A Grid-based Approach to the Body Correspondence Problem in Robot Learning by Imitation

In the learning by imitation framework, a module is required to translate the visually perceived behaviour of the demonstrator’s body to a represen tation where the imitator perceives its own movements, either visually, propioceptively, or both. This transformation must take into account the different body configurations of imitator and d emonstrator. In this work, a solution to translation of the visually perceived end-effector motio n is presented. The proposed approach is included in a system which allows a robot to learn the behaviour of a human demonstrator from stereo visual information. Real-time results ar e presented and discussed.