Dancing Humanoid Robots Recognition and Generation of Primitive Motions for Dance Imitation

Shinichiro Nakaoka, CVL Motion Group

(Collaborated with Humanoid Robotics Group, AIST) The goal of this study is to develop a humanoid robot which can master dance

performances by observing performances of a human. This is one of our attempts to develop a total technology of digital motion archive for traditional dance preservation.

Now we use a motion capturing system to acquire dance motion from a human, and we focus attention to how to import the captured motion data into a robot. A robot body is different from the human body in joint structure, mass distribution, athletic ability, etc. It becomes constraints when the captured motion is applied to the robot. In this study, robot motion is not directly converted from the original motion, but generated from symbolic representation recognized from the original motion. Symbolic representation is based on the concept ‘Primitive Motion’, which is a minimal unit of dance performance. Robot motion is recreated from a scratch according to a primitive sequence extracted from the original motion. This approach can simply solve the constraints of the robot, and enables various applications such as choreography editing. We have developed a software system to generate a feasible robot motion which satisfy mechanical constraints and dynamic balance. Generated dance motions were successfully performed on an actual robot, HRP-1S. Papers:

“Recognition and Generation of Leg Primitive Motions for Dance Imitation by a Humanid Robot”,

Shinichiro Nakaoka et al. 2nd International Symposium on Adaptive Motion of Animals and Machines,

2003, Kyoto, Japan.

“Generating Whole Body Motions for a Biped Humandid Robot from Captured Human Dances”, Shinichiro

Nakaoka et al. 2003 IEEE lInternational Conference on Robotics and Automation.

Captured Motion and Generated Robot Motion

Dance Demonstration by HRP-1S

Extracted Primitive Sequence