LAUSR

Robotic compliant manipulation is a very challenging but urgent research spot in the domain of robotics. One difficulty lies in the lack of a unified representation for encoding and learning of compliant profiles. This work aims to introduce a novel learning and control framework to address this problem: i) we provide a parametric representation that enables a compliant skill to be encoded in a parametric space and allows a robot to learn compliant manipulation skills based on motion and force information collected from human demonstrations; and ii) the updating laws of the compliant profiles including impedance and force profiles are derived from a biomimetic control strategy based on the human motor learning principles. Our approach enables the simultaneous adaptation of impedance and feedforward force online during robot’s reproduction of the demonstrated tasks to deal with task dynamics and external interferences. The proposed approach is verified based on both simulation and real-world task scenarios.