Compensating Joint Configuration through Null Space Control in Composite Weighted Least Norm Solution

Avik Chatterjee, S. Majumder & I. Basak
International Journal of Robotics and Automation (IJRA), Volume (4) : Issue (1) : 2013 31
Compensating Joint Configuration through Null Space Control
in Composite Weighted Least Norm Solution
Avik Chatterjee avik@cmeri.res.in
CSIR-CMERI
MG Avenue
Durgapur, 713209, India
S. Majumder sjm@cmeri.res.in
CSIR-CMERI
MG Avenue
Durgapur, 713209, India
I. Basak indrajit.basak@me.nitdgp.ac.in
Department of Mechanical Engineering NIT
MG Avenue
Durgapur,713209,India
Abstract
We have presented a methodology for compensating joint configuration by composite weighting
in different sub spaces. It augments the weighted least norm solution by weighted residual of the
current joint rate and preferred pose rate in null space, so that we can arrive at a solution which is
able to handle both joint limits and preferred joint configuration simultaneously satisfying the
primary task. The null space controller is formulated in conjunction with the work space controller
to achieve the objective. The contribution of null space has been discussed in the formulation in
two different situations including joint limits, workspace and near configuration singularities.
Keywords: Null Space Controller, Weighted Least Norm, Joint Limit, Singularity, Joint
Configuration.
1. INTRODUCTION
A robotic manipulator in general sense or an articulated serial structure in particular is
kinematically redundant when the number of operational space variables necessary to specify a
given task, is less than the number of joints. Redundancy leads to infinite solutions for the joint
space but offers greater flexibility and dexterity in motion as different constraint based or goal
based criteria can be formulated as sub tasks in the solution. Two kinds of approaches have been
reported in the literature to deal with this situation. One is set to exploit the null space of the
Jacobian matrix in the homogeneous solution that infuses self motion of joints without affecting
the task space. Typical method of this kind is gradient projection method (GPM) [1][2]. In GPM
the anti-gradient of a quadratic cost function, is projected in the null space of the task Jacobian,
which is reminiscent of the projected gradient method for constrained minimization. The other
approach is weighted least norm (WLN) approach [3][4], which minimizes the weighted norm of
joint rate. In both the cases the primary task is to follow the prescribed trajectory and there may
be multiple secondary tasks or nested subtasks with priority fixation [5] [6].
GPM has been used in Joint Limit Avoidance (JLA), obstacle avoidance [7], visual servoing [4].
WLN which was introduced in JLA in [3], has been successfully exploited by others with single or
multiple criteria and Close Loop Inverse Kinematics (CLIK)[8]. Null space based motion control [9]
has been studied with configuration optimization [10], influence of un-weighted and inertia
weighted pseudoinverse [11], proportional-integral-derivative (PID) controller considering
passivity [12], task priority implementation based on behavioral scheme [13]. An elaborate

