You are on page 1of 5

I

Adaptive Control of Flexible-Joint Manipulators


Fathi Ghorbel, John Y. Hung, and Mark W. Spong
ABSTRACT: This paper presents an adaptive control result for flexible-joint robot manipulators. Under the assumption of weak joint elasticity, a singular perturbation argument is used to show that recent adaptive control results for rigid robots may be used to control flexible-joint robots provided a simple correction term is added to the control law to damp out the elastic oscillations at the joints. In this way, fundamental properties of rigid robot dynamics may be exploited to design adaptive control laws for flexible-joint robots that are robust to parametric uncertainty. Experimental results are given to illustrate the theory. be derived that is of the same order as rigid robot dynamics and which can be linearized by nonlinear feedback using only position and velocity information. The chief drawback to this approach to date is its lack of robustness to parametric uncertainty. In this paper, we present an adaptive control result for flexible-joint robots. In a typical scenario, such as a revolute joint robot actuated by direct-current (DC) motors with harmonic drive transmission, the joint stiffness is relatively large compared with other parameters in the system. At the same time, the joint damping is small. This can lead to strongly resonant behavior, even instability, using rigid control schemes unless the control bandwidth is severely restricted, as shown in [ l ] . Intuitively, a control scheme for a robot manipulator should control the rigid-body dynamics of the robot while effectively damping out the elastic oscillations at the joints. The control algorithm in this paper makes this intuitive idea rigorous. The overall control law consists of an adaptive control law designed based on a rigid model, together with a linear correction term to damp out the elastic oscillations at the joints. A singular perturbation argument is used to provide rigorous justification of this approach. The particular adaptive algorithm used in this paper was first derived for rigid robots by Slotine and Li [7] and exploits, in a fundamental way, the passivity properties of rigid robot dynamics. However, as we point out later, any of a number of control schemes developed for rigid robots can be used as part of our overall design. The adaptive control scheme presented here has several attractive features with respect to its design and implementation. First, the overall complexity of the scheme is roughly the same as the rigid adaptive control scheme of [7]. Second, our control scheme exploits the two-time scale behavior in the system due to the relatively large joint stiffness, while, at the same time, it exploits the fundamental passivity properties of rigid robot dynamics. This is significant since the flexible-joint robot dynamics do not possess the required passivity properties themselves. Third, the implementation of the full controller requires only joint position and ve0272-1 70818911200-0009 $ 1 00 0
December 1989

locity information. Thus robustness to parametric uncertainty is achieved without the need for acceleration and jerk measurements as would be required by the global feedback linearization results of [2] and [4].

Adaptive Control of Rigid-Joint Manipulators


Consider the standard model for the dynamics of an n-link rigid robot [8], with q the n-dimensional vector of joint angles, D(q) the n X n inertial matrix, J the diagonal matrix of actuator inertias reflected to the link side of the gears, C(q, q)q the Coriolis and centrifugal terms, g(q) the gravitational terms, and U the n-dimensional vector of input torques.

Introduction
Compared with the large volume of literature available on rigid robot control, relatively little has been published on the control of flexible-joint robots. On the other hand, experimental evidence [ 11 indicates that joint flexibility should be taken into account in both the modeling and control of manipulators if high performance is to be achieved. One approach to control flexible-joint robots is based on the idea of feedback linearization [2]-[4] and generalizes the notion of inverse dynamics for rigid robots. As pointed out in [4], this technique can provide robustness to parametric uncertainty only if the link acceleration and jerk are available for feedback. In addition, the computational burden of this approach is correspondingly greater than the related computed torque scheme for rigid robots. Another approach, which overcomes the need for feedback of link acceleration and jerk, is based on the concept of integral manifold [4]-[6]. In this approach, the flexible-joint robot dynamics are restricted to a suitable integral manifold in state space, after which a reduced-order model can
Presented at the 1989 IEEE International Conference on Robotics and Automation, Scottsdale, Arizona, May 15-19, 1989. Fathi Ghorbel and Mark W. Spong are with the Coordinated Science Laboratory at the University of Illinois at UrbanaChampaign, Urbana, IL 61801. John Y . Hung was with the University of Illinois. He is now with the Department of Electrical Engineering, Auburn University.

