input yaw angle
feedback yaw
+ PID + dynamics
-

Figure 6: Block diagram of the yaw controller, considering two inputs, one output, a feedback controller
and a feedforward controller.
The controller is developed according to the following specifications: rise time of 1 sec., settling time
smaller than 5 sec. and overshoot between 5% and 8%. The design is made focusing on disturbance
rejection and the actuator’s ranges. The controller is designed using the sisotool of Matlab and

-8-

ISBN-xxx-y-zz-yyyyyyy-x 425
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

implemented and tested in simulation with Simulink.

The amplitudes for the collective and pedals step signals are taken from the usual maximum
variations that the inputs have in flight. The wind gust step value (load perturbation) is deduced with the
help of the pilot. A step with amplitude of 5 produces 45 degrees per second change and is considered a
high strength wind. A breeze corresponds to a value of 1 and causes an angular change of 9 degrees per
second. The amplitude selected for the simulation is 2.5.
The feedforward controller is

. (7)
This part was first modeled as a pure static gain (FFgain=0.0147). It showed a very poor performance
and its effect was barely noticeable, so the complete transfer function of the feedforward controller is
used.
The feedback controller transfer function is:

. (8)
Its rise time is 0.201s, the overshoot is 61.3%, the settling time is 2.85s, the gain margin is 9.5 dB and
the phase margin is 26.1˚.
Fig. 7 shows the response of the system under control. The overshoot is higher than expected because
this simulation includes delay terms not accounted for in the design.
Step simulation without FF Step simulation FF-TF
30 30

Collective Collective
Pedals Pedals
25 25
Yaw Yaw

20 20
Amplitude

Amplitude

15 15

10 10

5 5

0 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time Time

Figure 7: Closed loop response. At time 1s there is a yaw set point change and at time 10s there is a
collective change. Left figure corresponds to a feedback controller and right figure corresponds to a
feedforward-feedback controller.
Figure 8 shows the feedforward and feedback controllers working in the helicopter in flight. The
controller is activated at time 20. After it is working we can see that the yaw angle stays around the
reference of -165°; from this point until time 105 the “pedals” input is only modified by the controllers
output. If we disregard the peaks that take place at time 60, that correspond to a bounce with the ground,
the angle does not exceed 5 degrees of error. If we take close attention to the two input signals, we can
see that the pedals’ input reacts to the movements of the collective; it is more noticeable at time 30, 55

-9-

ISBN-xxx-y-zz-yyyyyyy-x 426
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

(before the bump) and 75. Although Wind and air gusts were present during the experiment, they could
not be measured, but their effects are reflected in the tracking of the reference. A simplified, but similar
effect was seen in the simulation of the wind gust.
Inputs

120

100
Collective
Pedals
80
10 20 30 40 50 60 70 80 90 100 110

Filtered and processed pressure

162

160
altitude(m )

158

156

154
10 20 30 40 50 60 70 80 90 100 110

Yaw
-150
-155
angle(degrees)

-160
-165
-170
-175
10 20 30 40 50 60 70 80 90 100 110
time(s)

Figure 8: Experimental results of yaw controller. Top figure shows the inputs of the system; middle
figure shows the altitude of the helicopter; and bottom figure shows the yaw angle and the set point.

7. CONCLUSIONS
This paper resents the control design of the yaw of an unmanned helicopter system using multi-loop
control approach. System identification experiments are applied to model the yaw motion of the UAV
then ARMAX and SS models are identified and validated. Accuracy of the model is verified by
comparing real flight and simulation data obtained during the experiments.
The controller for the yaw system consists of a feedforward and a feedback controller. Essentially the
feedforward section consists in a mathematic cancellation of the signal that affects the system. In the
simulations this controller shows good attenuation of the perturbation but not a complete cancellation
due to the delays of the signals.
The feedback controller is obtained by tuning the PID controller parameters taking into account the
specifications proposed. The control design is validated experimentally showing good performance.

ACKNOWLEDGMENT

- 10 -

ISBN-xxx-y-zz-yyyyyyy-x 427
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

This work was supported in part by the Spanish Exploration and Innovation Ministry under project
DPI2008-00403 (TEOMAPIN) and DPI2008-01996 (HYFA). The authors also want to thank Carlos
Sanlley for his dedication and skilled piloting; without him most of the identification experiments
would not had been able to be carried out.

