On-Line Modifications of Robotic Trajectories: Learning, Coaching and Force Vs. Position Feedback

Robotic coaching is a process of modifying the motion of a robot during execution through human intervention, in a manner of a coach. Thus, parts of the motion are changed depending on the instructions which can be provided through different approaches. Methods and analyses of robotic coaching have been described in the literature, including methods relying on physical human-robot interaction. In this paper we analyze one such method, which uses a compliantly controlled robot, to different variations of position and force feedback. Our method is based on changing the motion by changing the dynamic motion primitives’ (DMPs) weights. We qualitatively evaluate how such DMP based coaching methods can be made more responsive and what are the benefits of position vs. force feedback. Finally, through the use of position feedback on compliant robots, we show that the inherent properties of DMP learning itself can be used to achieve the same results as when coaching the robot. We present the results of a real-world experiment on a Kuka LWR-4 robot.