discussion with illustrations on various pros and cons of different approaches for operational
space control with null space contributions has been reported in [14].
In practice, many subtasks are often needed for the control of manipulator. For example both the
joint limits and the joint configuration became the basic requirements where human motion
analysis is concerned. In many cases, local optimality of GPM may not provide good performance
to all prioritized subtasks. WLN method is effective only for the joints limits but direct optimization
of the weighted norm sum of all tasks may lead to the poor performance for all tasks. The ability
of WLN to effectively handle joint limits and the self motion from null space, motivate us to
presents a methodology of composite weighted least norm (CWLN) solution in conjunction with
GPM. It is so called because the formulation tries to minimize the primary task objective of
weighted norm of joint rate in range space and the weighted residual of the current joint rate ( )q&
and preferred pose rate ( )rq& in null space (hence composite weighting in different sub spaces) so
that we can arrive at a solution which is able to handle both joint limits and preferred joint
configuration simultaneously satisfying the primary task.
This paper is organized as follows: Section II formulates the CWLN method from classical
redundancy control methods. Section III discusses stability of the CWLN method and its
regularized version. The case studies are illustrated in Section IV. Section V concludes the paper.
2. COMPOSITE WEIGHTED LEAST NORM
We focus on first and second order kinematics for the time variant task space defined as
1
( ) m
x t ×
∈ℜ and joint space 1
( ) n
q t ×
∈ℜ related by the direct kinematic non linear and
transcendental vector function ( )tk q , whose time differentiation will define the non square
analytic Jacobian matrix ( ) ( ) / ;ij j m n
t t iJ q J q k q n > m×
∂ ∂ ∈ℜ ∀ , with its assumption of bounded
higher order terms and linearization. We denote the desired task space positions, velocities, and
accelerations as ,d d dx x and x& && respectively and reference or preferred joint configuration as rq
Dropping the subscript t for brevity, the classical forward kinematics differential relationships can
be expressed as
( ) ; ( ) ( , )x J q q and x J q q J q q q= = + && & && && & & (1)
and inverse kinematics least norm (LN) general solution as
† † † †
1 2( ) ; ( ) ( )p h dq q q J x I J J q J x Jq I J Jξ ξ= + = + − = − + −&& & & & && && & (2)
where ( )pq J∈ℜ& is particular solution, ( )hq J∈ℵ& is homogeneous solution, † 1
( )T T
J J JJ −
is
the right pseudoinverse of the Jacobian, 1
1 2
n
andξ ξ ×
∈ℜ are arbitrary vectors and †
( )I J J− is the
null space projector. The Weighted Least Norm (WLN) solution formulates the problem as
2
1 1( )[ ( )] ( ) ( )[ ],T
min q q min q q min q q W q= =& & & & & & &H st ( ) 0x Jq− =& & , 1
n n
W ×
∀ ∈ is the symmetric positive
definite weighing matrix. To stabilize the ill posed condition of LN or WLN solution near
singularities, Tikhonov like regularization has been used, which makes a trade off between
tracking accuracy and the feasibility of the joint velocities, known as classical Damped Least
Square (DLS) solution. The trade off parameter is the damping factor α . If the objective is
specified through a configuration rate dependent performance criteria 2 ( )q&H , set to be the closest
to some particular pose, hence forth called the reference configuration ( )rq the problem can be

reformulated as 2 2)[ ( )] ( )[(1/ 2)( ) ( )];T
r rmin(q q min q q q W q q= − −& & & & & & &H . ;s t Jq x=& & 2
n n
W ×
∀ ∈ .In our
approach an augmented objective function has been formulated by combining configuration rate
dependent performance criteria 2 ( )q&H for pose optimization and 1( )q&H for joint limit avoidance,
subjected to the requirement of primary task space ( ) 0x Jq− =& & , as
2 1 2( ) ( ) ( ) ( , ) ,n n
q q q and W W ×
∀ = + ∀ ∈& & &3 1H H H henceforth called as Composite Weighted Least
Norm Solution (CWLNS) as,
1 2( ) ( ) ( )[(1/ 2) (1/ 2)( ) ( )]; .T T
r rmin q q min q q W q q q W q q s t Jq x= + − − =& & & & & & & & & & &3H (3)
To solve this optimization problem with equality constraint, it should satisfy both the necessary
condition 0q L∇ =& and sufficient condition 2
0q L∇ >& , where the Lagrangian is
( , ) ( ) ( )L q q Jq xλ λ= + −& & & &3H and we can directly evaluate 2
1 2( ) 0q L W W∇ = + >& , which is true for
minimization. Putting the value of q& from 0q L∇ =& in the expression 0Lλ∇ = , we get λ .
Substituting λ back in q& from 0q L∇ =& , and 1 1 1
( ) ,T T
J W J JW J− − −
∀ h
1 2( )W W W∀ + , 1 ,rqξ∀ &
the general solution of CWLS reduces to [Appendix-I.A]
1
2 1( )q J x I J J W W ξ−
= + −h h
& & (4)
It is trivial to show 1
2( )I J J W W−
− h
is the null space projector of reference joint rate vector rq& and
hence no impact on task space as JJ I=h
. The optimization in the direction of the anti-gradient
of scalar configuration dependent performance criteria 3 ( )qH can also be set up by minimizing
3 ( )qH for weighted reference configuration ( )rq as
3 2 3 2( ) (1/ 2)( ) ( ); ( ) ( )T
r r q rq q q W q q q W q q= − − ⇒ ∇ = −H H (5)
and for a positive scalar Hk and 1
1 1 2 2 3( ) ( )H qk W W W qξ −
′∀ − + ∇ H the GPM flavor of CWLS
formulation is
1
2 1 2( ) ; ( ) ( )dq J x I J J W W q J x Jq I J Jξ ξ−
′= + − = − + −h h h h&& & && && & (6)
Using Eq.(2), the relation h h
JJ JJ= −& & and after simplification we can establish the relation
between 2ξ and 1ξ as.
2 1 1( )h
J J qξ ξ ξ′= − + && & (7)
The diagonal elements 1( )i
w of 1W has been utilized to implement JLA [2][3] with a modified
sigmoid function to vary smoothly from -1 to 1. If τ is the threshold parameter for each joint, the
activation limits are defined as , ,( )th
i max i maxq q τ= − and , ,( )th
i min i minq q τ= + . If , ,( )th th th
i i max i minq q q∆ = − is
the activation range of i
th
joint, then 1 | ( ) |, wherei iw h q a large positive gainµ µ= + = ,
, ,
,
0 0 ( )( )/
1
( ) 0 ; , 0
1
th th th
i max i i i min i
th
i i i min
i i i a q q q q q
i
q q
h q a
eotherwise
ϕ ε
ε ϕ ε ϕ
ϕ
− − ∆
− ∀ < +

