Human-Robot Variable Impedance Skills Transfer Learning Based on Dynamic Movement Primitives

Abstract

Endowing robots with human-like abilities to perform motor skills smoothly and naturally is one of the important goals of robotics. Learning from demonstration (LfD) has been successfully applied for learning tasks on robots, for which the human tutor can demonstrate a successful execution. Learning human stiffness schedule strategy is one of the promising approaches for lightweight collaborative robots to execute physical in-contact tasks while remaining compliant when possible, as it enables robots to show human-like adaptive impedance behavior in an effective and efficient manner. In this letter, we develop a robot skill learning framework considering both movement and impedance features. Electromyography (EMG)-based method is used to estimate human upper limb stiffness such that the information of impedance can be obtained. Dynamic movement primitives (DMPs) model is employed to model movement and impedance features simultaneously. Throughout this procedure, the robot skills learning from a human tutor can be achieved. Besides, the learned skills can be generalized. An impedance controller imitating the human impedance mechanism is utilized in the task reproduction phase to achieve both task successful execution and safe physical interaction. The effectiveness of the proposed methods is verified in a water pumping task on a Kinova robot.