@(q)

+ J)ii +

C(q, q)4 + g(q) =

(1)

Although these equations are complex, nonlinear equations for all but the simplest robots, they have several fundamental properties that can be exploited to facilitate control system design. W e state these properties as follows:

(1) The inertia matrix D(q) is symmetric, positive definite, and both D(q) and D(q)- are uniformly bounded as a function of the vector q.

(2) There is an independent control input for


each degree of freedom.

(3) All of the constant parameters of interest, such as link masses, moments of inertia, etc., appear as coefficients of known functions of the generalized coordinates. By defining each coefficient as a separate parameter, a linear relationship results so that we may write the dynamic equations [Eq. ( l ) ] in terms of an n x r matrix of known functions, Y(q, q, q), called the regressor, and an r-dimensional vector 0 of parameters as follows:

(D(q)
=

+ J)ii + C(q, q)q + g(q)


Y(q, q, q)e
= U

(2)

(4) For a suitable choice of C(q, q) (see 191, [lo]), the matrix b(q) - 2C(q, q) is
skew symmetric.

1989 IEEE 9

Recently, Slotine and Li [7] used properties 1-4 above to derive a globally convergent adaptive control algorithm for rigid robots. We briefly discuss their results in this section. Their design scheme will be incorporated into the next section as part of an adaptive control law for flexible-joint robots. Given a twjce continuously differentiable reference trajectory qd(t) in joint space, whose first and second derivatives are bounded, and a constant diagonal positive definite matrix A, define the quantities q, v , r, and a as follows:

q=q-qd, r=q-v,

. = a d -

Aii

a = v

(3)

and consider the following control law, and g represent the terms in where B, j , Eq. (2) with estimated values of the paramD eters and K is a diagonal matrix of positive gains:

e,

U =

(B(q)

+ .f)a + Qq,
- KDr

q)v
(4)

+ g(q)
q

Substituting Eq. (4) into Eq. (2), since = r a and q = r v, we can write the combined system as follow, where ( ) = (.) - ( . ) .

(D

+ J)r + Cr + KDr = (b + J)a + Cv + g


=

Y(q, q, v , a)e

(5)

The quantity 8 = 8 - 0 thus represents the parameter error. Note that the regressor function Y in Eq. (5) does not depend on the manipulator acceleration. The parameter update law is defined in terms of a given symmetric, positive definite matrix r as

the vectors q , and q2 represent the link angles and motor angles, respectively, and we define a diagonal matrix K of joint stiffness coefficients. For notational simplicity, we will assume that all joint stiffness constants are the same, in which case K may be taken as a scalar. As stated previously, it is realistic to assume that the joint stiffness is large relative to the other parameters in the system, whereas the joint damping is small. We idealize the assumption of large joint stiffness and small joint damping by assuming that K is on the order of I/, where E is a small parameter, and we neglect altogether the joint damping terms. The parameter E can be interpreted as follows. First, it is obvious that E is inversely proportional to the square root of the joint stiffness. The choice of the proportionality constant is dictated by design considerations. It is well known in singular perturbation theory that the proper scaling of parameters is crucial to the successful implementation of the theory. See [ I l l for a detailed treatment of scaling in physical systems. Roughly speaking, E should be chosen so that this proportionality constant is in the same range as other parameters (inertias, etc.) in the system. At the same time, E should be small enough to ensure that the settling time of the boundary-layer system, defined in the next section, is sufficiently rapid. The techniques in this paper are thus applicable to robots whose joint stiffness is large enough so that an E can be found satisfying both of the preceding specifications. Under the preceding assumptions, the dynamic equations [Eq. (I)] can be modified as follows [4]:

and Li [7], do not extend immediately to Eqs. (7) and (8).

