STABILIZATION OF AN INVERTED PENDULUM BY A LAYERED NEURAL-NETWORK

This paper discusses the stabilization of an inverted pendulum by a ne ural network. The inverted pendulum is a typical example of an unstabl e and nonlinear system. Fujitsu has developed a self-learning method t hat enables a neural network to control an inverted pendulum by using heuristic rules. A system based on this method has been developed whic h generates teaching data for inversion by trial and error. Then, the neural network learns the teaching data, and acquires the control rule s after a period of trial and error. The method proved its effectivene ss by stably inverting a pendulum in an experimental inverted pendulum system after a period of trial and error.