ARHITECTUR DE CONTROL

ADAPTIV PENTRU ROBOI

REDUNDANI

ADAPTIVE CONTROL
ARCHITECTURE FOR
REDUNDANT ROBOTS

Aurel FRATU*, Laurent VERMEIREN**

*

"Transilvania" University of Brasov, Romania

**
University of Valenci ennes, France

Rezumat. n aceast lucrare est e propus o arhitectu r

de control cu structur vari abil pentru comanda unui
robot redundant. Coop erarea actu atorilor robotului,
pentru efectu area unei sarcini robotice, se realizeaz prin
alocarea dinamic de sub-s arcini robot, pentru fiecare
sub-controler al arhitecturii. Algoritmul de control al unui
sub-controler autonom se bazeaz pe matricea Jacobian a
robotului. Algoritmul de estimare a elementelor matricei
Jacobian utilizeaz matricea de covarian.
n lucrare este folosit tehni ca facto rizrii matricei
de cov arian pentru identi ficarea direciilor redund ante
din spaiul de micare al actuato rilor.

Abstract. In this pap er, a co ntrol architecture with

variabl e structure fo r red undant robot is propos ed. The
cooperation o f the robots actu ators, to realis e a robot
task is made by the sub-tas ks dynami cally allo cation for
each sub -cont roller o f the archit ectu re. The control
algorithm o f a sub -cont roller with sel f-gov erning is
based on robot Jacobian matrix.
The algorithm for the estimation o f Jacobian
matrixs elements uses th e cov ari ance matrix. In this
paper, th e factorization technique o f th e covarian ce
matrix fo r the identi fication of the redund ant directions
inside the actuator motion space is used.

Cuvinte cheie: sub-controlere, alocarea dinamic de

sub-sarcini, redundant

Key words: sub-controllers, sub-t asks dynamical

allocation, redund ancy

1. Introducere

1. Introduction

Un robot redundant are mai multe grade de

libertate dect cele necesare pentru a-i poziiona
efectorul final.
O schem de control pentru un robot redundant
este destinat s asigure evitarea obstacolelor n
spaiul de lucru n timpul micrii i s permit
robotului s i ndeplineasc misiunea n cazul
funcionrii defectuoase a unor actuatori. n astfel de
situaii sunt activate gradele de libertate redundante.
Pentru controlul micrii unui robot redundant
autorii propun o arhitectur de control cu structur
variabil, format dintr-un set cu numr variabil de
sub-controlere adaptive, autonome.
Fiecare sub-controler comand un actuator, ce
deservete un singur grad de libertate.
Fiecare sub-controler funcioneaz ca un subsistem autonom i asigur controlul micrii la
nivelul unui grad de libertate.
Prin funcionarea sub-controlerelor, ca subsisteme cooperante, actuatorii sistemului robot vor
coopera pentru ndeplinirea sarcinii robotice.
Pentru coordonarea funcionrii actuatorilor se
propune tehnica alocrii dinamice de sub-sarcini
robotice, pentru fiecare sub-controler n parte [1].
Numrul de grade de libertate active la un
moment dat determin tipul de configuraie a
arhitecturii de control i nivelul de redundan.

A redundant robot has more degrees of freedom

than those needed to position its end-effector.
A redundant robot control scheme is provided
for the obstacles avoiding in the workspace, during
of the movement and allows accomplishing its
mission in the case of malfunction of several
actuators. In such situations, the redundant degrees
of freedom are activates.
For the motion control of a redundant robot we
propose a control architecture with variable
structure, formed by a set with variable number of
adaptive, self-governing sub-controllers.
Each sub-controller commands an actuator
which deserves a de gree of freedom.
Each sub-controller works as an autonomous
sub-system and assure the motion control at the
level of one degree of freedom.
By the working of the sub-controllers as
cooperative sub-systems, the actuators of the robot
system will cooperate, to accomplish the robotic task.
For the coordination of the actuators working
one propose the sub-tasks dynamical allocation
technique, for each sub-controller [1].
The number of the active degrees of freedom,
at a given instant, determines the type of
configuration for the control architecture and the
level of redundancy.

