Jacobian: Velocities and Static Forces

Kinematics Relations - Joint & Cartesian Spaces

• A robot is often used to manipulate object attached to its tip (end effector).

• The location of the robot tip may be specified using one of the following
descriptions:

• Joint Space  θ1  {N}

θ 
θ = 2
M
 
θ N 
• Cartesian Space

[ ]
 0 PN 
X = 0 
[ ]
 rN 
Euler Angles
 Kinematics Relations - Forward & Inverse

• The robot kinematic equations relate the two description of the robot tip location

X = FK (θ )
 θ1 
θ 
θ = 2 [ ]
 0 PN 
X = 0 
M
 
[ ]
 rN 
θ N 

θ = IK ( X )
Tip Location in Tip Location in
Joint Space Cartesian Space
 Kinematics Relations - Forward & Inverse

 vx 
 θ&1  v 
&   y
d θ
θ& = [θ ] =  2 
d [ ]
 v  v 
X& = [ X ] =  N  =  z 
dt M
& 
dt [ ]
 ω N  ω x 
θ N  ω y 
 
ω z 

Tip Velocity in Tip velocity in

Joint Space Cartesian Space
 Jacobian Matrix - Introduction

• The Jacobian is a multi dimensional form of the derivative.

• Suppose that for example we have 6 functions, each of which is a function of 6
independent variables

y1 = f1 ( x1 , x2 , x3 , x4 , x5 , x6 )
y2 = f 2 ( x1 , x2 , x3 , x4 , x5 , x6 )
M
y6 = f 6 ( x1 , x2 , x3 , x4 , x5 , x6 )

• We may also use a vector notation to write these equations as

Y = F(X )
 Jacobian Matrix - Introduction

• If we wish to calculate the differential of yi as a function of the differential xi

we use the chain rule to get

∂f1 ∂f ∂f
δy1 = δx1 + 1 δx2 + K + 1 δx6
∂x1 ∂x2 ∂x6
∂f 2 ∂f 2 ∂f 2
δy 2 = δx1 + δ x2 + K + δx6
∂x1 ∂x2 ∂x6
M
∂f 6 ∂f ∂f
δy 6 = δx1 + 6 δx2 + K + 6 δx6
∂x1 ∂x2 ∂x6

• Which again might be written more simply using a vector notation as

∂F
δY = δX
∂X
 Jacobian Matrix - Introduction

• The 6x6 matrix of partial derivative is defined as the Jacobian matrix

∂F
δY = δX = J ( X )δX
∂X

• By dividing both sides by the differential time element, we can think of the
Jacobian as mapping velocities in X to those in Y

Y& = J ( X ) X&

• Note that the Jacobian is time varying linear transformation

 Jacobian Matrix - Introduction

• In the field of robotics the Jacobian

matrix describe the relationship
between the joint angle rates ( θ& N )
and the translation and rotation
velocities of the end effector ( x& ).
This relationship is given by:

x& = J (θ )θ&

θ& = J (θ ) x&
−1
 Jacobian Matrix - Introduction

• This expression can be expanded to:

 x&     θ&1 
 y&    & 
    θ 2 
 z&   J (θ )  
 =  
ω x   L
ω y    
    & 
ω z    θ N 
6x1 6xN Nx1
• Where:
– x& is a 6x1 vector of the end effector linear and angular velocities
– J (θ ) is a 6xN Jacobian matrix
– θ& N is a Nx1 vector of the manipulator joint velocities
– N is the number of joints
 Jacobian Matrix - Introduction

• In addition to the velocity

relationship, we are also interested in
developing a relationship between
the robot joint torques ( τ ) and the
F
forces and moments ( F ) at the
robot end effector (Static
Conditions). This relationship is
given by:

τ = J (θ ) F
T
 Jacobian Matrix - Introduction

• This expression can be expanded to:

T
τ1     Fx 
τ    F 
 2    y
   J (θ )   Fz 
 =   
L    M x 
    M y 
     
τ N    M z 
6x1 6xN Nx1
• Where:
– τ is a 6x1 vector of the robot joint torques
– J (θ ) is a 6xN Transposed Jacobian matrix
T

– F is a Nx1 vector of the forces and moments at the robot end effector
– N is the number of joints
 Jacobian Matrix - Calculation Methods

Differentiation the
Iterative
Forward Kinematics Eqs.

