Evolution of locomotion controllers for snake robots

Abstract

Snake robots are redundant structures, that are able to traverse many unstructured environments unlike wheeled and legged mobile robots. This research presents a novel method of designing efficient movement control systems for snake robots using artificial neural networks, optimized by a genetic algorithm. This approach outperforms the common control methods in terms of diversity, using no a-priori knowledge about snake movements and ease of implementation. Research was conducted to design two gaits namely sidewinding and turning for the snake robot on a planer terrain in a 3D Physics simulation environment. The locomotion controllers obtained were evolved quickly for the two gaits and efficient desired behaviors were obtained. The proposed evolutionary method, introduced embodied intelligence in snake robot and the trained controllers can be employed for following a trajectory easily by a high level controller.