Carnegie Mellon

An Introduction to
Robot Kinematics

© Howie Choset, 2007

 Carnegie Mellon

An Example - The PUMA 560

2
3

4
1
There are two more
joints on the end
effector (the gripper)
The PUMA 560 has SIX revolute joints
A revolute joint has ONE degree of freedom ( 1 DOF) that is
defined by its angle
 Carnegie Mellon

Other basic joints

Revolute Joint
1 DOF ( Variable - )

Prismatic Joint
1 DOF (linear) (Variables - d)

Spherical Joint
3 DOF ( Variables - 1, 2, 3)
 Carnegie Mellon

We are interested in two kinematics topics

Forward Kinematics (angles to position)

What you are given: The length of each link
The angle of each joint

What you can find: The position of any point

(i.e. it’s (x, y, z) coordinates

Inverse Kinematics (position to angles)

What you are given: The length of each link
The position of some point on the robot

What you can find: The angles of each joint needed to obtain
that position
 Carnegie Mellon

Quick Math Review

Dot Product: a x 
a 
Geometric Representation:  y
b x 
A  B  A B cosθ A b 
θ  y
Matrix Representation: B

a x  b x 
A  B        a xb x  a y b y
a y  b y 

Unit Vector
Vector in the direction of a chosen vector but whose magnitude is 1.
B
uB 
B B
uB
 Carnegie Mellon

Quick Matrix Review

Matrix Multiplication:
An (m x n) matrix A and an (n x p) matrix B, can be multiplied since
the number of columns of A is equal to the number of rows of B.
Non-Commutative Multiplication
AB is NOT equal to BA

a b   e f  ae  bg  af  bh

c d    g  
h  ce  dg  cf  dh
  

Matrix Addition:
a b   e f   a  e  b  f 
c d    g  
h  c  g  d  h
  
 Carnegie Mellon

Basic Transformations
Moving Between Coordinate Frames

Translation Along the X-Axis

Y O

(VN,VO)

VO

P
N
X VN
Px

Px = distance between the XY and NO coordinate planes

V X  V N  Px 
Notation: V XY
  Y V NO
  O P 
V  V  0
 Carnegie Mellon

Writing V XY in terms of V NO

Y O

VO
P 
N
X VN

PX  V N 
V XY
 O   P V NO

 V 
 Carnegie Mellon

O
Translation along the X-Axis and Y-Axis

VO

N
VN

 Px 
X P XY
 
PY 

PX  V N 
V XY  P V NO  O
 Y
P  V 
 Carnegie Mellon

Using Basis Vectors

Basis vectors are unit vectors that point along a coordinate axis
O

n Unit vector along the N-Axis

o Unit vector along the N-Axis
VO
V NO Magnitude of the VNO vector
o
N
n VN

V N   V cosθ  V NO cosθ  V NO  n 
NO

V NO   O    NO    NO    NO 
V   V sinθ   V cos(90  θ)  V  o 
 Carnegie Mellon

Y
Rotation (around the Z-Axis)
Z X
Y

VY

 X
VX

 = Angle of rotation between the XY and NO coordinate axis

V X  V N 
V XY   Y V NO   O
V  V 
 Carnegie Mellon

x
Y Unit vector along X-Axis
V
Can be considered with respect to
the XY coordinates or NO coordinates
V
V XY  V NO
VY
 

X
VX

V X  V XY cosα  V NO cosα  V NO  x
(Substituting for VNO using the N and O
V  (V  n  V  o)  x
X N O
components of the vector)

V X  V N (x  n )  V O (x  o)
 V N (cosθ)  V O (cos(θ  90))
 V N (cosθ)  V O (sinθ)
 Carnegie Mellon

Similarly….

V Y  V NO sinα  V NO cos(90  α)  V NO  y
V Y  (V N  n  V O  o)  y
V Y  V N (y  n)  V O (y  o)
 V N (cos(90  θ))  V O (cosθ)
 V N (sinθ)  V O (cosθ)
So….
V  V (cosθ)  V (sinθ)
X N O V X 
V XY   Y
V Y  V N (sinθ)  V O (cosθ) V 
Written in Matrix Form

V X  cosθ  sinθ V N  Rotation Matrix about the z-axis

V XY
  Y     O
V   sinθ cosθ  V 
 Carnegie Mellon
Y1

(VN,VO)
Y0
VNO

VXY

X1
P
Translation along P followed by rotation by 
X0

V X  Px  cosθ  sinθ V N 

V XY   Y       
V  Py   sinθ cosθ  V O 
(Note : Px, Py are relative to the original coordinate frame. Translation followed by
rotation is different than rotation followed by translation.)

