A Stable Guidance Method for Teleoperation-based Robot Learning from Demonstration

Robot learning from teleoperation-based demonstration has been proven to be an effective method for robot skills learning. However, the human demonstration process is mentally and physically challenging due to the dexterous and continuous interaction control. Especially, for contact-rich tasks human operators need to control the pose and contact force between the end-effector and the environment, only based on visual and haptic feedback. To tackle the problem, this work studies a stable guidance method for teleoperation-based demonstration to improve the experience and performance of human demonstration and reduce the workload of human operators. The guidance design is based on a dynamic system-based imitation learning model, which is achieved by human demonstration. The virtual guidance force will decrease as the virtual guidance’s iterative learning. The efficiency of human demonstration is improved, and human operators’ physical and mental workload is reduced. Two typical experiments, robot-assisted polishing and robot-assisted ultrasound scanning, are conducted to validate the performance of the guidance method.