AAS 95-319

An Update on a Monotonic Learning Control Law and Some Fuzzy Logic Learning Gain Adjustment Techniques

H. S. Jang and R. W. Longman, Columbia University, New York, NY

Abstract

In a previous papers, several learning control algorithms were developed that make use of a nominal model of the system to pick the learning control gains. Provided that the system model is correct, these algorithms have guaranteed good transient behavior that has monotonic decay of the RMS error or monotonic decay of the error frequency component amplitudes. Periodic reidentification can be used to tune the model when these properties are violated, indicating that the data contains information to correct the model. Both of these algorithms have been shown to be very effective in experiments, one of which produced a decrease in tracking error of a robot by a factor of 1000. This paper develops an understanding of the relationships between these algorithms, and an understanding of various methods of improvement. In addition, the ultimate error level in learning control in the presence of noise is studies, and fuzzy logic concepts are applied to improve the final error levels.