Research Article

International Journal of Advanced

Robotic Systems
September-October 2016: 1–14
An online real-time path compensation ª The Author(s) 2016
DOI: 10.1177/1729881416663366
system for industrial robots based arx.sagepub.com

on laser tracker

Xiaojia Shi, Fumin Zhang, Xinghua Qu, and Bailing Liu

Abstract
Industrial robots have been widely used in many areas for the properties such as low cost, high flexibility and relatively large
working space. The low absolute accuracy and low stiffness, however, limit the application in high precision areas. This article
presents an approach for industrial robots that overcomes the inaccuracy of robotic path in offline programming. Robot Sensor
Interface of KUKA robot controller makes the error compensation available. First, the compensation performance of the KUKA
KR5 arc robot is investigated and the straight-line and circular arc path compensation strategies are proposed. Besides, an error
compensation algorithm is proposed to generate the compensation values as well as to deal with the oscillation of real-time
compensation. Finally, experimental results using a KUKA KR5 arc robot verify the correctness and accuracy of the approach.

Keywords
Industrial robots, robotic path, offline programming, real-time compensation

Date received: 22 February 2016; accepted: 17 July 2016

Topic: Robot Manipulation and Control

Topic Editor: Andrey V Savkin
Associate Editor: Alexander Pogromsky

Introduction programmer-safe environment in programming 3 as

the robot can be programmed in virtual environment,4
Industrial robots have been widely used in the field of industry
while the low absolute accuracy restricts the application.
so far due to their lower price, the larger working space, high
To solve these aforementioned problems and improve
degree of automation and so on. Most of the robot applica-
the absolute accuracy, a common approach that can elim-
tions, however, are not involved in high precision fields, such
inate the errors caused by the mechanical errors is to cali-
as handing, welding, painting, assembling and so on. The high brate the robots using kinematic models.5,6 The most
repeatability of industrial robots represents for those tasks.1
widely used model is the Denavit–Hartenberg (DH)
Nevertheless, the industrial robots contain lots of errors, such
model.7 Similarly, to compensate the robot deformation
as mechanical errors, thermal errors and so on.
that is caused by force acting on the end of the robot manip-
Moreover, the stiffness of today’s industrial robots is very
ulator, a stiffness model is used8–10 to apply robot stiffness
low. Consequently, the error is particularly large especially
identification.
when the robots load weights. The dominant contribution
factor for this is the joint compliance, such as gear transmis-
sion elasticity.2 The industrial robots, hence, have limitation
on the fields of precision path control. All these factors State Key Laboratory of Precision Measuring Technology and
mentioned earlier account for their lower absolute accuracy. Instruments, Tianjin University, Tianjin, People’s Republic of China
Due to the advantage of the reduction of the downtime
Corresponding author:
on the production line, the offline programming has been
Fumin Zhang, Tianjin University, 92 Weijin Road, Nankai District, Tianjin
more and more extensive. Some programming methods 300072, People’s Republic of China.
based on virtual reality are proposed to provide Email: zhangfumin@tju.edu.cn

Creative Commons CC-BY: This article is distributed under the terms of the Creative Commons Attribution 3.0 License
(http://www.creativecommons.org/licenses/by/3.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 International Journal of Advanced Robotic Systems