RECENT, Vol. 8, no. 2 (20), July, 2007

97

Arhitectur de control adaptiv pentru roboi redundani

Alocarea dinamic a sub-sarcinilor robot, pentru
fiecare sub-controler, este o cerin esenial pentru
sistemele robot cu actuatori cu funcionare autonom.
Metoda de alocare dinamic a sub-sarcinilor,
permite robotului s se adapteze mai uor la
schimbrile mediului de operare i s i
ndeplineasc misiunea n cazul apariiei unor
disfuncionaliti.

The dynamical sub-tasks allocation, for each

sub-controller, is an essential demand for robot
systems with autonomous actuators.
The dynamical sub-tasks allocation method,
allows to robots to easily adapt at the
environmental changes and to accomplish its
mission in the malfunction cases.

2. Arhitectur de control pentru roboi

redundani

2. Control architecture for redundant

robots

2.1. Sub-controle re autonome

2.1 Autonomous sub-controlle rs
S considerm cea mai simpl structur de
Let us consider the simplest controller i.e., P
controler, controlerul proporional, P. Robotul poate
controller. T he robot can find out the redundancy
dobndi redundan n raport cu o sarcin robotic
with respect to one given sub-task, if the control
dat dac structura de control folosete tehnica
structure uses the dynamical sub-tasks allocation
alocrii dinamice a sub-sarcinilor.
technique.
Prin combinarea sub-sarcinilor atribuite subBy combining the sub-controllers offered to the
controlerelor se obine o arhitectur de control n
sub controllers, he is obtained a control architecture,
care actuatorii coopereaz ntre ei.
in which the actuators cooperate between them.
Fie sub-sarcina i, a unui sub-controler, definit
Let us the i-th sub-task for a sub-controller who
n spaiul de operare propriu actuatorului comandat
is defined in the own space of the commanded
[2]. Tipic, acest spaiu este un spaiu de acoperire al
actuator [2]. T ypically, this sub-space is an output
unui senzor exteroceptiv de prelevare de informaii.
space of an external sample sensor. T he sensor
Semnalul de ieire al senzorului este dependent de
output xi is a function of actuator displacement:
deplasarea actuatorului:
(1)
xi = xi () .
Difereniind relaia (1), n raport cu rata
Differentiating the relation (1) with respect to
prelevrii, se obine:
the sampling rate, we can get
(2)
xi (k ) = Ji (k ) (k )
n pasul k al procesului de prelevare.
in the k -th sampling step.
Matricea Jacobian este dat de:
The Jacobian matrix is defined as:
x ( k )
J i (k ) = i
.
(3)
(k )
Sub-sarcina (k), definit n spaiul articular
The sub-task (k) in the joint space, must be
trebuie exprimat n funcie de deplasarea dorit,
define in terms of a the desired shifting value, xid and
xid , i deplasarea curent, xi, din spaiul operaional.
the current shifting, xi inside the operational space.
From relation (2) one obtain the following
Din relaia (2) se obine pentru relaia
equation:
urmtoare:

(4)
( k ) = J i+ ( k ) ( xid (k ) xi (k )) + I Ji+ ( k ) Ji (k ) 0 .
+
+
Matricea J reprezint pseudo-inversa matricei
The matrix J denotes the pseudo-inverse of a
Jacobian, J.
Jacobean matrix, J.
Acest rezultat este de o importana
This result is of fundamental importance for
fundamental pentru stabilirea nivelului de
redundancy level establishing since solution (4)
redundan, deoarece soluia (4) evideniaz
evidences the possibility of choosing the vector 0 ,
posibilitatea de alegere a vectorului 0, pentru a
so as to exploit the services of the redundant
exploata serviciile gradelor de libertate redundante.
degrees of freedom.
Contribuia lui 0 const n a genera un spaiu
The contribution of 0 is to generate null space
nul de micare pentru a permite manipulatorului s
motions for allow to the manipulator, to put himself
se plaseze n postura din care are o mai mare
in the posture from which has more dexterity for the
dexteritate pentru executarea sarcinii date [3].
execution of the given task [3].
98

RECENT, Vol. 8, nr. 2 (20), Iulie, 2007

Adaptive Control Architecture for Redundant Robots

Dac un actuator se deplaseaz n direcia dorit
xid , utiliznd un control proporional al vitezei,
pentru sub-sarcina i avem comanda de forma:

If an actuator move in the desired direction, xid ,

using the proportional velocity control, for i-th
sub-task, we have the control low:

(5)
u (k ) = K i Ji+ (k ) ( xid ( k ) xi ( k )) + I J i+ (k ) J i ( k ) ,
unde I, Ki i sunt matricea identitate, matricea de
where I, Ki and are an identity matrix, a feedback
ctig a buclei de reacie i respectiv, un vector
gain matrix, and an arbitrary vector. T he arbitrary
vector denotes the contribution of the sub-controller
arbitrar. Vectorul arbitrar reflect aportul
command, to the robot system redundancy [4].
comenzii
sub-controlerului
la
asigurarea
redundanei pentru sistemul robot [4].
2.2. Estimare a matricei Jacobian
Dac nu se cunosc caracteristicile mediului de
operare i nu se cunosc toi parametrii dinamici ai
robotului, matricea Jacobian Ji(k) este necunoscut.
Pentru a realiza un control adaptiv la schimbrile de
mediu, este necesar un mecanism de estimare a
matricei Jacobian.
Utiliznd metoda celor mai mici ptrate, n
lucrarea [5] este propus o relaie de calcul pentru
matricea Jacobian estimat, :

)(

2.2 Estimating the Jacobian matrix

If there is no a priori knowledge on the
environment characteristics and the robot dynamic
parameters, the Jacobean matrix, Ji(k) is unknown.
To make the adaptive control to the change of
environment, a mechanism to estimate the matrix
Jacobian is needed.
Using the least-mean-square method, in the
paper [5] a computing relation is proposed for the
Jacobian estimate matrix, :

)(

(6)
J i ( k ) = J i ( k 1) + xi (k ) J i (k 1) (k ) ( k )T P( k 1) / + ( k )T P(k 1) ( k ) ,
unde , 0 < < 1, este un factor de uitare i P(k)
where , 0 < < 1, is a forgetting factor, and P(k)
este o matrice de covarian calculat cu:
is a covariance matrix calculated as:
1
P( k 1) (k ) ( k )T P( k 1)
(7)
P( k ) = P(k 1)
.

+ (k )T P( k 1) (k )
Mecanismul de estimare a matricei Jacobian se
The mechanism to estimate of the matrix
bazeaz pe observarea procesului robot-mediu i
Jacobian is made on the observations of the robotconst n prelevri aleatoare de informaii din
environment process and consist in random samples
mediul n care opereaz robotul.
about the environment in which the robot operate.
Numrul observaiilor, ca valoare statistic, este
The number of observations, as statistical
folosit la determinarea valorilor probabile ale
quantities, is used to determine the expected values
elementelor matricei Jacobian.
of Jacobian matrix elements.
Matricea de covarian P se folosete pentru
The covariance matrix P is used to define the
definirea valorilor probabile ale mrimilor estimate.
expected values, of the estimated variables. The
Covariana ofer o msur a puterii de corelare ntre
covariance provides a measure of the strength of the
seturile de prelevri aleatoare de informaii [6].
correlation between the sets of random samples [6].
2.3. Estimare a re dundanei
Caracteristicile estimatorului (6) permit
identificarea direciei deplasrii redundante.
Pentru a realiza acest lucru matricea de
covarian P se transcrie sub forma factorial astfel:

2.3. Estimating the re dundancy

The characteristics of the estimator (5), allows
us to identify the redundant shifting direction.
To make this thing the covariance matrix P is
written as factorial mode:
(8)
P = U D U T ,
unde U este o matrice superior triunghiular cu
where U is an upper triangular matrix whose
elementele de pe diagonala principal egale cu unu,
diagonal elements are all 1, and D is a diagonal
iar D este o matrice diagonal:
matrix:
d1 0

D = 0 O 0 .
(9)

0 d n
RECENT, Vol. 8, no. 2 (20), July, 2007

99

Arhitectur de control adaptiv pentru roboi redundani

Dac matricea P este slab pozitiv, un element
al diagonalei matricei D este apropiat de zero.
Nivelul de redundant al robotului se poate
determina prin observarea matricei D.
Dac un element d i = 0 sau este foarte mic,
direcia corespunztoare a lui este direcia
redundant n raport cu sub-sarcina dat.