In other words, knowing the coordinates of a point (V N,VO) in some coordinate

frame (NO) you can find the position of that point relative to your original
coordinate frame (X0Y0).
 Carnegie Mellon

HOMOGENEOUS REPRESENTATION
Putting it all into a Matrix

V X  Px  cosθ  sinθ V N  What we found by doing a

XY
  Y       
cosθ  V O 
V translation and a rotation
V  Py   sinθ

V X   Px  cosθ  sinθ 0  V N  Padding with 0’s and 1’s

   
 V Y   Py    sinθ cosθ 0 V O 
 1   1   0 0 1  1 
 

V X  cosθ  sinθ Px  V N 
   
 V Y    sinθ cosθ Py  V O  Simplifying into a matrix form
 1   0 0 1   1 
 
cosθ  sinθ Px 
H   sinθ cosθ Py  Homogenous Matrix for a Translation in
XY plane, followed by a Rotation around
 0 0 1  the z-axis
 Carnegie Mellon

Rotation Matrices in 3D – OK,lets return from

homogenous repn
cosθ  sinθ 0
Rotation around the Z-Axis
R z   sinθ cosθ 0
 0 0 1

 cosθ 0 sinθ 
R y   0 1 0  Rotation around the Y-Axis

 sinθ 0 cosθ

1 0 0 
R z  0 cosθ  sinθ Rotation around the X-Axis

0 sinθ cosθ 

 Carnegie Mellon

Homogeneous Matrices in 3D
H is a 4x4 matrix that can describe a translation, rotation, or both in one matrix
O
1 0 0 Px 
Y
0 1 0 Py 
N H
P
0 0 1 Pz 
X A  
Translation without rotation 0 0 0 1
Z

Y
n x ox ax 0
n oy ay 0
H y
O
N
n z oz az 0
 
X 0 0 0 1
Rotation part:
Z Rotation without translation Could be rotation around z-axis,
A x-axis, y-axis or a combination of
the three.
 Carnegie Mellon

Homogeneous Continued….
V N 
 O The (n,o,a) position of a point relative to the
V
V XY
 H A  current coordinate frame you are in.
V 
 
 1 

n x ox ax Px  V N 
n   O
oy ay Py  V  V X  n x V N  o x V O  a x V A  Px
V XY  y
n z oz az Pz  V A 
  
0 0 0 1   1 

The rotation and translation part can be combined into a single homogeneous
matrix IF and ONLY IF both are relative to the same coordinate frame.
 Carnegie Mellon

Finding the Homogeneous Matrix

EX.

Y W N 
 O
W 
W A 
 
T
P
X

Z
W X   WI  W N 
 Y  Point relative to the  J  Point relative to the  O  Point relative to the
 W  X-Y-Z frame  W  I-J-K frame  W  N-O-A frame
W Z  W K  W A 
     

 W   Pi   n i
 WI   ni oi ai Pi   W N 
I
oi ai  W  N
 J   
 J    
a j   W O 

 W   n j oj aj Pj   W O 
 W    Pj    n j oj
 W K  Pk  n k  W K  n k Pk   W A 
ok a k   W A    
ok ak
   
 1   0 0 0 1   1 
 Carnegie Mellon

W N 
Y  O
W 
W A 
 
T
P
X

Z
 W X  i x jx kx Tx   W I 
 W  Tx  i x
X
jx k x  W  I
 Y   
 Y      Ty   W J 
k y   W J   W   i y jy ky
 W   Ty   i y jy
 W Z  i z jz kz Tz   W K 
 W Z  Tz  i z jz k z   W k      
   1   0 0 0 1   1 

 W X  i x jx kx Tx   n i oi ai Pi   W N 
W   Y   
Ty   n j Pj   W O 
I

 
Substituting for  W J   W   i y jy ky oj aj
W K   W Z  i z jz kz Tz  n k ok ak Pk   W A 
 
     
 1   0 0 0 1  0 0 0 1   1 
 Carnegie Mellon

W X  W N  i x jx kx Tx   n i oi ai Pi 
 Y  O i
 W   H W  jy ky Ty   n j oj aj Pj 
WZ  W A  H y
i z jz kz Tz  n k ok ak Pk 
      
 1   1  0 0 0 1  0 0 0 1
Product of the two matrices

Notice that H can also be written as:

1 0 0 Tx  i x jx kx 0 1 0 0 Pi   n i oi ai 0
0 1 0 Ty  i y jy ky 0 0 1 0 Pj   n j oj aj 0
H
0 0 1 Tz  i z jz kz 0  0 0 1 Pk  n k ok ak 0
    
0 0 0 1  0 0 0 1  0 0 0 1  0 0 0 1

H = (Translation relative to the XYZ frame) * (Rotation relative to the XYZ frame)
* (Translation relative to the IJK frame) * (Rotation relative to the IJK frame)
 Carnegie Mellon

The Homogeneous Matrix is a concatenation of numerous

translations and rotations

W N 
Y  O
W 
W A 
 
T
P
X

Z
One more variation on finding H:

H= (Rotate so that the X-axis is aligned with T)

* ( Translate along the new t-axis by || T || (magnitude of T))
* ( Rotate so that the t-axis is aligned with P)
* ( Translate along the p-axis by || P || )
* ( Rotate so that the p-axis is aligned with the O-axis)

This method might seem a bit confusing, but it’s actually an easier way to
solve our problem given the information we have. Here is an example…
 Carnegie Mellon

Forward Kinematics
 Carnegie Mellon

The Situation:
You have a robotic arm that
starts out aligned with the xo-axis.
You tell the first link to move by 1
and the second link to move by 2.

The Quest:
What is the position of the
end of the robotic arm?

Solution:
1. Geometric Approach
This might be the easiest solution for the simple situation. However,
notice that the angles are measured relative to the direction of the previous
link. (The first link is the exception. The angle is measured relative to it’s
initial position.) For robots with more links and whose arm extends into 3
dimensions the geometry gets much more tedious.

2. Algebraic Approach
Involves coordinate transformations.
 Carnegie Mellon

Example Problem:
You are have a three link arm that starts out aligned in the x-axis.
Each link has lengths l1, l2, l3, respectively. You tell the first one to move by 1
, and so on as the diagram suggests. Find the Homogeneous matrix to get the
position of the yellow dot in the X0Y0 frame.
Y3

3

Y2 2 X3 3
2 X2

H = Rz(1 ) * Tx1(l1) * Rz(2 ) * Tx2(l2) * Rz(3 )

Y0 i.e. Rotating by 1 will put you in the X1Y1 frame.

1
Translate in the along the X1 axis by l1.
Rotating by 2 will put you in the X2Y2 frame.
1
X0
and so on until you are in the X3Y3 frame.

The position of the yellow dot relative to the X3Y3 frame is

(l1, 0). Multiplying H by that position vector will give you the
coordinates of the yellow point relative the the X0Y0 frame.
 Carnegie Mellon

Slight variation on the last solution:

Make the yellow dot the origin of a new coordinate X4Y4 frame
Y3

Y4
3

Y2 2 X3 3
2 X2
X4

H = Rz(1 ) * Tx1(l1) * Rz(2 ) * Tx2(l2) * Rz(3 ) * Tx3(l3)

Y0
1
This takes you from the X0Y0 frame to the X4Y4 frame.
1
X0
The position of the yellow dot relative to the X4Y4 frame
is (0,0).
 X 0 
Y  0 
   H  Notice that multiplying by the (0,0,0,1) vector will
Z 0  equal the last column of the H matrix.
   
 
1 1 
 Carnegie Mellon

More on Forward Kinematics…

Denavit - Hartenberg Parameters

 Carnegie Mellon

Denavit-Hartenberg Notation

Z(i - 1)

Y(i -1) Yi Zi

Xi ai
a(i - 1 ) di
X(i -1)
i
( i - 1)

IDEA: Each joint is assigned a coordinate frame. Using the Denavit-

Hartenberg notation, you need 4 parameters to describe how a frame (i)
relates to a previous frame ( i -1 ).

THE PARAMETERS/VARIABLES: , a , d, 
 Carnegie Mellon

The Parameters

You can
Z(i - 1) align the
two axis
Y(i -1) Yi Zi just using
Xi ai the 4
a(i - 1 ) di parameters
X(i -1)
i
( i - 1)

1) a(i-1)
Technical Definition: a(i-1) is the length of the perpendicular between the joint
axes. The joint axes is the axes around which revolution takes place which are the
Z(i-1) and Z(i) axes. These two axes can be viewed as lines in space. The common
perpendicular is the shortest line between the two axis-lines and is perpendicular
to both axis-lines.
 Carnegie Mellon

a(i-1) cont...
Visual Approach - “A way to visualize the link parameter a(i-1) is to imagine an
expanding cylinder whose axis is the Z(i-1) axis - when the cylinder just touches the
joint axis i the radius of the cylinder is equal to a(i-1).” (Manipulator Kinematics)

It’s Usually on the Diagram Approach - If the diagram already specifies the
various coordinate frames, then the common perpendicular is usually the X(i-1)
axis. So a(i-1) is just the displacement along the X(i-1) to move from the (i-1) frame
to the i frame.

