Spatial Descriptions and

Transformation
(Chapter 2)

Yunhui Liu
Dept. of Mechanical and Automation Engineering
The Chinese University of Hong Kong
 Introduction
• The configuration of a robot manipulator is determined
by its joint variables such as joint angles or joint
distances.
• Task specification is usually given in a Cartesian space,
i.e. the desired position and orientation of the end-
effector are given in a Cartesian coordinates frame.
• Problem: How do we relate the joint variables to the
position and orientation of the end-effector?

2 3
Object Position and
1 Oreintation:
 
( x, φ )

Cartesian Coordinate Frame

 Forward Kinematics
• Forward kinematics:
 Given a set of joint
variables  , determine the position and
orientation x of the end-effector with respect to a
task (world, inertia) coordinate frame.

x = f (q )
Kinematics equation


 Forward Kinematics
x
 Inverse Kinematics
• Definition: Given a position and orientation of
the end-effector, find the corresponding joint
variables of the robot manipulator.
 1 
f ( x)
Inverse kinematics equation

 
 Inverse kinematics
x
 Questions

• How to represent a position?

• How to represent an orientation?
• How to define joint variables?
 Representation of Position
• How to represent the position of a point?
– Define a coordinate frame
– Use a vector of coordinates with respect to
the coordinate frame to represent the position
• What is a coordinate frame?
• How to define a coordinate frame?
P

x
 Coordinate Frame
• Definition: A set of n linearly independent
vectors in R n is a coordinate system (frame) in .
n
R
x2 x
3 x4

x1

Which two axes form a coordinate frame?

 Orthonormal Coordinate Frame
• A complete orthonormal set of linearly
independent vectors { x1, x2 ,..., xn } forms an
orthonormal coordinate frame in R n .
- Orthonorma l :
T 
xi x j  0 i j
T 
xi xi  1

- Completene ss :
T  
if y xi  0 i, then y  0
 Representation of a Coordinate Frame
z
y1
y2
x2
C


45  x1 y
A


B
x

x2 y2 x y z
x1 y1 ì 2 ü
2 ì1 0 0ü
ï - ï
ì1 0ü ï ï
 í ý  B
ï 2
í
ï 2
2 ï
ý
ï

C í0 1 0ý
ï0 1ïþ
A 2
î0 1þ ïî 2 2 ïþ
î 0

Rotation Matrix
 Rotation Matrix
• Rotation matrix represents the orientation of the
coordinate frame.
• By listing the directional vectors of the axes as the
column vectors, we obtain the rotation matrix of
the coordinate frame.

R  x, y, z
• The rotation matrix R is an orthogonal matrix, i.e.

1
R R
T
 Coordinates
• What are the coordinates of a point with
respect to a coordinate frame?
The coordinates of the point w.r. t.  A yA
yB
xB
(1,2) 2
45  xA
The coordinates w.r.t.  B ? 1

 2  2
    
B
p1  (1 2)  2 , B
p2  (1 2)  2 
 2  2 
   
 2   2 
 Coordinates
• The i-th coordinate of a vector p w.r.t an
orthonormal coordinate frame  :
S
pi = pT xi Unit vector of the axis

Example: The coordinates of Vector

a point w.r.t.  A are (1,2). The yA

coordinates w.r.t.  B yB
xB


45  xA
2  2
    
B
p1  (1 2)  2 , B
p2  (1 2)  2 
 2  2 
   
 2   2 
 Coordinate Transform
• Given two coordinate frames with the same origin
point: y2
y1
x2
 A : O  x1 y1 z1 , {I } 45 
o x1
 B : O  x2 y2 z2 , A
{ x1 A
x2 A
x3 }

transform the coordinates of a point w.r.t. one frame to

those w.r.t. another frame.
æ3 2 ö
ç ÷ A
æ1 ö
B
p= ç 2 ÷
? p =ç ÷
ç
ç 2
÷
÷ è 2ø
è 2 ø
 Coordinate Transform
• The rotation matrix Coordinate Transform
of frame B w.r.t.
 2   
frame A is   2 
3 3 2   2  2   1 
 2 2  2  2  2  2   2
     
A
RB   2 2   2   2 
 2 2   2 2  3 3 
   
 2 2   2 2  2   1 
 2 2  2   2 
  
yA  2 2  2 
yB
xB

45  A B A
o xA RB p = p
Inverse Coordinate Transform
A B A
RB p = p

B 
p R
A
B
1 A  A T A B A
p RB p RA p

B
RA  R
A
B
1
 RB
A T
 Relation between 3 Frames
A
RB : the rotation matrix of frame B w.r.t. frame A
B
RC : the rotation matrix of frame C w.r.t. frame B
A
RC : the rotation matrix of frame C w.r.t. frame A
The direction of the x-axis of frame C w.r.t. zA
frame A can be calculated as follows zC yB
A A B
xC = R B xC
yC
zB

yA

A
RC = A R B éë B xC B
yC B
zC ùû = A R B B RC
xA xB
xC
 Representation of Position and
Orientation of a Rigid Body
• Set up two frames zo

