Sie sind auf Seite 1von 7

196

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO. 2, MARCH 2005

Experimental Results on Adaptive Output Feedback Control Using a Laboratory Model Helicopter
Ali T. Kutay, Anthony J. Calise, Senior Member, IEEE, Moshe Idan, and Naira Hovakimyan, Senior Member, IEEE
AbstractExperimental results are presented that illustrate a recently developed method for adaptive output feedback control. The method permits adaptation to both parametric uncertainty and unmodeled dynamics, and incorporates a novel approach that permits adaptation under known actuator characteristics including actuator dynamics and saturation. Only knowledge of the relative degree of the controlled system within the bandwidth of the control design is required. The controller design was tested by controlling the pitch axis of a three degrees-of-freedom (DOF) helicopter model, using attitude feedback through a low-resolution optical sensor. Index TermsAdaptive control, neural networks (NNs), output feedback, uncertain systems.

I. INTRODUCTION

ESEARCH in adaptive control is motivated by many applications, modeling for which may vary from having accurate low-frequency models in the case of rigid bodies, to having no reasonable set of model equations in the case of active control of ows and combustion processes. Moreover, growing interest in the use of novel actuation devices introduces additional uncertainty into the problem. Regardless of the extent of the model accuracy that may be present, an important aspect in any control design is the effect of parametric uncertainty and unmodeled dynamics. Intensive research efforts have been devoted to adaptive control of uncertain nonlinear systems. Universal approximation capabilities of neural networks (NNs) have been widely employed to model complex nonlinear physical phenomena. The majority of papers on this subject conclude with numerical simulations that illustrate the advantages inherent in adaptive approaches. Very few of such papers include experimental results. Successful experimental evaluations of NN-based adaptive control methods reported in the literature, several of which are detailed below, date back to the early 1990s. A NN-based adaptive state feedback controller has been tested on a unicycle robot in [1], with stability guaranteed based on Lyapunov analysis. An output feedback direct NN-based adaptive controller was tested

Manuscript received August 15, 2003; revised April 19, 2004. Manuscript received in nal form June 9, 2004. Recommended by Associate Editor S. Kim. A. T. Kutay and A. J. Calise are with the School of Aerospace Engineering, Georgia Institute of Technology, Atlanta, GA 30332 USA (e-mail: ali_kutay@ae.gatech.edu; anthony.calise@ae.gatech.edu). M. Idan is with the Faculty of Aerospace Engineering, Technion-Israel Institute of Technology, Haifa 32000, Israel (e-mail: moshe.idan@technion.ac.il). N. Hovakimyan was with Georgia Institute of Technology, Atlanta, GA 30332 USA. She is now with the Department of Aerospace and Ocean Engineering, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061 USA (e-mail: nhovakim@vt.edu). Digital Object Identier 10.1109/TCST.2004.839563

on a force control servomechanism in [2], with stability shown for a linear positive real plant. A full state feedback neural identier and controller has been tested on a four-bar linkage load system in [3], where the structure of the nonlinearity is assumed known. A disk with a high-friction load mounted to a motor has been used as a test bed for evaluation of a full state feedback direct adaptive control method in [4], in which the structure of the nonlinearity is assumed known. A full state feedback neural dynamic adaptive controller with no stability guarantee has been tested on a selective compliant assembly robot arm (SCARA) robot in [5]. The method in [6] uses a NN trained online with a recursive least squares training algorithm without any stability analysis to approximate the inverse model of the plant. An output feedback method that uses a high-gain observer to estimate states and integrates NNs into an adaptive robust control method has been tested on a linear motor drive system in [7]. Stability and transient performance are guaranteed in this paper, and asymptotic tracking is shown for the case where only parametric uncertainty exists in the system. A NN-based output feedback variable structure control method with guaranteed stability evaluated on a four-bar linkage system is presented in [8]. Most of the earlier experimental works either lack a stability analysis, or assume that the structure of the nonlinearity is known. Also, to our knowledge, no experimental results of an adaptive output feedback control method that can be applied to systems with unmodeled dynamics and actuator nonlinearities such as saturation have been reported. In this paper, we evaluate experimentally the theoretical results of a recently developed adaptive output feedback method [9] in conjunction with a method for protecting the adaptive element from nonlinear actuator characteristics such as saturation and possibly neglected dynamics [10]. The model used in the experiments is a laboratory-scale bench-top three degrees-of-freedom (DOF) helicopter produced by Quanser Consulting Inc. We consider control of the 1-DOF pitch motion of the helicopter to evaluate the single-inputsingle-output (SISO) design approach of [9]. There are signicant nonlinearities in the system due to friction and the aerodynamics of the propellers. Only the angular position of the helicopter is used for feedback. The paper is organized as follows. In Section II, we describe the experiment setup. Section III formulates the problem and briey describes the controller design. Experimental results are presented in Section IV. Section V summarizes the paper. II. EXPERIMENT SETUP The laboratory model helicopter used to evaluate the adaptive output feedback control method is shown in Fig. 1. It consists