Although the usage of these models and offline compen-

sation methods improves the absolute accuracy of robots, the
error varies from different positions and postures in the work-
ing space. It is impossible to obtain the accuracy requirements
for the whole working space. These methods can only take as
much as dispersed points to calibrate in the workspace. Con-
sequently, the selection of calibration lattice in the workspace
has a great influence on the robot final accuracy. The accuracy
out of these selected points is not particularly high. Addition-
ally, this calibration method is relatively time-consuming,
which needs large amount of computations.
There are also some methods to solve the problem. Chiba
et al.11 have proposed a method for rapid manipulator path
planning that is able to satisfy the design time constraints. A
flexible multisensor positioning system was proposed in the
study of Roebrock.12 The system puts forward a multisensor Figure 1. System set-up.
frame structure to realize measurement using different kinds
of sensors and solves the positioning problem of the tool of System overview
robots rather than solving the path accuracy when performing
path tasks. An offline path correction is presented in the study System structure
of Roebrock and BÃ} uhnke13 to improve the path accuracy. The online real-time compensation system is comprised of an
The offline method is divided into a measurement run and an industrial robot, a laser tracker, a computer and the measure-
application run, which is time-consuming and obtains a low ment compensation software program. The overall structure is
accuracy as well. In the study of Zi et al.,14 the sliding mode shown in Figure 1. The personal computer (PC) communicates
control is utilized to control robots, through which higher with the robot via Ethernet. Simultaneously, PC communi-
trajectory accuracy can be obtained after the control system cates with the FARO Laser Tracker XI, FARO Technologies,
is stable. Zhao et al.15 proposed a method to correct a pre- Inc. via Ethernet. The software program was made in Cþþ
planned path through an iterative learning control method. language based on Microsoft Foundation Class (MFC). The
Wang et al.16 proposed an approach used for trajectory or open laser tracker interface for secondary development allows
contour tracking with the goal of supporting noncontact robot us to realize continuous measurement. The function of the PC
machining applications. An error analysis of cooperative software program is mainly to receive data from the robot, read
cable parallel manipulators is analysed in the study of Qian the measured values of the laser tracker, perform compensa-
et al.,17 which provides a way of solving the effect of force. tion algorithm and simultaneously transmit compensation data
In this article, an online real-time compensation method to the robot in real time. If the robot reaches the required
based on a laser tracker is presented. This method aims at accuracy that we expect to achieve during performing path
increasing the accuracy without precise calibration. By com- tasks, the software doesn’t execute the computational compen-
paring the real-time data of robot measured by an external sation program. That is to say, we don’t give the robot com-
measurement system (laser tracker) with the theoretical pensation values. The laser tracker runs in continuous
coordinate values in robot offline programming when measurement mode. Figure 2 presents the relationship
executing path tasks, the error values of the robot can be between different coordinate systems. Ob Xb Yb Zb , Ot Xt Yt Zt
calculated through the proposed path compensation strategy. and Oe Xe Ye Ze are, respectively, the robot base coordinate
By a real-time mechanism, compensation values at each system, the laser tracker coordinate system and the robot tool
moment can feed back to the robot in real time. Industrial coordinate system at the robot end effector. The compensation
robots can compensate for these errors through the compen- values calculated in the laser tracker coordinate system should
sation mechanism relatively quickly. The errors resulting be converted to the robot base coordinate system. The flow
from robot loads can also be compensated during executing block diagram is shown in Figure 3.
path tasks, such as robot manufacturing and machining. The program runs using multiple threads because it
This article is divided into six sections. In section ‘System simultaneously receives the information from the robot and
overview,’ the online measurement and control compensa- laser tracker. After processing these data, it sends the infor-
tion system are presented and the KUKA Robot Sensor mation with the Extensible Markup Language (XML) for-
Interface (RSI) is introduced. In section ‘Robot path com- mat to the robot.
pensation’, researches on the compensation performance of
the KUKA robot are presented. Besides, the path compen-
sation algorithm and strategy are presented. Experimental Robot sensor interface
design and validation are presented in ‘Experiments’ section. KUKA RSI enables the robot controller to communicate
Finally, the last section gives the conclusions. with external systems,18 which can exchange data via
Shi et al. 3

SMR

Laser
tracter

Robot

Figure 2. Relationship between different coordinate systems.

Figure 4. Robot Sensor Interface execution diagram.

Ethernet. The movement or path of robots can be affected or

guided by the corresponding data received from external sys-
tems. The exchange data format and the communication must
be configured in the robot controller. And data transmission
must be strictly in accordance with XML format, which
enables data between different systems to be compatible.
The characteristic of RSI is that the signal processing
must be in the 12 ms interpolation cycle and is parallel to
robot program execution (see Figure 4). If the robot fails to
receive data within 12 ms, an error of late packets occurs.
When the error counts reach the set threshold in the con-
figuration file, the communication interrupts. Thus, the
data exchange between the robot and the external devices
is relatively real time.

Robot path compensation

Single error compensation performance
of KUKA robots
Through KUKA RSI, the errors can be compensated. First,
the single error compensation experiment aims at investi-
gating the compensation performances of KUKA robots. In
this section, we set 0.5 mm, 0.7 mm and 1.0 mm compen-
sation values, respectively, in the directions of X, Y and Z
with respect to the laser tracker coordinate system. The
experimental results obtained are shown in Figure 5.
It is found that the error compensation of the robot has
hysteresis. This is because the robot’s electronic and
mechanical components are hysteretic. Hence, it is not
able to compensate the error with an extremely high
speed. It can be found that the compensation cycle is
Figure 3. Flow block diagram. approximately 500 ms.
4 International Journal of Advanced Robotic Systems

Figure 5. Results of single error compensation.

