A complete fault diagnostic system for automated vehicles

14th World Congress oflFAC

A COMPLETE FAULT DIAGNOSTIC SYSTEM FOR AUTOMATED V ...

B-Ie-02-2

Copyright © 1999 IFAC

14th Triennial World Congress, Beijing, P.R. China

A COMPLETE FAULT DIAGNOSTIC SYSTEM FOR AUTOMATED VEHICLES R. Rajamani I , A. Howell 2 , C. Chen J , J.K. Hedrick 2 and M. Tomizuka 2

of Minnesota, MinneapoJis, MN 55455 ([email protected]) California PATH, University of California, Berkeley, CA 94720 3 Applied Materials, Santa Clara, CA 95054

J University 2

Abstract: A "complete" fault diagnostic system is developed for automated vehicles operating as a platoon on an automated highway system. The diagnostic system is designed to monitor the complete set of sensors and actuators used by the lateral and longltudinal controllers of the vehicle, including radar sensors, magnetometers and intervehicle communication systems. A fault in any of the twelve sensors and three actuators is identified without requiring any additional hardware redundancy. The diagnostic system uses parity equatIOns and several reduced-order nonlinear observers constructed from a simplified dynamic model of the vehicle. Nonlinear observer design techniques are used to guarantee asymptotically stable convergence of estimates for the nonlinear dynamic system. Different combinations of the observer estimates and the available sensor measurements are then processed to construct a bank of residues. The paper analytically shows that a fault in any of the sensors or actuators creates a unique subset of these residues to grow so as to enable exact identification of the faulty component. This conference paper is a brief summary of the full paper submitted for journal publication. Both simulation and experimental results that demonstrate the effectIveness of the fault diagnostic system in the presence of various faults are included in the journal version of this paper. Copyright © 1999lFAC 1. INTRODUCTION The Automated Highway Systems (AHS) Pro~am at California PATH] aims to reduce congestIOn on highways by closer packing of automatically controlled vehicles into platoons. Studies of automatic control of the longitudinal and lateral motion of cars have been undertaken to establish feasibility of the AHS concept ([7}, [11], [14], [15J, [16] and [19]). These experimental studies have demonstrated the viability of automatic driver-less control of cars so as to achieve high traffic throughput on highways. Statistical studies of highway accidents have shown that over 90% of accidents occur due to driver-related errors. The AHS system eliminates these accidents by drastically reducing the burden of the driver. The reliability and safe operation of the hardware is, however, of increased lmportance. The present paper deals with automated monitoring and diagnostics of all the sensors and actuators used by the lateral and longitudinal control systems. A summary of important results on fault diagnostics using analytical redundancy and fault diagnostics for automated vehicles is included in the journal version of this paper. 2. VEHICLE DYNAMICS MODEL 2.1 Simplified Longitudinal Vehicle Model I Partners for Advanced Transit and Highways

The reader is referred to Cho and Hedrick [2] and Hedrick, et al. [71 for a detailed model of the car's longitudinal dynamics. We present here the simplified model used very 'effectively for control deSIgn in [7]. Under the assumptions that there is no slip between the tire and the road and that the torque converter is locked, the longitudinal velocity of the j th vehicle in the vehicle platoon can be related to the angular velocity of the engine through the gear ratio and tire radius as follows : x} = Vj = (Rhw e )/ (1) where R = gear ratio and h = tire radius. The dynamics relating engine speed We to the pseudo-inputs "net combustion torque" Tnel and brake torque Thr and aerodynamic losses can be mode led by

ciJ

Tnel - CaR 3h3 W e 2

-

R(hFf + Tor)

e= - - - : . - - - - - - - - - - " - - - ' - ' - -

Je

(2)

where the effective inertia reflected on the engine side is given by (3) J" = le + (Mh 2 + i(,,)R 2 A description of all the variables and their symbols can be found in Appendix A. The pseudo-input Tne! is related to the throttle angle a (the actual control or actuator input) by the following dynamics . Steadystate engine maps define Tne , as a nonlinear function

922

Copyright 1999 IFAC

ISBN: 0 08 043248 4

A COMPLETE FAULT DIAGNOSTIC SYSTEM FOR AUTOMATED V ...