1063-6536/$20.00 2005 IEEE

KUTAY et al.: EXPERIMENTAL RESULTS ON ADAPTIVE OUTPUT FEEDBACK CONTROL

197

Fig. 1.

Three-DOF model helicopter.

of a rectangular frame and two propellers mounted at its two ends with axes normal to the frame. The helicopter frame is free to roll about its center where it is connected to the end of a long arm, which is free to rotate in pitch and yaw. The system has a total of three DOF with two control inputs as the voltages applied to the electric motors driving the propellers. Control voltage is applied equally to both motors, keeping the helicopter frame horizontal. Regulated output is the pitch angle of the arm, denoted by , measured by a 12-b encoder. Design of the adaptive controller requires the relative degree of the regulated output , and the sign and upper bound on the , where is the magnitude of the control effectiveness control. Open-loop experimental analysis presented in [11] refor veals that the relative degree of the regulated output is frequencies up to 3 rad/s and the control effectiveness is approx. Open-loop frequency analysis also imately shows that the system has nonlinear characteristics, mainly due to friction, nonlinear actuation system, and nonuniform mass distribution. III. DESCRIPTION OF THE ADAPTIVE OUTPUT FEEDBACK METHOD The adaptive output feedback method employs feedback linearization, coupled with an online NN-based adaptive element to compensate for modeling errors, and a linear compensator for the ideally linearized dynamics. The compensator also generates an auxiliary signal for training the linearly parameterized NN by ltering the tracking error. For this, a stable low-pass lter is used to meet a strictly positive real (SPR) condition of a transfer function associated with the error dynamics. This condition is utilized in the Lyapunov stability analysis to construct the NN adaptation law using only available measurements [9]. Pseudocontrol hedging (PCH) is incorporated within the above adaptive control setting to address unmodeled actuator dynamics and saturation. PCH was rst introduced for the state feedback case [10], and later applied in an output feedback setting [12]. The main idea behind the PCH approach is to limit or hedge the reference model (RM) of a model reference adaptive control architecture to prevent the adaptive element from attempting to adapt to the actuation anomalies, while not affecting adaptation to other sources of inversion error, for which compensation is possible. Conceptually, PCH moves the RM backward by an estimate of the amount the controlled system did not move due to selected actuator characteristics. PCH has been successfully tested on various simulation and

Fig. 2. Controller architecture with PCH loop (shown with dashed lines).

experimental applications in both full state and output feedback settings [13]. The overall control architecture, including the PCH loop, is depicted in Fig. 2. To briey summarize the controller design process, let the actual dynamics of the model helicopter be given by (1) is the system state, , are the where system input (achieved control) and output (measurement) sig, are unknown funcnals, respectively, and tions. We assume that the dynamic model in (1) satises the inputoutput feedback linearization condition with relative degree , i.e., (2) Here, , such that for and . Our aim is to design an output feedback control law that utilizes the available measurement , to achieve . output tracking of a bounded trajectory A. Feedback Linearization and Model Inversion Error Feedback linearization is performed by introducing the transformation (3) where is the commanded control, is commonly referred to as a pseudocontrol signal and is an approximain which only is used in the approximation. tion of Consequently, the approximation may be very crude. Using (3), the output dynamics (2) can be expressed as (4) where (5) The control input to be commanded is obtained from (3) as (6) The pseudocontrol is chosen to have the form (7)

198

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO. 2, MARCH 2005

Fig. 3. Computation of the PCH signal. Fig. 4.


r th-order RM with PCH signal.

