Robotics 1

Inverse kinematics
Prof. Alessandro De Luca

Robotics 1

Inverse kinematics
what are we looking for?

Robotics 1

direct kinematics is always unique;

how about inverse kinematics for this 6R robot?

Inverse kinematics problem

!

given a desired end-effector pose (position +

orientation), find the values of the joint variables
that will realize it
a synthesis problem, with input data in the form
!

T=

R p
000 1

classical formulation:
inverse kinematics for a given end-effector pose

" r =

p
!

, or any other task vector

generalized formulation:
inverse kinematics for a given value of task variables

a typical nonlinear problem

!
!
!

Robotics 1

existence of a solution (workspace definition)

uniqueness/multiplicity of solutions
solution methods
3

Solvability and robot workspace

!

primary workspace WS1: set of all positions p that can be

reached with at least one orientation (! or R)
!
!

secondary (or dexterous) workspace WS2: set of positions p

that can be reached with any orientation (among those
feasible for the robot direct kinematics)
!

out of WS1 there is no solution to the problem

for p " WS1 and a suitable ! (or R) there is at least one solution

for p " WS2 there is at least one solution for any feasible ! (or R)

WS2 # WS1

Robotics 1

Workspace of Fanuc R-2000i/165F

section for a
constant angle $1

WS1

(! WS2 for spherical wrist

without joint limits)

rotating the
base joint angle $1
Robotics 1

Workspace of planar 2R arm

2 orientations
y
l2
l1

p
q2

|l1-l2|

q1

(WS1

x
!

if l1 % l2
!
!

l1+l2

inner and outer

boundaries

1 orientation

WS1 = {p " R2: |l1-l2| & !p!& l1+l2}

WS2 = '

if l1 = l2 = "
!
!

Robotics 1

WS1 = {p " R2: !p!& 2"}

WS2 = {p = 0} (infinite number of feasible orientations at the origin)
6

Wrist position and E-E pose

inverse solutions for an articulated 6R robot

LEFT DOWN

Unimation PUMA 560

RIGHT DOWN

4 inverse solutions
out of singularities
(for the position of
the wrist center only)

LEFT UP

Robotics 1

8 inverse solutions considering

the complete E-E pose
(spherical wrist: 2 alternative
solutions for the last 3 joints)

RIGHT UP

Counting and visualizing the 8 solutions

to the inverse kinematics of a Unimation Puma 560

RIGHT UP

RIGHT DOWN

LEFT UP

LEFT DOWN

Robotics 1

Multiplicity of solutions
some examples

E-E positioning of a planar 2R robot arm

!
!
!

singular solutions

E-E positioning of an articulated elbow-type 3R robot arm

!

2 regular solutions in WS1

1 solution on (WS1
for l1 = l2: ) solutions in WS2
4 regular solutions in WS1

spatial 6R robot arms

!

& 16 distinct solutions, out of singularities: this upper bound of

solutions was shown to be attained by a particular instance of
orthogonal robot, i.e., with twist angles *i = 0 or +/2 (,i)
analysis based on algebraic transformations of robot kinematics
!

Robotics 1

transcendental equations are transformed into a single polynomial

equation of one variable
seek for an equivalent polynomial equation of the least possible degree
9

A planar 3R arm

workspace and number/type of inverse solutions

q3

y
l2
l1

q1

q2

l1 = l2 = l3 = "

l3

WS1 = {p " R2: !p!& 3"}

WS2 = {p " R2: !p!& "}
x

any planar orientation is feasible in WS2

1. in WS1 : )1 regular solutions (except for 2. and 3.),

at which the E-E can take a continuum of
) orientations (but not all orientations in the plane!)
2. if !p!= 3" : only 1 solution, singular
3. if !p!=

" : )1 solutions, 3 of which singular

4. if !p!<

" : )1 regular solutions (never singular)

Robotics 1

10

Multiplicity of solutions
summary of the general cases

if m = n
!
!
!

if m < n (robot is redundant for the kinematic task)

!
!
!

! solutions
a finite number of solutions (regular/generic case)
degenerate solutions: infinite or finite set, but anyway
different in number from the generic case (singularity)
! solutions
)n-m solutions (regular/generic case)
a finite or infinite number of singular solutions

use of the term singularity will become clearer when dealing

with differential kinematics
!
!

instantaneous velocity mapping from joint to task velocity

lack of full rank of the associated m"n Jacobian matrix J(q)

Robotics 1

11

Dexter robot (8R arm)

!
!
!