REFERENCES
[1] G. D. Padfield, Helicopter Flight Dynamics: The theory and Application of Flying Qualities and
Simulation Modeling, Second Edition. AIAA Education Series, Oxford Blackwell Publishing,
2007
[2] R. Austin, Unmanned Aircraft Systems: UAVS Design, Development and Deployment. John
Wiley & Sons, 2010
[3] X. Wang, X. Zhao and M. Tan, Modeling, Identification and Robust Control of Yaw Dynamics of
Small-Scale Unmanned Helicopters, The Key Laboratory of Complex Systems and Intelligence
Science Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China.
[4] D. McFarlane and K. Glover, “A Loop Shaping Design Procedure Using H∞ Synthesis,” IEEE
Transactions on Automatic Control, vol. 37, Jun. 1992, pp. 759-769, doi:10.1109/9.256330.
[5] E. N. Sanchez, H. M. Becerra and C. M. Velez, “Combining fuzzy, PID and regulation control for
an autonomous mini-helicopter,” Information Sciences, vol. 177, issue 10, pp 1999-2022, May
2007
[6] B. Mettler, Identification Modeling and Characteristics of Miniature Rotorcraft, Kluwer
Academic Publishers, Boston, MA, USA, 2003.
[7] L. Ljung, T. Glad, Modeling of Dynamic Systems, Prentice Hall, Englewood Cliffs, New Jersey,
2004
[8] D. H. Shim, H. J. Kim and S. Sastry, “Control System Design for Rotorcraft-based Unmanned
Aerial Vehicles using Time-domain System Identification,” Proceedings of the 2000 IEEE
International Conference on Control Applications, Anchorage, 2000
[9] J. del Cerro, “Arquitectura Abierta para el Control Autónomo y Teleoperado de un
Mini-Helicòpter”, PhD Dissertation, Universidad Politécnica de Madrid, Spain, 2001.
[10] C. P. Sanders, P.A. DeBitetto, E. Feron, H. F. Vuong and N. Leveson, "Hierarchical control of
small autonomous helicopters," Proceedings of the 37th IEEE Conference on Decision and
Control, vol.4, pp.3629-3634, 1998.
[11] Z. Fang, J. Wu, P. Li, "Control system design and flight testing for a miniature unmanned
helicopter," Intelligent Control and Automation, 2008. WCICA 2008. 7th World Congress on ,
vol., no., pp.2315-2319, 2008.
[12] G. Cai, B. M. Chen, X. Dong, T. H. Lee, “Design and implementation of a robust and nonlinear
flight control system for an unmanned helicopter”, Mechatronics, In Press, 2011.

- 11 -

ISBN-xxx-y-zz-yyyyyyy-x 428
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Towards ATM Automation:

A Flight Management System for Unmanned Aircraft Systems
(ATLANTIDA Project)

JUAN JOSÉ CALVENTE PLAZA JULIO ANTEQUIERA GALIANO

AERTEC Ingeniería y Desarrollos, Avda. Descubirmientos 11 Bajo, 41092 Sevilla, Spain
[email protected]

Abstract
This paper covers all the aspects related to a Flight Manager System (FM) for Unmanned
Aerial Vehicles (UAV) within ATLANTIDA project (Application of Leading Technologies
to Unmanned Aerial Vehicles for Research and Development in ATM). In contrast with the
current definition of the Flight Management System (FMS)0, which is a multipurpose
system with trajectory management, navigation and communication functionality among
others, the FMS in the ATLANTIDA context is defined as the automation tool that
implement only the FM logic leaving out of the system some capabilities, such as:
communication, navigation and surveillance (CNS).
The main objective of the ATLANTIDA FMS is to allow testing and validating new Traffic
Management and Flight Management operational concepts.
Within the ATLANTIDA context, the FM is responsible for negotiating trajectories with the
Traffic Manager System (TM), guiding the aircraft according to the contracted trajectory,
and monitoring the execution of the contracted trajectory. To that end, the FM shall keep
internal subsystems up-to-date with external information, such as the Operational Context
Model (OCM), navigation data, and the Weather Model (WM).

Keywords: FM, UAV, TP, SWIM, TM.

1. INTRODUCTION
Nowadays, the Air Transportation sector faces one of the greatest challenges of all its history. In a
time when technological advances in many areas (such as computational methods, computer aided
design and production, materials, engines, information technologies, communications and automation)
are revolutionizing air vehicle aeronautics, paradoxically, the system responsible for their sustainable,
efficient, and safe operation is reaching the limit of its capabilities.
The ATM system of today is the result of more than 60 years of continuous evolution, along which

-1-

ISBN-xxx-y-zz-yyyyyyy-x 429
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

key technological advances as well as procedural and regulatory improvements introduced have
enabled air transportation attaining the highest safety record ever achieved by any other transportation
means. However, some difficulties arise today [2]:
 The central responsibility of the human element as sole source of predictability
 the reduced look-ahead time of the predictions, which confers decision making an eminently
tactical character
 The inability of third-party agents (human or systems) other than the ones in charge of
predicting the traffic situation to take part in the decision-making process
 The need to fit aircraft trajectories to the ATM system limitations instead of being the ATM
systems what accommodates the user-preferred trajectories.
On the other hand, Unmanned Aerial Systems (UAS) sit at the technological fore-front with respect to
automation of air vehicle operations, and so for this reason, the integration of these operations in the
ATM system is often perceived as the most difficult problem that the solution for the future ATM
system needs to address. However, these emerging users that confront the ATM system with the
greatest difficulties do also provide a unique opportunity to develop the solution – and this idea is
exactly the rationale behind the ATLANTIDA initiative (2007-2011).
Aligned with SESAR (Single European Sky ATM Research) and NextGen Programmes, with a budget
of 28,9 million euro, led by BOEING Research & Technology and encompassing another 16 leading
aerospace, information technology and communications companies and 16 prestigious public R&D
entities; ATLANTIDA project has explored an approach for automation in the management of air
traffic seamlessly applicable to any air vehicle operations, including conventional aviation, civil and
military UAVs (Unmanned Aircraft Vehicles).
ATLANTIDA project and has allowed the development, validation and evaluation of innovative
concepts and technologies for air vehicle trajectory management in a net-centric environment, in line
with the following premises:
• Flexible, capable of accommodating user preferences
• Focused on strategic trajectory-based control (TBO paradigm)
• Based on datalink communications
• Separation assurance performed by advanced automation supervised by humans
• Improved predictability leading to reduced safety buffers
• Integration of the FMS with the ground-based automation
• Interoperable solutions based on international standards
Within this context, AERTEC has been responsible for designing and developing the Flight Manager
System, core of the required automation.