where is the output of a RM that denes the desired response is the output of a linear dynamic of the closed-loop system, compensator designed to stabilize the ideally linearized system, is the adaptive control signal designed to approximately and ), the RM cancel . In the case of perfect actuation ( is designed as resulting in the following error dynamics: (8) where B. PCH When the applied control is different from the commanded due to, e.g., actuator saturation, PCH modies the control RM by introducing a signal . To compute the PCH signal , a measurement or an estimate of the actuator position is required. If actuator position measurements are not available, a model that includes the characteristics we want to protect the adaptive system from can be used to generate actuator position estimates. The process is illustrated in Fig. 3 for an actuator model that has position limit, rate limit, actuator dynamics, and time delay. Any of these characteristics can signicantly change the system dynamics and even render the system unstable. is dened as the difference between the The PCH signal commanded pseudocontrol and the estimated achievable pseudocontrol (9) is introduced as an additional input into the RM, Next, forcing it to move back. If the RM update without PCH was of the form (10) where is the external command signal, then the RM update with PCH is (11) The instantaneous pseudocontrol output of the RM that is used as an input to the linearized plant model is not changed by the use of PCH and remains (12) is the tracking error.

The effect of the PCH signal on a linear th-order RM is shown schematically in Fig. 4. With the modied RM (11), the error dynamics become

(13) where (14) Equation (13) is in the same form as (8), the error dynamics without PCH, with a very important difference that appears in in the expression for the modeling error. This the place of in effect prevents the adaptive process from seeing the portion and of the modeling error that is due to difference between . Without PCH, the modeling error that the NN is designed to cancel contains all the effects of the actuation anomalies. Since the other sources of nonlinearities and modeling error still appear in the error signal, the NN continues to compensate for them. An extreme case can exist during periods of full saturation in which the adaptive process is no longer in control, but is still capable of tracking the modeling error. Thus, PCH also permits the introduction of limited authority adaptive control in practical applications. Such limited authority control might extend to the limit of having no authority, in which case the adaptive process is simply monitoring the model error, and can be engaged only when it is determined that the modeling error is sufciently large (for example, in the event of an actuator failure). More general methods for accommodating saturation in adaptive systems have recently been reported in [14]. In the experimental results reported in the next section, we use PCH to shield the adaptive system from the dc motor dynamics and the voltage limit that is enforced to protect the motor windings. C. Linear Compensator and Adaptive Signal Design A single-inputtwo-output dynamic compensator denoted by

(15)

is used to generate the linear control signal together with that is required to ensure an implean auxiliary error signal

KUTAY et al.: EXPERIMENTAL RESULTS ON ADAPTIVE OUTPUT FEEDBACK CONTROL

199

Fig. 5.

Block diagram of the error dynamics.

mentable adaptive law. The stability analysis requires that the to given by transfer function from

(16) be SPR. This transfer function is depicted in Fig. 5. For , SPR, so a stable low-pass lter it is not possible to make is introduced in (16) so that can be made SPR (17) The adaptive signal is the output of a linearly parameterized NN, represented by (18) where is the NN weight vector, is a suitably chosen vector of radial basis functions ltered through , and is the input vector composed of current and past inputoutput data. NN weights are initialized as zero and updated using (19) where the positivedenite matrix adaptation gains. and are the

Fig. 6.

Output tracking of the model helicopter with sinusoidal inputs.

. function is approximated by the linear model The approximate model inversion function introduced in (3) where the constant is selected as is chosen in accordance with the assumption stated in [9] that . guarantees existence and uniqueness of a solution for Consequently, the adaptive element of the controller must compensate for nonlinearities of the system, parameter uncertainty, and unmodeled actuator dynamics. The linear compensator (15) that stabilizes the linear system is designed as

(20) The above compensator places the poles of the closed-loop , . The SPR lter is chosen as system at (21)

