Dribble Optimization Of Humanoid Soccer Robot Based on Overlapping Layered Learning

In recent years, the research of humanoid robots has developed rapidly, especially in the RoboCup3D soccer competition. With the reinforcement of each team's defensive strategy, it's hard to kick a ball successfully through the surounded opponents sometimes. However, fast and accurate dribbling skill is more effective in such critical situation. This paper mainly proposes an extended Overlapping Layered Learning method to optimize the parameters of dribbling. Firstly, we used parameters of sprint to optimize a fast dribbling skill. Then, the hierarchical optimization tasks with multifarious target points were designed to construct a accurate dribble model. Eventually, we enhanced a training method to coordinate the new dribbling skill with original kicking skills. The training results demonstrate the effectiveness of the proposed method, therefore a fast and accurate robot dribble behavior was obtained.