Path compensation strategy

Robots are increasingly using offline programming, which
requires high absolute accuracy. Because of industrial robots’
errors, the absolute accuracy is low and there are deviations
between the actual path and theoretical path while executing
tasks. When the robot loads weights, the low stiffness leads to
considerable deviations as well. Therefore, the deviations caused
by these factors must be detected and be compensated either.
As the straight-line and circular arc paths are the most basic
components of robotic path tasks during robot running programs
in practice, we will propose a straight-line path and a circular
arc path compensation strategy in the following section.

Straight-line path compensation strategy. The straight-line

path compensation strategy proposed in this system is
shown in Figure 6. The straight-line path is assumed as
the theoretical path we need. The end points of LIN (it is
the line path command of KUKA robot) command in
robot controller using offline programming is PA and
PB . However, due to the low absolute accuracy and low Figure 6. Straight-line path compensation strategy.
Shi et al. 5

stiffness of industrial robots, the robot tool centre point

(TCP) can’t execute the path exactly according to the
coordinates in robot controller, namely line PA -PB . While
executing path tasks, the position at each moment is
assumed to be PðtÞ, which can be measured using the laser
tracker. The perpendicular foot P 0 ðtÞ on the theoretical
line path PA -PB is considered as the theoretical position
at each moment. Because the theoretical position P0 ðtÞ is
on the straight line PA -PB , we can obtain equation (1).
Equation (2) describes the property of the perpendicular
foot P 0 ðtÞ, namely, PðtÞP 0 ðtÞ ⊥ PA PB .
x 0 ðtÞ  xA y 0 ðtÞ  yA z 0 ðtÞ  zA
¼ ¼ (1)
xB  xA yB  yA zB  zA

ðxB  xA Þðx 0 ðtÞ  xðtÞÞ þ ðyB  yA Þðy 0 ðtÞ  yðtÞÞ

(2)
þðzB  zA Þðz 0 ðtÞ  zðtÞÞ ¼ 0 Figure 7. Circular arc path compensation strategy.
By solving equations (1) and (2) simultaneously, the
theoretical position at each moment can be described as

8
>
> dx 2 xðtÞ þ dy 2 xA  dx dy ðyA  yðtÞÞ þ dz 2 xA  dx dz ðzA  zðtÞÞ
>
> x 0 ðtÞ ¼
>
> dx 2 þ dy 2 þ dz 2
>
>
< dy ðx0 ðtÞ  xA Þ
y 0 ðtÞ ¼ þ yA (3)
>
> d x
>
>
>
> dz ðx 0 ðtÞ  xA Þ
>
>
: z0 ðtÞ ¼ dx
þ zA

where, dx , dy and dz are the coordinate difference between where R is the radius of the circle. According to equations
two end points, namely (5) and (6), the coordinate values of the centre and the
8 radius of the circle can be described as
< dx ¼ xB  xA
2 3 2 31 2 3
d ¼ yB  yA (4) x0 A1 B1 C1 D1
: y 4 y 0 5 ¼ 4 A2 B 2 C2 5 4 D 2 5
dz ¼ z B  z A (7)
z0 A3 B3 C3 D3
Circular arc path compensation strategy. In the circular arc qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
path tasks, the start point, middle point and end point at R ¼ ðx 1  x 0 Þ 2 þ ðy 1  y 0 Þ 2 þ ðz 1  z 0 Þ 2 (8)
the CIRC (it is the Circular arc path command of KUKA
robot) program are, respectively, C 1 ðx 1 ; y 1 ; z 1 Þ, where, A1 -D1 , A2 -D 2 and A 3 -D 3 are a series of parameters
C2 ðx 2 ; y 2 ; z 2 Þ and C3 ðx 3 ; y 3 ; z 3 Þ (see Figure 7). We can cal- as follows
culate the coordinate values of the centre of the circle, 8
namely C0 ðx 0 ; y 0 ; z 0 Þ. First, the centre point is in the plane, >
> A1 ¼ ðy 2  y 1 Þðz 3  z 1 Þ  ðz 2  z 1 Þðy 3  y 1 Þ
>
> B1 ¼ ðz 2  z 1 Þðx 3  x 1 Þ  ðx 2  x 1 Þðz 3  z 1 Þ
which is composed of C1 , C 2 and C3 ; hence, the property is >
>
>
> C1 ¼ ðx 2  x 1 Þðy 3  y 1 Þ  ðy 2  y 1 Þðx 3  x 1 Þ
expressed as equation (5).Equation (6) describes the prop- >
>
>
>
erties of the circle. >
> D 1 ¼ A 1 x 1  B 1 y 1  C 1 z 1
>
>
  >
> A 2 ¼ 2ðx 2  x 1 Þ
 x0  x1 y0  y1 z0  z1  <
  B2 ¼ 2ðy 2  y 1 Þ
 x2  x1 y2  y1 z2  z1  ¼ 0 (9)
  (5) >
> C2 ¼ 2ðz 2  z 1 Þ
 x3  x1 y3  y1 z3  z1  >
> D2 ¼ x1 2 þ y1 2 þ z1 2  x2 2  y2 2  z2 2
>
>
>
> A3 ¼ 2ðx 3  x 1 Þ
8 >
>
>
>
< ðx 1  x 0 Þ 2 þ ðy 1  y 0 Þ 2 þ ðz 1  z 0 Þ 2 ¼ R 2 >
>
>
B3 ¼ 2ðy 3  y 1 Þ
ðx  x 0 Þ 2 þ ðy 2  y 0 Þ 2 þ ðz 2  z 0 Þ 2 ¼ R 2 >
>
: C3 ¼ 2ðz23  z 12Þ
(6) >
: 2
ðx 3  x 0 Þ 2 þ ðy 3  y 0 Þ 2 þ ðz 3  z 0 Þ 2 ¼ R 2 D3 ¼ x1 þ y1 þ z1 2  x3 2  y3 2  z3 2
6 International Journal of Advanced Robotic Systems