∀ − < < ∀ = ∀ >
+

(8)
In general µ should be large enough to make the 1/ iw near to zero when JLA is activated, so
that 0iq →& as in this case ( )ih q is bounded between 1± . In this case the role of ε is to

smoothen ( )ih q when changes from iϕ to iϕ− . Away from the joint limits when 0iϕ ≈ , iw may
still have oscillations due to large gain µ and oscillatory iq , which is smoothened by
implementing 0 1 4eε ≈ − .
The role of the term 1
1 2 2( )W W W−
+ in the null space of Jacobian needs to be discussed. Starting
with 2 1[ , ] n nW W I ×∈ , if we increase 1W ( which will occur during JLA activation), keeping 2W constant
then since 1|| ||W → ∞ , 1
1 2 2|| ( ) || 0W W W−
+ → resulting diminishing contribution from null space. On
the other hand, if we increase 2W , keeping 1W constant, which will occur most of the time when the
joint is away from its limits, 1
1 2 2|| ( ) || 1W W W−
+ → , since 1 2 2( )W W W+ ≈ .
3. CONTROL SCHEMES AND STABILITY
Introducing Proportional ( )PK and Derivative ( )DK error control in Eq.(6) by positive definite
diagonal gain matrices and task space error ( )d de x x x qκ− = − , we can arrive at the second
order close loop kinematic scheme (Figure-1[a]) with error system [9][11][13][14]
2( ) ( ) ; 0h h
d D P D Pq J x Jq K e K e I J J e K e K eξ= − + + + − + + =&&& && & & && & (9)
FIGURE 1: [a] Schematic implementation for 2nd order resolution in CWLS solution. [b]
Null space controller schematic. Nφ is the null space contribution.
[a]
[b]

