Performance Comparisons of Evolutionary Algorithms for Walking Gait Optimization

Abstract

To investigate the performance of different evolutionary algorithms on walking gait optimization, we designed an optimization framework. There are four bio-inspired methods in the framework, which include Genetic Algorithm (GA), Covariance Matrix Adaption Evolution Strategy (CMA-ES), Particle Swarm Optimization (PSO) and Differential Evolution (DE). In the learning process of each method, we employed three learning tasks to optimize the walking gait, which are aiming at generating a gait with higher speed, stability and flexibility respectively. We analyzed the gaits optimized by each four methods separately. According to the comparison of these results, it indicates that DE performs better than the other three algorithms. The comparison also shows that the gaits learned by CMA-ES and PSO are acceptable, but there exist drawbacks compared to DE. And among these methods, GA presents weak performance on gait optimization.