IV. EXPERIMENT RESULTS In this section, we present experimental results obtained using the laboratory helicopter model depicted in Fig. 1. The arm carrying the helicopter is balanced so that it remains horizontal when the propellers are not actuated, and this position is dened as zero pitch angle. Positive and negative commands in the following results correspond to positions above and below the horizontal, respectively. The PCH loop contains an actuator model that includes the characteristics that are desired to be hedged from the adaptive system. The most important characteristic that can destabilize the system is saturation of the control voltage introduced to protect the control hardware. In addition to saturation, a rst-order dynamic model of the dc motors that was identied in [11] is also included in the PCH loop, reducing the relative degree of the plant as seen by the adaptive system to two. Motivation for removing the actuator dynamics from the adaptive system is to simplify the design of the linear compensator, and to demonstrate the effectiveness of PCH against actuator dynamics. Hence, the controller design is performed , implying that the ideally inverted plant transfer using A linearly parameterized NN with ten neurons is used with the and . Pitch angle of adaptation gains the arm is commanded to follow the output of a linear thirdorder RM, designed with a pair of complex conjugate poles with rad s and damping , and a natural frequency s . A second-order RM would be sufcient real pole at to generate for the design. A third-order RM is is continuous in case input to the RM is used to ensure that not. Results for two cases are presented. In the rst case, the RM is driven by a sum of three sinusoids with frequencies 1.5, 3, and 4.5 rad/s, and in the second one it is driven by a series of positive and negative step commands. Fig. 6 shows the results for the rst case. The upper two plots show the tracking performance without PCH; the rst plot depicts experimental results performed without NN compensation and the second with NN compensation. The dotted curves show the output of the RM and the solid curves show the actual pitch angle. The lower two plots show the responses with PCH turned on. As explained earlier, PCH modies the output of the RM. The desired pitch attitude is

200

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO. 2, MARCH 2005

Fig. 7.

Commanded and applied control voltages for the sinusoidal inputs.

Fig. 8. Output tracking and control activity of the model helicopter with step inputs.

the unmodied RM command, shown with dashed lines. If this command is not achievable with the supplied actuator and inverted plant models (just a double integrator for this example), PCH modies it and generates an estimate of the best achievable command, shown with dash-dotted lines. This would be the best achievable command if the actuator and the plant models were exact. Since they are not, the dash-dotted lines shown in the bottom two plots are estimates. Note that the dash dotted lines are nearly indistinguishable from the actual response. The top two plots show that the helicopter cannot track the command without PCH, even when the NN is incorporated: the system is unstable both with and without NN. The sharp corners in the response curves correspond to instants when the helicopter arm hits the table. Experiments were stopped around s to protect the helicopter. The instability encountered here is not unexpected. The adaptive method requires knowledge of the ignores relative degree. The assumed relative degree of the dynamics of the actuators. When PCH is off, the actual rela, which vitive degree as seen by the adaptive controller is olated a basic assumption in the overall approach. Nonetheless, the most signicant factor causing the instability is probably control saturation observed in the following gure. The bottom two plots in Fig. 6 with PCH show that modied RM commands are tracked almost perfectly both with and without NN adaptation. However, there is a signicant improvement with NN adaptation since the RM commands, both with and without PCH, and the helicopter response are all nearly identical. Fig. 7 shows the commanded and applied controls corresponding to the results in Fig. 6. Vertical axis limits are adjusted to focus on the applied voltage variations rather than the commanded voltage variations. Without PCH, voltage applied to the motors is almost always saturated. It is not apparent in the gure, but the amplitudes of the commanded voltages without PCH gradually increase, a further indication that the closed-loop system is unstable. The value of the voltage limit V) is treated as a known quantity, and used to model the ( actuator limits in the PCH loop. In general, if it is not known, then it is preferable to use a conservative value for the limit.

With PCH, the durations of periods of saturation are reduced, and when adaptation is on, the saturation is almost completely eliminated. One striking feature in Fig. 7 is the excessive noise in the control activity when both NN and PCH are on. The reason for the noisy control activity will be discussed at the end of this section. Results for the second case of step inputs are presented in Fig. 8. Adaptation is on at the beginning, then it is turned off at s and turned on again at s. PCH is on throughout the experiment. The upper plot shows tracking performance, and the lower plot shows the commanded and applied voltages. The dashed and dotted curves show the RM outputs with and without hedging. The response tracks the hedged RM output closely, except when adaptation is turned off. When adaptation is off, the response is very oscillatory, and the helicopter strikes the table when commanded in the downward direction. The overshoot and oscillations are completely eliminated when adaptas disturbances are applied (by manution is on. After ally pushing on the helicopter frame) to demonstrate disturbance rejection. Note that the second disturbance causes the control signal to saturate for a while. The large error that occurs when a disturbance is applied is due to the control voltage limit. Both Figs. 6 and 7 exhibit a noisy control activity when adaptation is on. This is due to the fact that a 12-b resolver (4096 counts per revolution) was used to measure the pitch angle of the arm, which corresponds to approximately 0.1 deg. resolution. The staircase form of the feedback signal causes the NN to generate sudden peaks to compensate the sudden changes in the error signal, and leads to a noisy control signal as observed in Figs. 6 and 7. In effect, the adaptive process views quantization error as modeling error, and attempts to adapt to it. The development of a method for hedging that can be applied at the sensor level remains a topic for future research. A higher resolution sensor would eliminate the noise in the control signal. This fact has been veried in simulation. Fig. 9 shows simulated control activity in which we employed models of a 12 b and 24 b (16 777 216 counts per revolution) angular resolver, which