If the matrix P is weak-positive, one of the

diagonal elements of D is near to zero.
The redundancy level of the robot one can find
out by observing the matrix D.
If the element d j = 0 or very small, the
corresponding direction of is the redundant
direction with respect to the given sub-task.

3. Concluzii

3. Conclusions

Funcionarea unui robot n condiii de siguran

n medii necunoscute presupune existena unei
capaciti de adaptare ridicat. Adaptabilitatea
robotului la condiii reale de lucru este posibil prin
activarea gradelor de libertate redundante.
Pentru un robot cu grade de libertate redundante
se propune n lucrarea de fa o arhitectur de control
cu structur variabil. Algoritmul de control se
bazeaz pe matricea Jacobian a robotului.
Pentru a identifica elementele matricei Jacobian
a fost propus un algoritm de estimare bazat pe
corelarea ntre seturile de prelevri aleatoare de
informaii (covarian).
Forma factorial a matricei de covarian
permite identificarea direciilor redundante din
spaiul de micare al actuatorilor.

The robot working in the safety conditions,

inside unknown environment supposes a high
adaptability capacity. The robot adaptability at real
work conditions is possible if are activate the
redundant degrees-of-freedom.
For a robot with redundant degrees-of-freedom,
he is proposed in this paper a control architecture
with variable structure. T he control algorithm is
based on the Jacobian matrix of robot.
For identify the Jacobian matrix elements, an
algorithm based on the covariance, has been
proposed.
The factorial shape of the covariance matrix
allows identifying the redundant direction inside of
the motion space of actuators.

Refe rences
1. P arker, L.E.: Distributed Algorithms for Multi-Robot Observation of Multiple Moving Targets. Journal Autonomous Robots, Vol.
2, No. 3, May 2002, p. 231-255, ISSN 0929-5593, Springer Netherlands P ublisher
2. Zavala-Rio, A., Brogliato, B.: On the Control of a One Degree-of-Freedom Juggling Robot. Journal Dynamics and Control, Vol.
9, No. 1, January 1999, p. 67-90, ISSN 0925-4668, Springer Netherlands Publisher
3. Yoshida, K.; Nenchev, D.N.: Reaction null-space based control of under-actuated manipulators. P roceeding of International
Conference on Intelligent Robots and Systems, Vol. 2, p. 1358-1363, ISBN 0-7803-4465-0, October 1998, Victoria, BC,
Canada
4. Hosoda, K., Igarashi, K., Asada, M.: Adaptive hybrid control for visual servoing and force servoing in an unknown environment.
IEEE Robotics and Automation Magazine, Vol. 5, No. 4, p. 3943, December 1998, ISSN 1070.99321, University of South
Florida, USA
5. Heejin, L., Hyunseok, S., Euntai, K., Seungwoo K., Mignon, P .: Variable structure control of manipulator using linear timevarying sliding surfaces. P roceeding of International Conferences on Intelligent Robots and Systems, Vol. 2, p. 806-811, ISBN
0-7803-4465-0, October 1998, Victoria, BC, Canada
6. Yuh, J.; Nie, J., Lee, W.C.: Adaptive control of robot manipulators using bound estimation. P roceeding of International
Conferences on Intelligent Robots and Systems, Vol. 2, p. 1126-1131, ISBN 0-7803-4465-0, October 1998, Victoria, BC,
Canada
7. Vainio, M., Appelqvist, P ., Halme, A.: Generic control architecture for a cooperative robot system. P roceeding of International
Conferences on Intelligent Robots and Systems, Vol. 2, p. 1119-1125, ISBN 0-7803-4465-0, October 1998, Victoria, BC,
Canada
8. Fratu, A., Fratu, Mariana: The problem of operational space control for redundant robots. RECENT, Vol. 6, No. 3(15), December
2005, p. 52-55, ISSN 1582 0246, Brasov, Romania
9. Fischer, J., Breithaupt, R., Bode, B.: Adaptive and economic data representation in control architectures of autonomous realworld robots. Artificial Life and Robotics, Vol. 6, No. 4, December 2002, p. 200-204, ISSN 1433-5298, Springer Japan
P ublisher

Lucrare primit n Aprilie, 2007

(i n form revizuit n Iunie, 2007)

100

Received in April, 2007

(and revised form in June, 2007)

RECENT, Vol. 8, nr. 2 (20), Iulie, 2007