m = 6 (position and orientation of E-E)

n = 8 (all revolute joints)
)2 inverse kinematic solutions (redundancy degree = n-m = 2)

video

exploring inverse kinematic solutions by a self-motion

Robotics 1

12

Solution methods
ANALYTICAL solution
(in closed form)
" preferred, if it can be found*
" use ad-hoc geometric inspection
" algebraic methods (solution of
polynomial equations)
" systematic ways for generating a
reduced set of equations to be
solved
*

sufficient conditions for 6-dof arms

3 consecutive rotational joint axes are

incident (e.g., spherical wrist), or
3 consecutive rotational joint axes are
parallel

NUMERICAL solution
(in iterative form)
" certainly needed if n>m (redundant
case), or at/close to singularities
" slower, but easier to be set up
" in its basic form, it uses the
(analytical) Jacobian matrix of the
direct kinematics map

(fr (q)
Jr(q) =
(q

" Newton method, Gradient method,

and so on

D. Pieper, PhD thesis, Stanford University, 1968

Robotics 1

13

Inverse kinematics of planar 2R arm

y
py

l2
l1

direct kinematics

px = l1 c1 + l2 c12

q2

py = l1 s1 + l2 s12

q1
px

data

q1, q2 unknowns

squaring and summing the equations of the direct kinematics

px2 + py2 - (l12 + l22) = 2 l1 l2 (c1 c12 + s1 s12) = 2 l1 l2 c2
and from this
c2 = (px2 + py2 - l12 - l22)/ 2 l1 l2, s2 = -1 - c22
must be in [-1,1] (else, point p
is outside robot workspace!)
Robotics 1

in analytical form

q2 = ATAN2 {s2, c2}

2 solutions
14

Inverse kinematics of 2R arm

y
py

(contd)

by geometric inspection

q1 = * - .

q2
.
q1

*
px

2 solutions
(one for each value of s2)

x
q1 = ATAN2 {py, px} - ATAN2 {l2 s2 , l1 + l2 c2}

q2

{q1,q2}UP/LEFT

q2
q1
q1

Robotics 1

note: difference of ATAN2 needs

to be re-expressed in (-+ , +]!

{q1,q2}DOWN/RIGHT
q2 e q2 have same absolute
value, but opposite signs
15

Algebraic solution for q1

another
solution
method

px = l1 c1 + l2 c12 = l1 c1 + l2 (c1 c2 - s1 s2)

py = l1 s1 + l2 s12 = l1 s1 + l2 (s1c2 + c1s2)
l1 + l2c2 - l2s2
l2s2

l1 + l2c2

c1
s1

det = (l12 + l22 + 2 l1l2c2) > 0

linear in
s1 and c1

px
py
except for l1=l2 and c2=-1
being then q1 undefined
(singular case: )1 solutions)

q1 = ATAN2 {s1, c1} = ATAN2 {(py(l1+l2c2)-pxl2s2)/det, (px(l1+l2c2)+pyl2s2)/det}

Robotics 1

notes: a) this method provides directly the result in (-+ , +]

b) when evaluating ATAN2, det > 0 can be eliminated
from the expressions of s1 and c1

16

Inverse kinematics of polar (RRP) arm

pz

px = q3 c2 c1

q3

q2 is NOT a
DH variable!

py = q3 c2 s1
pz = d1 + q3 s2

q2
d1

px

px2 + py2 + (pz - d1)2 = q32

py
q1

q3 = + -px2 + py2 + (pz - d1)2

our choice: take here only the positive value...

if q3 = 0, then q1 and q2 remain both undefined (stop); else

(if it stops,
q2 = ATAN2{(pz - d1)/q3, -(px +
}
a singular case:
)2 or )1
if px2+py2 = 0, then q1 remains undefined (stop); else
solutions)
q1 = ATAN2{py/c2, px/c2} (2 regular solutions {q1,q2,q3})
2

Robotics 1

py2)/q32

we have eliminated q3>0 from both arguments!

17

Inverse kinematics
for robots with spherical wrist
j6

first 3 robot joints

of any type (RRR, RRP, PPP, )

j5

z0
y0
x0

j1

z4

z3

z5

j4

d6

y6

x6
O6 = p
z6 = a

find q1, , q6 from the input data:

p (origin O6)
R = [n s a] (orientation of RF6)

1. W = p - d6 a / q1, q2, q3 (inverse position kinematics for main axes)

2. R = 0R3(q1, q2, q3) 3R6(q4, q5, q6) / 3R6(q4, q5, q6) = 0R3T R / q4, q5, q6
(inverse orientation
known, Euler ZYZ or ZXZ
given
kinematics for wrist)
after 1. rotation matrix
Robotics 1

18

6R example: Unimation PUMA 600

spherical
wrist

n = 0x6(q)

s = 0y6(q)
a function of
q1, q2, q3 only!

a = 0z6(q)
p = 06(q)

Robotics 1

here d6=0,
so that 06=W directly

8 different inverse solutions

that can be found in closed form
(see Paul, Shimano, Mayer; 1981)

