Modular controls design for robot manipulators using CMAC neural networks

This paper presents a practical method of achieving improved performance from existing industrial robot controllers using intelligent control techniques. While there have been many solutions to dealing with the unmodeled dynamics and disturbances in a robotic system, these results have not found much acceptance in the industrial community as they require a complete redesign of the controller. The strategy presented here uses concepts and results from the intelligent control literature, to build on the existing linear controller. It is shown that the existing conventional linear controller can be augmented by a control component that can compensate for the unmodeled dynamics and disturbances. This component can be manufactured by neural networks, fuzzy logic controllers, or adaptive controllers as long as they satisfy key approximation properties. In this paper, this concept is demonstrated by implementing the design using CMAC neural networks (CMAC NN). The adaptation laws for this controller are designed and the closed-loop performance of the overall system is rigorously proved