Shared Impedance Control Based on Reinforcement Learning in a Human-Robot Collaboration Task

In this work a shared impedance control scheme for a hybrid human-robot team is designed for transporting a rigid workpiece to a desired position. Within the scope of proposed control structure, both human and robot are regarded as mechanical impedance and their parameters are adapted continuously in real-time. Reinforcement learning is used to find an impedance parameter set for the whole team to optimize a task-orient cost function. Then the learned parameters are further adjusted by taking human's disagreement into consideration. The proposed method is aimed to reduce human's control effort during collaboration and be flexible to variation of the task or environment. Experimental results are presented to illustrate the performance.