Main Result
In this section, we show that the control law [Eq. (4)] that achieves robust tracking of the reference trajectory under the assumption of perfect rigidity can be modified easily to control the flexible-joint system of Eqs. (7) and (8). Specifically, we choose a control law of the form
U =

u,(q,,

ql)

+ KfiQI - i12)

(9)

where U, is given by Eq. (4) in terms of q , , and Kr,is a constant diagonal matrix whose diagonal elements are on the order of l / t . The remainder of this section is devoted to showing why the simple modification of Eq. (9) to Eq. (4) achieves the desired result. Let us substitute the control law [Eq. (9)] into Eq. (8), subtract J q , from both sides of the equation, and define e = q2 - q l . The result is

Je

+ K,.e + Ke = U,

Jq,

(10)

Next, define the variable z = Ke, and rewrite Eq. (IO) as

JZ

+ Kf,Z + KZ = K(u,

Jql)

(11)

By our assumption of K and our choice of Kc,,we may write

K = Kl/t2 Kt, = K ~ / E

(1 2 4

(12b)

where K, and Kz are on the order of 1 . Thus Eq. (11) may be written as

D(q,)ql + C(q,, qJ41 + g(qJ

E J ~ E K ~ Z K,z = K,(u, - Jql) +


(7)

8=

-r-lyTr

(6)
Jq2

+ K(qi

q2) = 0

(13)
Equations (7) and (13) may now be written together as

The control law (4) and the update law (6) ensure the convergence of q and q to zero, with all signals bounded (see 191, [IO]). The parameter update law (6) is a standard-gradient-type update law. Its choice can be justified by considering either a straightforward Lyapunov analysis or by appealing to the so-called passivity theorem. See [9] for more details and also a discussion of other update laws.

- K(qi - q2) =

(8)

Flexible-Joint Case
Experimental investigations of industrial manipulators [I] indicate that joint flexibility contributes significantly to the overall dynamics of the system. If one models the joint flexibility, there are now twice the number of degrees of freedom in the system. We let

This model of a flexible-joint robot reduces to Eq. ( I ) as the joint stiffness K tends to infinity and shares many of the properties of Eq. ( I ) , such as linearity in the parameters, a positive definite, symmetric inertia matrix, and the skew-symmetry property 4. However, there is no longer an independent control input for each degree of freedom. The immediate consequences of this are ( I ) that one cannot simply cancel the nonlinearities in the system as in the computed torque scheme for rigid robots, and (2) the system of Eqs. (7) and (8) does not possess the same passivity properties as the rigid model of Eq. (I). As a result, recent adaptive control results for rigid robots, e.g., [12]-[14], including the preceding algorithm of Slotine

D(ql)ql

+ a q , , 4l)ill + g(qJ

=z

(14)

EJZ

+ EK~Z K,z = K,(u, - J q l ) +


(15)

The system of Eqs. (14) and (15) is singularly perturbed. The variables z and z have the interpretation of fast variables, whereas the link variables q1 and ql are slow variables. Using standard results from singular perturbation theory [ 1I], we may approximate the system of Eqs. (14) and (15) by using a quasi-steady-state system and a boundary-layer system as follows. With E = 0, Eq. (15) becomes

10

IEEE Control Systems Magazine

U ,

JGI

(16)

where the overbars indicate that the variables are defined at E = 0. Substituting Eq. (16) into Eq. (14) yields

(mil)+ J ) G + CCSl, GJGI


+ &SI)
=

U ,

(17)

Equation (17) is called the quasi-sready-state system. Note that Eq. (17) is just the rigid model [Eq. (I)] in terms of From Tichonov's theorem [ 1 I], the joint force z(t) and the link angle ql(t) satisfy

cl.

z(t) = Z ( t )

+ q ( ~+ O ( t ) )

(18a)

4I(t)

Sl(4 + W E )

(1 8b)

for t > 0, where T = t / t is the fast time scale, @(E) denotes terms of order t and higher, and q satisfies the boundary-layer

equation: J(d2q/dT2)+ K2(dq/d.r) + Klq = 0


(19)

It follows that the flexible-joint system [Eqs. (14) and (15)] can be written up to order E as

