Adaptation of bimanual assembly tasks using iterative learning framework

The paper deals with the adaptation of bimanual assembly tasks. First, the desired policy is shown by human demonstration using kinesthetic guidance, where both trajectories and interaction forces are captured. Captured entities are portioned to absolute and relative coordinates. During the execution, small discrepancies in object geometry as well as the influence of an imperfect control can result in large contact forces. Force control can diminish the above mentioned problems only to some extent. Therefore, we propose a framework that iteratively modifies the original demonstrated trajectory in order to increase the performance of the typical assembly tasks. The approach is validated on bimanual peg in a hole task using two KUKA LWR robots.