2016 Volume 29 Issue 8 Pages 346-354
Recently, the framework of shared control between a human operator and a robot system using a cooperative controller with human operator to decide control input to a robot system has been drawing much attention for dexterous telemanipulation. However, desiging a suitable controller for shared control scheme for the given task is difficult due to the difficulties in modeling of operator's behavior and environment. In this paper,we proposed a model-free approach using reinforcement learning to learn a shared control policy through interactions with the operator, robot and environment. To validate our method,we adopted a page turning task by telemanipulation and developed an experimental platform with a physical simulator.Experimental results suggest that our method is able to learn task-relevant shared control for flexible and enhanced dexterous manipulation.