Lecture 9: Introduction to Kinematics and Rotation Matrices

Based on Chapter 6, Bullo and Smith

Vaibhav Srivastava
Department of Electrical & Computer Engineering
Michigan State University

Feb 13, 2017

ECE 818: Robotics
http://www.egr.msu.edu/∼vaibhav/teaching/robotics.html

1 / 17
 Introduction to Kinematics

• Kinematics is study of motion, independent of the force causing it

• Concerned with describing the motion of objects by relating positions to

velocities, i.e., via differential equations of the first order
• In kinematics we study how to define and manipulate:
(i) body positions and orientations,
(ii) reference frames and changes of coordinates, and
(iii) rigid body motions and their composition

2 / 17
 Rotation Matrices

• the set of rotation matrices is the configuration space or part of the configuration
space of any object that rotates

• rotation matrices one can then use the motion planning algorithms to plan the
motion of three-dimensional objects such as aerial, space and underwater vehicles

• rotation matrices can be understood as transformations that rotate objects and

play a central role in modeling reference frames in space
• rotation matrices are a fundamental technology in

(i) robotics and aerospace engineering: orientation of robots, aircraft, satellites

(ii) computer vision: orientation of cameras and objects in the scene, and
(iii) video games and 3D virtual worlds

3 / 17
 Geometric objects and their algebraic representation
The 3-dimensional Euclidean space contains various useful geometric objects:
z0
• points
p

• free vectors: described by a length and a direction, but

without a base point !
O0 p
y0
• vectors fixed at a point: with length direction and base O0

points

• fixed vectors defined by ordered pair of points x0

Figure 6.1: Basic geometric objects

• Reference frames
• reference frame is a pair Σ0 = (O0 , {x0 , y0 , z0 })

• O0 is a point called the origin of the reference frame

• {x0 , y0 , z0 } is a triplet of unit-length orthogonal vectors fixed at the origin and

satisfying the right-hand rule
4 / 17
 Products between free vectors
• The dot product between free vectors (or scalar product or inner product):

v · w = kv kkw k cos(θ)

k · k = length of vector θ = angle between v and w

• The cross product between free vectors (or outer product) is a vector with length

kv × w k = kv kkw k sin(θ),

and direction to both v and w as given by the right-hand-rule

• Both the dot product and the cross product are defined without any reference
frame

• An axis is a unit-length vector. The dot product of a vector with an axis is the
length of the projection of the vector onto the axis

5 / 17
 Representing vectors and points as arrays

• a free vector can be written as an column array

• entries of array are called the coordinates of the free vector

• Let p be a point in 3-dimensional space, v be a free vector, and

Σ0 = (O0 , {x0 , y0 , z0 }) be a reference frame. Let pq ~ from O0 to p. The
coordinate representation of v and p with respect to Σ0 are
   
v · x0 O~0 p · x0
 
v 0 = v · y0  , p 0 = (O~0 p)0 = O~0 p · y0  ,
v · z0 O~0 p · z0

• In other words v = (v · x0 )x0 + (v · y0 )y0 + (v · z0 )z0

• Although the representations of points and free vectors are identical, but
• it makes no sense to rotate a point or to translate a free vector

6 / 17
 Product between vectors using coordinates

Given two vectors v and w with coordinates v 0 = [v1 , v2 , v3 ] and w 0 = [w1 , w2 , w3 ],

then

v · w = (v 0 )> w 0 = v1 w1 + v2 w2 + v3 w3 ,
(v × w )0 = [v2 w3 − v3 w2 , v3 w1 − v1 w3 , v1 w2 − v2 w1 ]>

7 / 17
 Geometric and algebraic viewpoints
Geometric Viewpoint

• points and free vectors in space, as basic objects,

• properties independent of reference frames, and

• concepts and tools from Euclidean geometry.

