Robot learning from human demonstration with remote lead hrough teaching

Industrial robots are playing increasingly important roles in factories. Many production applications require both position and force control; however, tuning the positionforce controller is nontrivial. To simplify this process, the learning from demonstration (LfD) is proposed to transfer the human skills directly into robot applications. However, the current teaching methods, such as direct demonstration, lead through teaching, and teleoperation, all have their own drawbacks. Hence, Remote Lead Through Teaching (RLTT) is proposed to robot learn some tasks from human knowledge and skill. To implement the human skill model, the demonstration data is firstly synchronized by dynamic time warping (DTW), then decomposed into several actions by a support vector machine (SVM) based classifier. Lastly, the learning controller is trained by the Gaussian mixture regression (GMR). The experimental validation is realized on FANUC LR Mate 200ÍD/7L in a H7/h7 peg-hole insertion task and a surface grinding task.