Spatial Descriptions & Transformations

Berke Gür

Robotics 3/15/2021
Berke Gür 1
Chapter Objectives
• Our aim in this lecture is to learn how the position & orientation of an object can be
represented in the 3D space
• For this purpose, we will use tools and methods from linear algebra
• Material covered in this lecture will form the foundation of the kinematic, static, &
dynamic analysis of manipulators that we will study in the subsequent weeks

Robotics 3/15/2021
Berke Gür 2
Position of a Point

• A point p in 3-D space is a geometric

{0} ẑ0
entity
• The position of this point p can p
mathematically be expressed in terms
of a 3×1 coordinate vector 0p with pz
respect to a reference frame {0} o0 ŷ0
• The frame {0} is defined by origin px
point o0 & 3 orthogonal unit vectors py
x̂0
xˆ0 , yˆ 0 , zˆ0
T
0
p =  px py pz 

{0} ≡ [ o0 zˆ0 ]
T
xˆ0 yˆ 0
Robotics 3/15/2021
Berke Gür 3
Position of a Point
• The position of the point can 
p =  px
T
0
py pz 
geometrically be expressed in terms of
the position vector = px ⋅ xˆ0 + p y ⋅ yˆ 0 + pz ⋅ zˆ0
• We will use the coordinate & position 
vectors interchangeably
0
p≡ p 0

• The same point will have different

position vectors for different frames ŷ1 x̂1
{0} & {1}
p
ŷ0
1 
p= p
1

 0
0
p= p
x̂0
Robotics 3/15/2021
Berke Gür 4
Position of a Point

• Note that textbooks use slightly different notations to represent point p with respect
to frame {0}
– Class notes notation 0p is equivalent to Spong’s book notation p0
Be careful about these notational differences!
• Also note that the origin of a frame {1} is described by the point o1
• If you want to represent the origin of frame {1} with respect to frame {0}, this is
shown by the vector 0o1

Robotics 3/15/2021
Berke Gür 5
Orientation of an Object
• It is sufficient to define the position
vector for a point ẑ1
• For an object, we need to define both {0} ẑ0 {1}
ŷ1
the position & the orientation
• To describe the orientation of an  o1
0
object, we attach a frame {1} to the o1
object
o0 x̂1
• This object frame {1} is different than ŷ0
the reference frame {0}

x̂0

Robotics 3/15/2021
Berke Gür 6
Orientation of an Object
• Describing the object in space is
equivalent to describing the frame {1} ẑ1
attached to the object with respect to {0} ẑ0 {1}
the reference frame {0} ŷ1
• Frame {1} is described relative to
frame {0} by the position vector 0o1 0  o1
o1
and by the rotation matrix (e.g., 01R)
o0 x̂1
ŷ0

x̂0
0
{1} = { 0 o1 0
1 R}
Robotics 3/15/2021
Berke Gür 7
The Rotation Matrix
• The rotation matrix is a 3×3 matrix that describes the unit coordinate vectors of a
frame with respect to the coordinate vectors of the reference frame

 r11 r12 r13 

=0
R =0
ˆ
x 0
ˆ
y 0
ˆ
z  r r22 r23 
1  1 1 1  21 
 r31 r32 r33 

• The rotation matrix can be computed from the scalar product of the unit vectors of
the two frames

 xˆ1  xˆ0 yˆ1  xˆ0 zˆ1  xˆ0 

0
R =  xˆ  yˆ yˆ1  yˆ 0 zˆ1  yˆ 0 
1  1 0 
 xˆ1  zˆ0 yˆ1  zˆ0 zˆ1  zˆ0 
Robotics 3/15/2021
Berke Gür 8
 Properties of the Rotation Matrix

1. The transpose of the rotation matrix 01R gives 10R

T
 xˆ1  xˆ0 yˆ1  xˆ0 zˆ1  xˆ0 
=0 T
R = ˆ
x  yˆ ˆ
y  yˆ ˆ
z  yˆ   0
xˆ 1 0
yˆ 1 0
zˆ 1 
T
1  1 0 1 0 1 0  
 xˆ1  zˆ0 yˆ1  zˆ0 zˆ1  zˆ0 
 xˆ0  xˆ1 yˆ 0  xˆ1 zˆ0  xˆ1 
=  xˆ0  yˆ1 =yˆ 0  yˆ1 zˆ0  yˆ1  =
1
xˆ 0 1 yˆ 0 1 zˆ 0  1
R
  0