Continuous time stability can be analyzed by Lyapunov second or direct method for Eq.(9) by
selecting Lyapunov candidate function 2( ) (1/ 2) T
V e e Ke V= + , ( ) 0V e∀ > and 2
2 (1/ 2) T
NSV q K qβ= & &
resulting 2( ) T
V e e Ke V= +& && . 2V is included to ensure that the system does not go unstable in the
Null Space Motion. K , NSK are symmetric positive definite diagonal matrices for task space and
null space respectively. Substituting the value of e and e& in expression of ( )V e& and after
simplification and substitution of 1; ( ) = Oh
n nJJ I and J I J J ξ×= − h
we can establish
2( ) T T
pV e e K K e V= − +& & .Considering the case of a constant reference ( 0)dx =& , the function ( )V e& is
negative definite, under the assumption of full rank for J and β is so chosen such that 2V& is
negative , indicates solution is stable in Lyapunov sense. If we consider the regularized version
[8] of CWLN solution, *
;h
JJ I≠Q ( - )h
J I J J O= , and *
( - )h
I JJ O≠ the error system reduces to
*
[ ]; ( )h
D P d D Pe K e K e N x Jq K e K e I JJ O+ + = − + + ∀ − ≠N&&& & && & & (10)
In defining the null space controller (Figure-1[b]), the first question that has to be answered is how
many sub tasks the null space can simultaneously handle? If we choose k sub tasks each of
rank kr , the limit is
1
k
i
i
r n
=
=∑ . Once all the dof’s are exhausted, it is useless to put additional low
priority tasks, as their contribution will be always projected in to null space or they can even
corrupt the primary task. Dropping the regularizing term for the time being and defining the null
space error Ne , the null space contribution as Nφ is
1 1 1( )[ ( )]; ( )( )h h h
N N N NI J J K e J J q e I J J qφ ξ ξ ξ= − + − − ∀ − −& & & & (11)
Defining a Lyapunov positive definite candidate function ( ) (1/ 2) T
N N NV e e e= , or ( ) T
N N NV e e e=& & ,
substituting the values of Ne& in ( )NV e& and after simplification we can establish T
N N NV K e e= −& ,
[Appendix-I.B] which is negative definite for positive definite symmetric null space proportional
gain matrix NK , which implies that the proposed controller in Eq.(11) stabilizes null space motion
as long as the Jacobian is full rank.
4. RESULTS AND DISCUSSION
To illustrate the performance, we discuss the results of null space optimized _ ( )cwls optq t form in
Eq.(6) and its canonical _ ( )cwls refq t form in Eq.(4), for a planar serial 3RRR manipulator following
two distinct types of trajectories, namely, the trajectory resembling the motion of finger tip ( 1Γ )
and lamniscate trajectory ( sΓ ). The particular solution _ ( )cwls pq t and CWLN solution with joint
limit activation _ _ ( )cwls opt jlaq t are also plotted to understand the contribution of null space and self
motion.
In both the cases the link parameters in Denavit Hardenberg standard convention
is 1 2 3[1.5,0.9,0.7] , [0,0,0], [0,0,0] [ , , ]i i i il cm d and q q qα θ= = = = . 1Γ is analytically generated by joint
space vector 2 2 2
0.3 0.2;0.5 0.5;0.7 0.3( ]) [tq t t t t= + + + and the reference joint space vector is
0.4 0.2;1.0 0.5( ) [ ;1.0 0.5]rq t t t t= + + + with values far away from ( ).tq t JLA parameters in Eq.(8)
are 0.1,τ = 100,a = 0.3,ε = 0 1 4,eε = − T
[0.8 1.8 2.6] ,maxq = T
[ 0.5 0.5 0.5] ,minq = − − − 1 7.eµ = +
Initial values of 1 3 3W I ×= and 2 [45.0 45.0 45.0],W diag= resulting 1
1 2 2( ) 0.978.W W W−
+ = The task
space controller parameters are [1 1]*0.07 / ;PK diag dt= , [1 1]*0.9DK diag= and

[1 1]*0.1.IK diag= The null space controller parameters are
0.95,Hk = [45 45 45],NPK diag= [4 4 4]NDK diag= [10 10 10]NIand K diag= .
_ ( )cwls optq t solution for 1Γ recovers the joint configuration better than _ ( )cwls refq t and it is in good
agreement with ( )tq t Figure-2[a]-[c]. The particular solution _ ( )cwls pq t (range space) fails to
follow ( )tq t after 0.5 .t s≈ The null space error Ne for 1q , rapidly converges from -0.7 at 0t s= to
-0.01 at 0.02t s= and remains steady with a peak response at 1.7t s= after which it again
converges to zero ( Figure-2[e]). The peak in Ne time history corresponds near configuration
singularity in joint space between 1.3 1.7s t s≤ ≤ , in which minσ ( min svd(J)) drops from 1.2 to
0.56. The effect of ND NK e& term is more prominent in contributing to 2ξ and finally in null space
acceleration aφ .
The contribution of
1
NI NK e dτ
τ ∫ is insignificant here and contribution from 1( )h
J J qξ −& & is difficult to
interpret in this case as its value is seen rising only during the configuration singularity period.
The net effect of these terms is reflected in aφ . Here Vφ is used to evaluate ( )q t& as a 1st order
resolution and from which we can evaluate Ne and subsequently aφ in the 2nd order resolution.
Thus the null space interaction between 1.3 1.7s t s≤ ≤ , which raises Vφ and aφ shifts the
recovered joint space trajectory towards rq and tq in _ ( )cwls optq t . This response can be utilized
FIGURE 2: [a]-[c]: Time history of joint configurations with null space contribution for finger tip
trajectory. Horizontal dotted lines represents joint activation threshold values th
maxq and th
minq .[d]
Time history of task space error norm || ( ) ||e t .[e] Null space response for 1q with out JLA, left Y-
axis for variables N Ve and null space velocity φ .

