A stochastic optimization method of CPG-based motion control for humanoid locomotion

In this paper, we propose a motion control method for bipedal humanoid locomotion. The motion control is based on the central pattern generator (CPG), and is optimized by simulated annealing. The motion control for robot locomotion is a multi-objective optimization problem. The aim of this research is to generate various gaits so as to meet the objectives; there may be a bias, such as stability or speed. Therefore, we consider plural evaluation functions for each of the objectives, and the motion control and the parameter optimization are designed to be conscious of the bias. To optimize all parameters simultaneously may cause explosion of search space. The optimization process, thus, hits two phases; optimizing the parameters of lower body and then optimizing the parameters of upper limb. In the experiments, our method performed some typical gaits, which respects stability, speed and both the two elements.