Cop\Tight © IF.

-\C Robot COlltrol

Karlsruhe. FR(;. IljHH

TIME DECOMPOSITION APPROACH OF

ROBOT CONTROL
G. Stavrakakis
Sationa/ Techni((l/ l:nil·ersity of Athens, Control and Alltol/lation C;/{JIIP.
3 Zakyntholl Str., GR-11361 Athens. Greece

ABSTRACT: Recent progress in the use of singular perturbation and two-time-scale methods
of modeling and design for control systems is applied for Robot Control design.
In this paper the two-time scale behavior of the full order Robot systems is first shown.
Then. recent two-time scale results are exploited to design slow and fast controls and
to combine them into a composite control for Robots. As an illustration we present a cor-
rective design to compensate for fast actuator dynamics modeled as singular perturbations.
KEYWORDS: Robot Control Two Time Scale Robot models Singular Perturbation methods.

1. INTROIlJCTION 2. SINGULAR RERTURBATION ANALYSIS Of EUll ORDER

ROBOT MODELS
Singularly perturbed systems and. more generally.
myltitime-scale systems. often occur naturally In many industrial robot applications a task must
due to the presence of small "parasitic" parameters. be accomplished and the task coordinates. which
typically small time constants. masses. etc. multi- are convenient to the task description are trans-
plying time derivatives. The two-time scale beha- rormed to the joint coordinates. which are con-
vior of the full order Robot system - arm and trolled uirectly. In the most of such cases robot
actuator dynamics - arises from the fact that the motions are achieved through the joint actuators.
actuating systems have fast dynamics when compared that is the motors which apply the conveninet
with the Robot arm dynamics. control torques on the robot arms.
feedback Control system design for Robots is often It is known that a large class of commercially
simplified by neglecting sensor and actuator dyna- available motors for robotic applications has
mics. This is tacitly assumed in almost all the extremely low inductance (typically less than
classical Robot control investigations O.2.3.6L 85 mH). Assuming for simplicity. that all the in-
In these cases the control law is given in terms ductances are equal. we can write the n-dof robot
of the generalized forces. that is the motor tor- full dynamic model (including motor dynamics) as:
ques. as if it was possible to assign these tor-
ques instantaneously.
When the performance requirements for industrial
robots are increased and require accurate. fast
l; 1,[ :J1(w'G'flH\jI (1)

and versatile manipulations. dynamic effects be- LI=-RI-Bw+V{8.w)

come significant. This requires the control of We denote: J(8)= {D{8)+N 2J ml . C=NK t • B=NK b
a multi-input multi-output system described by a
set of highly nonlinear. strongly coupled diffe- where:
rential equations. An attractive idea of how to aT_{- ~t 8 Z • .. ·,8 n ), is the vector of joint angles
compensate for coupling and nonlinearities. when T
w= 8 • is the vector of joint veloci-
actuator dynamics are neglected. through a sui-
table feedback was introduced by Nicosia and Tomei T ties
{3}. for Robot systems whose number of inputs I =(Il 12 ,,,, ,In)' is the actuator motor armature
currents.
equals the number of outputs. they obtained con- W{8,w), G(8), f{w),represent the centrifugal dlld
vergence to a suitable reference decoupled model coriolis, gravitational tor-
by means of a nonlinear PI feedback. But the ques and frictional torques
method suffers from the restrictions indroduced respectively
by the fast actuator dynamics. The motor torques
calculated in {3} for a trajectory tracking case D(8) , is the (n X n) generalized
present many dramatic changes. thus it is not inertia matrix
possible to assign these torques to the Robot arm J m=di ag {J m1 ,· .. ,J mn}, is the motor inertia matrix
instantaneously. Control schemes based on the full
order Robot model have solved this problem and Kb=diag{ Kb1 •··· ,Kbn ! · is the motor back emf matrix
they have already been presented by Tzafestas and
Stavrakakis in {4} . Kt =d iag{ Kt!, ... ,K tn : . is the motor torque constant
matrix
In the present work a design to compensate for , motor armature inductance
fast Robot actuator dynamics is presented by ex- matrix
ploiting the two-time scale behavior of full order ,~ =diag { i11'N2"" ,N ; , NI > 1, motor gear ratio
Robot systems. A composite control for Robot arm n
is derived and its performance is tested by simu- matrix
lation. R=diag{ Rl'R 2, ... ,R n}, motor armature resistance
matrix
RC-e·
60 G. Sta\Takakis