for an event where some preferred poses are desired in joint space, keeping the task space error
minimum. Increase in the value of the scalar Hk , results in initial oscillations in the solution as
reflected in the Figure-2[d].
To observe the response of 1W near joint limits, its normalized value is additionally plotted in
Figure-2[a]-[c]. When 1q , reaches its joint threshold limit 1, 0.7th
maxq rad= , at 1.26t s= , the normalized
value of 1
1w in Eq.(8) increases from zero at 1.17t s= to 0.6 at 1.26t s= . The first diagonal element
of 1
1 2( ) 0W W −
+ → , arresting further motion. The null space controller contribution is drastically
scaled down as 1
1 2( ) 0W W −
+ → , and the solution finally dominated by the particular solution.
Arresting of motion near 1,
th
maxq results oscillations in joint accelerations in second order
formulation which amplifies oscillations in 1
ϕ , by the term µ . This is because we have formulated
the JLA algorithm based on the joint configuration as ( , , , , )th th
t t max max min minf q q q q qϕ = . This will only
occur when 1q over shoots 1,
th
maxq in th
k time step gets damped and returned back to lower value
in ( 1)th
k + time step, until it is gradually damped out. This behavior has been reduced by
implementing the term 0ε in Eq.(8). For joint 2, the _ ( )cwls pq t solution overshoots the limit and
_ ( )cwls refq t touches the maximum limit. For joint 3, JLA is not actuated for _ ( )cwls optq t as it is well
under actuation threshold limit.
For the Regularized Composite Least Square (RCWLS) solution, the lamniscate trajectory ( sΓ )
simulates the condition of reaching workspace singularity condition, crossing it and then moving
away form it as the trajectory is closed and has two distinct lobes which results in multimodal
joint space trajectories. Moreover this particular case is extreme as ( )tq t and ( )rq t differs both
in amplitude and phase. The iteration started with 1 3 3 ,W I ×= 2 [75.0 75.0 75.0],W diag=
, [-1.5 -0.5 -0.5],i minq = =[2 2.3 2.3],i,maxq 0.25 ,radτ = 075; 0.4; 1 4; 1 7,a e eε ε µ= = = − = +
[45 45],PK diag= [0.45 0.45],DK diag= and [0.1 0.1].IK diag= The null space controller
parameters are 0.95,Hk = [45 45 45],NPK diag= [2.5 2 . ,.5 2 5]NDK diag= and
[1.0 1.0 1.0]NIK diag= .
The first workspace singularity crossing occurs between 0.08 0.3s t s≤ ≤ when the tip crosses
from A to B in sΓ (Figure-3[d]) and second workspace singularity occurs between 1.1 1.5s t s≤ ≤
when the tip crosses from C to D. In between these two, the solution faces near configuration
singularity when it crosses from P to Q between 0.6 0.8s t s≤ ≤ and from R to S
between1.6 1.8s t s≤ ≤ . It is to be mentioned here that initial high oscillating acceleration between
0.0 0.05s t s≤ ≤ in || ||e is due to the task space gains. In the near configuration singularity cases
(pq and rs) in Figure-3[e] which lowers ( )m tσ between 0.6 0.8s t s≤ ≤ and 1.6 1.8s t s≤ ≤ , the
damping parameter ( )tα does not interfere 0.5ε∀ = , the threshold value to initiate damping and
( ) ( , ).mt fα σ ε=
_ _ ( )cwls opt jlaq t solution increased || ||e between 1.3 1.5s t s≤ ≤ due to the simultaneous
occurrences of JLA for 3q and singularity crossings from C to D. It should be noted that in the
expression of * 1 1 2 1
1 2 1 2( ) ( ( ) )h T T
m mJ W W J J W W J Iα− − −
×= + + + , increase of ( )tα to maxα during
singularity keeping 1 2W and W to its initial values , will reduce the over all value of *h
J . On the
contrary, during JLA, increase of the diagonal element 1,3w of the weighing matrix 1W to a very
high value (Oe+7), will only make the third row of *h
J approaching to zero in order to make that

