Effective humanoid motion generation based on programming-by-demonstration method for entertainment robotics

Abstract

Generating a motion for a humanoid robot is very cumbersome task for robot programmers who use machine level program codes to develop new behaviors of the humanoid robot. Especially in the case of a motion that is usually consisted of complex postures with long length such as dances and martial arts, the difficulty of the generation process is very high. In this paper, we propose a method to generate motions of humanoid robots based on the user demonstration. A wearable interface is designed and developed to capture the users demonstration. A curve simplification algorithm and a clustering method are applied to convert an original motion sequence into an editable motion. Transition probabilities are calculated to construct the motion primitive from the editable motion and motions are regenerated by rearranging and transforming the motion primitives. A humanoid robot, AMIO is used to test the generated motion.