the relative yaw angle of the vehicle with respect to road co-ordinates. The front wheel steering angle /5 is the control input to regulate lateral and yaw motion of the automated vehicle. Note also that the road reference co-ordinates rotate on curves. This effect is represented as the desired yaw rate id in equation (6a). The other variables and symbols used in the model are described in Appendix 1.

of engine speed (Oe and the mass of air in the intake manifold T"e' = Tnel ({tJe, mu)' These steady-state maps are available for each car from the manufacturer and are obtained from standard dynamometer tests conducted after the engine has been designed and built. The mass flow rate of air in the manifold is defined by (4) where

In implementing the lateral control system, vehicle lateraI displacement can be measured by an on-board machine vision system [3], [9] or by a magnetic sensor system which measures displacement by measuring the magnetic field from discrete ma~nets buried every 1.2 meters in the center of the road l21]. Magnetometers mounted on the car serve as sensors to measure the magnetic field. The output equation for lateral displacement is YI = Cl x with C[=[l 0 d., 0] (7) where d s is the longitudinal distance between the

mai = M4X TC(a) PRl(m a ) (5) MAX is a constant dependent on the size of the throttle body, TC(a) is nonlinear invertible function of the throttle angle, P RI (ma) is the pressure influence function which describes the choked flow relationship which occurs through the throttle valve and mar) is the mass flow rate into combustion chamber (nonlinear function of Pm and We available from the manufacturer in the form of a table)

magnetometer and the vehicle c.g.

A complete combined simulation model including a realistic representation of both the lateral and longitudinal dynamics is presented in Peng and Tomizuka [15]. A simplified lateral dynamics model incorporating only the lateral translation and yaw degrees of freedom is used for controller design and is available in [I], [15]. The simplified lateral dynamics model is derived by linearizing vehicle lateral dynamics with respect to the road centerline reference co-ordinates and is shown below: x=Ax+BIt5+B28d (6 a) with

= [Yeg

Y c'g

D.li

In addition, on-

board inertial sensors such as a yaw-rate sensor and a lateral accelerometer are available and are typically used by the lateral control system. The outputs equations for these sensors are given below:

2.2 Simplified Lateral Dynamics Model

x

14th World Congress ofIFAC

Yaw-rate sensor: Y2 = C 2 x with C 2 = [0 0 0 1] (8) and lateral accelerometer: Y3 = C3X + bl2 t5 + bn id with (9)

2.4 Sensors and actuators A review of the vehicle dynamics model and the controllers shows that the following sensors are needed by the longitudinal and lateral control systems

t1i[

Steering angle sensor, Yaw rate sensor, Magnetometer, Lateral accelerometer, Wireless intervehicle radio communication, Radar, Longitudinal accelerometer, Wheel speed sensor, Throttle angle sensor, Brake pressure sensor, Manifold temperature sensor, Manifold pressure sensor, Engine rpm sensor.

(6b)

The steering actuator, throttle actuator and the brake actuator are the three actuators used by the control system which need to be monitored. Throttle angle, brake torque and steering angle are the corresponding actuator inputs.

3. ANALYTICAL REDUNDANCY 3.1

A

= o

l~

a22 0

a:3 0

a~4 j

o

a42

Q43

G44

I

Overview of the Fault Diagnostic System Design Procedure If three sensor sisnals are algebraically related so that there exist three mdependent parity equations relating these signals, then the three residues obtained from these parity equations can be used to determine exactly which of the three sensors is at fault (assuming that not more than one sensor becomes faulty at the same time). This fact is used in section 3.3 to determine if either of the wheel speed, engine speed or range rate sensors are at fault. Once it is ensured that the engine speed, wheel speed and range rate sensors are all operational, several reduced order observers are deSigned using these sensor signal measurements. These reduced-order observer are used to diagnose the health of the other sensors and actuators. For example, a second order

(6c)

Here Y cg is the lateral dIsplacement of the c.g. of the vehicle with respect to road co-ordinates and ,1.£ is

923

Copyright 1999 IF AC

ISBN: 008 0432484

A COMPLETE FAULT DIAGNOSTIC SYSTEM FOR AUTOMATED V ...

14th World Congress ofIFAC

x

x)

observer that utilizes engine speed measurement and commanded throttle angle as inputs is used to estimate engine manifold pressure and engine speed. A comparison of the estimated and measured engine speeds is used to detennine if the throttle actuator is at fault. The estimated manifold pressure is used to determine if the manifold pressure sensor is at fault.

i = A + w(x, u) + L lY - C (11) converge exponentially to the states of the system defined by (10).