A fixed reference frame: z yo

W : O  xyz C

An object frame fixed on the object: object

xo
 O : C  xo y o z o O
y

The position and orientation of the moving frame can represent

the position and orientation of the object
Position: vector OC
Orientation: rotation matrix of the object frame w.r.t.
the reference frame
 Fundamental Rotation
• Rotation about a single axis is a fundamental
rotation

z z
 
 y y


x
Rotation about x Rotation about y Rotation about z
 Rotation about x
• Rotating a frame about its x-axis by an angle leads
to a new frame. The rotation matrix of the new
frame w.r.t. the original one is
z


1 0 0 
 
Rot x ( )   0 c  s  y

 0 s c 

x
Rotation about x
c  cos , s  sin
 Rotation about y
• Rotating a frame about its y-axis by an angle
leads to a new frame. The rotation matrix of
the new frame w.r.t. the original one is


 c 0 s 
Rot y ( )   0 0 
y
1
 s c 
  0
Rotation about y
 Rotation about z
• Rotating a frame about its z-axis by an angle
leads to a new frame. The rotation matrix of
the new frame w.r.t. the original one is
z

 c  s 0

Rot z ( )   s c 0 
0 1 
 0

Rotation about z
 Composite rotation
• Rotate frame A about zA by angle  => frame B
• Then rotate frame B about yB by angle  =>frame C
zA zA
 yB zC yB

yA yA
xC
xA xB xA xB

A
RB  Rot z ( ) B
RC  Rot y ( )

RC  ARB B RC  Rot z ( ) Rot y ( )

A
Rotation about non-reference frame
• Frame A: A reference frame
• Frame B: A frame to be rotated
• Rotate frame B about its own k-th axis by angle 
 frame C z A

zB yB
B
RC  Rotk ( )
A
RC  RB RC  RB Rotk ( )
A B A
xC
yA

xB
xA
Post-multiplication of
rotation matrix
 Rotation about Reference Frame
• Frame A: A reference frame
• Frame B: A frame to be rotated
• Rotate frame B about the k-th axis of frame A by
angle   frame C zA
Rotation of frame A about its 
own k-th axis by   frame A’ zB yB
A
RA'  Rotk ( )
as both C and A’ are
A'
RC  ARB results of the same
rotation
xC
yA

xB
RC  RA' RC  Rotk ( ) RB
A A A' A xA xA’

Pre-multiplication of rotation matrix

 Example 1
• Frames A and B are coincident initially. Rotate frame B
about its x-axis by 90 degrees and then rotate the new frame
about the y-axis of frame A by 90 degrees. Finally rotate the
frame obtained after the previous rotation about its own z
axis by -90 degrees. Calculate the rotation matrix of the final
frame w.r.t. frame A. z A

Initially, the rotation matrix of frame B w.r.t frame A.

zB

1 0 0 
RB   0 1 0 
yA
yB
 0 0 1 xB
 
xA
 Example 1
zA

• After the rotation about xB

yB’
1 0 0 
   zB’
RB'  RB Rot x (90 )   0 0  sin 90  yA
 0 sin 90  0 
 
xB’
• After the rotation about yA xA zA

 0 0 1 1 0 0 
R B ''  Rot y (90 ) RB '   0 1 0  0 0  1

xB’’’
  1 0 0  0 1 0 
   zB’’’
zB’’
yA
• After rotation about zB’’ yB’’ yB’’’

xB’’
RB'''  RB'' Rotz (90 )  ... xA
Representation of a rotation matrix by three
angles
 xT y  0
 y T z  0
R  x, y, z  subject to T
x z  0
 x  y  z  1
• A rotation matrix has 3 independent variables
because
• It is possible to represent a rotation matrix by three
angles, such as the Euler angles, Roll-Pitch-Yaw
angles, etc.
 Euler Angles z

• Rotation about z-axis by 

• Rotation about the new y-axis by 
• Rotation about the new z-axis by 

R  Rot z ( ) Rot y ( ) Rot z ( ) Y’

 c c c  s s  c c s  s c c s 
 
  s c c  c s  s c s  c c s s 
  s c s s c 
 Z’’
 Roll-Pitch-Yaw Angles
z

• Rotation about z-axis by 

• Rotation about the new y-axis by 
• Rotation about the new x-axis by 

Y’

R  Rot z ( ) Rot y ( ) Rot x ( )