The ðA 1 ; B1 ; C 1 Þ, here, is the normal vector of the plane. The equation of the circle is derived from the aforementioned
equations. As shown in Figure 7, in the circular arc path tasks, the actual position at each moment is described as PðtÞ, and
Pp ðtÞ is the projection of PðtÞ in the plane composed of C1 , C 2 and C3 . Pp ðtÞ can be calculated from the following equations
xp ðtÞ  xðtÞ yp ðtÞ  yðtÞ zp ðtÞ  zðtÞ
¼ ¼ (10)
A1 B1 C1

A 1 ðxp ðtÞ  x1 Þ þ B 1 ðyp ðtÞ  y 1 Þ þ C 1 ðzp ðtÞ  z1 Þ ¼ 0 (11)

The Pp ðtÞ can be described as

2 3 2 31 2 3
xp ðtÞ A1 B1 C1 A1 x1 þ B1 y1 þ C1 z1
4 yp ðtÞ 5 ¼ 4 B 1 A 1 0 5 4 B 1 xðtÞ  A 1 yðtÞ 5 (12)
zp ðtÞ C1 0 A 1 C 1 xðtÞ  A 1 zðtÞ

As shown in Figure 7, in the circular arc path tasks, the intersection P0 ðtÞ of the straight line Pp ðtÞ-C0 and the planed
_
circular arc path C1 C2 C3 is assumed as the theoretical position at each moment, which is the closest position in the
theoretical path. The P 0 ðtÞ can be derived from the following equations

ðx0 ðtÞ  x 0 Þ2 þ ðy 0 ðtÞ  y 0 Þ2 þ ðz 0 ðtÞ  z 0 Þ 2 ¼ R 2 (13)

x 0 ðtÞ  x 0 y 0 ðtÞ  y 0 z 0 ðtÞ  z0

¼ ¼ (14)
xp ðtÞ  x 0 yp ðtÞ  y 0 zp ðtÞ  z 0

Hence, the theoretical position during the circular arc path tasks at each moment, namely P0 ðtÞ, can be described as
8
>
> R
>
> x0 ðtÞ ¼ x 0 + vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!2 ! 2ffi
>
> u
>
> u y ð tÞ  y z ð tÞ  z
>
> t1 þ p 0
þ
p 0
>
> x p ð tÞ  x0 x p ð tÞ  x0
< qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 
> ðx 0 ðtÞ  x 0 Þðyp ðtÞ  y0 Þ ðxp ðtÞ  x 0 ðtÞÞ2 þ ðyp ðtÞ  y0 ðtÞÞ2 þ ðzp ðtÞ  z 0 ðtÞÞ2 < R (15)
>
> y0 ðtÞ ¼ þ y0
>
> xp ðtÞ  x0
>
>
>
> ðx0 ðtÞ  x0 Þðzp ðtÞ  z 0 Þ
>
>
: z0 ðtÞ ¼
>
xp ðtÞ  x 0
þ z0

As the robot deviation is usually less than the radius of planned circular arc path tasks, hence the condition in the
bracket, here, makes the P0 ðtÞ in a position, which is closer intersection to the PðtÞ.