Jacobian Matrix
 Jacobian Matrix by Differentiation - 1R - 1/4

• Consider a simple planar 1R robot

y

Py Fee ,Vee
r
θ ,θ&
x
Px

• The end effector position is given by

0
Px = x = r cosθ
0
Py = y = r sin θ
 Jacobian Matrix by Differentiation - 1R - 2/4

• The velocity of the end effector is defined by

0
Vx = 0P&x = x& = −θ& r sin θ = −ω r sin θ
0
V y = 0P&y = y& = θ& r cosθ = ω r cosθ

• Expressed in matrix form we have

x& = J (θ )θ&

 x&   − r sin θ  &

[]
 y&  =  r cosθ  θ
   
2x1 2x1 1x1
 Jacobian Matrix by Differentiation - 1R - 3/4

Py Fee ,Vee
r
θ ,θ&
x
Px

• The moment about the joint generated by the force acting on the end effector is
given by

τ = − rFx sin θ + rFy cos θ

 Jacobian Matrix by Differentiation - 1R - 4/4

• Expressed in matrix form we have

τ = J (θ ) F
T

 Fx 
[τ ] = [− r sin θ r cos θ ] 
 Fy 
 Jacobian Matrix by Differanciation - 3R - 1/4

• Consider the following 3 DOF Planar manipulator

y0
α

y x3

x2
y3
y2
y1
x1

x x0
 Jacobian Matrix by Differanciation - 3R - 2/4

• Problem: Compute the Jacobian matrix that describes the relationship

x& = J (θ )θ& τ = J (θ ) F
T

• Solution:

• The end effector position and orientation is defined in the base frame by

x
x =  y
 
α 
 Jacobian Matrix by Differanciation - 3R - 3/4

• The forward kinematics gives us relationship of the end effector to the joint
angles:
0
P3 org , x = x = L1c1 + L2c12 + L3c123
0
P3 org , y = y = L1s1 + L2 s12 + L3s123
0
P3 org , α = α = θ1 + θ 2 + θ 3
• Differentiating the three expressions gives

( )
x& = − L1s1θ&1 − L2 s12 θ&1 + θ&2 − L3s123 θ&1 + θ&2 + θ&3 ( )
= −(L1s1 + L2 s12 + L3s123 )θ&1 − (L2 s12 + L3s123 )θ&2 − (L3s123 )θ&3
2 12 ( 1 2 )
y& = L c θ& + L c θ& + θ& + L c θ& + θ& + θ&
1 1 1 3 123 ( 1 2 3 )
= (L1c1 + L2 c12 + L3c123 )θ&1 + (L2 c12 + L3c123 )θ&2 + (L3c123 )θ&3
α& = θ&1 + θ&2 + θ&3
 Jacobian Matrix by Differanciation - 3R - 4/4

• Using a matrix form we get

x& = 0J (θ )θ&

 x&  − L1s1 − L2 s12 − L3s123 − L2 s12 − L3s123 − L3s123  θ&1 

 y&  =  L c + L c + L c  
L2c12 + L3c123 L3c123  θ&2 
   11 2 12 3 123 
α&   1 1 1  θ&3 

• The Jacobian provides a linear transformation, giving a velocity map and a force
map for a robot manipulator. For the simple example above, the equations are
trivial, but can easily become more complicated with robots that have additional
degrees a freedom. Before tackling these problems, consider this brief review of
linear algebra.
 Singularity - The Concept

• Motivation: We would like the hand of a robot (end effecror) to move with a
certain velocity vector in Cartesian space. Using linear transformation relating
the joint velocity to the Cartesian velocity we could calculate the necessary joint
rates at each instance along the path.

θ& = J (θ ) x&
−1

• Given: a linear transformation relating the joint velocity to the Cartesian velocity
(usually the end effector)

• Question: Is the Jacobian matrix invertable? (Or) Is it nonsingular?

