Robot control with a fully tuned Growing Radial Basis Function neural network

Abstract

A fully tuned Growing Radial Basis Function (GRBF) neural network controller for the control of robot manipulators is proposed. In addition to the weights, the centers and the standard variations are adapted online. Furthermore, we present an algorithm in which nodes of the network are appended based on sliding window performance criteria. Lyapunov analysis is used to show uniform ultimate boundedness and a discretization method is used to derive the growing algorithm. Simulations of a 2-DOF planar robot arm are presented to illustrate the method.