-2-

ISBN-xxx-y-zz-yyyyyyy-x 430
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2. OBJECTIVES
The main technical objectives in the frame of the ATLANTIDA project are:
 Design a Flight Management framework that enable testing and validating new Flight and
Traffic Management concepts using the ATLANTIDA experimental UAV setup.
 Define an adequate architecture that allows testing the Flight Management Logic either in a
real flight (Flight Manager as an embedded system in the air platform) or in a Fast Time
Simulation environment (simulated system for software in the loop tests).
 Implement and test a Flight Management Systems that operate according to trajectory
negotiation paradigm proposed within ATLANTIDA Project. That means, an automate way
for exchanging preferred and conflict-free trajectory information between aircrafts and
Traffic Manager in order to come agreement.
Next chapters summarize the needed approach for the analysis and design in to comply with the
objectives aforementioned and provide a systematic approach to comply with the expectations
regarding “Automation” for such a complex scenario.

3. FLIGHT MANAGER RESPONSABILITIES

In order to meet the objectives mentioned above, the FMS shall comply with the following
responsibilities:
 Initialize properly according to a set of user-defined configuration parameters, obeying to the
ATLANTIDA Simulation Control System and setting up it’s communications with the SWIM
platform.
 Switch between different management modes in order to accommodate the requirements of
other ATLANTIDA systems and different operational concepts.
 Generate the guidance reference and the predicted trajectories which are going to be
negotiated, commanded and monitored during the flight. Negotiate the preferred trajectory
with a centralized Traffic Manager according to the negotiation protocol defined in the
ATLANTIDA project. [8]
 Command the contracted guidance reference to the UAV Flight Control System (FCS) and
forward the monitoring information to ground systems. Also, it must ensure that, in case of
negotiation time window expiration, a guidance reference must be commanded always, being
imposed by the TM, or using a safety pre-generated guidance reference.
 Keep log of all actions and events taken during the execution, as well as reporting this
information and module performance data to the Emergency Flight Termination System
(EFTS) and Flight Data recording System (FDRS).
 Refresh predicted trajectory whenever the models that affect the trajectory prediction
functionality change in such a way that the models’ new predictions affect and require to

-3-

ISBN-xxx-y-zz-yyyyyyy-x 431
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

modify the current and active predicted trajectory.

 Monitor the flown trajectory to ensure that those predictions are within a predefined threshold.
Datasets to monitor are supplied by Airborne Navigation System (NAVAIR) and by the
Trajectory Prediction Module (TPM).
 Update the OCM and WM data supplied to the TPM to ensure that the generated guidance
reference comply with the actual flight operational requirements and environmental
conditions.
 Serve as bridge between the Remote-Flight-Manager Human-Interface-Machine (RFMHMI)
ground system and the airborne FCS to monitor the system remotely during laboratory tests
and flight tests.

4. ARCHITECTURE
In order to achieve the aforementioned responsibilities maintaining the flexibility enough that allow
testing different implementations of the FMS functionality, the following logical architecture, showed
in Error! Reference source not found., was purposed.

Figure 1: Flight Manager Logical Architecture.

-4-

ISBN-xxx-y-zz-yyyyyyy-x 432
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

The FM architecture is organized into several cooperating subsystems or sub-modules illustrated on

Error! Reference source not found. and Error! Reference source not found.:
 Data Logger (DL): is responsible of receiving all data to be log from every module and log
it on file and/or screen, and sending the information to the FDRS system as well. The
information filtering is based on verbosity levels and white lists of events, which are
configurable by the user. It also has multiple configuration options for text format, file sizes
and multiple files.
 Flight Manager Logic (FML): is the core of the application, responsible of the trajectory
negotiation with TM, of triggering the trajectory and guidance requests to the TPM, as well as
supplying negotiation results to the Flight Control Manager (FCM) module for commanding.
Also, is responsible of publishing the aircraft information to airborne surveillance systems
(SURAIR). If it’s informed by the ME that the negotiation window has expired, it will first
expect a trajectory imposition from the TM, and if the Expected Time of Arrival (ETA) to the
Terminal Manoeuvre Area (TMA) is reached, it will order the FCM to command the safety
guidance reference that was calculated in the first steps of the execution.
 Flight Control Manager (FCM): is responsible of feeding data to the FC by order of the
FML, as well as receiving reports from it and broadcast them to SURAIR and RFMHMI.
 Functional Status Reporter (FSR): is responsible of collecting the functional status of other
modules and sending them to the EFTS airborne system. In each execution cycle (see Design),
each module must report its functional status to the FSR by rising or lowering the
corresponding flag in the FSR. Then this module collects all the flag data and sends the report
to the EFTS.