Path real-time compensation algorithm. From the aforemen- where n, o and a are, respectively, the normal vector, the
tioned discussion, errors at each moment can be obtained direction vector and the approach vector, which represent the
during both straight-line and circular arc path tasks. rotation relation between the robot base coordinate system and
Assume that the homogeneous transformation matrix laser tracker coordinate system. The vector p shows the trans-
between the robot base coordinate system and laser tracker lation relation. The relationship can be expressed as follows
system is tb T , as shown in Figure 2, which converts the robot t
coordinates to laser tracker coordinates. Here, the transfor- bT  P rob ¼ P tracker (17)
mation matrix is obtained according to the study of Liu where P rob and P tracker are, respectively, the position at the
et al.19 The transformation matrix tb T can be expressed as robot coordinate system and laser tracker system. There-
2 3 fore, the position at the laser tracker coordinate system can
nx ox ax px
6 ny oy ay py 7 be converted to robot base coordinate system through the
t 6 7
bT ¼ 4 n oz az pz 5
(16) following equation
z
0 0 0 1 P rob ¼tb T 1 P tracker (18)
Shi et al. 7

Figure 8. Robot control algorithm structure.

Hence, the errors at each moment during the path ongoing compensation. Thus, a proportion-integral-deriva-
tasks in the laser tracker coordinate system can be con- tive (PID) control algorithm is used to make the real-time
verted to the errors in the robot base coordinate system, compensation possible.
as follows First, the robot control algorithm structure (see Figure 8)
2 3 was proposed. Each moment in the process of continuous
Dx rob ðtÞ
trajectory tasks, the real-time measurement values of the
6 Dy ðtÞ 7
6 rob 7 t 1 laser tracker are compared with the theoretical coordinate
DP rob ðtÞ ¼ 6 7 ¼ T DP tracker ðtÞ
4 Dz rob ðtÞ 5 b values, and the compensation values are obtained by calcu-
0 lating. With the coordinate system transformation matrix,
2 3 (19) these values can be converted to the robot base coordinate
x 0 ðtÞ  xðtÞ
6 y ðtÞ  yðtÞ 7 system, which are sent to the KUKA robot every 12 ms.
6 0 7 The control algorithm is as follows
¼tb T 1 6 7
4 z 0 ðtÞ  zðtÞ 5
0
eðtÞ ¼ DP rob ðtÞ (20)
where Dx rob ðtÞ, Dy rob ðtÞ and Dz rob ðtÞ are, respectively, the X
real-time errors in the path tasks in X, Y and Z directions uðtÞ ¼ KP  eðtÞ þ KI  eðtÞ þ KD  ½eðtÞ  eðt  1Þ
with respect to the robot base coordinate system.
(21)
Because the robot executes a number of continuous
tasks, it is necessary to compensate the errors in the process where KP , KI and KD are, respectively, the proportion,
of performing continuous trajectory tasks. However, due to integral and differential coefficient of the PID controller;
the hysteresis of error compensation, if real-time error com- eðtÞ is the path error in the robot coordinate system and uðtÞ
pensation values are sent to robot directly via RSI, it will is the output value of the PID controller, namely the feed-
make the robot work with oscillation. It is because that the back value to robot controller.
compensation values sent to robots will take about 500 ms Substituting equations (3) and (15) into equation (21),
to complete the compensation process, and the errors mea- the compensation value at each moment is obtained as
sured next will accumulate to the former unfinished follows

2 3 2 3
x 0 ðtÞ  xðtÞ x 0 ðtÞ  xðtÞ
6 7 X6 7
6 y ðtÞ
 yðtÞ 7
t 1 6 0
6 y 0 ðtÞ  yðtÞ 7
uðtÞ ¼ KP  7 t 1
þ KI  bT 6 7
bT 6 z ðtÞ
 zðtÞ 7
4 0 5
6 z ðtÞ  zðtÞ 7
4 0 5
0 0
2 3 (22)
x 0 ðtÞ  x 0 ðt  1Þ  xðtÞ þ xðt  1Þ
6 7
6 y ðtÞ  y 0 ðt  1Þ  yðtÞ þ yðt  1Þ 7
t 1 6 0
¼ KD  bT 6 7
7
4 z 0 ðtÞ  z 0 ðt  1Þ  zðtÞ þ zðt  1Þ 5
0
8 International Journal of Advanced Robotic Systems

Figure 9. Results of real-time error compensation.

