ADAPTIVE RBF NEURAL NETWORK FOR A BIPED WALKING MACHINE
João Bosco Gonçalves, Daniel Danelli, Douglas Eduardo Zampieri · 2003
Abstract. It is necessary to consider the robot dynamic model in order to design a biped walking machine in the dynamic locomotion form. To project a high-performance control system is a very difficult task, which arises from the complexity of the dynamic model of the biped walking machine. Nevertheless, the robot abilities can be improved by employing a dynamic locomotion gait similar to the walk of a human being. The main objective of this paper is investigate a high-performance control system for a biped walking machine, by employing techniques developed for linear systems, applied to the highly non-linear robot mathematical model. We have emulated some unknown non-linearities by using RBF neural network, whose radial base function vector is fixed and with the vector weight tuned by an adaptive law. The strategic control law employed could ensure the closed-loop stability in the Lyapunov sense, even with the presence of neural network approximation errors. We tested a strategic control by using the experimental data of a MIMO biped walking machine. The computer simulation results proved to be very good.