Parameter Optimization of a Reduced Model for Multi‐Contact Locomotion of Humanoid Robots

Abstract

In this study, we investigate a reduced model intended for approximating the complex multi-body dynamics of our full-sized humanoid robot LOLA. In particular, we compare various optimization algorithms to find an optimal set of model parameters. For rating parameter sets, we define a cost function which penalizes differences between the simulated motion of the fulland reduced model. Due to the complex motions of a walking humanoid robot, which includes cyclic opening and closing contacts, one is easily trapped in local minima. Thus the focus of this study lies on techniques for global optimization based on meta-heuristics. We plan to use the reduced model for real-time stability analysis during multi-contact locomotion, i. e. walking with additional arm support.