基于案例学习的排球机器人运动规划及其支持向量回归实现

The Case-Based Learning of Motion Planning and Its SVR Implementation for Volleyball Robot

针对混合型控制问题,以排球任务为例研究机器人的运动规划.模拟人类球员通过经验积累而采取相应动作的行为学习模式,采取案例学习的方式解决球的初始状态微小变化(仅发球速度和角度变化)时的运动规划问题.由于支持向量回归(SVR)在处理小样本问题的优越性并受局部学习思想的启发,采用局部加权SVR(LW-SVR)实现案例学习.结果证明,LW-SVR的学习精度较RBF神经网络和SVR明显提高.

This paper takes volleyball task as an example to explore motion planning of hybrid control problem. Emulating the learning mode of human players to perform motion by experience, case based learning method is applied to the motion planning where ball＇s initial state is partially changed （only velocity and angle is changed）. Inspired by the advantage of local learning and support vector regression （SVR） dealing with finite data, locally weighted SVR （LW-SVR） is adopted to implement the case-based learning. The results show that the precise of LW-SVR is distinctly improved relative to RBF network and SVR.