 xˆ0  zˆ1 yˆ 0  zˆ1 zˆ0  zˆ1 

Robotics 3/15/2021
Berke Gür 9
Properties of the Rotation Matrix

2. The rotation described by the matrix 10R reverses the rotation associated by the
0 R matrix
1

0
1 R ⋅ 01 R =
I
3. Combining the first and second properties yields

=
1
0R =
0
RT
1
0
1 R −1

4. Multiplication of (rotation) matrices are not commutative

0
1 R ⋅ 21 R ≠ 21 R ⋅ 01 R

Robotics 3/15/2021
Berke Gür 10
Properties of the Rotation Matrix
ẑ0 ẑ0

ŷ0 ŷ0

x̂0 x̂0
ẑ0 ẑ0

Rzˆ0 (π 2 )
Rxˆ0 (π 2 )

ŷ0 ŷ0

x̂0 x̂0
ẑ0 ẑ0

Rzˆ0 (π 2 )
Rxˆ0 (π 2 )

ŷ0 ≠ ŷ0

x̂0 x̂0
Robotics 3/15/2021
Berke Gür 11
Properties of the Rotation Matrix

5. The columns of the rotation matrix R = [x y z] are orthonormal unit vectors

xˆ=
yˆ xˆ=
zˆ yˆ=
zˆ 0
xˆ= yˆ= zˆ= 1
6. Since 10R = 01RT, the rows of the rotation matrix R are also orthonormal unit
vectors

Robotics 3/15/2021
Berke Gür 12
Properties of the Rotation Matrix
7. Any column (or unit vector) of the rotation matrix can be computed from the other
two columns. Thus, knowing only 2 columns (6 entries) of the rotation matrix is
sufficient

xˆ =
yˆ × zˆ , yˆ =
zˆ × xˆ , zˆ =
xˆ × yˆ

Robotics 3/15/2021
Berke Gür 13
Properties of the Rotation Matrix
8. The rotation matrix can also be specified with 3 elements using Cayley’s Formula

(I − S) −1 (I + S)
R=

where S is a 3×3 matrix defined as

 0 − sz sy 
=S  sz 0 − sx 
 
 − s y sx 0 

Robotics 3/15/2021
Berke Gür 14
Mapping
• Mapping is changing the description of a point from one frame to another frame
• There are three types of mapping
1. Translational Mapping: If the two frames
have the same orientation, the
{1}
mapping between these two frames is expressed as a vector sum
ẑ1 p

{0} 1 
ẑ0 0  p
p ŷ1
o1
=
0
p 0
o1 + 1 p 
o0
0
o1 x̂1
ŷ0

x̂0
Robotics 3/15/2021
Berke Gür 15
Mapping
2. Rotational Mapping: If the origins of the two frames are coincident but the frames have
different orientations, the mapping is expressed as a matrix product

{0}
{1} ẑ
=
0
p 0
1 R⋅ p
1 1 ẑ0
p

ŷ1
ŷ0
o0,1

x̂0
x̂1

Robotics 3/15/2021
Berke Gür 16
Example: Rotational Mapping

• Given 1p = [0 2 0]T , find 0p where frame {1} is obtained by rotating frame {0} by 30
degrees about the positive z0 axis

Robotics 3/15/2021
Berke Gür 17
Mapping
3. General Mapping: Any general mapping between two arbitrary positioned and oriented frames
can be expressed as a rotational mapping followed by a translational mapping

{1}
ẑ1
0
p= 0
1 R ⋅ 1 p + 0 o1
p
1 ŷ1
{0} ẑ0 p
ẑ1 0  o1
p
 0
o1
o0 ŷ1 x̂1
ŷ0

x̂0
x̂1
Robotics 3/15/2021
Berke Gür 18
Mapping
• The mapping between two arbitrary frames can compactly be expressed in terms
of the 4×4 homogeneous transform matrix

 0 p   01 R 0
o1   1 p 
=   ⋅ 
 1  0 0 0 1  1
 1
p
= 1A ⋅  
0

1

• The homogeneous transformation matrix is depicted with (A)

Robotics 3/15/2021
Berke Gür 19
Example: General Mapping

• Given 1p = [3 7 0]T , find 0p where frame {1} is obtained by rotating frame {0} by 30
degrees about the positive z0 axis and translated in the x0 direction by 10 units and
the y0 direction by 5 units

