Whole body Imitation of human motion with Humanoid robot via ZMP Stability Criterion

Abstract

In this paper, we deal with optimal mimicking of human motions by humanoid robot, KIBO (by Korea Institute of Science and Technology, KIST) in real-time. Particularly, the key issue is to investigate stability discriminants for preventing humanoid robot, KIBO, from falling down when imitation human body. First of all, we replace skeleton data of human body collected from 3D depth camera, KINECT (Microsoft) with robot angles in joint space. In this case, the translated data is unsuitable for robot movements since frequently dissatisfying ZMP (Zero Moment Point) criterion for humanoid robot. Hence, we have proposed EFS (Estimation Function for Stability) which estimates COG (Center of Gravity) of humanoid robot through simple model. The proposed stability discriminant is used for data effectiveness based on the ZMP criterion. In case of invalidity data, we have additionally introduced optimization methods corresponding with EFS, which minimizes the amount of change to maintain similarity with the original human movements.