particular joint immobile but the other two rows of *h
J may increase or decrease as per the
action of the task space controller.
So the combined effect is overall damping of *h
J due to ( )tα and the third row is approaching
zero. This increases the task space error between 1.3 1.5s t s≤ ≤ in comparison to _ ( )cwls optq t ,
where only singularity avoidance is active. The null space contribution from ( )v aandφ φ has been
considerably diminished as high gain of 1W during JLA makes 1
1 2( ) 0W W −
+ → and 2W remains
constant in the null space. Further increase of value of 2 HW and k and null space gain
parameters results in increased oscillation in initial joint velocity and acceleration and also
increases || ||e .The task space and null space gains are kept on the higher side in the simulation
which causes initial oscillations in joint space in some cases. It has been verified that reducing
these gains eliminates these initial oscillations except during near singular or singularity
crossings. The role of the weighing matrices 1 2W and W has been defined with a bias to higher
gain of 2W which will amplify the null space contribution and in doing this the 1
1 2 2( )W W W−
+ term
is advantageously used.
In a hypothetical situation, we want to see the response when the reference signal ( )rq in joint
space approximates the analytical joint trajectory ( 0.9 ),r tq q= as in the earlier cases rq is
generated with considerably deviation from tq . For trajectory 1Γ , ( Figure-4: Top row) both
FIGURE 3: [a]-[c]: Time history of joint configurations with null space contribution for
lamniscate trajectory sΓ . Horizontal dotted lines represents joint activation threshold values
th
maxq and th
minq .[d] Trajectory trace for the solutions. Analytical trajectory generating workspace
singularity sΓ is OABPQDCRSO. [e] Time histories for || ||e , α and ( )m min svd Jσ values.

_ ( )cwls optq t and _ ( )cwls refq t solutions remain in between tq and rq , and the difference between
them can be neglected where as the particular solution deviates significantly tq as before.
Similar responses obtained from trajectory sΓ for lamniscate path (Figure-4: Bottom row) for
0.9r tq q= . In this situation, the null space error ,Ne for trajectory 1Γ , remains stable at 0≈ until
it briefly oscillates in near configuration singularity period between 1.3 1.7s t s≤ ≤ (Figure-5: Left)
and between 0.6 0.7s t s≤ ≤ and 1.5 1.7s t s≤ ≤ for lamniscate trajectory sΓ (Figure-5: Right).
During these time periods there is a surge in Vφ and aφ injecting the null space contribution in
the solution. For the remaining time in all cases , the null space contribution is 0≈ , which is
desired as the recovered joint space trajectory is in between tq and rq (Figure-4: Top Row).
FIGURE 4: Top: Time history of joint configurations for trajectory 1Γ with the special case
of 0.9 .r tq q= Bottom: Time history of joint configurations for trajectory sΓ with the special
case of 0.9 .r tq q= All results are for _ ( )cwls optq t solution. Columns from left represent
joints 1 2 3,q q and q respectively.
FIGURE 5: Left : Null space response for 1Γ when ( 0.9 )r tq q= Left Y-axis for variables
N Ve and φ . Right: Null space response for variables N Ve and φ for lamniscate trajectory
sΓ . All results are for _ ( )cwls optq t solution.
Trajectory: sΓ
Trajectory: 1Γ

