Composite adaptive control of robot manipulators

Abstract

Adaptive control of linear time-invariant single-input single-output systems has been extensively studied, and a number of globally convergent controllers have been derived. While extensions of the results to non-linear or multivariable systems have rarely been achieved, similar global convergence properties can indeed be obtained in the case of robot manipulators, an important and unique class of non-linear multi-input multi-output dynamic systems. Based on the observation that the parameter uncertainty is reflected in both the tracking error in joint motion and the prediction error in the joint torques or the power input, we recently proposed a new class of adaptive robot controllers, the parameter adaptation of which is driven by both tracking error and prediction error. In this paper, following a brief review of our earlier globally convergent direct adaptive controller, we provide a detailed analysis of these “composite” adaptive controllers. Results on global asymptotic and exponential tracking convergence are established and confirmed by simulation.