Autonomous evolution of high-speed quadruped gaits using particle swarm optimization

This paper presents a novel evolutionary computation approach to optimize fast forward gaits for a quadruped robot with three motor-driven joints on each limb. Our learning approach uses Particle Swarm Optimization to search for a set of parameters automatically aiming to develop the fastest gait that an actual quadruped robot can possibly achieve, based on the concept of parameterized representation for quadruped gaits. In addition, we analyze the computational cost of Particle Swarm Optimization taking the memory requirements and processing limitation into consideration. Real robot experiments show that the evolutionary approach is effective in developing quadruped gaits. Satisfactory results are obtained in less than an hour by the autonomous learning process, which starts with randomly generated parameters instead of any hand-tuned parameters.