Is the Jacobian invertable for all values of θ ?
If not, where is it not invertable?
 Singularity - The Concept

• Answer (Conceptual): Most manipulator have values of θ where the Jacobian

becomes singular . Such locations are called singularities of the mechanism
or singularities for short

Singularities of the mechanism

Workspace boundary singularities Workspace interior Singularities

- Stretched out End Workspace

- Two or more joints are lining up
- Folded back Effector Boundary
 Singularity - The Concept

Manipulator

Singular General
Configuration Configuration

Losing All DOF

One or More DOF Are Available

• Losing one or more DOF means that there is a some direction (or subspace) in
Cartesian space along which it is impossible to move the hand of the robot (end
effector) no matter which joint rate are selected
 Brief Linear Algebra Review - 1/

• Inverse of Matrix A exists if and only if the determinant of A is non-zero.

A−1 Exists if and only if

Det ( A) = A ≠ 0

• If the determinant of A is equal to zero, then the matrix A is a singular matrix

Det ( A) = A = 0

A Singular
 Brief Linear Algebra Review - 2/

• The rank of the matrix A is the size of the largest squared Matrix S for which

Det ( S ) ≠ 0

 2 − 1  2 − 1
• Example 1 - A=  A=S =  A = S =3 Rank ( A) = 2
− 1 2  − 1 2 

1 1 1 1
1 1 1 1
• Example 2 - A= S = [1] S =1 Rank ( A) = 1
1 1 1 1
 
1 1 1 1
 Brief Linear Algebra Review - 3/

• If two rows or columns of matrix A are equal or related by a constant, then

Det ( A) = 0

• Example

 2 0 − 1
A =  6 − 3 − 3
10 − 6 − 5

−3 −3 6 −3 6 −3
det( A) = A = 2 −0 −1 = 6+0−6 = 0
−6 −5 10 − 5 10 − 6
 Brief Linear Algebra Review - 4/

• Eigenvalues

AX = λX

( A − λI ) X = 0

• Eigenvalues are the roots of the polynomial

Det ( A − λI )

• If X ≠ 0 each solution to the characteristic equation λ (Eigenvalue) has a

corresponding Eigenvector
 Brief Linear Algebra Review - 4/

1 2
A= 
2 1
1 − λ 2  X1 
( A − λI ) X =   X  = 0
 2 1 − λ  2 

1− λ 2
Det ( A − λI ) = = λ2 − 2λ − 3 = 0
2 1− λ

λ1 = −1
λ2 = 3
 Brief Linear Algebra Review - 4/

λ1 = −1

 2 2  X 1  − 1
 2 2  X  = 0 X = 
  2  1

λ2 = 3

− 2 2   X 1  1
X = 
 2 − 2  X  = 0 1
  2 
 Brief Linear Algebra Review - 5/

• Any singular matrix ( Det ( A) = 0 ) has at least one Eigenvalue equal to zero
 Brief Linear Algebra Review - 6/

• If A is non-singular ( Det ( A) ≠ 0 ), and λ is an eigenvalue of A with

corresponding to eigenvector X, then

A−1 X = λ−1 X
 Brief Linear Algebra Review - 7/

• If the n x n matrix A is of full rank (that is, Rank (A) = n), then the only
solution to
AX = 0

is the trivial one

X =0

• If A is of less than full rank (that is Rank (A) < n), then there are n-r linearly
independent (orthogonal) solutions

xj− 0 ≤ j ≤ n- r
for which

Ax j = 0
 Brief Linear Algebra Review - 8/

• If A is square, then A and AT have the same eigenvalues

 Properties of the Jacobian -
Velocity Mapping and Singularities

• Example: Planar 3R

− L1s1 − L2 s12 − L3s123 − L2 s12 − L3s123 − L3s123

det( J (θ )) = L1c1 + L2c12 + L3c123 L2c12 + L3c123 L3c123 = L1L2 s2
1 1 1

det( J (θ )) = L1L2 s2 = 0

• Note that det( J (θ )) is not a function of θ1 ,θ 3

 Properties of the Jacobian -
Velocity Mapping and Singularities

 θ 2 = 0 Stretched Out
singular configuration 
θ 2 = π Fold Back

• The manipulator loses 1 DOF. The end effector can only move along the tangent
direction of the arm. Motion along the radial direction is not possible.
 Properties of the Jacobian -
Force Mapping and Singularities

• The relationship between joint torque and end effector force and moments is
given by:

τ = J (θ ) F
T

The rank of J (θ ) is equals the rank of J (θ ) .

