Transfer of contact skills to new environmental conditions

Robots operating in contact with the environment should typically take into account the knowledge of both position/orientation trajectories as well as the accompanying force/torque profiles for successful execution. Pure position control is not appropriate because even small errors in the desired trajectory can cause significant forces at the contact points. In this paper we present a method that computes an appropriate control policy for a given condition of a contact task, with the peg-in-hole (PiH) assembly tasks as example use-cases. Our method is based on statistical generalization of successfully recorded executions at different values of the external condition. The major novelty of the method is that it provides not only generalized position and orientation trajectories, but a complete skill, consisting of desired position/orientation trajectories and the accompanying force/torque profiles. To improve the execution of the skill after generalization, we combine the proposed approach with an adaptation method to refine the newly generated movement. The versatility of the proposed approach was shown by applying it to firstly, two different types of robot arms: a humanoid 7-axis Kuka LWR-4 arm and a 6-axis industrial Universal robot UR5 arm and secondly, two different peg-in-hole problems: insertion of a square peg and insertion of a round peg.