The operational space formulation with neural-network adaptive motion control

Abstract

An operational space controller that employs three-layer neural network (NN) adaptive motion control is presented in this paper. It is shown that the trajectory tracking errors and the NN weight errors are bounded and consequently the controller is shown to be stable. Comparative study is made between the performance of the proposed NN adaptive motion control strategy and the conventional inverse-dynamics control and PD plus gravity compensation controller. The simulation results show that the NN adaptive motion control can be as effective as the inverse-dynamics without the need for textit{a priori} knowledge of the system dynamics. Real-time implementation of the strategy was also conducte..