-5-

ISBN-xxx-y-zz-yyyyyyy-x 433
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 2: FM sub-module communications with other ATLANTIDA Systems

 Trajectory Prediction Module (TPM): is responsible of generating the guidance reference

and a trajectory prediction starting from a set of trajectory requirements, considering the
restrictions given by the Operational Context Module, and taking into account the influence
of the atmospheric conditions modelled within the Weather Model. As some modules of the
FMS were developed by different partners, some wrapper functionality was developed for
integration purpose. For example: the Trajectory Prediction Module Wrapper (TPMW) is
responsible of receiving all data requests for TPM and doing the library initialization,
configuration and updating tasks, as well as actually requesting TPM calculations and keep
results available for FML and ME.
 Monitoring Engine (ME): is responsible of checking that the navigation data received from
NAVAIR, representing the flown trajectory, corresponds to the predicted trajectory generated
by TPM, issuing the convenient warning signals if a deviation is detected. This monitoring
process must be restarted if a WM data change is detected and the model updated, after
asking for a new trajectory prediction to TPM. Also, ME must keep count of the time left on
the negotiation window, and notify the FML if it expires.
 Operational Context Model Data Container (OCMDC): is responsible of getting the
OCM update notifications and choose which data parts of the available catalogue shall be
requested to perform an update on the TPM information feeds.
 Weather Model Data Container (WMDC): is responsible of getting the WM update

-6-

ISBN-xxx-y-zz-yyyyyyy-x 434
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

notifications and updated data to perform an update on the ME, which will then update the
TPM WM feed. The data feed subscription and receiving is done using the Model
Synchronizer library.
 Model Synchronizer (MS): this subsystem is responsible to receive the new predictions
calculated by the external service DMET each time it publishes a new weather prediction.
This subsystem was designed to minimize the communication’s bandwidth consumed
because there could be more than one internal user of the weather information, and in this
way all clients would subscribe to this centralized service.
 Simulation Signal Container (SIMSINC): is responsible of receiving the instructions from
the ATLANTIDA Simulation Control System which governs the simulation execution.

Figure 3: FM internal data and orders traffic.

5. DESIGN
Due to the necessity of sharing some common functionalities of the DL with the TPM library for
integration purposes, a process of abstraction resulted in the creation of a class hierarchy, being the
FMModule (see Error! Reference source not found.) the parent class, grouping all functionalities
and methods common to all modules, serving some of them as virtual methods, in application of the
template method pattern.
The children classes FMLogger and FMWorker are specializations of the FMModule, being the
former subclasses used by the latter’s. Also, in order to ensure consistency and uniqueness of
processes and data, every concrete module-class is declared as singleton.

-7-

ISBN-xxx-y-zz-yyyyyyy-x 435
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Each module acts as a states machine after its initialization, in which changes between statuses
happen depending on internal data processing results, other modules’ actions or external
communications data arrival. Each states machine, and in the case of TPMW and ME, other complex
calculations are launched in separate threads (using boost implementation) to allow the continuous
execution of all processes and no external data is missed on-arrival due to the module being attending
to other tasks.

5.1. Communications

A parent Communication class was created with all the SWIM integration code, offering abstract
methods for the children subclasses to complete with concrete data types and treatment used in each
module’s communications.
The FMModule class, Communication class and FMLogger, along with several common data type
implementations and auxiliary methods (omitted from the diagram for clarity purposes) were grouped
into the common_fm library, in order to be distributed along the datalogger library, which packs the
DL and DLComms classes, as well as some data types and implementations needed, to make
integration with the TPM development possible.
The OCMDC and SIMSIN use the same ContainerComms class due to the simplicity of their
communications (just one topic each), trying to avoid some code redundancy. Apart from this, WMDC
uses the Model Synchronizer library communications system, and TPMW has no communications at
all. Both of them act as façade for their corresponding external libraries.
ATLANTIDA Communications are based on a SWIM (System Wide Information Management) API.
This API acts as a façade and controller between the FM Communications classes and the RTIDDS
and TAO CORBA systems, used for publish-subscribe and request-reply communications.

-8-

ISBN-xxx-y-zz-yyyyyyy-x 436
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Figure 4 : Simplified class diagram for FM.

6. CONCLUSSIONS
A FMS was successfully implemented and tested within a simulated scenario, in real time and in fast
time simulation, demonstrating the capacity of the system to fully automate the flight management
tasks including the capacity to negotiate the preferred trajectory with a ground automated and
centralized Traffic Manager.
The states-machine based design and thread management has been proved to be key to fully
manage the communications with such a great number of external systems, integrate other libraries
inside the FM and carry out all the internal tasks expected.
The ATLANTIDA project was officially closed last month by means of a Workshop. Several
demos were performed: individual and integrated demos, analysis demos, real-time and fast-time
simulations, hardware in the loop scenarios and flight test.
As part of future work, the negotiation protocol could be improved, by adding the capability to
the FM of starting negotiations, and to perform a tactical renegotiation based on actual flight
conditions of non-nominal scenarios. Another future work is to analyze errors reported by FCS
dynamically and triggering the appropriate actions.
Managing the whole flight, not only the approaching phase is also an interesting challenge. The
update of operational constrains in-flight and assessment of the impact in trajectory terms could also
be a future work to undertake.

-9-

ISBN-xxx-y-zz-yyyyyyy-x 437
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Finally the adaptation of the FM to use real time data, signals and thread management would
definitely be an improvement in performance and memory usage in combination to adding the
capability of fully pausing/resuming the whole system if the Simulation Signal system issued the
corresponding signals.
All these improvements would, in the end, add flexibility to the FM making it more responsive,
adaptable and usable in different situations which differ from the simulated scenarios which drove the
FM development and feature implementations

7. ACKNOWLEDGEMENTS
This research has been supported and funded under the project CENIT ATLANTIDA, led by Boeing
Research & Technology, and has counted with the cooperation of several ATLANTIDA Consortium
partners, subcontractors, and universities such as Indra, Vaxtor, Integrasys, TCP, AICIA, UPM and
Boeing Research and Technologies Europe.