5. CONCLUSION
By composite weighting the range and null space we can arrive at a solution which is able to
handle both joint limits and preferred joint configuration, simultaneously satisfying the primary
task. The solution lies between t rq and q also shifts the recovered joint space to wards the
reference configuration rq without JLA. In this formulation the role of 2W and rq is of paramount
importance as it controls the contribution form null space along with scalar Hk . It has been
observed that null space velocities Vφ and acceleration aφ are shooting up antagonistically to Ne
which signifies that the null space controller is working and there is self motion contribution form
null space when Ne is facing a drift from asymptotic stability. This enables the CWLS framework
to retrieve the desired joint configuration given the desired task space and preferred joint rate
( )rq& without considering any joint dependency.
The response can be utilized for an event where some preferred poses are desired in joint space,
keeping the task space error minimum, which can be exploited for recovering various human
postures where the motion workspace is limited and there is practical difficulty in mounting optical
markers or inertial motion sensors due to limited space availability or hindrance in natural
articulation. A typical application in this regard is recovering human palmer grasps (full closure of
fist) postures which are currently under study. The task is challenging, as in human palmer grasp
motion, apart from it’s high dimensionality, the problem is much more aggravated by limited
workspace space availability, cross finger occlusion, constraints in finger joint motion and full
traversal of joint motion ranges. The state of the art motion tracking technologies using optical or
inertial sensing for retrieving position and orientation data from each joint sometimes becomes
infeasible for this particular grasp mode, due to space limitations and slip, which results in
restricting natural articulation.
The limitation with 1
1 2( )W W −
+ term is with the activation of JLA, it reduces the null space
contribution. Sensitivity of Hk parameter is another issue and hence its bound has been kept in
between 0.75-0.95 for most of the cases as it is additionally coupled with the term 1
1 2 2( )W W W−
+ .
The other important limitation observed in CWLS scheme is its dependency on initial
configuration. Hence it will require an initial configuration close to the analytical solution.
6. REFERENCES
[1] A. Liëgeois. “Automatic supervisory control of the configuration and behavior of
multibody mechanisms.” IEEE Transactions on Systems, Man, and Cybernetics, SMC-
vol.7(12), pp.868–871, 1977.
[2] N. Mansard and F. Chaumette. “Directional redundancy for robot control.” IEEE
Transactions on Automatic Control, vol.54(6), pp.1179–1192, 2009.
[3] T. F. Chan and R. V. Dubey. “A weighted least-norm solution based scheme for avoiding
joint limits for redundant joint manipulators.” IEEE Transactions on Robotics and
Automation, vol.11(2), pp. 286–292, 1995.
[4] F. Chaumette, E. Marchand. “A redundancy-based iterative approach for avoiding joint
limits: Application to visual servoing.” IEEE Trans. On Robotics and Automation,
vol.17(5), pp.719-730, October 2001.
[5] Y. Nakamura, H. Hanafusa, T. Yoshikawa, “Task-Priority Based Control of Robot
Manipulators.” International Journal of Robotics Research, vol. 6(2), pp. 3 - 15, 1987.

[6] Oussama Kanoun, Florent Lamiraux, Pierre-Brice Wieber. “Kinematic Control of
Redundant Manipulators: Generalizing the Task-Priority Framework to Inequality Task.”
IEEE Transactions on Robotics vol.27(4), pp. 785-792, 2011.
[7] J. Baillieul. “Avoiding obstacles and resolving kinematic redundancy. In Proceedings of
IEEE International Conference on Robotics and Automation, Washington, UWA, 1986, pp
1698–1704.
[8] G. Antonelli. “Stability analysis for prioritized closed-loop inverse kinematic algorithms for
redundant robotic systems,” IEEE Trans. Robot., vol. 25(5), pp. 985–994, Oct. 2009.
[9] Hsu, P., Hauser, J. and Sastry, S. “Dynamic control of redundant manipulators.” Journal
of Robotic Systems, vol.6(2), pp. 133–148, 1989.
[10]Senda, K. “Quasioptimal control of space redundant manipulators.” AIAA Guidance,
Navigation, and Control Conference, 1999, pp. 1877–1885.
[11]Bojan Nemec and Leon Zlajpah, “ Null space velocity control with dynamically consistent
pseudo-inverse.” Robotica, vol.18, pp. 513–518, 2000.
[12] Tsuyoshi Shibata and Toshiyuki Murakami, “Null Space Motion Control by PID Control
Considering Passivity in Redundant Manipulator.” IEEE Trans. on Industrial Informatics,
vol. 4(4), pp.261-270, November 2008.
[13]G. Antonelli, F. Arrichiello, S.Chiaverini. “The null-space-based behavioral control for
autonomous robotic systems.” Intel Serv Robotics, vol.1, pp.27–39, 2008.
[14]J.Nakanishi, R. Cory, M. Mistry, J. Peters and S. Schaal. ”Operational Space Control: A
Theoretical and Empirical Comparison.” The International Journal of Robotics Research,
vol.27(6), pp. 737-757, 2008.
APPENDIX- I.A CWLS derivation
Objective : 1 2( ) ( ) ( )[(1/ 2) (1/ 2)( ) ( )]; .T T
r rmin q q min q q W q q q W q q s t Jq x= + − − =& & & & & & & & & & &3H
1 2( , ) n n
W W ×
∀ ∈ and positive diagonal positive definite ,
1 2
2
1 2 1 2
: ( , ) ( ) ( ) [(1/ 2) (1/ 2)( ) ( )] ( )
( ) 0 0 ( ) 0
T T
r r
T
T
q r q
Lagrangian L q q Jq x q W q q q W q q Jq x
L
L W q W q q J and L Jq x with L W W
q
λ
λ λ λ
λ
= + − = + − − + −
∂
∇ = ⇒ + − + = ∇ = − = ∇ = + >
∂
& &
& & & & & & & & & & & &
& & & & &
&
H
1
1 2 1 2 2 1 2 2
1
1 2 2
1 1 1 1 1
1 2 1 2 2 1 2
( ) 0 ( ) ; ( ) ( )
0; ( ) ( ) ;
( ( ) ) [ ( ) ] ( ), ( ) [
T T T
r r r
T
r
T T
r
W q W q q J W W q W q J q W W W q J
Putting the value of q in L J W W W q J x
J W W J J W W W q x and W W W JW J JW
λ
λ λ λ
λ
λ λ
−
−
− − − − −
⇒ + − + = ⇒ + = − ⇒ = + −
∇ = + − =
⇒ = + + − ∀ + =
& & & & & & &
& & &
& & 1
2 ]rW q x−
−& &
1 1 1 1 1 1 1
2( ) ( ( ) )T T T T
rq W J JW J x I W J JW J J W W q− − − − − − −
⇒ = + −& & &
1 1 1 1
1 1 2 2( ) ; ( ) ;T T
rJ W J JW J W W W qξ− − − −
∀ ∀ +h
&

