Cellular Neural Network Trainer and Template Optimization for Advanced Robot Locomotion, Based on Genetic Algorithm

Abstract

A new learning algorithm for advanced robot locomotion is described in this paper. This method involves both cellular neural networks (CNN) technology and evolutionary algorithms. Learning is formulated as an optimization problem. CNN templates are derived by genetic algorithms after an optimization process [1]. A template generates a specific wave on CNN that leads to the best motion of a walker robot. Details of the algorithm and several applications and simulation results are shown and commented. It is shown that an irregular and even a disjointed walker robot can move with the highest performance due to this method.