(WSl) + J ) %
= U,

C(q1,

ir1)Sl

+ g(qJ

+ V(f/E) (20) J(d2q/dT2)+ K,(dq/dr) + Klq = 0


(21) Substituting the control law [Eq. (4)] into Eq. (20) and using the update law [Eq. (6)], the system of Eqs. (20) and (21) can be written, after a little algebra, as

estimates are not guaranteed to converge without additional conditions on the reference input, so-called persistency of excitation 1161. However, as shown in 171, the position and velocity tracking errors converge to zero for all suitably smooth reference trajectories. Second, as can be seen from (23), the boundary layer system is globally asymptotically stable in the fast-time scale. The above two properties, however, do not guarantee stability of the flexible joint system by themselves. Several instability mechanisms and methods for overcoming them have been investigated for similar systems in the adaptive control literature [17]. For example, even though the joint resonant modes have been strongly damped by our scheme, limits to the bandwidth of the reference trajectory are still necessary to avoid instability caused by their excitation (recall the examples of Rohrs, et al. [18]). For robustness to external disturbances, one can modify the parameter update law as proposed in 1171 or use persistently exciting inputs [16] to guarantee parameter convergence, and, hence, exponential stability. Other methods, such as the addition of dead zones, can also be used for problems caused by parameter drift.

&?>k&ngrand encoder Motor

Fig. 1 . Sketch o experimental hardware. f

/q, + Mgl sin (4,) + k ( q , - q2) = 0 Jq2 + Bq, - k ( q , - q 2 ) = U

Fig. 2. Model o single-link jexiblejoint f am.


is not sufficient to stabilize the elastic oscillation of the joint, and the correction term in Eq. (9) is therefore required. The related rigid model (Fig. 3), obtained in the limit as k approaches infinity, is

(I

+ J ) q I + Bql + Mgl sin ( 4 , ) = U


(27)

Experimental Results
Next we demonstrate the effectiveness of the proposed control with results from two experiments performed on a specially constructed, single-link. flexible-joint arm. The hardware is shown in Fig. 1. The joint consists of two aluminum plates joined by extension springs. The actuator is a large DC motor directly driving one plate. A hollow aluminum tube (1.5 in. in diameter) about 18 in. long is connected to the second plate. Two incremental encoders provide feedback of the motor and link positions to within rad, while velocity informa1.534 x tion is obtained by filtering the position feedback data. Parameter uncertainty is introduced by clamping payloads to the end of the arm. We model the dynamics of this system (see Fig. 2) as

The damping coefficient B in Eq. (27) is known with sufficient precision that we can simply cancel it in the control law to follow. Only the inertia parameters are affected by varying payloads. Therefore, we define the parameter vector 0 as

(WqJ + J ) i . + C(ql, irdr + KDr


=

Ye

+ q(t/E)

(22) (23)

J(d2q/d?)

+ K2(dq/dT) + Klq = 0

The design of the rigid control law U, is now based on the model of Eq. (27). Using the

Table 1 Nominal Values of the Arm Parameters Parameter


Link inertia, I Rotor inertia, J Rotor friction, B Nominal load, Mgl Joint stiffness, k

We note that only the parameters of the rigid model are updated in this scheme. The joint stiffness and motor inertia need be known only with sufficient precision to determine K 2 to stabilize the boundary-layer system in the fast time scale. Typically, these parameters can be identified with sufficient accuracy off-line and will not change with varying payloads. Although a stability analysis is not provided here, a few remarks are in order regarding the stability (and instability) properties of the above adaptive system. First off, in the absence of external disturbances, the quasi-steady-state system, Eq. (22), at E = 0, is stable-but not, in general, asymptotically stable, since the parameter

Value
0.031 kg-m2 0.004 kg-m2 0.007 N-m-sedrad 0.8 N-m 3 1.O N-m/rad

Iql

sin ( q l ) + k(ql - q2) = 0


(25)

Jij2

+ Bq2

- k(ql - 42) =

(26)

Nominal values for the arm parameters without a payload are shown in Table 1. This system is of the form of Eqs. (7) and (8), except for the nonzero damping at the joint. However, as we will see, this damping

(I

+ J)q + Bq + Mgl sin (q) = U

Fig. 3. Model o single-link rigidjoint f arm.