T _ · q
V (S.w)-{V I V2 ... Vn). is the vector of the input time scale. denoted from t he prevlous as xql' XZ.
voltages. The only fast variations are the dqviations of
It is assumed an ideal gear box with perfect gear X3 from its "quasi -steady state" X3.
mating (no backlash).
Denoting them by xi as in (4). and setting X3=
Equations (I) can now be written as: Xf+xq and £=0 in (7) we obtain the fast subsys-
3 3
XI =X 2 tem
X2=-J-I{XI)H{XI'X2)+J-I{XI)CX3= gI (X) (2) dxi / dt=-RXi+{V-vq) (8)
EX 3=-BX 2-RX 3+V = g2{X,V) using also that dXj / dt=O and xj is given from eq.
where : (5) .
T T T T T T T
X = XI{ £.t) X2{£.t) X3{£.t)= S (£.t) w (£.t) I ( £ .~ It is important to see from eq. (8) that the fast
and £ = L. actuator dynamics are represented by linear state
equations with stable homogeneous part. Thus. even
In the equations (2) the dependence of the involved with a control voltage V=V q designed for the re-
robot variables on the scalar parameter L is shown duced order slow subsystem. "parasitic" actuator
explicitly.The case when inductances are not equal. dynamics will vinish asymptotically. This is an
and hence there are several parameters LI .L 2 •.•.• Ln important result which can justify the frequently
adopted assumption in the Robot control literature.
multiplying the derivatives of the currents. can that the Robot actuator dynamics can be neglected
also be similarly treated . {I.Z.3 }. However. when robust control schemes are
investigated. as the adaptive control schemes. un-
Equations (2) is the standard form of a class of modeled fast dynamics can destroy stability and
Nonlinear Singularly Perturbed Systems. the Two boundedness properties ( 5}.
Time Scale (TTS) systems. which often occur natu-
rally due to the presence of small parameters The aim of the next section is the design of a
(small time constants. mass inductances. etc.) corrective feedback in order to correct compensate.
multiplying time derivatives {7}. that is to accelarate as much as possible. the
disappearance of the fast actuator dynamics ef-
The main physical characterictic of singularly fects.
perturbed systems is the presence and coexistence
of two (interacting) modes: The fast modes and the
slow ones. Under rather weak conditions. it is 3. CCJo1POSITE aJNTROL FOR ROBOTS
possible to decompose the original system into two
subsystems: a fast one and a slow one. that is a The decomposition of the full order Robot system
Two Time Scale decomposition, The chief purpose of into separate slow and fast subsystems suggests
the singular Perturbation approach is the allevia- that separate slow and fast control laws be de-
tion of the high dimensionality and ill-condition- signed for each subsystem and then combined into
ing resulting from the interaction for slow and a composite control of the original system. If the
fast dynamic modes. fast control is not needed.then only the slow re-
duced control is used.
Let us introduce the fast time scale
t=t/£. t=O at t=O. with to=O (3) a. Slow subsystem regulator
State X3 in eq. (2) is formed of a "fast transient" The slow Robot subsystem (6) are the well known
X; and a "quasi-steady state" xj. defined as a so- dynamical equations for Robotic manipulators re-
sulting from classical mechanics and from the as-
lution of the third of eq. (2) with £):3=0. This sumption that the actuator dynamics can be negle-
motivates the definition of the fast "parasitic" cted. Thus. the slow reduced order control voltage
state as : Vq can be calculated from any known {I.2.3.6}. as
X;=X3-R-I{vq-BX~)=X3-xj (4) Computed Torque. Model Reference Adaptive Control
(MRAC). Nonlinear feedback decoupling Control ( 6}.
where the "quasi-steady state" part of X3 in the etc., or new Robot control scheme. based on the
fast time scale is model of eq. (6).
xj=R-I{vq-BX~) (5)
In this paper the slow subsystem control law is
The superscript q denotes the corresponding vari- derived using the MRAC approach. as it is described
ables in the "slow" time scale t. in ( 3} by Nicosia and Tomei.
In the 1imit £ -+ 0 the reduced-order Robot model in
the t time-scale defined by the "quasi-steady sta- b. fast subsystem regulator
tes" xi. x~. xj and the "slow reduced" control
from the analysis of the previous section. the
voltage vq. that is the I.ltow I.lu. bl.lljl.l tem, is ~slow" variables are constant during the transient
xq=x q states of the "fast" subsystem that is Xq : const .•
1 2
vq : const. and it is non linear function of xq.
xq=-J-I{Xq)H{X q • xq)_J-I {xq)CR-IBXq+J-I{xq)CR-Ivq
2 1 1 2 1 Z 1 ( 6) Defining vf=v-v q the fast subsystem regulator can
To obtain the fast parts of the states. Xl' x2 • X3 be derived from the following LQ design problem:
we rewrite eq. (Z) in the fast time scale
dXI / dt= £XZ
dXZ/dt=£9I(XI,X Z) (7)
with
dX 3/ dt=gZ{X,V) .f f
£X3=- RX 3+ Vf' (9)
and again examine the 1imit as £-+ O. Then dXI / dt=O.
The pair (- R.I) is completely controlable. then
dXZ/dt=O. that is Xl' X2 = constant in the fast the solution of the fast subproblem exists in a
 Tilll c Dcc()lllp()sit iOIl :\ ppro;tch of Ro bot Co III rol 61