REFERENCES
[1] Eurocontrol, EATM glossary of terms, June, 2004.
[2] http://www.atlantida-project.org/portal/index2_en.html
[3] ATLANTIDA D4.5.1 Analysis of automation technologies for flight management.
[4] ATLANTIDA D4.5.2 – Flight Management Automation CoOp Definition.
[5] ATLANTIDA D11.9.1 – FMS Design.

- 10 -

ISBN-xxx-y-zz-yyyyyyy-x 438
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

APPENDIX

A ACRONYMS
Table 1: Acronyms.
API Application Programming Interface
ATM Air Traffic Management
CNS Communication, Navigation and surveillance
DL Data Logger
EFTS Emergency Flight Termination System
FC Flight Control
FCM Flight Control Manager
FDRS Flight Data Recording System
FM Flight Manager
FML Flight Manager Logic
FMS Flight Management System
FSR Functional Status Reporter
HIL Hardware In The Loop
ME Monitoring Engine
NAVAIR Airborne Navigation System
OCM Operational Context Model
OCMDC Operational Context Model Data Container
RFMHMI Remote-Flight-Manager Human-Interface-Machine
R&D Research & Development
SESAR Single European Sky ATM Research
SIMSINC Simulation Signal Container
SWIM System Wide Information Management
TBO Trajectory-Based-Operations
TM Traffic Manager
TMA Traffic Manoeuvring Area
UAS Unmanned Aircraft Systems
UAV Unmanned Aircraft Vehicle
WM Weather Model
WMDC Weather Model Data Container

- 11 -

ISBN-xxx-y-zz-yyyyyyy-x 439
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Proposal for PELICANO “block 0” demo

GONZALO PACIOS
Indra Sistemas, [email protected]

Abstract
The PELICANO system is reduced maintenance and logistics UAS with the capability to
perform medium range missions with payloads up to 50 Kg. It incorporates a fully
autonomous take and landing system and it can be operated from land (Army
configuration) or from a vessel (Naval configuration).
The system is designed to fit with Defence and Security Forces, taking into consideration
Airworthiness Certification requirements as well as the Spanish Navy requirements and
CONOPS.
Keywords: AVTOL, UAV, certification, naval applications

1. SYSTEM DESCRIPTION
The PELICANO system is being developed with the following type of missions in mind:
 Navy deployment: Taking-off from a vessel and reaching the area of interest in very short
time, while the system remains long time over the area, using the payload to identify/follow
objects. 24hours operations are granted up to a period of 30 days.
 Search and rescue missions: The system carries a payload devoted to provide support in
rescue missions and to allow recovery of people involved in catastrophes.
 NBQ missions: The PELICANO is equipped with a set of sensors to measure NBQ levels and
even to pick up samples for further analysis.
Nowadays, the UAS are required to be certified in order to allow them to achieve their maximum
potential in civil and non-segregated airspace missions. Therefore the PELICANO system design
takes into consideration the future Airworthiness Certification of the system by following the CS-VLR,
D0254 and DO178/B as applicable to the design while Indra has developed qualified equipment &
payloads for PELICANO System.
The PELICANO project follows an incremental approach where three different prototypes are to be
generated which allows easier system integration. This demo will be performed by the PELICANO
“block 0” which does not include payload (to be integrated in the next prototype) but it does perform
fully autonomous missions.

-1-

ISBN-xxx-y-zz-yyyyyyy-x 440
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

2. TECHNICAL CHARACTERISTICS
Table 1: Main technical features of the system.

Weights
MTOW 200 kg
Empty 120 kg
Useful load 80 kg
Maximum fuel capacity >50 kg
Maximum payload weight 50 kg
Performance (Standard day, sea level, MTOW, 30 minutes fuel reserve, unless
otherwise noted ) (All speeds are indicated air speed)
Maximum operational velocity 150 km/h
Speed for best endurance 90 km/h
Maximum operating altitude 3.600 m
Mission examples (Standard day, sea level, unless otherwise noted )
Maximum endurance (15 kg payload) > 6 hours
Maximum range (15 kg payload) TO and landing at the same place 270 km

3. DEMO FLIGHT DESCRIPTION

Before the demo flight will start, a safety check on the fuel pumps is required to guaranty safety of
operation. Then, the helicopter will take off autonomously, it will performed a programmed mission in
auto mode and it will stop at one point where the control will be switched to assisted and the GCS
operator will control the helicopter form the GCS. Finally, the system will land automatically.

4. REQUIREMENTS
 Permit to flight for Civil Aviation authority; due to the PELICANO’s weight and type of missions
it is required to have a certification for Civil Aviation which will ensure the safety of operation
and that no interference can occur.
 A device to measure the wind speed and direction
 The area where the demo will be executed shall have minimum dimensions of 300m x 20m.
 A map of the area should be provided in advance so it can be calibrated in advance.
 Access to the test filed shall be provided days in advance to the demo day, o Indra’s crew can
make some missions to become familiar with the site, winds, etc.