In order to compare the results of single error compen-

sation, we also set 0.5 mm, 0.7 mm and 1.0 mm compen-
sation values, respectively, in the directions of X, Y and Z
with respect to the laser tracker coordinate system. The
results of real-time compensation are shown in Figure 9.
As shown in Figure 9, different control coefficients
result in different effects. Here, a simple proportional
controller is utilized. The results show that when the pro-
Target base
portional coefficient KP is set to 0.05, the oscillation is
small enough and the compensation time is short enough SMR Robot
as well.
In the path of the offline programming, the usage of the Laser tracker Mental plate
compensation algorithm mentioned above at each moment
during the path tasks can achieve very good accuracy.
Experimental design and validation will be stated in the
‘Experiments’ section.

Experiments
From what has been discussed earlier, the error compensa-
tion performance of the KUKA KR5 arc industrial robot is Figure 10. The set-up of experiment.
Shi et al. 9

known. The following discussion focuses on the error com-

pensation on the process of the path tasks.
As we all know, the robots’ absolute positioning accu-
racy is low. So, when robots execute trajectory tasks, they
cannot perform accurately in accordance with the coordi-
nates in the offline program, hence, resulting in deviation.
In this section, the experimental results will be analysed
by comparing the online real-time compensation method
with the process without compensation. A KUKA KR5 arc
Industrial Robot and FARO XI Laser Tracker are used for
experimental validation. A laser tracker target base with
magnetic force is fixed at a metal plate, and the metal plate
is installed at the robot flange. The laser tracker SMR target
is mounted at this base by magnetic force. The absolute
distance measurement accuracy of the laser tracker is
20 mm þ 1.1 mm/m. The measurement frequency here is set
Figure 11. Comparison of positional error between online to 100 Hz, that is, we can obtain 100 data per second. The
compensation and offline calibration. set-up of experiment is shown in Figure 10.

Figure 12. Results of straight-line path real-time compensation without load.

10 International Journal of Advanced Robotic Systems

Comparison experiment of online compensation

and offline calibration
In order to compare with the accuracy of online compen-
sation approach proposed in this article and the offline
calibration approach, the online compensation and offline
calibration experiment are executed, respectively. The off-
line calibration method, here, refers to the study of Liu
et al.20 Fifty-one points in the robot workspace are selected
to calibrate the robot. For comparison, the selected 51
points are also executed according to the online compen-
sation. The results are illustrated in Figure 11.
Without compensation, the maximum positional error is
about 1.783 mm and the error can be reduced to 0.600 mm
after offline calibration. After online compensation method
proposed in this system, the maximum positional error is
Figure 13. Straightness errors of line without compensation. 0.016 mm, which meets the high accuracy requirements of

Figure 14. Results of straight-line path real-time compensation with load.

Shi et al. 11

Figure 15. Results of circular arc path real-time compensation without load.

the fields. Hence, the accuracy obtained in the system is Here, we draw a straight line whose start and end
better than offline calibration. points are the points robot actually executes and we
define the distance errors deviated from this line as the
straightness errors. As shown in Figure 13, without
Experiment of robot executing LIN command
compensation, the straightness error of the straight-
without load line path is relatively good. However, the maximum
When KUKA robot executes LIN commands, straight-line straightness error also reaches 0.206 mm at a length
paths are generated. The coordinate values of the robot TCP of about 210 mm.
in the robot program are executed according to the robot Besides, as shown in Figure 12, the maximum errors in
kinematic model, so there are errors between the actual path the X, Y and Z directions are 0.337 mm, 0.562 mm and
that the robot executes and the path we actually planned. In 0.404 mm, respectively, and the distance error is 0.758
this section, the KUKA robot executes a LIN command to mm. The error of Y direction is larger than other directions
perform straight-line path and to validate the accuracy of the in the planned path, and such low accuracy limit robots are
compensation method. The start and end points of the planned to be used widely in precise fields. During the compensa-
straight-line path in the external measuring coordinate system tion stage, the errors are extremely declined. After compen-
(laser tracker) are PA (2256.000, 122.000, 40.000) and sation, the maximum errors in the X, Y and Z directions are
PB (2444.000, 201.000, 90.000). The results are illustrated 0.042 mm, 0.061 mm and 0.087 mm and the distance error
in Figures 12 and 13. is less than 0.107 mm.
12 International Journal of Advanced Robotic Systems

Figure 16. Results of circular arc path real-time compensation with load.