4. SIMULATION RESULTS
( 10 ) In the application, the 3 dof robot, which is des-
where Kf is the solution of the Riccati equation cribed in {4l , is considered. The non-zero ele-
ments of the inertia and cor io lis- centrifugal
for the LQ design problem of eq. ( 9 ) and P a posi- matrices and those of the gr avitational vector as
tive definite matrix. well as the dynamics of the actuator s are analyti-
cally descrioed in { 41. They are the well known
c. Composite Control Lagrangian and DC motor dyna mic s used in Robotics.
The tests were carried ou t so as to follow smooth
In order to use v= vq+vf as a composite control for trajectories in joint coordinate s with a maximum
the full order system (2) , we mus t express the velocity 900 / sec and maximum accelaration 1800/
fictitious subsystem state s xi, xj and x; in xi, sec 2 . The 3dof Robot arm is s imulated as a full
order model including actu ator dyna mi cs . In fig. 1
terms of actual state s X. Thus, the composite con- the desired trajectory tracking of the second
trol becomes: joint angle is shown when the Robot arm is control-
V=Vq+V f=Gf 3 Xq+Vq( Xq ) .
Xf+Vq (xq )=Gf X3-G f3 led using the reduced order cont rol l aw, that is
G =0 in (11). In fig. 2 similar result s are shown
Using eq. (5 ) we can obtain a realizable composite when the Robot arm is controlled us ing t he Compo-
control: site controller of eq . (11 ) .
q
V(X )=G fX3fu f R- l BX 2+(I -GfR- l )V (X ) ( 11) 13.3,---,.,-- r - ---r-- - - , - - -- r - -- _
·1 f t he approxlmatlOn
., Xq1 "-X l' Xq <:'- X2 and X -- Xq+Xf3 11.
2 3 3
is used ( 7l . The symbol Vq (X) denotes that the ~ .4

slow subsystem ~R AC control law is calculated

using appro xi mately the real s t ates Xl ,X2 ,X3· 7

5 .6
3~ ,1:3
// " 3. 6
~
33 . ft
'"
~
1. 7
3 1.a ,/
/ - 1o:L2
29 .0 w
~
/ ~ -2 . 2
·I
,2 7 .il
f /
/ ~
2'5.0
'" - ·LI

,/ -6 .1 -
I .~ 5 3 .6 h1t. . 2 1'58 . 8 211 .4 264 . 6
23.e

el.1!)
/ T II'lE I 4 ftI $. E- ( J

/
/ Fig.lc
19.13
, , l/
17 . e
---/'
15 . 0
~3 .6 11;)6 . 2 1'58 . 8 211 ,4 264 . 1l

Tl l'lE I 4 r~ s.H J

Fig.la . The d e si r e d and actu a l

d ~ s pl acemen ts c o i nc i d e

4S .t:.