December 1989

11

algorithm of Slotine and Li [7] for this term, the complete control law with correction is given as
U =

1.8

16 .-

1.41.2U

Table 2 Control Law Gain Values Gain


A KD K,'
g1
g2

8,a

+ 8, sin (q,) + Bql - KDr


42)

6 1.0v 20.8 -

-Link Response -Reference Position

Value
10.0 0.6 0.1 0.001 10.0

+ kJ4,

(29)

0.60.4 ~~

where a and r a r e given by Eq. (3). Choosing r-' = diag (g,, g2), the parameter update law is given by Eq. (9) as

Modified Adaptive Control Experiment


(including correction term) 0.5
1 .o
-

Time (sec)

1.5

8,

-glar

(30)

8,

Fig. .5. Response of the jexible-joint system using mod$ed control.

= -g2 sin ( q , ) r

(3 1)

than the minimum resolution of the encoder after approximately 1 sec. Finally, the control torque is plotted in Fig. 9.
1.6-1
t I I

The computer controller is built around a Motorola 68000 microprocessor and includes a floating-point math coprocessor. The control algorithm is implemented in Pascal with a sampling period of 5 msec (200 Hz). A payload that is approximately 40 percent of the nominal gravitational load of the arm is used in both experiments. The reference trajectory qd(t)is a smooth 90-deg rotation with the arm initially pointing straight down. First, we demonstrate that the rigid-jointbased adaptive algorithm (without modification) is unstable when link variables are used for feedback. Figure 4 shows the reference position trajectory and the unsatisfactory link and rotor position response. The response of the flexible-joint system is bounded only because the joint deflection is limited by mechanical stops. The effectiveness of the modified control law [Eq. (29)] in the second experiment is shown in Figs. 5 and 6. The gains used in the control law [Eqs. (29)-(31)] are shown in Table 2. The reference trajectory and response of the link are shown in Fig. 5 , whereas tracking error is plotted in Fig. 6. The behavior of the estimated parameters is shown in Figs. 7 and 8. By modeling the added payload as a point mass at the end of the arm, the true values of gravitational loading ( M g l ) and inertia (I J ) are analytically estimated to be

(including correctionterm)

Modified Adaptive Control Experiment


(including correction term)

-0.0320'

' '

0.5

'

' '

10 . Time ( s e c )

' I

' '

1.5

'

'

'

2.0

" I

"

'

'

"

Fig. 6. Tracking error.


0.040
0.038NL

,
-

Conclusions

F
Y

0.036

F
Modified Adaptive Control Experiment

90.034 0
0.032 0.030
I l l l l 1 8

(including wrrection term) 8 8 1 1 ' ' l ' 8 1 j 1

f 1.0
1.81.61.4-

5 0.8
0.6

Modified Adaptive Control Experiment


(including correction term)
I '

$2$1.0"0.80.6 0.4 -

---

Link Response Rotor Response Reference Position

0.4i'

'

0.5

' '

I .o Time (sec)

"

' '

'

'

1.5

'

"

2.o

Adaptive Control Experiment


(without correction term 1 0.5
1.0

Fig. 8. Estimate of parameter t12 = Mgl.


1.13 N-m and 0.05 kg-m2, respectively. In general, the parameter estimates are not guaranteed to converge to their true values without a sufficient richness condition on the reference trajectory. We see in this experiment, however, that the position error is less

0.2

Time (sec)

1.5

2 .o

Fig. 4. Link and rotor response of the jexible-joint system without correction term.

