Attractor design and prediction-based adaption for a robot waltz dancer in physical human-robot interaction

Physical human-robot interaction between a human leader and a robot follower in waltz is studied in this paper. The dancers´ body dynamics in single-support phase are modeled as inverted pendulums. On the robot side, an ankle torque control method is proposed and applied. The control law forms a time-dependent vector field, which makes the nominal orbit of the robot to be an attractor. To physically interact with human, the human leader´s state is estimated from range image data by using an extended Kalman filter. Parameters of the robot´s orbit are then adjusted according to the leader´s estimated and predicted state. The proposed method is verified by simulation results.