-2-

ISBN-xxx-y-zz-yyyyyyy-x 441
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Proposal for MANTIS demo

CESAR MENDEZ
Indra Sistemas, [email protected]

Abstract
MANTIS is our third generation low weight and very low acoustic trace mini UAV
multirole system designed for a wide range of operations for civilian and military
applications, from surveillance, reconnaissance to border control and intelligence missions
It is a mature system that has evolved from previous designs improving acoustic track,
flight performances, endurance and reliability
It provides real-time IMINT (Imagery Intelligence) information to support tactical level
units (including Special Forces battalion level) running patrol, urban areas research, force
protection or secure perimeter missions
The system has been designed to operate autonomously during all phases of flight,
allowing operators to obtain real time recognition data at different weather conditions, at
day or night.

Keywords: ATOL, IMINT, UAV, STANAG 4586, surveillance, reconnaissance, border control
applications

1. SYSTEM DESCRIPTION
Air Segment is composed by up to four simultaneously operated UAV`s, powered by LiPo or LiFe
batteries (application dependant), with electric propulsion motor, autopilot with GPS, inertial,
magnetic, and barometric navigation sensors, visible or infrared EO payloads with video processing
unit for video stabilization compression and video-tracking, LPL subsystem for precision landing, and
Security Switch with removable pin-flag for disabling motor running when inserted

Ground Segment is composed by a GCS (Ground Control Station) for control the system, together
with GDT (Ground Data Terminal) both may be placed static on ground or on a surface moving
platform, on ground or at sea.

The system has been designed in a modular basis to facilitate transport and maintenance tasks, also to
facilitate upgrades as may be needed in the future
It has the following main operative capabilities:
• Simultaneously control of four UAV’s from a single Ground Control Station (GCS).

-1-

ISBN-xxx-y-zz-yyyyyyy-x 442
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

• Capability to hand-over control from one GCS to another GCS on flight time.
• Up to 30 Km operational range.
• Designed to operate on climatic conditions from -25º to +50º and up to 30Km/h wind force.
• Landing capability over non conditioned fields.
• Easy interchange of multifunctional payloads on line flight.
• GCS operation in accordance with STANAG 4586.

2. TECHNICAL CHARACTERISTICS
Table 1: Main technical characteristics.

Wing span 2.10 m

Length 1.47 m
Maximum height 0.28 m
Wing area 0.52 cm2
Mean aerodynamic chord 0.24
Aspect ratio 8.75
Wing load 120g/dm2
MTOW (EO camera configuration) 5075 gr.
MTOW (IR camera configuration) 5150 gr.
Maximum endurance 120 min.
Maximum range (LOS) + 30 km
Service ceiling 5000 m (MSL)
Maximum not exceeding speed 150 Km/h
Maximum speed at stable altitude 110 km/h
Stall speed 34.5 km/h
Maximum range speed 65 km/h
Maximum endurance speed 65 km/h
Maximum rate of climb 6 m/s

3. DEMO FLIGHT DESCRIPTION

A real operational demonstration with the following steps:
 Mission Plan edition and uploading
 Single UAV launch
 Mission Way-point execution monitoring

-2-

ISBN-xxx-y-zz-yyyyyyy-x 443
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

 EO video images exploitation demonstration

 Automatic Landing

4. REQUIREMENTS
 Flight permission as required by pertinent authorities
 100 by 50 meters test field
 Local coordinates for mission planning (a map will be helpful for it)

-3-

ISBN-xxx-y-zz-yyyyyyy-x 444
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Proposal for demonstration by Unmanned Solutions in RED-UAS 2011

JESÚS GARCÍA MUÑOZ

Unmanned Solutions, C/Milán 34 bajo – 28043 Madrid, [email protected]

Abstract
This is a proposal for demonstration by Unmanned Solutions in RED-UAS 2011.
Unmanned Solutions is a technology based company originated inside the Universidad
Politécnica de Madrid. Its founders are researchers from the Aeronautical and Space
engineering School with the vision of becoming one of the few manufacturers of certified
civil UAS in the world.
Unmanned Solutions would like to propose the execution of a flight demonstration in
RED-UAS. K2B6 UAV is the platform that USol will use in this flight demonstration. The
Cloudcap Piccolo Autopilot will be used for autonomous flight, and one Tase Duo Gimbal
will be the payload in use during the flight demonstration.

Keywords: Unmanned Solutions, K2B6, Piccolo, TASE-DUO, flight, autonomous

1. DESCRIPTION
Unmanned Solutions would like to propose the execution of a flight demonstration in RED-UAS
using the current UAS demonstrator. The vehicle of this system is the K2B6, and the Cloudcap
Piccolo Autopilot will be used in autonomous flight. One Tase Duo Gimbal will be the payload in use
during the flight demonstration.
The demonstration includes:

− Remote engine start

− Remote taxi to the runway
− Take off (autonomous or manual; it will depend on the conditions)
− Climb to 200 m AGL
− Autonomous circuit at 200m AGL
− Autonomous surveillance circuit at 200m AGL, using the Tase Duo retractable gimbal.
− Descent and approach
− Landing (autonomous or manual; it will depend on the conditions)

-1-

ISBN-xxx-y-zz-yyyyyyy-x 445
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

− Remote taxi to the platform

− Flight end

2. CHARACTERISTICS OF THE SYSTEM

2.1. K2B6

The K2B6 aerial platform is Unmanned Solutions fourth generation UAV. It is based upon the
experience gathered since 2004 in the design and manufacture of high performance tactical range
unmanned aircrafts for civil and security applications.

2.2. Piccolo II

PiccoloII Autopilot provides a complete, off the shelf avionics system solution, including the core
autopilot, flight sensors, navigation, wireless communication, and payload interfaces. It will be
integrated in the demonstration with a Microair-T2000UAV-L transponder and differential GPS.

