Adaptive neural networks for adaptive nonlinear robot control

Abstract

A new neural network (NN) architecture is introduced, and the following concepts are employed: (i) combinations of input, interaction and output activation functions, (ii) input and interaction time-varying signal distributions, (iii) combinations of feedforward and recurrent neural networks, (iiii) time-discrete domain synthesis and (iiiii) one-step learning iteration approach. The proposed NN synthesis procedures are useful for applications to identification and control of dynamical systems. In this sense an adaptive neural network for an adaptive nonlinear robot control is proposed.