Adaptive neural network control of flexible link robots based on singular perturbation

Abstract

An adaptive neural network controller is presented for flexible link robots (FLRs) using the singular perturbation technique. The complex full model of FLRs can be decomposed into two separate time scale subsystems by modelling the elastic forces as the fast variables and the joint variables as the slow variables. A composite control strategy is adopted to control the two-time scale model with the slow subsystem being controlled by a neural network controller. The weights of the NNs are updated online based on direct adaptive techniques. A robust control term is also added for closed-loop stability. The fast subsystem is stabilized by a simple LQR control around the equilibrium trajectory defined by the slow subsystem.