Responsive Gesture Planning based on Integration of Motion Database and On-line Planning

Abstract

Generating natural body gestures is one of the essential functions in human-robot interaction. Various approaches, such as motion database approach and on-line planning, have been developed. However, these approaches cannot adapt to user's interruption, which usually occurs. Therefore, we propose a novel method for generating motions flexibly and immediately. The proposed method integrates Motion Graph and Probabilistic Roadmap Method. The former helps generation of human-like motion, while the latter helps rapid reaction due to user-intended interruption. We conducted the experiments to verify that the proposed method can maintain naturalness of body gestures, which we define based on key-poses and velocity profiles of the original motion. In addition, we implemented the proposed method on the android robot Actroid-SIT as a human-robot interaction system. Through subject experiments, we confirmed that the proposed method can make human-robot interaction more smooth and durable.