Apply Acceleration Sampling to Learn Kick Motion for NAO Humanoid Robot

Abstract

In the current level of evolution of Soccer 3D, kick motion control plays a vital role in team’s performance. Keyframe Sampling, Optimization, and Behavior Integration (KSOBI) is an effective method for NAO robot learning to generate kick motion, which is proposed by MacAlphine. However, we observe that without strong computing power, KSOBI can prematurely shrink the exploration variance, which resulting in slow progress and may make the algorithm prone to getting local optima. This paper proposes a method for learning kick skill from demonstration, which is based on Acceleration Sampling (AS) to create and use this kicking action in the following ways: (i) observing the acceleration of the joints of another robot and calculate objective angle for each joint; (ii) optimizing the skill begin from this seed. This method is fully tested in RoboCup 3D simulation platform. With minor changes to KSOBI, our methodology considerably improves performance in generating kick motion.