A Model of Acquisition of Discrete Bimanual Coordination Skill for a Humanoid Robot

There are several motivations for adapting the concepts of bimanual coordination originating in human motor control to humanoid robotics. Amongst the most important reasons we can highlight are the following: 1) guarantee adequate performance of humanoids in dual-arm tasks; 2) make their movements more effective and natural-looking for humans; and 3) provide the basis for more complex forms of interlimb coordination, e.g., quadro-manual coordination between robotic/human collaborators in the case of joint task execution. Robot Programming by Imitation proved to be a useful methodology for transferring various skills from a human to a robot through direct interaction. Although learning of manipulation tasks has received a lot of attention in this framework, none of the works address the problem of inter-limb coordination explicitly by extracting high-level features of coordination. This work attempts to shed some light on what are the main aspects of bimanual coordination to guarantee satisfactory robot’s performance in simple manipulation tasks and how these aspects can be learnt. We investigated two types of coordination constraints: spatial constraints (e.g. two arms must