Robotics 3/15/2021
Berke Gür 20
Spatial Operations
• Spatial operators are mathematical tools for moving points and rotating vectors
• Spatial operations are different than mappings in the sense that all operations are
done in the same reference frame (ie., no transformation between frames)
• There are three types of spatial operators
1. Translational Operator (D): Moving a point 0p in the direction of a vector 0q by an amount of q
= |q| can be expressed as a vector sum

0
=
v 0
p + 0q

or a matrix operation

=
0
v Dq (q ) ⋅ 0 p

Robotics 3/15/2021
Berke Gür 21
Spatial Operations
where the translation operator is defined as

1 0 0 qx   qx 
0  q = q 
1 0 qy  y
Dq (q ) =  
 qz 
0 0 1 qz 
 
0 0 0 1  q = q = qx2 + q y2 + qz2

Robotics 3/15/2021
Berke Gür 22
Spatial Operations
2. Rotational Operator (R): Rotating a vector 0p about a vector 0k by an angle of θ can be
expressed as a matrix product

=
0
v 0
R ⋅ 0p

or a matrix operation

=
0
v R k (θ ) ⋅ 0 p

3. Transformation Operator (A): The combination of rotating and translating a vector 0p to obtain
a new vector 0v can be also be expressed as a matrix operation

=
0
v 0
A ⋅ 0p

Robotics 3/15/2021
Berke Gür 23
Example: Compund (Multiple) Rotations

• Find the rotation matrix 01R that describes frame {1} obtained first by a rotation of
π/6 about the x0 axis and then by a rotation of π/6 about the z0 axis

Robotics 3/15/2021
Berke Gür 24
Example: Compund (Multiple) Transformations

• Given 2p, find 0p if the intermediate transformations 01A and 12A are given.

Robotics 3/15/2021
Berke Gür 25
Example: Compund (Multiple) Transformations

• Given the relations 01A = 02A · 21A and 01A = 03A · 34A · 41A find 34A.

Robotics 3/15/2021
Berke Gür 26
Example: Inverse Homogenous Transform
• Assume that you are given the homogenous transform

 0
1R
0
o1 
1A = 
0

0 0 0 1 

The inverse homogeneous transform can be obtained by taking the matrix inverse
1 A = (0 A)−1
0 1
• Prove that it also possible to compute 10A from 01A without any matrix inversions

Robotics 3/15/2021
Berke Gür 27
Parameterization of Rotations

• A simple rotation is a single rotation about the x-, y-, or z-axes

• A compund rotation involves multiple simple rotations about different axes
• A compound rotation may involve
– Multiple rotations about fixed axes (fixed axis rotations)
• The rotation matrix is computed from the individual rotation matrices multiplied in reverse
order (pre-multiplication of rotation matrices)
– Multiple rotations about current axes (current axis rotations)
• The rotation matrix is computed from the individual rotation matrices multiplied in the order
of rotation (post-multiplication of rotation matrices)

Robotics 3/15/2021
Berke Gür 28
Simple Rotations

• Rotations w.r.t. a single axis (either x-, y-, or z-)

1 0 0 
R x (ψ ) 0 cosψ
= − sinψ 
 
0 sinψ cosψ 

 cosθ 0 sin θ 
R y (θ ) =  0 1 0 
 
 − sin θ 0 cosθ 

cos φ − sin φ 0
R zˆ0 (φ ) =  sin φ cos φ 0
 
 0 0 1 

Robotics 3/15/2021
Berke Gür 29
Parameterization of Rotations
• It is difficult to visualize the physical rotation represented by the 3×3 rotation matrix
• The 9 element rotation matrix can be specified using only 3 variables (see Cayley’s
formula)
• However, there are many more physically meaningful ways to specify a compund
rotation using only 3 variables
• Any rotation can be specified in terms of the
– X-Y-Z Fixed Angles
– Z-Y-X Euler Angles
– Z-Y-Z Euler Angles
– Axis and Angle Representation
– X-Y-Z-W Quaternions

Robotics 3/15/2021
Berke Gür 30
X-Y-Z Fixed Angles
• Consider that we have two coincident frames with the same orientation
• Assume that frame {0} is the reference frame and frame {1} is the rotating frame
• The X-Y-Z fixed angles consist of three rotations performed as follows:
1. First, rotate frame {1} about the fixed x0 axis by an angle ψ
2. Then, rotate frame {1} about the fixed y0 axis by an angle θ
3. Finally, rotate frame {1} about the fixed z0 axis by an angle ϕ