In the study of Tao et al.,21 an online preprocessing robot. When robots are used to assemble parts, the weight
implementation to calibrate robots is proposed. The accuracy of different parts is acted on them. Therefore, robot load is
of the maximum position error is 0.974 mm. The accuracy in a very important factor in the use of robots. The following
our system is higher, whose maximum position error is 0.107 experiments are designed to verify the accuracy of this
mm during path tasks. Moreover, the approach here is system while executing path in the case of robot with load.
obtained in the robot path tasks; that is, the accuracy is The rated payload of KUKA KR5 arc robot is 5 kg. We
obtained in the continuous path process, rather than in the install 4 kg load at the end of the robot and the robot
discrete process. Such accuracy is more valuable in the perform LIN command path tasks. The results are shown
actual use of industrial robots. in Figure 14.
Without compensation, mainly due to the low stiffness
of the robot and the robot’s mechanical errors, the accuracy
Experiment of robot executing LIN command
of path tasks under load condition is relatively low. The
with load trend of errors is similar to the condition without load and
The aforementioned experiments are carried out under the there is a similar linear increase. The maximum errors in X,
condition of no load. Industrial robots, however, are often Y and Z directions, as shown in Figure 14, are 0.541 mm,
used as a carrier for other tools. For example, when we use 0.803 mm and 0.863 mm, respectively, and the distance
robots to process parts, the machining tool is installed at the error is 1.251 mm, which are much larger than the case
end of the robots. The machining force is acted on the of no load. After compensation, the accuracy of path is
Shi et al. 13

improved with maximum errors of 0.046 mm, 0.061 mm Conclusions

and 0.092 mm in X, Y and Z directions, respectively. The
This article presents an online real-time compensation
maximum distance error is reduced to 0.110 mm. The accu-
approach for industrial robots to execute precise path tasks.
racy after compensation is quite close to the case without
The laser tracker was used to measure the position each
load. Hence, the performance of the proposed system above
time, and the straight-line and circular arc path compensa-
is good enough in the case of robot performing straight-line
tion strategies were proposed and the PID control algorithm
path with load.
was used to calculate compensation values that the robot
can compensate for every 12 ms. The oscillation of real-
time compensation is solved through the PID control. The
Experiment of robot executing CIRC command method didn’t need accurate offline calibration and can
without load reach relatively high accuracy. Besides, this system also
reaches high accuracy when the robot loads weights. Con-
In this section, the robot runs CIRC command to perform a
sequently, the approach didn’t need robot stiffness identi-
circular arc task. The planned circular arc path in the laser
fication when considering the load.
tracker coordinate system is calculated from C1 (2256.000,
In further work, the compensation method will be tested
122.000, 40.000), C2 (2306.000, 72.000, 70.000) and C3
in more complex conditions, and the attitude errors should
(2356.000, 130.000, 100.000) and the centre can be cal-
be considered and compensated.
culated according to equation (7), which is (2305.728,
130.630, 69.837). The results are illustrated in Figure 15.
Without the real-time compensation strategy, the errors Declaration of conflicting interests
are large. The errors without compensation vary more tre- The author(s) declared no potential conflicts of interest with
mendously compared with the straight-line path. The errors respect to the research, authorship, and/or publication of this
in X and Y directions are larger than Z direction. The max- article.
imum errors in X, Y and Z directions are 0.530 mm, 0.393
mm and 0.281 mm, respectively, and the distance maxi- Funding
mum error is 0.575 mm. The author(s) disclosed receipt of the following financial support
The compensation values in circular arc path at each for the research, authorship, and/or publication of this article: This
moment can be calculated according to equations (15) and work is supported by the National Natural Science Foundation of
(22). With the path compensation algorithm, the accuracy China (no. 51275350) and the Marine Science and Technology
of the circular arc path is improved greatly. The errors can Project from the Tianjin Marine Bureau (no. KJXH2014-08).
be reduced to 0.114 mm, 0.121 mm, 0.163 mm and 0.200
mm. Furthermore, the mean errors decrease significantly as References
well. From Figure 15, we can also see that the compensa- 1. Schneider U, Diaz Posada JR, Drust M, et al. Combining
tion errors are larger when the uncompensated path errors holistic programming with kinematic parameter optimisa-
change significantly. The spikes of errors in both uncom- tion for robot machining. In: Schraft RD and Berns K
pensated and compensated conditions in Figure 15 (eds) ISR/Robotik 2014; Proceedings of 41st International
explain that. Symposium on Robotics, Munich, Germany, 2–3 June
2014, pp. 1–7. Munich, Germany: VDE.
2. Pan Z, Zhang H, Zhu Z, et al. Chatter analysis of robotic
machining process. J Mater Process Technol 173(3):
Experiment of robot executing CIRC command
301–309.
with load 3. Yap HJ and Taha Z, Dawal SZM, et al. (2014) Virtual reality
In this section, robot performs the circular arc command based support system for layout planning and programming
mentioned earlier and the real-time path compensation is of an industrial robotic work cell. PloS One 2006; 9(10):
executed with a load of 4 kg. The results are illustrated in e109692.
Figure 16. As the straight-line path tasks, the error during 4. Liu Z, Bu W and Tan J. Motion navigation for arc welding
circular arc path tasks is also larger in the condition of with robots based on feature mapping in a simulation environment.
load than without load, and there is a similar linear Robot Com-Int Manuf 2010; 26(2): 137–144.
increase. The low stiffness is responsible for that. After 5. Nguyen HN, Zhou J and Kang HJ. A calibration method for
compensation, the errors can also be eliminated dramati- enhancing robot accuracy through integration of an extended
cally. The maximum errors in the directions of X, Y and Z Kalman filter algorithm and an artificial neural network. Neu-
are 0.099 mm, 0.141 mm and 0.162 mm, respectively, and rocomputing 2015; 151: 996–1005.
the distance error is reduced to 0.218 mm. Hence, the errors 6. Wu Y, Klimchik A, Caro S, et al. Geometric calibration of
after online real-time compensation are similar in both industrial robots using enhanced partial pose measurements
conditions, which means the system can achieve high accu- and design of experiments. Robot Com-Int Manuf 2015; 35:
racy with load during circular arc path tasks. 151–168.
14 International Journal of Advanced Robotic Systems

