Locomotion Imitation of Humanoid Using Goal-directed Self-adjusting Adaptor

We propose a novel framework for imitation learning that helps a humanoid robot achieve its goal of learning. There are apparent discrepancies in shapes and sizes among humans and humanoid robots. It would be advantageous if robots could learn their behavior from different individuals. Toward this end, this paper discusses appropriate behavior generation method through imitation learning considering that demonstrator and imitator robots have different kinematics and dynamics. As part of a wider interest in behavior generation in general, this work mainly investigates how an imitator robot adapts a reference locomotion gait captured from a demonstrator robot. Specifically, a goal-directed adaptation process that we call self-adjusting adaptor is proposed to achieve stable locomotion of the imitator. The proposed adaptor has an important role that the perceived locomotion patterns are modified to keep the direction of lower leg contacting the ground identical between the demonstrator and the imitator, sustaining the dynamic stability by controlling the position of the center of mass. The validity of the proposed scheme is evaluated through simulations employing various imitator models on OpenHRP and then verified on a real robot