KUTAY et al.: EXPERIMENTAL RESULTS ON ADAPTIVE OUTPUT FEEDBACK CONTROL

201

Fig. 9. Control activity in simulation with 12-b and 24-b resolvers.

corresponds to approximately 2 resolution. Control activity with a 12-b resolver is very noisy, similar to the experimental results. Increasing the resolution of the sensor to 24 b without changing any other parameter in the design eliminates the noise-like activity in the control response.

[6] C. Pereira, J. Henriques, and A. Dourado, Adaptive RBFNN versus conventional self-tuning: Comparison of two parametric model approaches for nonlinear control, Control Eng. Practice, vol. 8, no. 1, pp. 312, Jan. 2000. [7] J. Q. Gong and B. Yao, Neural network adaptive robust control with application to precision motion control of linear motors, Int. J. Adaptive Control Signal Process., vol. 15, no. 8, pp. 837864, Dec. 2001. [8] C.-L. Hwang and C.-Y. Hsieh, A neuro-adaptive variable structure control for partially unknown nonlinear dynamic systems and its application, IEEE Trans. Contr. Syst. Technol., vol. 10, no. 2, pp. 263271, Mar. 2002. [9] A. J. Calise, N. Hovakimyan, and M. Idan, Adaptive output feedback control of nonlinear systems using neural networks, Automatica, vol. 37, no. 8, pp. 12011211, 2001. [10] E. N. Johnson and A. J. Calise, Limited authority adaptive ight control for reusable launch vehicles, J. Guid. Control Dyn., vol. 26, no. 6, pp. 906913, Nov.-Dec. 2003. [11] A. T. Kutay, A. J. Calise, M. Idan, and N. Hovakimyan, Experimental results on adaptive output feedback control using a laboratory model helicopter, in Proc. AIAA Guidance, Navigation, and Control Conf., 2002, pp. 20024921. [12] N. W. Kim, Improved methods in neural network based adaptive output feedback control, with applications to ight control, Ph.D. dissertation, Georgia Tech., Atlanta, GA, 2003. [13] A. J. Calise, E. N. Johnson, M. D. Johnson, and J. E. Corban, Applications of adaptive neural-network control to unmanned aerial vehicles, in Proc. AIAA/ICAS Int. Air and Space Symp. and Exposition: Next 100 Years, Dayton, OH, Jul. 2003. [14] E. Lavretsky and N. Hovakimyan, Positive mu-modication for stable adaptation in a class of nonlinear systems with actuator constraints, in Proc. ACC04.

V. SUMMARY This paper presents experimental evaluation of a NN-based adaptive output feedback control method applied to a laboratory model helicopter. The method is based on model inversion with feedback linearization and uses a linearly parameterized NN to cancel modeling errors. Also, a method for protecting the adaptive element from the effects of unmodeled actuator dynamics and saturation has been employed. Experiments revealed robustness of the method to parameter uncertainty, unmodeled dynamics, and disturbances, and effectiveness of PCH in protecting the adaptive process during periods of control saturation.

Ali T. Kutay received the B.Sc. and M.Sc. degrees from the Aerospace Engineering Department, Middle East Technical University, Ankara, Turkey, in 1996 and 1999, respectively. He is currently working toward the Ph.D. degree at the Georgia Institute of Technology, Atlanta.