2.3. Tase Duo Gimbal

The TASE family of gimbals provides rich features such as built-in moving maps, geo-pointing and
geo-locating, object tracking, and sensor fusion. It will be equipped during the demonstration with the
324x256 FLIR Photon II IR Camera, and an FCB-EX980 Sony Block Camera.

3. NECESSARY REQUIREMENTS

3.1. Airfield and Flight area

Runway length: 500 m / Runway width: 20 m clear

Clear approach to the runway without electric power lines in less than 1500 m around the airfield
Active NOTAM in the Airfield and airfield officially closed to all other manned operations.
Air traffic and flight operations manager is required.
Civil Liability Insurance from the Event Organization is required.
Environment in accordance with the standard: BOE-A-2010-835 (January, 19/2009 – Sec1 p4649).

3.2. Time duration

The estimated flight time duration of the demonstration is about 30 minutes, from takeoff time to
landing time.

-2-

ISBN-xxx-y-zz-yyyyyyy-x 446
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

Proposal for demonstration of multi-UAS cooperation system

Carlos Albo, Allan Anderson Bell, Antidio Viguria and Anibal Ollero

Center for Advanced Aerospace Technologies (CATEC)

Parque Tecnológico y Aeronáutico de Andalucía
C/ Wilbur y Orville Wright 19 41309 - La Rinconada Seville (Spain)
{calbo, abell, aviguria, aollero}@catec.aero

Abstract
This proposal will set the flying demonstration with two CATEC’s UAS: Viewer and
SARAH. Both UAS will fly in a cooperative way in order to fulfill a surveillance mission
where, after a possible objective is detected using a long range UAS such as the Viewer,
the objective is confirmed by the SARAH helicopter.
.

Keywords: multi-UAS demonstration, rotary-wing UAV, surveillance, objective detection.

INTRODUCTION
In many surveillance missions it is important to confirm the possible detected objectives in order to
minimize false positives. In this demonstration, a surveillance mission will be simulated where the
fixed-wing UAV Viewer will be used as a long range aerial vehicle that will be responsible to detect
the possible objectives. Once they are detected, they will be confirmed by the SARAH helicopter
using a high resolution IR camera on-board.

The Viewer is a fixed-wing UAS with 2 kg payload capacity and an endurance of 1 hour. It has a
electrical engine and a some space to integrate new systems and processing units. It is completely
autonomous (Piccolo autopilot and GCS from Cloudcap) with a maximum range of 40 km (limited
by the data link). These vehicles lands over its belly and uses a launcher for taking-off. The default
payload is a gimbal with a visual or an IR camera.

Figure 1: Viewer fixed-wing UAV

On the other hand, the SARAH UAS has with 5-7 kg payload. It is completely electrical and has an
endurance of 30 minutes. As far as we know, this is the first UAV model completely electrical with
the above payload. This fact makes it very easy to maintain and reduce considerably the vibrations
of this small vehicle. It is completely autonomous (autopilot and GCS from Flying-Cam) including

ISBN-xxx-y-zz-yyyyyyy-x 447
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

take-off and landing. The default payload is a gimbal that can integrate many different cameras from
HD visual to cooled IR.

Figure 2: SARAH rotary-wing UAV

REQUERIMENTS
The requeriments for these demonstrations are a clear airspace of 150x150x30 meters for SARAH
and 1000x750x200 for Viewer (LxWxH) with a minimum flat terrain of 150x15 meters (LxW).
Approach slope should be clear of obstacles 500 from meters away. The Ground Control Station
(GCS) should be placed on the midpoint of one edges of the airspace, beside the landing area. As a
safety requirement only the vehicle operator and the safety pilot should stay at the GCS level. All
observers shall stay behind.

DEMONSTRATION
In this demonstration we will show a simulated surveillance mission where the fixed-wing UAV
Viewer will be used as a long range aerial vehicle that will be responsible to detect the possible
objectives. Once they are detected, they will be confirmed by the SARAH helicopter using a high
resolution IR camera on-board.

ISBN-xxx-y-zz-yyyyyyy-x 448
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ALO FLIGHT DEMO IN RED-UAS

Francisco Muñoz
Instituto Nacional de Técnica Aeroespacial

The ALO (Avión ligero de Observación, Light Surveillance Aircraft) is a small

size tactical unmanned aircraft system for mid-range operations. It incorporates
a payload based on a Mini Dome with integrated infrared and visible cameras
both with motion in elevation and azimuth.
The ground control station, which is currently integrated on an UNIMOG van,
which serves as well as transport and maintenance support vehicle, can be
downsized up to a small rack version containing all GCS features. Both TV
images (Vis+IR) sent by the air vehicle are monitored in real-time alongside with
the relevant mission and digimap data.

The system operation is carried out in automatic mode in all flight phases
including aircraft taxiing, takeoff, mission execution and landing, giving the
system a great capacity for night missions.

