关键词: 协作机器人, 示教学习, 回归路径, 高斯混合模型, 高斯噪声发散, 避障

Abstract: To solve the over-fitting problem of regression path in the learning from demonstration of collaborative robots,a path optimization method based on Gaussian noise divergence was proposed.To blur the unfavorable features caused by the jittery during human demonstration,the proposed method randomly added the noise that obeyed Gaussian distribution to the original demonstration data,which could reduce the retention of redundant bending features of the demonstration data,thereby improving the quality of the regression path.An adaptive obstacle avoidance strategy based on scattered data was proposed,which blanked the data in obstacle area to guide the regression path for avoiding the obstacle,so as to adaptively generate the obstacle avoidance path without re-demonstration.The UR5 robot was used to conduct an adaptive grasping experiment with obstacles.The experimental results showed that the method could effectively improve the smoothness of the path and achieve the result of adaptive obstacle avoidance.

Key words: collaborative robot, learning from demonstration, regression path, Gaussian mixture model, Gaussian noise divergence, obstacle avoidance