Dynamic Kick Optimization Of Humanoid Robot Based on Options Framework

Abstract

Robocup3D humanoid robot kicking skills has been a hot research topic in the project. Based on the options framework in hierarchical learning, this paper proposes a new dynamic model to optimize kicking. Firstly, keyframe reduction method was used to design seed skills of humanoid robot. Then, hierarchical optimization tasks were designed for robot kick time and ball drop position to construct a dynamic model. In the end, a computer cluster platform based on cmaes algorithm built by HTCondor software was used to optimize the kick parameters. Experimental results demonstrate the effectiveness of the proposed method. Among them, the analysis of the game results shows that this model greatly improves the team strength.