Performance of a neuro-model-based robot controller: adaptability and noise rejection

Abstract

Effective control strategies for robotic manipulators usually require the on-line computation of the robot dynamic model in real time. However, the complexity of the robot dynamic model makes this difficult to achieve in practice, and multiprocessor controller architectures appear attractive for real-time implementation inside the control servo loop. Furthermore, inevitable modelling errors, changing parameter values and disturbances can compromise controller stability and performance. In this paper, the performance of a neuro-model-based controller architecture is investigated. The neural network is used to adapt to unmodelled dynamics and parameter modelling errors. Simulation of the neuro-model-based control of a one-link robot demonstrates an improved performance over standard model-based control algorithm, in the presence of modelling errors and in the presence of disturbance and noise. >