X’’
 Euler Angles Rotation Matrix
• Given Euler angles  rotation matrix: This is easy (see
previous slides)
• Given rotation matrix  Euler angles: Solving the following
non-linear equation
 r11 r12 r13   c c c  s s  c c s  s c c s 
R   r21 r22 r23    s c c  c s  s c s  c c s s 
Given  r31 r32 r33    s c s s c 
      
2 
s   r132  r23     atan2( r13  r23 , r33 )
2 2
c  r33 
  atan2 (r23 , r13 )   atan2(b, a)
if s  0 
  atan2 (r32 ,r31 )
a

  r  atan2 (r , r )  b

if s  0
33 21 22

  90 
- 90 
r33
 Equivalent Angle-Axis
• A rotation matrix can be also obtained by a
  u1 
rotation of angle about an equivalent axis u  u2
u 
 3

(u ,  ) is called the equivalent 
angle - axis of the rotation matrix

Given (u ,  )  the rotation matrix is
 u12 v  c u u v  u s u u v  u s 
  1 2  3  1 3  2  
R(u ,  )   u1u 2 v  u3 s u 22 v  c u 2u3v  u1s 
 u u v  u s u u v  u s u 2v  c 
 1 3  2  2 3  1  3   
v  1  c
 Equivalent Angle-Axis
• Given a rotation matrix  how to calculate the
equivalent angle axis?
Solve
 u12 v  c u1u 2 v  u3 s u1u3v  u 2 s 
 r11 r12 r13  
r r23    u1u 2 v  u3 s u 22 v  c u 2 u3v  u1s 
21 r22
r   
 31 32 33  u1u3v  u 2 s
r r u 2 u3v  u1s u32 v  c
 
 Trace( R)  1 
  arccos 
 2 
 1  r32  r23 
u r13  r31
2 sin   r21  r12 
 
 Homogeneous Transformation

• A general approach for coordinate transformation

• Useful and important
 Mapping: Changing descriptions
from frame to frame
• Mapping involving translated frames
• Frames A and B have the same orientation
and their origins are different
Point Q
B
A r
r
A  B 
r r  p 
p Frame B
Frame A

A
r : coordinates of point Q w.r.t. frame A
B
r : coordinates of point Q w.r.t. frame B
p : vector from origin of frame A to that of frame B
 Mapping involving rotated frames
• Frames A and B have different orientations,
but their origins are the same
A
RB : rotation matrix of frameB w.r.t.frameA


A  Br

A B
r
r  RB r
A

Frame B
Frame A
 Mapping involving general frames

• Frames A and B have different origins and orientations

A
RB : rotation matrix of frame B w.r.t. frame A Frame B
p : vector from origin of A to origin of B.
B
• Introduce a frame C that is parallel to r Frame C
frame A and share the origin with frame B A
r 
A C  p
r r  p
C C B C B
r  R B r  RB r Homogeneous
 transformation
  A B matrix
r  p  RB r
Frame A
  
 A r   A RB p  B r 
   
   13
1 0 1  1  Homogeous coordinates
 Homogeneous Transformation Matrix
• A 4*4 matrix, always full rank

 RB A 
A
TB   p
 013 1
• Homogeneous coordinates transformation:

A
 r A  r B 
  TB  
1  1 
Operators: Translation, Rotation
and Transformation
 Translation Operator

• Translate a point Q by a vector p

r : the position before translation
r ': the position after translation 
   p
r' p  r 
r 
   
r'

 r '    I 33 p  r 
1   0 1 
   13 1  

D( p) : Translational operator
 Rotation Operator

• Rotate point about axis u by an angle 
Introduce a frame B, which is obtained

by rotating frame A about u by  Q
 
r u
frame A 
rotation
frame B  
point Q 

point Q'
 Q’
r'

Frame B

A B 
r r' Frame A

 A B    r '   R(u , ) 031  r 

r ' R r '  R(u , )r  
1 
B
 
0
 13 1 1 

Rot (u , ) : rotational operator
 Transform Operator
• For operation involving both translation and
rotation
– First rotate point about an axis by angle
– Then translate it by a vector

     u
 r '    R(u , ) p  r  r 
1   0  
   13 1 1   
p r'

T ( R, p) : transformation operator
Transformation Arithmetic
Compound Coordinates Transformation

• Coordinates in 3 frames

Frame C

Frame A Frame B
A B C 
r , r , r : homogeneous coordinates of a point w.r. t. the frames
A
TB , B TC , ATC : homogeneous transformation matrices among the frames
A B C A C
r  TB r  TB TC r ,
A A B
r  TC r A


A
TC  ATB BTC
 Compound Transformation
• Transformation w.r.t. mobile frame
– Transform frame A by T1 (w.r.t. frame A)frame B
– Transform frame B by T2 (w.r.t. frame B)  frame C

Frame A Frame B Frame C

A
TB  T1 , B
TC  T2

A
TC  T1T2
 Compound Transformation
• Transformation w.r.t. a reference frame
– Transform frame A by T1 (w.r.t. frame A)frame B
– Transform frame B by T2 (w.r.t. frame A) frame C

Frame A Frame B Frame C

A
TB  T1 , T2 is a transform operator


A
TC  T2 ATB  T2T1
 Summary on Compound Transformation

• Frame A: a reference frame

• Frame B: a moving frame
1) Transform frame B by an operator T w.r.t itself
A
TB'  T BT
A
Post-multiplication of T

2) Transform frame B by an operator T w.r.t. the

reference frame.