19

Numerical solution of
inverse kinematics problems
!

use when a closed-form solution q to rd = fr(q) does not exist

or is too hard to be found
(fr
(analytical Jacobian)
Jr(q) =
(q
Newton method (here for m=n)
!

rd = fr(q) = fr(qk) + Jr(qk) (q - qk) + o(!q - qk!2)

0 neglected

qk+1 = qk + Jr-1(qk) [rd - fr(qk)]

!
!
!
!

convergence if q0 (initial guess) is close enough to some q*: fr(q*) = rd

problems near singularities of the Jacobian matrix Jr(q)
in case of robot redundancy (m<n), use the pseudo-inverse Jr#(q)
has quadratic convergence rate when near to solution (fast!)

Robotics 1

20

Operation of Newton method

!
!

in the scalar case, also known as method of the tangent

for a differentiable function f(x), find a root of f(x)=0 by iterating as

x k+1

f(x k )
= xk "
f'(x k )

an approximating sequence

{x1 , x 2 , x 3 , x 4 , x 5 ,...}
animation from
http://en.wikipedia.org/wiki/File:NewtonIteration_Ani.gif

!
Robotics 1

21

Numerical solution of
inverse kinematics problems
!

(contd)

Gradient method (max descent)

!

minimize the error function

H(q) = # !rd - fr(q)!2 = # [rd - fr(q)]T [rd - fr(q)]

qk+1 = qk - * 1qH(qk)
from 1qH(q) = - JrT(q) [rd - fr(q)], we get

qk+1 = qk + * JrT(qk) [rd - fr(qk)]

!

the scalar step size * > 0 should be chosen so as to guarantee a

decrease of the error function at each iteration (too large values
for * may lead the method to miss the minimum)
when the step size * is too small, convergence is extremely slow

Robotics 1

22

Revisited as a feedback scheme

rd

+
-

Jr

T(q)

.
q

q(0)

2
3

r
fr(q)

rd = cost
(* = 1)

e = rd - fr(q) / 0 4 closed-loop equilibrium e=0 is asymptotically stable

V = # eTe 5 0 Lyapunov candidate function
.
.
.
T
d
(
T
T
V=e e=e
(rd - fr(q)) = - e Jr q = - eT Jr JrTe & 0
dt
.
in particular e = 0
V = 0 4 e " Ker(JrT)
asymptotic stability
Robotics 1

23

Properties of Gradient method

!

!
!
!

computationally simpler: Jacobian transpose, rather than its

(pseudo)-inverse
direct use also for robots that are redundant for the task
may not converge to a solution, but it never diverges
the discrete time evolution of the continuous scheme

qk+1 = qk + 6T JrT(qk) [rd - f(qk)]

!

(* = 6T)

is equivalent to an iteration of the Gradient method

scheme can be accelerated by using a gain matrix K>0

.
q = JrT(q) K e

note: K can be used also to escape from being stuck in a stationary point,
by rotating the error e out of the kernel of JrT (if a singularity is encountered)
Robotics 1

24

A case study

analytic expressions of Newton and gradient iterations

!
!

2R robot with l1 = l2 = 1, desired end-effector position rd = pd = (1,1)

direct kinematic function and error
"1%
" c1 + c12 %
e = p d - fr (q) = $ ' - fr (q)
'
fr (q) = $
#1&
# s1 + s12 &
Jacobian matrix
" fr (q) #-(s1 + s12 ) -s12 &
Jr (q) =
=%
(
c
+
c
c
$
"q
1
12
12 '
!
!
Newton versus Gradient iteration
det Jr(q)

k+1

=q +

!
Robotics 1

1
s2

Jr -1(qk )

" c12
%
s12
$
'
-(c
+
c
)
-(s
+
s
)
# 1 12
1
12 &

!#
-(s1 + s12 ) c1 + c12 &
" %
(
-s
c
$
'
12
12
!
Jr T (qk )

q=q

ek

"1 - (c1 + c12 )%

$1 - (s + s )'
#
1
12 &

q=q k

q=q k

25

Error function
!

2R robot with l1=l2=1, desired end-effector position pd = (1,1)

e = pd - fr(q)

plot of !e!2 as a function of q = (q1,q2)

Robotics 1

two local minima

(inverse kinematic solutions)
26

Error reduction by Gradient method

!
!

flow of iterations along the negative (or anti-) gradient

two possible cases: convergence or stuck (at zero gradient)

start
one solution

local maximum

(stop if this is the initial guess)

saddle point

(stop after some iterations)

another start...
...the other solution

(q1,q2) =(0,!/2) (q1,q2) =(!/2,-!/2)

Robotics 1

(q1,q2)max =(-3!/4,0) (q1,q2)saddle =(!/4,0)

e " Ker(JrT) !