7. Denavit J. A kinematic notation for lower-pair mechanisms based 14. Zi B, Sun H, Zhu Z, et al. The dynamics and sliding mode
on matrices. Trans ASME. J Appl Mech 1955; 22: 215–221. control of multiple cooperative welding robot manipulators.
8. Abele E, Weigold M and RothenbÃijcher S. Modeling and Int J Adv Robot Syst 2012; 9: 53.
identification of an industrial robot for machining applica- 15. Zhao YM, Lin Y, Xi F, et al. Calibration-based iterative
tions. CIRP Ann Manuf Techn 2007; 56(1): 387–390. learning control for path tracking of industrial robots. IEEE
9. Zhang H, Wang J, Zhang G, et al. Machining with flexible Trans Ind Electron 2015; 62(5): 2921–2929.
manipulator: toward improving robotic machining perfor- 16. Wang C, Zhao Y, Chen Y, et al. Nonparametric statistical
mance. In: Lee KM (eds) Proceedings, 2005 IEEE/ASME learning control of robot manipulators for trajectory or con-
International Conference on Advanced Intelligent Mechatro- tour tracking. Robot Com-Int Manuf 2015; 35: 96–103.
nics, Monterey, CA, 24–28 July 2005, pp. 1127–1132. New 17. Qian S, Zi B, Zhang D, et al. Kinematics and error analysis of
York, NY: IEEE. cooperative cable parallel manipulators for multiple mobile
10. Guillo M and Dubourg L. Impact & improvement of tool cranes. Int J Mech Mater Design 2014; 10(4): 395–409.
deviation in friction stir welding: weld quality & real-time 18. KUKA Roboter GmbH, KUKA. RobotSensorInterface 2.3,
compensation on an industrial robot. Robot Com-Int Manuf Version: KST RSI 2.3 V1 en, user manual. Augsburg, Ger-
2016; 39: 22–31. many: KUKA Roboter GmbH, KUKA, 2009.
11. Chiba R, Arai T, Ueyama T, et al. Working environment 19. Liu B, Zhang F, Qu X, et al. A Rapid coordinate trans-
design for effective palletizing with a 6-DOF manipulator. formation method applied in industrial robot calibration
Int J Adv Robot Syst 2016; 13: 1–8. based on characteristic line coincidence. Sensors 2016;
12. Roebrock P. A flexible multi-sensor positioning system for 16(2): 239.
industrial robots. WSEAS Transactions on Electronics 2008; 20. Liu B, Zhang F and Qu X. A method for improving the pose
3: 93–100. accuracy of a robot manipulator based on multi-sensor com-
13. Roebrock P and BÃ}uhnke K. Offline path correction system bined measurement and data fusion. Sensors 2015; 15(4):
for industrial robots. In: Demiralp M and Mastorskis N (eds) 7933–7952.
Proceedings of the 9th WSEAS International Conference on 21. Tao PY, Mustafa SK, Yang G, et al. Robot work cell
Automatic Control, Modeling & Simulation, Stevens Point, calibration and error compensation. In: Nee AYC (ed)
WI, pp. 276–280. Athens, Greece: World Scientific and Engi- Handbook of Manufacturing Engineering and Technology.
neering Acad and Soc. London: Springer 2015, pp. 1995–2034.