Algebraic Viewpoint

• arrays and matrices, as basic objects,

• computations in a specific reference frame, and

• concepts and tools from linear algebra and matrix theory

The two approaches are equivalent once a reference frame is selected

8 / 17
 Geometric properties of rotations
• Rotations are operations on free vectors;
• they preserve length of vectors and angles between vectors.
Robotic Planning and Kinematics, v0.91(e) (11 Apr 2016). Chapter 6, slide 230
• In three-dimensional space, there are three independent basic rotations
• In vehicle kinematics, three basic angles are roll, pitch and yaw
• Rotations can be composed and the order of composition is essential.
• Each rotation admits an inverse rotation.
• Each rotation leaves one axis unchanged called the rotation axis

roll x-axis

pitch y-axis
yaw z-axis

Figure 6.2: The roll-pitch-yaw conventions for aircra�. The yaw axis points downward towards the ground. The 9 / 17
 0 0

mposed rotation #2) = rotate about axis z0 by ⇡/2 and then about axis x0 by ⇡/2
Frames do not commute
ated graphically in Figure 6.3, it is easy to see that the composite rotation #1 is not e
• Composed rotation 1 = rotate about axis x0 by π/2 and then about axis z0 by π/2
omposite• rotation
Composed #2.
rotation 2 = rotate about axis z0 by π/2 and then about axis x0 by π/2
z0 z0 z0
2
C C B
` ` `

B A 2 D B D E

x0 x0 x0

z0 z0 z0
2

C A
` `

A
`

2
B A B E C B

x0 x0 x0

Figure 6.3: Rotations do not commute, i.e., the order of composition is essential.
10 / 17
 Rotations may not be composed as though they were arrays

• Imagine we represent a rotation as an array of the form

 
roll
pitch .
yaw

• Consider the seqjuence +90 pitch, + 90 roll, -90 pitch

• Does        
0 +90 0 +90
? 
+90 +  0  + −90 = 0 
0 0 0 0

11 / 17
out of the page). Let us assume that ⌃1 is obtained by rotating ⌃0 counterclockwise by an
angle ✓. The array representation in the frame ⌃0 of the basis vector x1 for the frame ⌃1 is
Representing reference frames
y0
y1
x1

✓
x0

Figure 6.4: Two example reference frames with the same origin

• Consider two reference frames Σ0 = (O0 , {x0 , y0 }) and Σ1 = (O1 , {x1 , y1 }) in the
 
plane with the same origin0O0 =xO
1 1· x0 cos ✓
x1 = = .
x1 ·counterclockwise
• Let Σ1 be obtained by rotating Σ y0 sin ✓ by an angle θ
0

Now, define
• Thea 2array reference-frame
⇥ 2 representation in rotation
the framematrix
Σ0 of (representing
the basis vector⌃1x1with respect
for the frametoΣ⌃
1 0is) by
     
0
⇥ 0 ⇤
x x01=· x0x1 y· 1x0· x=
0
θ
cos cos(✓) sin(✓)
0
R1 = x 1 | y 1 = 1 x · y =sin θ . (6.1)
x1 · y0 y1 · y0
1 0 sin(✓) cos(✓)
12 / 17
 Reference frame rotation matrix

• Define reference frame rotation matrix representing Σ1 wrt Σ0 by

   
 0
R10 = [x10  y1 ] = x1 · x0 y1 · x0 = cos θ − sin θ
x1 · y0 y1 · y0 sin θ cos θ

We can represent the frame Σ1 with respect to the frame Σ0 by either

• θ ∈ [−π, π) = S1 , or

• R10 a 2 × 2 matrix with special properties

13 / 17
 The fundamental advantage of the rotation matrix representation is that it naturally extends
to problems in three dimensions. Indeed, let us write down 3-dimensional reference frames and
Three-dimensional reference frames
understand 3-dimensional reference-frame rotation matrices. Given two reference frames ⌃0 =
z0
z1

y1

y0

x0 x1

Figure 6.5: Two example reference frames with the same origin in three dimensions
• Consider two reference frames Σ0 = (O0 , {x0 , y0 , z0 }) and Σ1 = (O1 , {x1 , y1 , z1 })
(O0 , {x0 , y0 , z0 }) and ⌃1 = (O1 , {x1 , y1 , z1 }) with the same origin O0 = O1 in 3-dimensions,
in the plane with the same origin O0 =⌃1O
the reference-frame rotation matrix describing 1 respect to ⌃0 is defined by
with
2 3
x1 · x0 y1 · x0 z 1 · x0
• reference frame rotation matrix
⇥ representing
⇤
R 0 = x0 | y 0 | z 0 = 4 x · y
Σ wrt Σ is given (6.2)
y · y1 z · y 5 .0
by
1 1 1 1 1 0 1 0 1 0
 y1 · z 0 z 1 · z 0 
x1 · z 0

  x1 · x0 y1 · x0 z1 · x0
R10 = [x10  y10  z10 ] = x1 · y0 y1 · y0 z1 · y0 
x1 · z0 y1 · z0 z1 · z0
• Also known as the “direction cosine matrix”