413 . 7
! 1\

3~ .S
I i \
1 I
1
\
·
~
1ra . 9

25 .9
I 1 I ! ,
,
\
~
':1 .~
, i ,,
\
\,
16 . I
.I I !
\

i' i \ ,
11. 2

6. 3
t i , \
1. 4
if I I
\1
-3.'5
rr I ,I "1
'5 3 .6 t ile:. . 2 ! 56 .t 211 , ..
l UIE (4
T H-lE [4 r',;:: H

Fig . lb Des ir ed and ac tual an gu l ar Fig . 2a . The desi r eJ a nJ actua l

\'cloci ti cs di sp l acemcnt s c o inciJc
62 C. Stavrakakis

CONCLUSION
48 . 6
In the present paper we have formulated and treated Af f\
the Robot control problem as a singular perturba- 43.5
'V
tion and especially as a Two Time Scale control
problem. 38 ••
\
33 .3
~ \
The simulated results have shown that the dramatic
changes in the control voltage can be reduced by a
corrective control design, to compensate for fast
.~.,
:!!
28 . 2
Il'
t'
\
\
23.8
actuator dynamics modeled a5 singular perturbations.
Moreover, this corrective control design may be ...~ \
U 17 .9
.c.ombined with every control strategy, which may
be adopted for the control of the reduced order
0

~ 12.13
l \
Robot model.
"'"5 7.7
if \
Acknowledgments. This work has been partly suppor- ~ 2 ••
If \\
ted by EEE under an ESPRIT cm project. Special ~
thanks are due to Prof. S. Tzafestas for advice -2. 0
and suggestions. 1.0 :53.6 hlb. 2 158.8 211.4 264.0

THE t4 ~se-c )

REfERENCES Fig. 2b. The desired and actua l angu l ar

v e l oc it ies
1. J.S. Luh, M. Walker, R.P. Paul (1980). "Resol-
ved accelaration control of mechanical manipu-
lators". IEEE Tr3ns. Aut. Contr. AC-25, vol.3.
2. A. Guez (1983). "Optimal Control of Robotic ma-
nipulators". PhD Dissertation. University of 14 .2

florida.
12 . 5

3. S. Nicosia, P. Tomei (1934). "Model Reference l B. 8

Adaptive Control Algorithms for Industrial Ro-
bots". Automatica, vol. 20, lb. 5, !=p. 635 - 644. 9.2

7 .5
4. S. Tzafestas, G.S. Stavrakakis (1986). "Model §
Reference Adaptive Control of Industrial Robots 5.8
wi th actuator dynamics". IfAC/lMACS symp. on
Robot theory. Vienna, Dec. 1986. • .1

5. P. Ioannou, P. Kokotovic (1984l. "Robust rede- 2.5

sign of Adaptive Control". IEEE Trans. on Atom. e .8

Contr. AC-29, No 3, pp . 202 - 21 1.
-~.9

6. T. Tarn, A. Bejczy, A. Isidori (1984). "Nonli-

near feedback inb Robot Arm Control". Proceed- - 2.6
1.0 53.6 106 .2 1513.8 ~ll .4 264.0
ings of 23rd Confer. on Decision and Control.
Las Vegas. NV, ~ 1SE4, W. 73h - 751. TIME (4 I.H'C )

Fig. 2c . Compo sit e co ntroller voltage

7. V. Saksena, J. O'Reilly, P. Kokotovic (1984).
"Sinaular Perturbations and Time-scale Methods
in Control Theorv: Survev 1976-1983". Automati-
ca. vol. 20, No 3, pp . 273 - 293 .