Robotics 3/15/2021
Berke Gür 31
X-Y-Z Fixed Angles
• The resulting compound rotation matrix is
0
1 R zˆ0 yˆ0 xˆ0 (ψ ,θ ,φ ) = R zˆ0 (φ ) ⋅ R yˆ0 (θ ) ⋅ R xˆ0 (ψ )
cos φ − sin φ 0   cosθ 0 sin θ  1 0 0 
=  sin φ cos φ 0  ⋅  0 1 0  ⋅ 0 cosψ − sinψ 
     
 0 0 1   − sin θ 0 cosθ  0 sinψ cosψ 
c φ cθ c φ sθ sψ − s φ cψ c φ sθ cψ − s φ sψ 
=  s φ cθ s φ sθ sψ + c φ cψ s φ sθ cψ − c φ sψ 
 
 − sθ cθ sψ cθ cψ 

• This rotation is termed the roll-pitch-yaw angles in Spong

Robotics 3/15/2021
Berke Gür 32
X-Y-Z Fixed Angles
• Given a rotation matrix
 r11 r12 r13 
0
R (φ , θ ,ψ ) = r r22 r23 
1 zˆ1 yˆ1xˆ1  21 
 r31 r32 r33 

the X-Y-Z fixed angles can be computed from (for cosθ ≠ 0 or −π / 2 < θ < π / 2)

θ =arctan(−r31 r112 + r212 )

φ = arctan[(r21 cos θ ) (r11 cos θ )]
ψ = arctan[(r32 cos θ ) (r33 cos θ )]

Robotics 3/15/2021
Berke Gür 33
X-Y-Z Fixed Angles

• For θ = ± π / 2, ϕ and ψ cannot be computed and the convention is to assume ϕ = 0

• Hence, for θ = π / 2, we have

θ = π / 2
ψ = arctan(r11 r22 )
φ =0 

• And also for θ = − π / 2

θ = −π / 2 
ψ = − arctan(r11 r22 )
φ =0 

Robotics 3/15/2021
Berke Gür 34
Z-Y-X Euler Angles
• The Z-Y-X Euler angles consist of three rotations performed as follows:
1. First, rotate frame {1} about the z1 axis by an angle ϕ
2. Then, rotate frame {1} about the current y1 axis by an angle θ
3. Finally, rotate frame {1} about the current x1 axis by an angle ψ

Robotics 3/15/2021
Berke Gür 35
Z-Y-X Euler Angles
• The resulting compound rotation matrix is
0
1 R zˆ1 yˆ1xˆ1 (φ ,θ ,ψ ) = R zˆ1 (φ ) ⋅ R yˆ1 (θ ) ⋅ R xˆ1 (ψ )
cos φ − sin φ 0   cosθ 0 sin θ  1 0 0 
=  sin φ cos φ 0  ⋅  0 1 0  ⋅ 0 cosψ − sinψ 
     
 0 0 1   − sin θ 0 cosθ  0 sinψ cosψ 
 c φ cθ c φ sθ sψ − s φ cψ c φ sθ cψ − s φ sψ 
= s φ cθ s φ sθ sψ + c φ cosψ s φ sθ cψ − c φ sψ 
 
 − sθ cθ sψ cθ cψ 

• Note that the resulting rotation matrix is the same as that derived for X-Y-Z fixed
axis rotations
• Hence, computing the Z-Y-X Euler angles from the rotation matrix is the same as in
the X-Y-Z fixed angles
Robotics 3/15/2021
Berke Gür 36
Z-Y-Z Euler Angles
• The Z-Y-Z Euler angles consist of three rotations performed as follows:
1. First, rotate frame {1} about the z1 axis by an angle ϕ
2. Then, rotate frame {1} about the current y1 axis by an angle θ
3. Finally, rotate frame {1} again about the current z1 axis by an angle ψ

Robotics 3/15/2021
Berke Gür 37
Z-Y-Z Euler Angles
• The resulting compound rotation matrix is

0
1 R zˆ1 yˆ1zˆ1 (φ ,θ ,ψ ) = R zˆ1 (φ ) ⋅ R yˆ1 (θ ) ⋅ R zˆ1 (ψ )
 c φ cθ cψ − s φ sψ − c φ cθ sψ − s φ cψ cφ sθ 
= s φ cθ cψ + c φ sψ − s φ cθ sψ + c φ cψ s φ sθ 
 
