Variable Impedance Learning from Human Demonstrations

Physical human-robot interaction (pHRI), allowing robots to cooperate with human beings to accomplish varies of tasks, requires safety and adaptability owing to the uncertainty and unpredictability of environment. This paper proposed a frame to transfer human impedance behaviors to robots through demonstrations. The impedance parameters of human estimated by RLS, along with interaction forces compose the dataset encapsulating desired impedance strategies. Learn and reproduce impedance strategies of demonstrations through GMM/GMR. This methodology implements force-based variable impedance control from demonstrations, enabling a safe and flexible human-robot interaction system. To validate the proposed approach, two simulation experiments are conducted in a MSD system.