If the link is prismatic, then a(i-1)

is a variable, not a parameter.
Z(i - 1)
Y(i -1) Yi Zi
Xi ai
a(i - 1 ) di
X(i -1)
i
( i - 1)
 Carnegie Mellon

2) (i-1)

Technical Definition: Amount of rotation around the common perpendicular so that

the joint axes are parallel.

i.e. How much you have to rotate around the X(i-1) axis so that the Z(i-1) is pointing
in the same direction as the Zi axis. Positive rotation follows the right hand rule.

3) d(i-1)
Technical Definition: The displacement
Z(i - 1)
along the Zi axis needed to align the a(i-1)
common perpendicular to the ai common Y(i -1) Yi Z
i
perpendicular. Xi ai
a(i - 1 ) di
X(i -1)
In other words, displacement along the i
Zi to align the X(i-1) and Xi axes. ( i -
1)

4) i
Amount of rotation around the Zi axis needed to align the X(i-1) axis with the Xi
axis.
 Carnegie Mellon

The Denavit-Hartenberg Matrix

 cosθ i  sinθ i 0 a(i 1) 

sinθ cosα cosθ i cosα (i 1)  sinα (i 1)  sinα (i 1)d i 
 i (i 1)

 sinθ i sinα (i 1) cosθ i sinα (i 1) cosα (i 1) cosα (i 1)d i 
 
 0 0 0 1 

Just like the Homogeneous Matrix, the Denavit-Hartenberg Matrix is a

transformation matrix from one coordinate frame to the next. Using a series of
D-H Matrix multiplications and the D-H Parameter table, the final result is a
transformation matrix from some frame to your initial frame.

Z(i -
Put the transformation here 1) Y(i - Y Z
1) i i X a
a(i - d
X(i - i i
1) i 
( i 1)
i
- 1)
 Carnegie Mellon

3 Revolute Joints -Y3

Z0,1 Z2

X3
d2
X0,1 X2

-Y2
-Y0,1
Denavit-Hartenberg Link
a0 a1 Parameter Table

i (i-1) a(i-1) di i
Notice that the table has two uses:
1) To describe the robot with its 1 0 0 0 1
variables and parameters.
2) To describe some state of the 2 0 a0 0 2
robot by having a numerical values
for the variables.
3 -90 a1 d2 3
 Carnegie Mellon

Y3
i (i-1) a(i-1) di i
Z0, Z2
1
1 0 0 0 1
X3
d2
X0, X2 2 0 a0 0 2
Y0, 1 Y2
1 3 -90 a1 d2 3
a0 a1

V X 2 
 Y2 
 V  T ( 0T)( 01T)(12T)
V X 0 Y0 Z 0
T Z
V 2 
 
 1  Note: T is the D-H matrix with (i-1) = 0 and i = 1.
 Carnegie Mellon

i (i-1) a(i-1) di i cosθ 0  sinθ 0 0 0

 sinθ cosθ 0 0 0
1 0 0 0 1
0T 
 0

 0 0 1 0
2 0 a0 0 2  
 0 0 0 1
This is just a rotation around the Z0 axis
3 -90 a1 d2 3

cosθ1  sinθ1 0 a0   cosθ 2  sinθ 2 0 a1 

 sinθ cosθ1 0 0   0 0 1 d 2 
0
T   1
2T 
1 
1
 0 0 0 0  sinθ 2  cosθ 2 0 0
   
 0 0 0 1  0 0 0 1
This is a translation by a1 and then d2
This is a translation by a0 followed by a
followed by a rotation around the X2 and
rotation around the Z1 axis
Z2 axis

T ( 0T)( 01T)(12T)
 Carnegie Mellon

Next Example
 Carnegie Mellon

Puma
 Carnegie Mellon

Alternative PUMA 560

 Carnegie Mellon

Inverse Kinematics
From Position to Angles
 Carnegie Mellon

A Simple Example

Revolute and
Finding :
Prismatic Joints
Combined θ  arctan( y, x)
More Specifically:
(x , y)
arctan2() specifies that it’s in the
θ  arctan 2( y, x) first quadrant

S
1 Finding S:
X

S  (x 2  y 2 )
 Carnegie Mellon

Inverse Kinematics of a Two Link Manipulator

(x , y) Given: l1, l2 , x , y

2 Find: 1, 2
l2 Redundancy:
A unique solution to this problem
does not exist. Notice, that using the
“givens” two solutions are possible.
1
l1
Sometimes no solution is possible.