Section 3.6 describes the design of a first order observer utilizing engine speed measurement and commanded brake torque to diagnose the health of the brake actuator. Section 3.7 describes an observer to estimate vehicle speed using an accelerometer and magnetic markers. This observer is used to diagnose the health of the peak detection ability of the magnetometer. Sec~ion 3.4 ?escribes the d~sign o~ an observer that estImates mter-car spacmg usmg magnetic markers and the difference of wheel speeds in the two cars. This is used to diagnose the health of the radar sensor.

SolveATP+PA+y2PP+I--2-
An explicit analytical solution for the observer gain matrix L can be provided as follows:

eTe

(12a)

y

and then choose p-IC T L=-2y 2

(l2b)

3.3 Speed Sensor Redundancy The longitudinal speed of the vehicle can be obtained by three different methods, as described in [5]. 1) From the wheel speed sensor, assuming no slip between the tire and the road. 2) From the engine speed sensor, assuming that the torque converter is locked. 3) From the range rate signal of the radar combined with the preceding car speed obtained through communication.

In the case of the lateral control system, the three sensors - yaw-rate sensor, lateral accelerometer and magnetometer - are used to design three different observers. Each observer uses two of these three sensor measurements and estimates the third sensor signal. This can be used to uniquely identify a fault in anyone of the three sensors. In addition 1) the wheel sp.eed sensor whose he
The following three residues are then calculated by different combinations of the above three longItudinal velocity signals. RI = wheel speed/engine speed residual R2 = wheel speed/radar range rate residual R3 = engine speed/radar range rate residual

usin~

The following truth table can then be used to detect a fault in anyone of the three speed sensors.

From the above summary, it is clear that observer design plays a key role in the fault diagnostic system design. Since the mathematical models for the vehicle dynamics are nonlinear, it is a challenge to ensure that the observers for the system are stable. robust and can be designed to have required rates of convergence. The observer design procedure used is described in section 3.2. Section 4 integrates the parity equations and observers designed in section 3 to create a systematic methodology for fault diagnostics that can uniq.uely identify the particular sensor or actuator that IS at fault.

3.2 Observer Design for Nonlinear Systems The design of exponentially stable observers for systems with non linear dynamics will be based on the following results from Rajamani and Cho ([17], [18]) Given a nonlinear system = A x + (x,u)

x

y=Cx

(l0)

where. 1) A is stable and
3.4 Inter-Car Spacing The following observer is proposed in this paper to obtain a redundant estimate of mter-car spacing. This observer uses a magnetometer measurement to count the number of magnetic markers passed by the two vehicles .

Si

o-a ;]

i

=V -V i _ 1 +k,[(nl -n 2 )L+,) (13) where n1 - n2 is the difference in the number of markers passed ~y the two vehicles and L is the mter-marker spacmg. 3.5 Throttle Actuator and Manifold Pressure Sensor Faults We propose using one second order non linear observer to estimate both the engine speed and manifold pressure utilizing commanded throttle angle This and engine speed measurement as inputs. observer

@

Tne/(we>ma)-caR3h3We 2 -RhFj

e= -~~~~~~~~~~~~-

Je

2) the distance to undetectability of the pair (A,C) is larger than the Lipschitz constant y of the nonlinear function
(14)

+f)(liJ e -We)

~a = M4X TC(a

des)

PRl(m a )

(15a)

-mao(me,ma)+f2(liJe -me) PmV =m"RgTm

there exists a matrix L such that the estimates from the following observer

05b)

924

Copyright 1999 IF AC

ISBN: 008 0432484

A COMPLETE FAULT DIAGNOSTIC SYSTEM FOR AUTOMATED V ...

14th World Congress ofIFAC

the car can be estimated by an observer using these two sensors. The observability matrix C/ (C}A)T {C 3 A)T j (19) has rank 4 which makes the states completely observable. The residue

can be designed to be asymptotically stable in the absence of throttle actuator faults by proper choice of gains f. 1 and f! 2. Since the engine speed can never

lcl T

physically exceed 4000 rpm, the nonlinearity me 2 can be regarded as locally Lipschitz and the Lipschitz constant of ca R3 h'" (7J c 2 can be calculated. A throttle actuator fault will cause the residue between estimated and measured engine speeds to grow. Assuming no fault in the engine speed measurement sensor, the groVoith in this residue can then be used to diagnose a throttle actuator fault. If a throttle actuator fault has not occurred, the residue between measured and estimated manifold pressure can be used for diagnostics of the manifold pressure sensor.

