Human Expert Skill Transferring and Imitation for Teleoperated Dual-Arm Manipulation

Abstract

Humanoid robots used for medical care, household, logistics are deployed more widely in a complex working context, and a great deal of current robot research focuses on teleoperated high degree-of-freedom control for delicate human skill transfer. However, severe vibration phenomenon caused by unwanted noise usually leads to non-smooth robot trajectory. Although advanced robot arms have been equipped with torque sensors to perform torque control, it is not feasible to deploy widely in different workspaces due to its cost. The work reported a model-free approach as an easy-realized and inexpensive method. First, noise elimination was realized by empirical mode decomposition and Savitzky-Golay filter. Second, regenerating motor controllable trajectory was done by down sampling and quadratic interpolation to re-construct a simpler, accurate trajectory with piecewise constant acceleration. After attenuating unwanted vibration noise and interpolating proper sample points, it was shown that the accuracy was improved by 60% with the proposed approach. The result also demonstrated remote human skill transferring, which matches teleoperated target signals for human motion to robotic arm kinematic motion.