We have presented the first adaptive control law for flexible-joint robots under the assumption of weak-joint elasticity. The advantage of the approach is its simplicity in the sense that the overall control law is not much more complicated than a rigid control law. Further work is under way to obtain precise estimates on the ,range of E for which stability is guaranteed. It would be interesting to determine experimentally for a large class of industrial manipulators if the joint flexibility is in the range of validity of this approach. It appears from the detailed treatment of joint compliance in [15] that our approach could be used for most, if not all, current industrial designs. Note, however, that many robots are designed with position and/or velocity sensors located on the motor shafts. Implementation of our approach requires a redesign to measure the link position and velocity as well as motor velocity. (Our experimental results on a single-link arm demonstrate good performance with just link and rotor position feedback.) We should also point out at this point that any control law for rigid robots that provides global tracking can be used as part of our scheme. For example, the adaptive inverse dynamics schemes of Craig et al. [12] and Middleton and Goodwin [13] could have been used instead. We have chosen to illus-

12

IEEE Control Systems Magazine

trate our approach using the algorithm of Slotine and Li because of its elegance and simplicity and because it does not require joint acceleration for its implementation. The approach in this paper i s most closely related to the integral manifold approach of 141 and 161. The reason that the integral manifold approach cannot be used in this context is that the flexible-joint system restricted to the integral manifold of [4] and [6] no longer satisfies the property of linearity in the parameters, which is crucial to virtually all adaptive control results. Our scheme amounts to dropping terms of order E and higher from the integral manifold expression of [4]. In this way, we retain linearity in the parameters, but our result no longer can be expected to hold independently of E , as it would if the exact integral manifold could be used.

171 J.-J. E. Slotine and W . Li, "On the Adaptive Control of Robot Manipulators." 1986 ASME Winter Annual Meeting. Anaheim. CA. vol. 3. pp. 51-56. Dec. 1986. [8] M. W. Spong and M. Vidyasagar, Robot D\namic~awl Control. New York: Wiley. 1989. 191 R. Onega and M. W . Spong, "Adaptive Motion Control of Rigid Robots: A Tutorial." Proc. o f I E E E Con$ OII DeciSi017 and Cotirrol. Austin. TX, pp. 1515-1584. Dec. 1988. IO] J:J. E. Slotine and W. Li, "On the Adaptive Control of Robot Manipulators." Intl. J . Robotics Res.. vol. 6. no. 3. pp. 49-59, 1987. 1 I ] P. V. Kokotovic, H. K . Khalil, and J. O'Reilly . Singufur P~rturhutio~i Merhody in Coiirrol: Anulysis and Desigii. London: Academic Press. 1986. 1121 J . J. Craig. P. Hsu, and S. Sastry, "Adaptive Control of Mechanical Manipulators."
Pror.
of

Assistant at the Coordinated Science Laboratory at the University of Illinois since 1987 and at the U.S. Army Construction Engineering Research Laboratory since 1988. His current interests include the application of adaptive control theory and singular perturbation theory to the control of flexible-joint robot manipulators. In addition, his interests include the design and construction of an in-pipe inspection robotic crawler for small-diameter pipes.

John Y. Hung is an Assistant Professor in the

Acknowledgments
The research for this paper was partially supported by the National Science Foundation under Grant 85-16091. The authors would like to thank Petar Kokotovic and Romeo Ortega for providing them with valuable insight into the subject matter of this paper.
[I31

f E E E Inrl. Cor$

ori

Rohorirs and

1141

References
L. M. Sweet and M. C. Good. "Redefinition of the Robot Motion Control Problem: Effects of Plant Dynamics, Drive System Constraints, and User Requirements."
Proc. of 23rd IEEE Conf on Decision und Control, Las Vegas, NV. pp. 724-730.

1151

1161

1171

1984. A. De Luca, "Control Properties of Robot Amis with Joint Elasticity," Proc. of I d
Sjwp. on Math. T h e o n of Neru,orks mid Sysr., Phoenix, AZ, June 1987.

