Adaptive neural network controller for robot manipulator systems

Abstract

A novel neural-network-based adaptive controller is presented for the trajectory control of robot manipulators. The new adaptive learning algorithm for training the weights of the multi-layer neural-network ensures good tracking performance, based as it is on the Lyapunov criterion, so that convergence to a stable solution and bounded weights is guaranteed. The new adaptive learning algorithm looks like B-P for simple cases, but the error signal for training the multi-layer neural-network compensator is directly derived from the controller design. This helps to explain why the widely used B-P algorithm may be an effective training algorithm as long as the error signal is suitably chosen. Much larger learning rates are also allowed by the new adaptive learning algorithm proposed. Simulation studies have been conducted with a view to corroborating the theoretical results.