Professional Documents
Culture Documents
and Manipulability
Howie Choset
(with notes copied from Rob Platt)
Forward Kinematics is in general many to one (if more joint angles than DOF in WS),
non linear
Inverse Kinematics is in general one to many (have a choice), non linear
Forward differential kinematics, linear, many to one, etc. based on # of DOF and WS
Inverse is easier because linear, but this many to one issue comes up
Cartesian control
y2
q3
q 1
x
y J q
2
q 1
1 x
q J y
2
z1
l2
x0
y0
x1
z0
l1
q1
y3
l3
z2
y1
q2
0
l
c
l
c
l
c
2 2
3 23
3 23
x2
z3
x3
Cartesian control
xd
y2
qd
q3
joint position
sensor
z1
l2
x1
z0
z3
l1
q1
x0
q 1
1 x
q J y
2
y3
l3
z2
y1
joint ctlr
q2
FK (q )
x2
y0
Columns span
Cartesian space, or
x3
Cartesian control
3
q 1
J q q 2
q 3
l1s1 l2 s12
l1c1 l2 c12
y
l3
l1s1
l1c1
l2
q3
1
q2
y
q1
multiple solutions
l1
Generalized inverse
3
what next?
y
l3
2
We have:
l2
q3
x Jq
x Jq
q2
y
q1
q J # x
l1
Generalized inverse
Two cases:
Underconstrained manipulator (redundant)
Overconstrained
Generalized inverse:
x s.t.
Jq
Is minimized
y
l3
2
l2
q3
1
q2
y
q1
Minimize f (z ) subject to g ( z ) 0 :
Use lagrange multiplier method: z f ( z ) z g ( z )
l1
q f (q ) q T
q g (q ) J
q T T J
q J T
Minimize
Subject to
Jq JJ T
JJ
JJ
T 1
T 1
q J T
q J JJ
T
T 1
J J JJ
q J # x
#
Jq
T 1
J # J T JJ T
J UV
m m m n
n n
1 0
0
2
J U 0
0
0
0
0
1
1
J# V 0
0
0
0
0
0
0
0
3 0 0
0 0
0 0 n
J # V 1U T
0
0
1
2
1
3
0
0
0
0
0 0
0 0
0 0 V T
0 0
0 0
0
0
0
0 UT
1
n
0
0
1. J # JJ # J #
2. JJ # J J
4. J J
3. JJ
# T
JJ #
J #J
J J
J J
# #
# T
T #
xd
J#
FK (q )
qd
joint ctlr
joint position
sensor
xerr xerr
Rd
Rc
Rcd Rc Rd
Rd
Rc
FK (q )
qd
joint ctlr
joint position
sensor
r
q
Rd
Rc
FK (q )
qd
joint ctlr
joint position
sensor
r
q
TA r J q
0 s
J A TA r J 0 c
1
0
y
l3
2
l2
q3
1
q2
y
q1
c s
s s J
c
l1
y
l3
2
l2
q3
1
rk J q
Axis angle delta
rotation
rerr rerr
q2
y
q1
l1
Rk e S k I S k sin S k 1 cos
2
axis
angle
p S b b p
Rt e
S b t
x Jq
#
J
x
2.
1.
e 12 xerr xerr
e
T x
xerr
q
q
Position error:
x
q xerr T
x
q
q
xerr xref x
xerr
T
q J v xerr
curr
k ref
q J
xerr
e 12 xerr xerr
q J T
or
q J #
xd
xd
Jq
JTw
Joint torques
Cartesian
wrench:
f
w
m
force
moment (torque)
minimizes
Singularities
Singularities
Manipulability
A measure of the Jacobian
Configuration dependent
Maybe it is a measure of how far away
from a singularity
Norms
Scalar
Vector
Matrix
Manipulability Ellipsoid
Can we characterize how close we are to a singularity?
Yes imagine the possible instantaneous motions are described by
an ellipsoid in Cartesian space.
Cant move much this way
Can move a lot this way
Manipulability Ellipsoid
The manipulability ellipsoid is an ellipse in
Cartesian space corresponding to the twists
that unit joint velocities can generate:
q T q 1
J x
T
J # x 1
x J JJ
JJ
x T JJ T
x T
JJ x 1
JJ x 1
T
T 1
T 1
JJ T
x 1
T 1
T 1
Manipulability Ellipsoid
You can calculate the directions and magnitudes
of the principle axes of the ellipsoid by taking
the eigenvalues and eigenvectors of JJ T
v1
v2
det JJ T
Manipulability Ellipsoid
Another characterization of the
manipulability ellipsoid: the ratio of the
largest eigenvalue to the smallest
eigenvalue:
v1
v2
Manipulability Ellipsoid
Isotropic manipulability
ellipsoid
NOT isotropic
manipulability ellipsoid
T 1
JTF
J F
T
JTF 1
F T JJ T F 1
Manipulability Ellipsoid
Principle axes of the force manipulability ellipsoid: the
T 1
eigenvalues and eigenvectors of:
JJ
JJ
T 1
vi
Velocity ellipsoid
l
c
l
c
l
c
2 12
1 1 2 12
12
J q
1
1
1
2
1
2
J q J q
T
1
2
Principle axes:
1
2
2 2
- 0.3029
- 0.1568
- 0.9530
2 v2
1.8411
1 v1
2 v2
1 v1
Eigenvectors
Ellipsoid analogy
Singular Values
Eigenvalues are for squares (remember
the determinant)