[I81

A. De Luca. A. Isidori. and F. Nicolo. "An Application of Nonlinear Model Matching to the Control of Robot Arms with Elastic Joints," 1st IFAC Symp. on Robot Cotitrol, Barcelona. 1985. M. W. Spong. "Modeling and Control of Elastic Joint Manipulators," J . Dyn. S w . , Meas., Conrr.. vol. 109. pp, 310-319, 1987. K. Khorasani and M. W. Spong, "Invanant Manifolds and Their Application to Robot Manipulators with Flexible Joints," Proc. of IEEE Inrl. Conf 011 Robotics and Automation, St. Louis, MO, pp. 978-983, Mar. 1985. M . W. Spong, K . Khorasani, and P. V . Kokotovic, "An Integral Manifold Approach to the Feedback Control of Flexible Joint Robots," IEEE J . Roborics and Auromarion. vol. RA-3, no. 4, pp. 291-300. Aug. 1987.

San Francixo. CA, Mar. 1986. R. H. Middleton and G. C. Goodwin. "Adaptive Computed Torque Control for Rigid Link Manipulators." Sysr. Coritr. Lett.. vol. IO. pp. 9-16. 1988. N . Sadegh and R. Horowitz. "Stability Analysis of an Adaptive Controller for Robotic Manipulators." IEEE I d . COI$ o r i Robotic\ u i i d Aurormrtiori, Raleigh. NC. 1987. E. 1. Rivin. Mecha~irralD e s i p ofRobots. St. Louis, MO: McGraw-Hill, 1988. B. D. 0. Anderson. "Exponential Convergence and Persistent Excitation." Proc. 21 f E E E Corif. Drcisiori ciitd Conrr. . Orlando, FL. pp. 12-17. Dec. 1982. P. A. loannou and P. V. Kokotovic, "Instability Analysis and Improvement of Robustness of Adaptive Control," Autoniarica. vol. 20. no. 5 . pp. 583-594, Sept. 1984. C. E. Rohrs, L. Valavani. M. Athans, and G. Stein. "Robustness of Adaptive Control Algorithms in the Presence of Unmodeled Dynamics." Proc. 2 / IEEE Conf: on Decisioii ~ n Cotif.. Orlando. FL. pp. 3-1 l , d Dec. 1982.
Autoniuriori.

Electrical Engineering Department at Auburn University. He received the B . S . degree from the University of Tennessee (1979), the M.S.E. degree from Princeton University (1981). and the Ph.D. degree from the University of Illinois at Urbana-Champaign (1989). As an engineer at Johnson Controls from 1981 to 1985, he designed microcomputer hardware and firmware for environmental control systems. Dr. Hung enjoys the design and control of mechatronic systems. and his current research involves the control of robots.

JIarh \V. Spong \ \ a h horn i n IUS2 i n \'men. Ohw. He received thc H..A Jegree. n1.ign.i i u n i

,1

Fathi Ghorbel was bom in 1962 in Sfax. Tunisia. He received the B . S . degree with honors in mechanical engineering from the Pennsylvania State University in 1985 and the

chanical engineering from Camegie-Mellon University in 1987. Presently. he is working toward a Ph.D. degree in the Department of Mechanical and Industrial Engineering at the University of Illinois at Urbana-Champaign. He has been a Research

lauds and Phi Bcts Kappii in niathcniatii\ and ph! ,ic, tram Hirarii c'~illc.;lc. Hir;Ini. Ohiti. in 1975. the 41,s. tlcgrcc i n m.ithciiiatic\ tnmi Sen hlcxi~o Ststc Univsr\ir! i n IY77. and thc h1.S. and L) Sc degrees in systems science and mathematics from Washington University in St. Louis, in 1979 and 1981, respectively. He has been on the faculty at Lehigh University and Comell University. and during the summer of 1984 he was with the Control Technology Branch. General Electric Company. Schenectady. New York. In August 1984. he joined the University of Illinois at UrbanaChampaign. where he currently holds the titles of Associate Professor, Department of General Engineering; Associate Professor. Department of Electrical and Computer Engineering: Research Associate Professor, Coordinated Science Laboratory; and Director, Department of General Engineering, Robotics Laboratory. His main interests are in robotics and nonlinear control theory. Dr. Spong is a past Associate Editor of the IEEE Coiirrol Systems Mugwine and is currently an Associate Editor of the IEEE Transactions on Autoniurir Control.

December

19x9

13

You might also like