Applying the Evolutionary Neural Networks with Genetic Algorithms to Control a Rolling Inverted Pendulum

Genetic Algorithms (GA) are applied to evolutionary neural networks to control a rolling inverted pendulum. The task of a rolling inverted pendulum is to control the driving force of a cart on which one side of a pole is jointed by a rotary shaft in order to roll the pole up from the initial state of hanging down and to keep the pole standing reversely. The controller is a multilayer perceptron (MLP) with three layers whose weight coefficients are evolved and optimized by GA. Experiments for evolving the weights of two types of MLPs are conducted and their results are compared. Simultaneously, the effect of the weight ranges of neural networks on evolutionary results is investigated. In these evolutionary experiments, MLPs are generated that successfully control the driving force of the cart to roll the pole up and stand it inversely. MLPs also gain the intelligent control patterns with a few swings that correspond to the variations in the maximum driving force of the cart.