REFERENCES
[1] D. W. Vos, L. Valavani, and A. H. von Flotow, Intelligent model reference nonlinear friction compensation using neural networks and Lyapunov based adaptive control, in Proc. IEEE Int. Symp. Intelligent Control, 1991, pp. 417422. [2] T. Yabuta and T. Yamada, Neural network controller characteristics with regard to adaptive control, IEEE Trans. Syst., Man, Cybern., vol. 22, no. 1, pp. 170177, Jan.-Feb. 1992. [3] S. I. Mistry, S. Chang, and S. S. Nair, Indirect control of a class of nonlinear dynamic systems, IEEE Trans. Neural Netw., vol. 7, no. 4, pp. 10151023, Jul. 1996. [4] D. Shukla, D. M. Dawson, and F. W. Paul, Real time adaptive control experiments with a multiple neural network based DCAL controller, in Proc. IEEE Conf. Control Applications, 1997, pp. 371376. [5] B. Bouzouia, M. Kadri, and N. Louam, Experimental neural decomposed dynamic adaptive control of robot manipulator, in IEEE Int. Conf. Intelligent Robots and Systems, vol. 1, 1999, pp. 488493.

Anthony J. Calise (S63M74SM04) was a Professor of Mechanical Engineering at Drexel University, Philadelphia, PA, for eight years prior to joining the faculty at the Georgia Institute of Technology, Atlanta. He also worked for ten years in industry for the Raytheon Missile Systems Division and Dynamics Research Corporation, where he was involved with analysis and design of inertial navigation systems, optimal missile guidance, and aircraft ight path optimization. Since leaving industry, he has worked continuously as a consultant for 19 years. He is the author of over 150 technical reports and papers. The subject areas that he has published in include optimal control theory, aircraft ight control, optimal guidance of aerospace vehicles, adaptive control using neural networks, robust linear control, and control of exible structures. In the area of adaptive control, he has developed a novel combination for employing neural-network-based control in combination with feedback linearization. Applications include ight control of ghter aircraft, helicopters, and missile autopilot design. He is a former Associate Editor for the Journal of Guidance, Control, and Dynamics. Dr. Calise was the recipient of the USAF Systems Command Technical Achievement Award, and the AIAA Mechanics and Control of Flight Award. He is a Fellow of the AIAA and an Associate Editor for the IEEE CONTROL SYSTEMS MAGAZINE.

202

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO. 2, MARCH 2005

Moshe Idan received the B.Sc. and M.Sc. degrees from the Aerospace Engineering Department at the Technion-Israel Institute of Technology, Haifa, Israel, in 1983 and 1986, respectively, and the Ph.D. degree from the Department of Aerospace Engineering, Stanford University, Stanford, CA, in 1990. Since 1991, he has been with the Department of Aerospace Engineering, TechnionIsrael Institute of Technology. From 2000 to 2001, he spent a sabbatical in the School of Aerospace Engineering, Georgia Institute of Technology, Atlanta. His research concentrates on robust and adaptive ight control system design techniques and applications.

Naira Hovakimyan (M01SM02) received the Ph.D. degree in physics and mathematics from the Institute of Applied Mathematics, Russian Academy of Sciences, Moscow, in 1992. After receiving the Ph.D. degree, she joined the Institute of Mechanics, Armenian Academy of Sciences, as a Research Scientist, where she worked until 1997. In 1997, she was awarded a governmental postdoctoral scholarship to work at INRIA, France. The subject areas in which she has published include differential pursuit-evasion games, optimal control of robotic manipulators, robust control, adaptive estimation, and control. In 1998, she was invited to the School of Aerospace Engineering, Georgia Institute of Technology, Atlanta, where she worked as a Research Faculty Member until 2003. In 2003, she joined the Department of Aerospace and Ocean Engineering, Virginia Polytechnic Institute and State University, Blacksburg, an Associate Professor. She has authored over 90 refereed publications. Her current interests are in the theory of adaptive control and estimation, neural networks, and stability theory. Dr. Hovakimyan received the SICE International Scholarship for the best paper of a young investigator in the VII ISDG Symposium, Japan, 1996. She is also the receipient of the 2004 Pride @ Boeing Award. She is a Senior Member of AIAA, a Member of AMS and ISDG, and is an Associate Editor for the IEEE Control Systems Society, the IEEE TRANSACTIONS ON NEURAL NETWORKS, and the International Journal of Control Systems and Automation.

Das könnte Ihnen auch gefallen