1
1 2 2 1( )( ) ( )rq J x I J J W W W q J x I J J ξ−
= + − + = + −h h h h
& & & &
APPENDIX- I.B Null space Lyapunov stability.
Differentiating the null space error term Ne in Eq.(11),
1 1( )( ) ( )( )h h h
Ne I J J q J J J J qξ ξ= − − − + −& & && && & ;
rewriting 1( )qξ − & and substituting in Ne& ,
1 1( )( ) ( ) ( );h h h h h h h
N N Ne I J J q J Je J Je J JJ J JJ J qξ ξ= − − − − − + −& & & & && && & 0 ( ),N NJe as e J= ∈Q and
,h
JJ I= 0h h
JJ JJ⇒ + =& & ; which after simplification
1 1( )( ) ( ) ( )h h h h
N Ne I J J q J Je I J J J J qξ ξ= − − − − − −& & && && &
Now 1 1( ) ( ) ( )( )h h h
N N NI J J J J q I J J K eξ ξ φ− − = − + −&& & from Eq. (11)
1 1, ( )( ) ( ) ( )h h h h
N N N N Nor e I J J q J Je I J J I J J K eξ ξ φ= − − − − − − − +& &&& &&
Substituting the value of q&& from Eq.(6)
1 1
1 1
1 1
( )[( (( ) )] ( ) ( )]
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
h h h h h
N d N N
h h h h h h h
d N N
h h h h h
N N N N N
h
N
e I J J J x Jq q J Je I J J J J q
I J J I J J J x Jq I J J q J Je I J J J J q
I J J I J J q J Je I J J I J J K e
J Je
ξ ξ
ξ ξ
ξ ξ φ
= − − − + − − − −
= − − − − − − − − − −
= − − − − − − − − +
= − −
& & & && && & && &
& & & &&& & && &
& &&&&
& ( ) ; ( - ) ( )h h
N N N N NI J J K e I J J q projects q in N J and φ− =&& &&Q
( )h h
N N N Ne I J J K e J Je∴ = − − − && .
Defining a Lyapunov positive definite candidate function :
(1/ 2) T T
N N N NV e e V e e= ⇒ =& & and substituting the values of Ne&
( ( ) ) ( ) )
( )
T h h T h T h
N N N N N N N N N
T T h T h T T h
N N N N N N N N N N N N N N
V e I J J K e J Je e I J J K e e J Je
e K e e J JK e e J Je K e e e J K J J e
= − − − = − − −
= − + − = − + −
& & &
& &
1 1 1[( )( )] ( ) ( ) ( ) ( ) 0T h h T h T h T h T h h
Ne J I J J q J q I J J J q I J J Jξ ξ ξ= − − = − − = − − =& & &Q
T
N N NV K e e⇒ = −&

which is negative definite for positive definite symmetric null space proportional gain matrix NK ,
which implies that the proposed controller stabilizes null space motion as long as the Jacobian is
full rank.

Compensating Joint Configuration through Null Space Control in Composite Weighted Least Norm Solution

More Related Content

What's hot (19)

Similar to Compensating Joint Configuration through Null Space Control in Composite Weighted Least Norm Solution (20)

More from Waqas Tariq (20)

Recently uploaded (20)

Compensating Joint Configuration through Null Space Control in Composite Weighted Least Norm Solution