T
•

• At a singular configuration there exists a non trivial force F such that

J (θ ) F = 0
T

• In other words, a finite force can be applied to the end effector that produces no
torque at the robot’s joints. In the singular configuration, the manipulator can
“lock up.”
 Properties of the Jacobian -
Force Mapping and Singularities

• Example: Planar 3R θ1 = θ ; θ 2 = θ 3 = 0 F

θ1

• In this case the force acting on the end effector (relative to the {0} frame) is
given by

 Fc1 
0
F =  Fs1 
 
 0 
 Properties of the Jacobian -
Force Mapping and Singularities

− L1s1 − L2 s12 − L3s123 L1c1 + L2c12 + L3c123 1  Fc1 

τ = 0J (θ ) 0 F =  − L2 s12 − L3s123 1  Fs1 
T
0
L2c12 + L3c123
  
 − L3s123 L3c123 1  0 

• For θ1 = θ ; θ 2 = θ 3 = 0 we get

− L1s1 − L2 s1 − L3s1 L1c1 + L2c1 + L3c1 1  Fc1 

τ = 0J (θ ) 0 F =  − L2 s1 − L3s1 1  Fs1 
T
0
L2c1 + L3c1
  
 − L3s1 L3c1 1  0 
 − Fs1c1 ( L1 + L2 + L3 ) + Fs1c1 ( L1 + L2 + L3 ) 0
 − Fs1c1 ( L2 + L3 ) + Fs1c1 ( L2 + L3 )  =  0
   
 − Fs1c1 ( L3 ) + Fs1c1 ( L3 )  0
 Properties of the Jacobian -
Force Mapping and Singularities

• This situation is an old and famous one in mechanical engineering.

• For example, in the steam locomotive, “top dead center” refers to the following
condition

• The piston force, F, cannot generate any torque around the drive wheel axis
because the linkage is singular in the position shown.
 Properties of the Jacobian -
Velocity Mapping and Singularities

• We have shown the relationship between joint space velocity and end effector
velocity, given by

x& = J (θ )θ&

• It is interesting to determine the inverse of this relationship, namely

θ& = J (θ ) x&
−1
 Properties of the Jacobian -
Velocity Mapping and Singularities

• Consider the square 6x6 case for J (θ ) .

• If rank < 6 ( Det ( J (θ )) = 0 ) , then there is no solution to the inverse

equation (see Brief Linear Algebra Review - 1,7).

Rank (J (θ )) < 6
θ& = J (θ ) x&
−1

• However, if the rank = 5, then there is at least one non-trivial solution to the
forward equation (see Brief Linear Algebra Review - 7). That is, for

x& = J (θ )θ& = 0
 Properties of the Jacobian -
Velocity Mapping and Singularities

• The solution is a direction (θ ) in the in joint velocity space for which joint
motion produces no end effector motion.

• We call any joint configuration θ = Q for which

Rank (J (θ )) < 6
a singular configuration.
 Properties of the Jacobian -
Velocity Mapping and Singularities

• For certain directions of end effector motion , x&i 1≤ i ≤ 6

x& = J (θ )θ& = λi (θ )ω i

where:
– λi are the eigenvalues of J (θ )
– ω i are the eigenvectors of J (θ )

• If J (θ ) is fully ranked (see Brief Linear Algebra Review - 6/ ), we have

ωi = J (θ ) x& = λi (θ ) x&
−1 −1
 Properties of the Jacobian -
Velocity Mapping and Singularities

• As the joint approach a singular configuration θ = Q there is at least one

eigenvalue for which λi → 0. This results in

x& x&
ωi = → →∞
λi (θ ) 0

• In other word, as the joints approach the singular configuration, the end effector
motion in a particular task direction x& jcauses the joint velocities to approach
infinity. However, there are task velocities that can have solutions.

• If J (θ ) loses rank by only one, then there are n-1 eigenvectors in the task
velocity space ( x& j ) for which solutions do exist. However, there can be multiple
solutions.