• rotation matrix representing Σ0 with respect to Σ1 : R01 = (R10 )>

14 / 17
 Changes of reference frame
Lemma: Given a point p and two reference frames Σ0 and Σ1 , the arrays p 0 and p 1
are related through
p 0 = R10 p 1

15 / 17
 entries in equation (6.2), one can immediately establish
R01 = (R10 )T .
Changes of reference frame
We continue to consider two frames ⌃0 and ⌃1 with the same origin.
Lemma 6.3
Lemma: (Coordinate
Given a 1pointtransformation) Given a frames
p and two reference point p and
Σ0 two
andreference
Σ1 , the frames
arrays ⌃p00 and
and⌃p1 ,1
the arrays p and p are related through
0
are related through
pp0 0==RR
0 01 1
1 p1 p
.
!
Proof. First, write the geometric decomposition of vector O0 p with respect to {x1 , y1 , z1 }:
! ! ! !
O0 p = O0 p · x 1 x 1 + O0 p · y 1 y 1 + O0 p · z 1 z 1 .
Second, express the le�-hand side and the right-hand side with respect to ⌃0 :
! ! ! !
p0 = (O0 p)0 = O0 p · x1 x01 + O0 p · y1 y10 + O0 p · z1 z10
2 ! 3
O0 p · x1
⇥ 0 ⇤ 6 ! 7
= x1 | y10 | z10 4 O0 p · y1 5 = R10 p1 .
!
O0 p · z 1
⌅
• The superscript of RHS (resp. LHS) p matches subscript (resp. superscript) of R

15 / 17
 The
Subsection 11: The three three
basic basic rotations
rotations
• Consider two reference frames Σ0 = (O0 , {x0 , y0 , z0 })
and Σ1 = (O1 , {x1 , y1 , z1 }) in the plane with the same
origin O0 = O1
Consider two frames ⌃0 = (O0 , {x0 , y0 , z0 })Robotic
and ⌃ 1 = (O1 , {x1 , y1 , z1 }) with the same origin
Planning and Kinematics, v0.91(e) (11 Apr 2016).

1 . Iftwo
O0•=IfOthe the two frames
frames share
share the the same
same z-axis,
z-axis, that thatis,is,if if z0 = z1 , then there must exist an
z0 = z1
anglez0✓ =
such z1 ,that
then there must exist an angle θ such that
(
x1 = cos(✓)x0 + sin(✓)y0 ,
y1
y1 = sin(✓)x0 + cos(✓)y0 ,
y0

2 3
0 x0 ✓
so that, since z10 = 405,
x1
1
Figure 6.6: The frame ⌃1 obtained by rotating about z0 by ✓
2 3
cos(✓) sin(✓) 0
R10 = 4 sin(✓)twocos(✓) 05and,
basis vectors . in summary, the three basic rotations about the th
0 0 1
2 3 2 16 / 17
 Thethethree
two basis vectors and, in summary, basicrotations
three basic rotations
about the three orthogonal axes are
The three basic rotations about the three orthogonal axes are
2 3 2 3
1 0 0 cos(✓) 0 sin(✓)
4
Rotx (✓) = 0 cos(✓) sin(✓)5 , Roty (✓) = 4 0 1 0 5,
0 sin(✓) cos(✓) sin(✓) 0 cos(✓)
2 3
cos(✓) sin(✓) 0
Rotz (✓) = sin(✓) cos(✓) 05 .
4 (6.3)
0 0 1

17 / 17
 Thethethree
two basis vectors and, in summary, basicrotations
three basic rotations
about the three orthogonal axes are
The three basic rotations about the three orthogonal axes are
2 3 2 3
1 0 0 cos(✓) 0 sin(✓)
4
Rotx (✓) = 0 cos(✓) sin(✓)5 , Roty (✓) = 4 0 1 0 5,
0 sin(✓) cos(✓) sin(✓) 0 cos(✓)
2 3
cos(✓) sin(✓) 0
Rotz (✓) = sin(✓) cos(✓) 05 .
4 (6.3)
0
Robotic Planning and Kinematics, v0.91(e) (11 Apr 2016). 0 1 Chapter 6, slide 241

Every one of the three matrices satisfies the following algebraic properties:

Rotz (0) = I3 ,
Rotz (✓) · Rotz ( ) = Rotz (✓ + ),
1
(Rotz (✓)) = (Rotz (✓))T = Rotz ( ✓),

each of which has a nice geometric interpretation.

17 / 17