Control Architecture for Robotic-Assisted Polishing Tasks Based on Human Skills

This paper describes a computed-torque control architecture for robotic-assisted polishing tasks based on human demonstration. An impedance control approach for both robot position and orientation is investigated without explicit force sensing, relying on compliant frames relative to the end-effector and on posture optimization in the null space. For the orientation control the approach relies on rotation matrix properties connected to the axis-angle formulation, having no singularity problems. Additionally, we also propose a robust technique for integral control on the orientation error. Regarding Cartesian positioning, we are using typical PID-based control techniques. A Panda robot from Franka Emika is used to validate the control architecture. Regarding human demonstration and skill transfer, two steps have been followed. The first one corresponds to robot teaching, where the user takes the manipulator by hand (co-manipulation), creating a pattern of positions and forces. The second step processes co-manipulation data to generate references for the controller. To evaluate robot performance, the teaching pattern is replicated several times throughout a wide area and assessed w.r.t. position and orientation tracking.