R14 = C 2

rl T

has rank 4 which makes the states completely observable. The residue

x - Y3

R I5 = C 3

(17)

,

C

= -

t:.E:

Vehicle

Speed

Estimation

using

044

t:.E:

24

14

YCg(t)

=

I

Ycg dt + Ycg(O)

(25)

The residue

The following observer using the accelerometer on the car and a magnetometer measurement to count the number of magnetic markers passed by the car can be used to estimate car velocity ~ n1 L , v=a+k ( -T- v ) (18) r v

3.9 Estimation of yaw-rate If we assume that the lateral acceleration and magnetometer sensors are working, the yaw-rate of

043

observable, Ycg can be estimated from the lateral accelerometer and yaw-rate sensor. If the initial condition Ycg(O) is known, then Ycg can be estimated as follows :

Accelerometers and Magnetic Markers

3.8 Communication fault 1) The car that communicates ensures that its sensors are not faulty. 2) In the receiving car, if no packet is received, the information from the last packet is frozen till the next packet arrives. 3) If no packet is received for more than 3 consecutive cycles, a communication fault is declared.

042

(24) is, however, completely observable from these two outputs. This means that while Ycg is not

than the Lipschitz constant of me 2 in order to ensure stability of the estimation error dynamics in the absence of faults. 3.7

lateral

~'{:~}~~'[,:~:ve~:, "::~ ]d{Y~:}i:'[~bb::]n:: :Y[!b::]S

R

and j..l (t) is nonzero le le only when there is a fault in the brake actuator. Since the engine speed can never physically exceed 4000 rpm, the nonlinearity We 2 can be regarded as locally Lipschitz. Since the observer has access to a measurement of Wc , the gain jI can be chosen larger Ca

if the

3.11 Estimation of Lateral Displacement If we assume that the lateral acceleration and yawrate sensors are working, one could try and estimate lateral displacement, usually measured by the magnetometer. However, the observability matrix lC 2 T C3 T (C 2 A)T (c)Al J (23) has a rank of only 3 which means the complete state is not observable with these measurements!

+f(me -we)

where a =

(22)

can then be used to determine acceleration sensor is faulty.

(16)

R 3h 3

(20)

3.10 Estimation of lateral acceleration If we assume that the yaw-rate and magnetometer sensors are working and that the steering angle sensor is not faulty, the lateral acceleration of the car can be estimated by an observer using these three sensors. The observability matrix C/ (C I A)1 (C 2Al J (21)

3.6 Brake actuator! sensor fauU Under the action of brakes, the throttle actuator is not used. The variable Tnel can therefore be set to zero in eqn. (2). The following observer can then be used to estimate the engine speed under the action of the braking actuator

The dynamics of the estimation error are then given by i\ = -awe 2 - fiJ - cp (t)

x- Y2

can then be used to determine if the yaw-rate sensor is faulty.

RI6

= Ycg(t)- Ycg(t)

(26)

can then be used to detennine if the magnetometer sensor is faulty.

3.12 Fault Diagnotics of Steering Angle Sensor J Steering Actuator If we assume that the steering wheel angle and the vehicle wheel angle are both measured, then the two sensors are related by a scaling factor. The steering wheel angle, vehicle wheel angle and commanded steering angle are related by three independent parity equations. The following three residues are then calculated by using different combinations of the above three signals : R 11 = commanded steering angle/ measured steering angle, R}2 = commanded steering angle! measured vehicle wheel angle, R 13 = measured steering wheel angle/ measured vehicle wheel angle

925

Copyright 1999 IF AC

ISBN: 008 0432484

A COMPLETE FAULT DIAGNOSTIC SYSTEM FOR AUTOMATED V ...

14th World Congress ofIFAC

Z27

The following truth table can then be used to detect a fault in anyone of the following three components : steering actuator, steering angle sensor, vehicle wheel angle sensor.

