Neuro-adaptive control with application to robotic systems

Abstract

This article presents a highly model-independent neural network (NN)-based adaptive control method for a class of nonlinear dynamic systems. Two NN units are incorporated into the control scheme which are shown to be effective in attenuating NN reconstruction error and other lumped system uncertainties. Because the control scheme is based on the worst case behavior of the NNs, it exhibits a “fail-safe” feature, which enhances the reliability of the NN-based control scheme. Stable on-line weight-tuning algorithms are derived based on Lyapunov stability theory. The control method is extended to robotic systems and simulation on a three-joint robot is presented. ©1997 by John Wiley & Sons, Inc.