 − sθ cψ sθ sψ cθ 

Robotics 3/15/2021
Berke Gür 38
Z-Y-Z Euler Angles

• Given a rotation matrix, the Z-Y-Z Euler angles can be computed from (for sinθ ≠ 0 or
0 < θ < π)

=θ atan 2( r312 + r322 r33 )

φ = atan 2[(r23 sin θ ) (r13 sin θ )]
=ψ atan 2[(r32 sin θ ) (−r31 sin θ )]

• For θ = 0 or θ = π, ϕ and ψ cannot be computed and the convention is to assume ϕ

=0

Robotics 3/15/2021
Berke Gür 39
Z-Y-Z Euler Angles

• Hence, for θ = 0 we have

θ = 0
= ψ arctan(− r12 r11 )
φ = 0

• Also, for θ = π

θ =π
= ψ arctan(r12 − r11 )
φ =0

Robotics 3/15/2021
Berke Gür 40
Z-X-Z Euler Angles

1) Rotation about the Z axis by ϕ

2) Rotation about the x′ axis by θ

3) Rotation about the z″ axis by ψ

Robotics 3/15/2021
Berke Gür 41
Z-X-Z Euler Angles
• The resulting compound rotation matrix is
0
1 R z − x − z (φ ,θ ,ψ ) = R z (φ ) ⋅ R x (θ ) ⋅ R z (ψ )
 c φ cθ cψ − s φ sψ − c φ cθ sψ − s φ cψ cφ sθ 
= s φ cθ cψ + c φ sψ − s φ cθ sψ + c φ cψ s φ sθ 
 
 − sθ cψ sθ sψ cθ 

Robotics 3/15/2021
Berke Gür 42
Axis-Angle Representation

• Any rotation can be expressed as a rotation of an angle θ about a vector k

 k x k x vθ + c θ k x k y vθ − k z s θ k x k z vθ + k y s θ 
R k (θ ) = k k vθ + k s θ k y k y vθ + c θ k y k z vθ − k x s θ 
 x y z 
 k x k z vθ − k y s θ k y k z vθ + k x s θ k z k z vθ + c θ 

where k = [kx ky kz]T and vθ = 1−cθ

• The unit vector along k and the rotation angle θ can be computed from a given
rotation matrix using

 r32 − r23 
1 
=θ acos[(r11 + r22 + r33 − 1) 2],=kˆ r13 − r31 
2sin θ  
 r21 − r12 
Robotics 3/15/2021
Berke Gür 43
Example: Axis-Angle Representation

• Find the transformation matrix that rotates frame {1} through an angle of 2π/3
about a unit vector k defined as the diagonal of a cube

Robotics 3/15/2021
Berke Gür 44
X-Y-Z-W Quaternions (Euler Parameters)
• Also termed to Euler parameters
• Very similar to Angle-Axis representations
• Mainly used in computer vision

x  ε1 = k x sin(θ 2),
y  ε 2 = k y sin(θ 2),
z  ε 3 = k z sin(θ 2),
w  ε 4 = cos(θ 2).

Robotics 3/15/2021
Berke Gür 45
X-Y-Z-W Quaternions (Euler Parameters)
• The resulting compound rotation matrix is

 1 − 2ε 22 − 2ε 32 2(ε1ε 2 − ε 3ε 4 ) 2(ε1ε 3 + ε 2ε 4 ) 
 
R=
ε 2(ε ε
 1 2 3 4 + ε ε ) 1 − 2ε 2
1 − 2ε 3
2
2(ε ε
2 3 + ε ε )
1 4 
 2(ε1ε 3 + ε 2ε 4 ) 2(ε 2ε 3 + ε1ε 4 ) 1 − 2ε12 − 2ε 22 

• The Euler parameters can be obtained from the rotation matrix entries using the
relations

r32 − r23 r −r
=ε1 = ε 2 13 31
4ε 4 4ε 4
r21 − r12 1
ε 3= ε 4= 1 + r11 + r22 + r33
4ε 4 2

Robotics 3/15/2021
Berke Gür 46
Next Lecture
• Lecture topics
– Kinematics
• Reading assignment: Ch. 3 in textbook
• Questions?

Robotics 3/15/2021
Berke Gür 47