27

Convergence analysis

when does the gradient method gets stuck?

!

lack of convergence occurs when

!
!

the Jacobian matrix Jr(q) is singular (the robot is in a singular configuration)

AND the error is in the null space of JrT(q)
p
"1%
pd = $ '
#1&
q2

q1

#1 - 2 &
e = pd " p = %
(
$1 - 2 '

pd

"( 2
" -(s1 + s12 ) c1 + c12 %
J (q) = $
=$
'
-s12
c12 & q=q
#
#( 2
saddle
!
T
r

pd

(q1,q2)saddle =(!/4,0)

2%
'
2&

e " Ker(JrT)

e " Ker(JrT)
p
Robotics 1

(q1,q2)max =(-3!/4,0)

pd

e Ker(JrT) !!
p

(q1,q2) =(!/9,0)
28

Issues in implementation
initial guess q0

!
!

only one inverse solution is generated for each guess

multiple initializations for obtaining other solutions

optimal step size * in Gradient method

a constant step may work good initially, but not close to the
solution (or vice versa)
an adaptive one-dimensional line search (e.g., Armijos rule) could
be used to choose the best * at each iteration

stopping criteria

Cartesian error

(possibly, separate for !rd

position and orientation)
!

f(qk)!

!qk+1-qk! $ #q

understanding closeness to singularities

$min{J(qk)} % $0
Robotics 1

$#

algorithm
increment

numerical conditioning
of Jacobian matrix (SVD)

(or a simpler test on its determinant, for m=n)

29

Numerical tests on RRP robot

!

RRP/polar robot: desired E-E position rd = pd = (1, 1 ,1)

see slide 17, d1=0.5

the two (known) analytic solutions, with q3 % 0, are:

q* = (0.7854, 0.3398, 1.5)
q** = (q1*- +, + - q2*, q3*) = (-2.3562, 2.8018, 1.5)

# = 10-5 (max Cartesian error), #q =10-6 (min joint increment)

norms

kmax=15 (max iterations), |det(Jr)| $ 10-4 (closeness to singularity)

numerical performance of Gradient (with different

! test 1: q0 = (0, 0, 1) as initial guess

*) vs. Newton

test 2: q0 = (-+/4, +/2, 1) singular start, since c2=0 (see slide 17)

test 3: q0 = (0, +/2, 0) double singular start, since also q3=0

solution and plots with Matlab code

Robotics 1

30

Numerical test
!

-1

test 1: q0 = (0, 0, 1) as initial guess; evolution of error norm

Gradient: * = 0.5

Gradient: * = 1
too large, oscillates
around solution

slow, 15 (max)
iterations

good, converges
in 11 iterations

0.5710-5

Newton
very fast, converges
in 5 iterations

Gradient: * = 0.7

Cartesian errors
component-wise

ex

ey

ez

Robotics 1

0.1510-8

31

Numerical test
!

-1

test 1: q0 = (0, 0, 1) as initial guess; evolution of joint variables

Gradient:

*=1

not converging
to a solution

Gradient:

* = 0.7

converges in
11 iterations

Newton
converges in
5 iterations

both to solution q* = (0.7854, 0.3398, 1.5)

Robotics 1

32

Numerical test
!

-2

test 2: q0 = (-+/4, +/2, 1): singular start

error norms

Newton

with check of
singularity:
blocked at start
without check:
it diverges!

Gradient
* = 0.7

starts toward
solution, but
slowly stops
(in singularity):
when Cartesian error
vector e Ker(JrT)

Robotics 1

joint variables

!!

33

Numerical test

test 3: q0 = (0, +/2, 0): double singular start

Robotics 1

0.9610-5

Gradient (with * = 0.7)

starts toward solution
exits the double singularity
slowly converges in 19
iterations to the solution
q*=(0.7854, 0.3398, 1.5)

Newton
is either
blocked at start
or (w/o check)
explodes (NaN)!!

joint variables

Cartesian errors

error norm

-3

34

Final remarks
!

an efficient iterative scheme can be devised by combining

!

joint range limits are considered only at the end

!

check if the found solution is feasible, as for analytical methods

if the problem has to be solved on-line

!

initial iterations with Gradient method (sure but slow, having linear
convergence rate)
switch then to Newton method (quadratic terminal convergence rate)

execute iterations and associate an actual robot motion: repeat steps

at times t0, t1=t0+T, ..., tk=tk-1+T (e.g., every T=40 ms)
the good choice for the initial q0 at tk is the solution of the previous
problem at tk-1 (gives continuity, needs only 1-2 Newton iterations)
crossing of singularities and handling of joint range limits need
special care in this case

Jacobian-based inversion schemes are used also for kinematic

control, along a continuous task trajectory rd(t)

Robotics 1

35