A
TB' T T B A
Pre-multiplication of T
 Example 2
• Let r be the homogenous coordinates of a point w.r.t.
a frame A. Rotate the point about zA by 30 degrees.
Then translate it along zA by -1 and yA by 2. Find the
new coordinates of the point w.r.t. frame A

 cos 30   sin 30  0 0 
  sin 30  cos 30 0 2 r

r' 
 0 0 1  1
 0 0 0 1 
 Example 3
• Let r be the homogenous coordinates of a point
w.r.t. a frame A. Translate it along zA by -1 and yA
by 2. Then, rotate the point about zA by 30 degrees.
Find the new coordinates of the point w.r.t. frame
A
 cos 30   sin 30  0 0 1 0 0 0 
  sin 30   0 1 0 2 r
r'  cos 30 0 0 
 0 0 1 0  0 0 1  1
 0 0 0 1  0 0 0 1 
 Example 4
• Let T0 be the position and orientation of an object
w.r.t. a frame A. Rotate the object about its own x-
axis by 60 degrees. Then translate it along zA by -1
and yA by 2. Furthermore, rotate the object about zA
by 30 degrees. Find the new position and orientation
of the object w.r.t. frame A
 cos 30  sin 30 0 0 1 0 0 0  1 0 0 0 
    
 sin 30 cos 30 0 0  0 1 0 2  0 cos 60  sin 60 0 
T'    1  0 sin 60 cos 60 0 
T0
 0 0 1 0  0 0 1

 0  1   0
 0 0 1  0 0 0 0 0 1 
 Inverse Transformation
• Inverse of a homogeneous transformation
matrix
Frame B
A
 r A  r
 
1 
 TB 
B 
 r
  
1  1 
B 
T
A
B
1  A r 
 
1  B
r
A  A B
r  p  RB r A
r 
 p
 A T A A T 
B
r  RB r  RB p

  A
 B r   A RBT p  r 
 ARBT
 
 1   013 1  1  Frame A

T
A
B
1
Inverse Transformation


T  R p

 013 1

1  R T
 R T 
p
T  
 013 1 
 Transformation Equation
• In robotics, many problems concern a set of
transformation forming a loop

A
TC  TB TC
A B

Given any two of transformation matrices, we

can calculate the other one.
 Example 5
• A camera mounted on a robot arm
sees the base of the robot as
0 0 1 1 
1 0 0  2 object
T1  
0 1 0 1 
0 0 1 
 0
And it sees the object as
1 0 1 10 
 5 
T2   0  1 0
 2
Robot base
0 1 1
0 0 0 1 

Find the position and orientation of the
object w.r.t. the robot base
 Example 5
• Solution:
b
To : the homogeneous transformation matrix of object w.r .t the base
b
Tc : the homogeneous transformation matrix of camera w.r.t the base
c
To : the homogeneous transformation matrix of object w.r .t the camera
c
Tb : the homogeneous transformation matrix of base w.r.t the camera

b
To  Tc T o Tb
b c
1
 
c 1 c 1
T o T1 T2
 0 0 1 1  1 0 1 10 
1 0 0  2  0 1 0 5 
b
To  
0 1 0 1  0 1 1  2
0 0 0 1  0 0 1 
   0
 0 1 0 2 1 0 1 10 
 0 0 1  1 0 1 0 5   ...

1 0 0 1  0 1 1  2
 0 0 0 1  0 1 
  0 0
 Summary on Homogeneous
Transformation Matrix
1) It can be used for coordinates transformation
A A B
r  TB r
2) It can be used to represent the relation between
two coordinate frames
zB
zA 5

xB
yA
xA- yB
0 1 0 0
 
 1 0 0 5
A
TB  
0 0 1 0
 
0 1 
 0 0
 Summary on Homogeneous
Transformation Matrix
3) Represent position and orientation
zo
of an object
z yo

To C

object
4) Represent translation and rotation xo
y
O

  I 33 p 
Pure translation : D( p)    x
 013 1 
Pure rotation : Rot ( R)  
R 031 
 013 1 
 R 
Translation & rotation : T ( R, p)   p 

 013 1 