Adaptive control of robot manipulators using multiple neural networks

Abstract

A new adaptive controller based on multiple neural networks for uncertain robot manipulator system was developed in this paper. The proposed multiple neuro-adaptive controller (MNAC) switches to a memorized control skill or blends multiple skills by using visual information on the given job to improve transient response at the time of task variation like change of manipulating object. MNAC is a kind of adaptive feedback controller where system nonlinearity terms are approximated with multiple neural networks. The proposed controller is effective for a job where some tasks are repeated but information on the load cannot be scheduled before the operation. During learning phase, MNAC memorizes a control skill for each load with each neural network. For a new task, most similar existing control skill may be used as a starting point of adaptation, which improves the performance of learning. Lyapunov function based design of MNAC guarantees the stability of the closed-loop system to be independent of switching or blending law. Simulation results on two-link manipulator or changing mass of the given load were illustrated to show the effectiveness of the proposed control scheme by the comparison with the conventional neuro-adaptive controller (NAC).