TECHNICAL DATA

SYSTEM MAIN PERFORMANCE

Endurance > 4 hours
Range 50Km (Optional up to100 km)
Ceiling > 12000 feet
Maximum speed 90 knots
Minimum surveillance speed 45 knots
Capacity ATOL (Automatic Take-Off and Landing)
Visible and IR Images

AIR VEHICLE DATA

Wingspan 3.48m
Length 2.30m
MTOW Maximum weight 55Kg
Maximum Payload Weight 8kg

POWERPLANT
Engine: 2T/150cc/16, 5hp
Alternator: 500 Watts
Fuel: 95 octane + 2% synthetic oil

AVIONICS
Data link: 900 MHz
Video link: 2.4 GHz
Navigation Lights / Position / Collision
Battery Back-Up
3-axis magnetometer
IMU

ISBN-xxx-y-zz-yyyyyyy-x 449
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

DIFF GPS and EGNOS

ADS
Laser Altimeter

PAYLOAD
MiniDome 2-axis (360 º azimuth/90 º elevation)
Extraction and retraction system
Daylight Camera PAL (25x optical zoom)
IR Camera PAL (7.5 to 13.5 microns)
Auto Targeting

SYSTEM FLEXIBILITY
Operation in short unprepared runways.
Completely automatic.
ATOL (automatic takeoff and landing)
Ramp launching capability
Emergency Parachute
Day and Night Operation
Quick readiness to operation and flight
Visible and IR images in real time
Capacity for different payloads
Missions ISR (Intelligence, Surveillance and Reconnaissance). EW (self
protection) pod available.

DEMONSTRATION BRIEF DESCRIPTION

Operation is based on a typical takeoff, flight and landing mode.

1 Automatic take off
2 Scheduled mission, flying at a selected altitude.
3 Flight along Programmed mission.
4 Surveillance over selected POI´s (points of interest). Switch on and off the
programmed route for watching specific POI´s selected in Real Time. Keeping
the automatic and semi automatic flight modes. Selection of preprogrammed
manoeuvres (circles, grid shape). Re-allocation of new waypoints in RT. At all
times the payload operator would obtain visible and IR images of the terrain,
focusing on specific points.
5 Proceed to approach and landing in automatic mode.

Constraints/ conditions:

o Total duration of demo can be fit in less than 30 minutes (40 including
warm up and taxi runs).

o Complete demo fit well inside a 4x2 km area (could be smaller if

needed). TO & L strip length < 400m

o Flight altitude 200m< h < 1000m.

ISBN-xxx-y-zz-yyyyyyy-x 450
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

o Demo site should be available the previous day in order to check about
EM ambient conditions in the area.

ISBN-xxx-y-zz-yyyyyyy-x 451
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

ELIMCO E100 & E300 UAS DEMOSTRATION

JOSE MARIA CALERO

Head of UAS at ELIMCO
PTA Aerópolis C/ Hispano Aviación 7-9 41300 La Rinconada (SPAIN)
[email protected]

Abstract
ELIMCO intent to make demonstration of full autonomous flight capabilities of the E100
micro and E300 mini fixed wing UAV.

1. UAS E100
1.1. System overview

Extremely light and easy transport aerial system, designed for surveillance and reconnaissance
missions.
E100 has day and night flight capability, a very noiseless electric engine with 5 km range and 60
min endurance.
The manual or launcher take off systems make possible the operation in wide type of fields and
situations.
 WING SPAN: 1,2 m
 ENGINE: Electric (brushless), high efficiency.
 CONSTRUCTION: Composite and Elapor
 AUTOPILOT: Paparazzi with IMU stabilization
 ENDURANCE: 60 min
 RANGE: 5 km
 MTOW: 1 kg
 PAYLOAD: 150 grams
 TAKE OFF: From hand.

1.2. Summary of demostration

The demonstration would consist in:

 Autonomous launch from hand
 Climbing to safe waypoint

-1-

ISBN-xxx-y-zz-yyyyyyy-x 452
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

 Waypoint orbiting
 Autonomous polygon survey
 Home return on signal fail
 Realtime video broadcast
 Autonomous approach and landing

1.3. Means required

50x10 m not prepared runway with no obstacles approach above 3 degrees.

2. UAS E300
2.1. System Overview

It is a sturdy, light and handy aerial system designed for strategic, observation and reconnaissance
missions.
E300 UAV stands out due to its payload and its low-sound impact (electric propulsion).
It can accomplish both night and daytime missions, with up to 180 minutes endurance and 45 km
operational range.
Due to its transportable gear-landing launch system can be operated at any place and situation. Its
recovery is made through the parachute recovery emergency system.

 WING SPAN: 4,8 m

 ENGINE: Electric (brushless), high efficiency.
 CONSTRUCTION: Composite (carbon and glass fiber)
 AUTOPILOT: CloudCap Piccolo II with DPGS RTK
 ENDURANCE: up to 180 min
 RANGE: 45 km
 MTOW: 15 kg
 PAYLOAD: 4 Kg
 TAKE OFF: From catapult.
 RECOVERY: Belly landing or emergency parachute.

2.2. Summary of demostration

The demonstration would consist in:

 Autonomous launch from catapult

-2-

ISBN-xxx-y-zz-yyyyyyy-x 453
Proc. of the 1 st Workshop on Research, Development and Education on Unmanned Aerial Systems (RED-UAS 2011)
Seville, Spain, Nov 30th - Dec 1 st , 2011

 Climbing to safe waypoint

 Waypoint orbiting
 Rectangle waypoint tracking
 Altitude change tracking
 Home return on signal fail
 Realtime video broadcast and gimbal pointing
 Autonomous approach and landing

2.3. Means required

100x15 m not prepared runway with no obstacles on 500 m approach above 3 degrees.

-3-

ISBN-xxx-y-zz-yyyyyyy-x 454