Human-robot Collaborative Manipulation through Imitation and Reinforcement Learning
This paper proposes a two-phase learning frame-work for human-robot collaborative manipulation tasks. A table-lifting task performed jointly by a human and a humanoid robot is considered. In order to perform the task, the robot should learn to hold the table at a suitable position and then perform the lifting task cooperatively with the human. Accordingly, learning is split into two phases. The first phase enables the robot to reach out and hold one end of the table. A Programming by Demonstration (PbD) algorithm based on GMM/GMR is used to accomplish this. In the second phase the robot switches its role to an agent learning to collaborate with the human on the task. A guided reinforcement learning algorithm is developed. Using the proposed framework, the robot can successfully learn to reach and hold the table and keep the table horizontal during lifting it up with human in a reasonable amount of time.
Ye Gu Anand Thobbi Weihua Sheng
Department of Electrical and Computer Engineering Oklahoma State University Stillwater, OK 74074, USA
国际会议
深圳
英文
151-156
2011-06-06（万方平台首次上网日期，不代表论文的发表时间）