Allen R. Stubberud

Professor & Chair, Electrical Engineering and Computer Science
The Henry Samueli School of Engineering
Co-Director, Center for High-Speed Image/Signal Pr

PH.D., University of California, Los Angeles

Phone: (949) 824-4664
Fax: (949) 824-3779

University of California, Irvine
305 EGR
Mail Code: 2625
Irvine, CA 92697
Research Interests
Control Systems and Digital Signal Processing
Research Abstract
An Artificial Neural Network-Based Extended Kalman Filter
Investigator: A.R. Stubberud
Support: National Science Foundation Center for High-Speed Image/Signal Processing

There exists a need in many engineering systems for a filter than can, in real time, estimate the state of a system modeled by nonlinear difference (or differential) equation. Currently, the extended Kalman filter (EKF) is used in many such applications. The EKF, however, has two major deficiencies for application to a much broader range of activities: 1) the model of the system whose state is being estimated must be quite accurate in order for the EKF to perform well; and 2) the level of mathematical sophistication necessary to design an appropriate EKF for a specific application is such that many technical personnel choose not to attempt to design such a filter. This project combines the attributes of an artificial neural network with an EKF, and develops a generic EKF that self-learns the dynamics of the system. As a result, the user can use directly the generic EKF in a state estimation application without having to develop a system model. Since the EKF learns the system model, it has potential for also eliminating accuracy problems associated with the standard EKF.

Neural Computing for Intelligent High-Speed Control Systems
Investigator: A.R. Stubberud
Support: National Science Foundation Center for High-Speed Image/Signal Processing

With increased requirements for performance and versatility being placed on engineering systems, far greater demands are being imposed on the design of control systems. The most crucial among these is the requirement for such systems to perform autonomously under uncertain environments. The causes for this uncertainty or imprecision may be several. First, the control process or the plant may be such that a precise mathematical model cannot be developed or is too costly to develop. Secondly, even if the model is know to a reasonable degree, there could be significant nonlinearities in the system dynamics, large parameter variations, faulty sensors and actuators, etc. A control system can cope with these problems by reconfiguring its control law so as to meet the performance requirements. To do this, the controller must first learn about the plant, the environment, and its own capabilities. A control system with this learning and decision-making capability can be thought of as being intelligent. This research proposes that the application of neural networks to some generic control problems could be the vehicle used as an application.
Last updated