4. A SYSTEM FOR AUTOMATED FAULT DIAGNOSTICS Table I summarizes 27 different signals to be used in the fault detection and identification scheme. Some of the signals are directly measured while others are estimates obtained from the observers discussed in the previous section. Table 2 summarizes 16 different residues calculated using combinations of the signals from the previous table. It is assumed that the failure of any sensor would cause a residue computed by subtracting this sensor measurement from an estimate of its signal using other measurements to grow.

z12

=ac

z13

=a

zI4 = ZI5

Tbr

=S

com

i

ZI6 =V i 217

= a syn

Zl8 = We br zl9

= Yd'

Z2CJ

=!ii

Z21

= YeI'[

zn

= y,.

z23

z25

= t:"i = Y cg =6

Z26

= Odes

Z24

commanded throttle angle throttle angle sensor commanded brake torque observer of eqn. (17) observer oteqn. (24) synthetic acceleratIOn (c~~culated by longitudinal controller observer of eqn. (22) lateral position magnetometer yaw rate sensor lateral acceleratIOn observer ot section 3.10 observer of section 3.8

from

observer of sectIOn 3.9 steermg angle trom encoder desired steenng angle calculated by lateral controller

I measured vehicle wheel angle

Table 2'. Calculation of residues RESIDUES SENSORS! ACTUATORS INVOLVED wheel speed, engme speed RI = ZI - Z2 R2 -z S -Z 4 -Z 1 ra~ar range rate, wheel speed communication radar range rate,engme speed, R3 "" z5 - z4 - z2 communication radar range, wheel speed, R4 =z3- z I5 magnetometer, communication acce lerometer Rs = Z17 - Zg accelerometer, markers, R6 = Z16 -zl wheel speed sensor throttle actuator, engine R7 = z10 - Z6 speed throttle angle sensor, R8 = Z12 - z13 throttle actuator manifold pressure, throttle R9 =zll -z7 actuator brake actuator, engine speed RIO = ZlB - Z6 steering angle sensor, Rll = Z25 - z26 steering actuator vehIcle wheel angle sensor, Rl2 = Z25 - z27 steering actuator steermg angle, wheel angle RI3 = Z26 - Z27 yaw-rate, steering angle, R14 = ZI9 - Z22 lateral accelerometer, wheel speed magnetometer, steenng angle, Rl5 = z20 - z23 lateral accelerometer, wheel speed Magnetometer, ?,:aw-rate, R 16 = Z21 - Z24 steering angle whee speed

Truth Table for Steering Angle I Steering Actuator Fault Detection

Table 1 : Bank of si~nals ~or f au I t d'I
=6w

By processing the above 16 residues, it is possible to identifY a fault in any of the sensors or actuators. Table 3 shows how a fault in any of the sensors or actuators causes a unique combination of residues to grow. To detect and identifY faults, a systematic algorithm obtained from Table 3 and presented in the journal version of this paper can be used. Table 3 : Behavior of residues under sensor!actuator faults FAULTY RESIDUES THAT SENSOR! ACTUATOR TURN HIGH wheel speed sensor 1,2,46,14,15,16 engine speed sensor 137910 radar range rate sensor 2 3 radar range sensor 4 long. accelerometer 5,6 4,6,15,16 maf-etometer 'pea -detection ability throttle actuator 7,8 9 throttle angle sensor 8 manitold pressure sensor 9 10 brake actuator steering angle sensor 11 12 13 14 15 16 magnetometer 4 6 15 16 yaw-rate sensor 14 16 lateral accelerometer 14, 15 steenng actuator 11, 12 wheel angle sensor 12 13 CONCLUSIONS

926

Copyright 1999 IF AC

ISBN: 008 0432484

14th World Congress oflFAC

A COMPLETE FAULT DIAGNOSTIC SYSTEM FOR AUTOMATED V ...

The diagnostic system developed in this paper provides a methodology to contmuously monitor all the sensors and actuators of the longitudinal and lateral controllers so as to ensure their health. Simulation and experimental results on the performance of the diagnostic system are included in the journal version of this paper. The fault diagnostic system works well when simulated with a detailed vehicle model incorporating realistic unmodeled dynamics. Experimental results show that the magnetic observer works extremely effectively in detecting radar faults and replacing the radar sensor in the event of a fault.

