Generating whole body motions for a biped humanoid robot from captured human dances

Recently, the technology of biped humanoid robots has progressed to the point where it has the potential to realize complex tasks with the whole body. The goal of this study is imitating human dance motions for preservation of traditional dances. The problem with dance imitation is the difference of body structure and physical properties between humans and robots. This study describes how we solve the problem by using filters to acquire feasible robot motions from the original human motions and developing a motion generation method based on motion primitives. Human dance motions are acquired by motion caputuring systems. Analyzing the trajectories of particular marker positions enables us to extract a high level structure of primitive motions. Initial arm motion is generated by inverse kinematics of the joint positions and is modified into a stable motion which statisfys the mechanical structure and capacity of actuators. Leg motion is generated from the sequence of primitive motions, considering constraints of the robot. Then, a waist trajectory which satisfies the desired balanced point between the body and the ground is calculated. This process enables the robot to keep its balance without falling down. Generated motions were tested on the OpenHRP system, which consists of a dynamics simulator and actual robots. In this test, the validity of this study has been certified.