(x , y)
 Carnegie Mellon
The Geometric Solution
Using the Law of Cosines:
(x , y)
l2 2 c 2  a 2  b 2  2ab cos C
( x 2  y 2 )  l1  l2  2l1l2 cos(180  θ 2 )
2 2

cos(180  θ 2 )   cos(θ 2 )
x 2  y 2  l1  l2
2 2
l1
 cos(θ 2 ) 
2l1l2
1
 x 2  y 2  l12  l2 2 
θ 2  arccos  

Using the Law of Cosines:
 2l1l2 
Redundant since 2 could be in the
sin B sin C
 first or fourth quadrant.
b c
sin θ1 sin(180  θ 2 ) sin(θ 2 )
  Redundancy caused since 2 has two possible
l2 x 2  y2 x 2  y2 values

θ1  α  θ1  l sin(θ ) 
θ1  arctan 2 y, x   arcsin  2 2 

y  x 2  y2 
α  arctan 2   
x
 Carnegie Mellon
The Algebraic Solution
(x , y)
l2 2

c1  cosθ1
c1 2  cos(θ 2  θ1 )
l1 (1) x  l1 c1  l2 c1 2
(2) y  l1 s1  l2 sin 1 2
1

(1) 2  (2) 2  x 2  y 2 
 2 2
 
 l1 c1  l2 (c1 2 ) 2  2l1l2 c1 (c1 2 )  l1 s1  l2 (sin1 2 ) 2  2l1l2 s1 (sin1 2 )
2 2 2 2

 l1  l2  2l1l2 c1 (c1 2 )  s1 (sin1 2 ) 
2 2

 l1  l2  2l1l2 c 2
2 2 Only Unknown Note :
 x 2  y 2  l12  l2 2  cos(ab)  (cos a)(cos b)  (sin a)(sin b)
 θ 2  arccos  
 sin(ab)  (cos a)(sin b)  (cos b)(sin a)
 2 l l
1 2 
 Carnegie Mellon

x  l1 c1  l2 c1 2 Note :
 l1 c1  l2 c1c 2  l2 s1s2 cos(ab)  (cos a)(cos b)  (sin a)(sin b)
 c1 (l1  l2 c 2 )  s1 (l2 s2 ) sin(ab)  (cos a)(sin b)  (cos b)(sin a)

y  l1 s1  l2 sin 1 2 We know what 2 is from the previous

 l1 s1  l2 s1c 2  l2 s 2 c1 slide. We need to solve for 1 . Now
 c1 (l2 s 2 )  s1 (l1  l2 c 2 ) we have two equations and two
unknowns (sin 1 and cos 1 )

 x  l1  l2 c 2  l2 s 2  c1 

 y   l s l  l c   s 
   2 2 1 2 2  1  Substituting for c1 and simplifying
many times

Notice this is the law of cosines

and can be replaced by x2+ y2

1  arctan 2(s1 , c1 )
 Carnegie Mellon

x  l1 c1  l2 c1 2 Note :
 l1 c1  l2 c1c 2  l2 s1s2 cos(ab)  (cos a)(cos b)  (sin a)(sin b)
 c1 (l1  l2 c 2 )  s1 (l2 s2 ) sin(ab)  (cos a)(sin b)  (cos b)(sin a)

y  l1 s1  l2 sin 1 2 We know what 2 is from the previous

 l1 s1  l2 s1c 2  l2 s 2 c1 slide. We need to solve for 1 . Now
 c1 (l2 s 2 )  s1 (l1  l2 c 2 ) we have two equations and two
x  s1 (l2 s2 ) unknowns (sin 1 and cos 1 )
c1 
(l1  l2 c 2 )
x  s1 (l2 s2 )
y (l2 s 2 )  s1 (l1  l2 c 2 ) Substituting for c1 and simplifying
(l1  l2 c 2 ) many times


1
(l1  l2 c 2 )
 
Notice this is the law of cosines
x l2 s2  s1 (l1  l2  2l1l2 c 2 ) and can be replaced by x2+ y2
2 2

y(l1  l2 c 2 )  x l2 s2 1  arctan 2(s1 , c1 )

s1 
x 2  y2