[14]

[15]

[16]

REFERENCES [I]

Chen, C. and Tomizuka, M., "Vehicle Lateral Control on Automated Highways : A Backstepping Approach", Proceedings of the IEEE Conference on Decision and Control, December 1997. [2] Cho, D. and Hedrick, J.K ., "Automotive Powertrain Modeling for Control", ASME Transactions on Dynamic Systems, Measurement and Control, 111 (4), December 1989. [3] Dickmanns, E.D. and Graefe, V., "Applications of Dynamic Monocular Machine Vision", Machine Vision and Applications, vol. I, pp. 241-261,1988. [4] Douglas, R.K., Sreyer, D.L. et ai, "Fault Detection and IdentIfication with Application to Advanced Vehicle Control Systems", California PATH Research Report UCB-ITS-PRR-95-26 , 1995. [5] Garg, V., "Fault Detection in Nonlinear Systems: An Application to Automated Hi~hway Systems", Ph.D. Dissertation, UniverSity of California at Berkeley, 1995. [6] Garg, V. and Hedrick, J.K., "Fault Detection Filters for a Class of Nonlinear Systems", Proceedings of the 1995 American Control Conference, pp. 1647-1651, June 1995. [7] Hedrick, J.K., McMahon, D., Narendran, V.K. and Swaroop, D., "Longitudinal Vehicle Controller Design for IVHS Systems", Proceedings of the 1991 American Control Conference, Vol. 3, 'pp. 3107-3112, June 1991. [8] Hingwe, P. and Tomlzuka, M., "Two Alternative Approaches to the Design of Lateral Controllers for Commuter Buses Based on Sliding Mode Control" , Proceedings of the 1995 ASME International Mechanical Engineering Congress and Exposition. [9] Luong, Q.T., Weber, J., Koller, D. and Malik, l ., "An Integrated Stereo-Based Approach to Autoglatic Vehicle Guidance", Proceedings of the 5 ICCV, 1995. [10] Swaroop, D ., Hedrick, J.K. , et aJ "A Comparison of Spacing and Headway Control Laws for Automatically Controlled Vehicles", Vehicle System DynamiCS Journal. [11] Tomizuka, M. and Hedrick, J.K., "Automated Vehicle Control for IVHS Systems", Proceedings of the IFAC Conference, Sydney, 1993. [12] Patwardhan, S . and Tomizuka, M., "Robust Failure Detection in Lateral Control for IVHS". Proceedings of the 1992 American Controi Conference, June 1992. [13] Patwardhan, S., Tomizuka, M. and Zhang, W., "Enhancing the Performance of Direction Sensitive Filters for Multiple Failures",

[17]

[18]

[19]

[20]

{21]

Proceedings of the 1992 American Control Conference, June 1992. Patwardhan, S. Tan, H.S. and Guldner, J, "A General Framework for Automatic Steering Control System Analysis", Proceedings of the American Control Conference, pp. ] 598- ]602, 1997. Peng, H. and Tomizuka, M., "Preview Control for Vehicle Lateral Guidance in Highway Automation", ASME Journal of Dynamic Systems, Measurement and Control, Vol. 115, No. 4, pp. 678-686, 1993. Pham, H., Hedrick, J.K. and Tomizuka, M. , "Combined Lateral and Longitudinal Control of Vehicles for IVHS", Proceedings of the 1994 American Control Conference, vol . 2, pp. 12056, Baltimore, 1994. Rajamani, R., "Observer Design for Lipschitz Nonlinear Systems", IEEE Transactions on Automatic Control, Vol. 43 , No. 3, pp. 397401, March 1998. Rajamani, R. and Cho, Y.M., "Existence and Design of Observers for Nonlinear Systems : Relation to Distance to Unobservability", International Journal of Control, Vol. 69, No. 5, pp. 717-730, May 1998. Tan, H.S., Guldner, 1., Chen, C. and Patwardhan, S., "Changing Lanes on Automated Highways with ~ook-Down Reference Systems", Proceedmgs of the 1998 IFAC Workshop on Advances In Automotive Contro!", pp. 69 - 74. White, J.E. and Speyer, J.L., " Detection Filter Design :Spectral Theory and Algorithms", IEEE Transactions on Automatic Control, 32, No. 7, pp.593-603, !987. Zhang, W. and Parsons, R.E., "An Intelligent Roadway Reference System for Vehicle Lateral Guidance IControl", in Proceedings of the American Control Conference, San Diego, CA, USA, 1990, pp. 281-286.

927

Copyright 1999 IF AC

ISBN: 0 08 043248 4