Role Adaptation of Human–Robot Physical Interaction Based on the Distribution of Learned Belief

Physical human–robot interaction (pHRI) poses great challenges to the role adaptation between the human and the robot in real time. The robot has several individual tasks attainable, while the human might switch between tasks unpredictably. The capability to estimate the confidence of the expected task is critical to empower the robot to behave properly in cooperation. In this paper, we construct a pHRI controller framework with real-time role adaptation. With the task beliefs learned by a belief estimation module, we mathematically analyze the task confusion index of the robot during its motion. The role adaptation module authorizes the robot to take proactive actions if the confusion index is minor, and vice versa. We deploy a reinforcement learning method on generating the robot assisting force. The proposed pHRI controller framework is validated in both simulation and experiment. More results and videos are available at https://sites.google.com/view/hrpcrl/ .