Robot Skill Learning in Human-Robot Collaboration with Dynamic Time Warping Iterative Learning Control

Human-robot collaboration is a common paradigm of physical human-robot interaction, typically involving human users transferring their expertise in performing repetitive tasks to the robot. However, conventional methods, such as programming by demonstration and learning from demonstration, are not able to yield accurate tracking performance and do not allow robot behavior modifications on an online basis. In order to overcome such deficiencies, iterative learning control is employed to attain improved tracking accuracy, while the trial-by-trial modification in accordance with the physical interaction is also allowed. Further, in terms of the varying trial lengths during the collaboration, dynamic time warping is integrated to align the multiple trial paths, while the convergence of spatial errors is ensured. Finally, simulation results are presented to demonstrate the effectiveness of the proposed method in collaborative tasks.