Teaching Robots How to Walk and Dance

Over the weekend, a new video on the internet took the world by storm. Not exactly news in itself, but who—or rather what—starred in the video makes it noteworthy.

Like most viral diversions, this latest video was a riff on another campy sensation—Gangnam Style, the K-pop music video featuring a slick-haired, doughy rapper who rides an imaginary horse across all manner of over-saturated backdrops. The new video starts with Gangnam Style's familiar bass beat, but instead of the Korean sensation Psy bouncing around the screen, there's a white clad, black-visored robot waving its arms and banging its head.

CHARLI, the robot in the video, isn't nearly as fluid as Psy, and his leg lifts are restrained compared with Psy's manic prancing. But amongst robots, CHARLI is a bona fide Michael Jackson.

Dennis Hong and his Robotics and Mechanisms Laboratory (RoMeLa) at Virginia Tech built CHARLI to study bipedalism in robots. "CHARLI's groovy dance moves were just done for fun in the lab during our 'free time,' " Hong says. In addition to dancing, the robot competes in the vaunted RoboCup, a soccer league where roboticists test the speed and agility of their creations.

To get the robot to move to the beat, Hong and his team scripted the entire dance. It was programmed frame by frame on a computer, not constructed by recording the captured motion of a human dancer. "If you simply do a 'motion capture' of a person dancing and 'playing that motion back' on a robot—which is often done in generating the motions for characters in video games or movies using computer graphics—it does not work. The robot will fall," Hong points out. That's because a robot's center of gravity, and the center of mass in each of its body parts, is different from a human being's, he says. A human moving his head, for example, will compensate differently from a robot doing the same thing.

CHARLI isn't dancing on its own yet, but the performance is still a tour de force of flexibility and dynamism. At five feet tall, CHARLI is not a small robot. Such size complicates matters greatly. For example, to flail its arms, CHARLI's actuators must be sufficiently powerful to quickly overcome the inertia. Balance is another challenge—all that mass moving around so rapidly could easily upset a less sophisticated robot, even one that's not following a motion captured human dancer.

"Balance is difficult, especially if it is moving its limbs around in high speed," Hong says. "Normally the inertial forces created by the upper body motion is considered as disturbances by the lower body, and without coordination, the robot will fall. The lower body needs to compensate for the forces created by the upper body, and vice versa." Compensating for such upper body motions is state of the art, meaning CHARLI won't be jumping around like Psy—a hallmark of the Gangnam Style video—anytime soon. But don't count out future generations of robots.

Ultimately, Hong would love to have a robot that could not only jump around, but learn to dance on its own. "For the robot to really dance—besides its capability to be able to 'enjoy' it—requires many things besides the hardware design," Hong says. He lists the challenges: It must listen to the music, track the beat, and "understand" the musical style enough to construct an appropriate dance (something even many humans can't do). Then the robot must remain balanced throughout all the motions. Finally, Hong says, "trying to create a robot that can actually 'enjoy' the dance itself, that would be the most challenging of all."

Hong, CHARLI, and some of RoMeLa's other robots will be featured in the November 14 episode of NOVA scienceNOW. Watch a sneak peek of the episode in which CHARLI scores a goal in robot soccer, another challenging feat roboticists are striving to perfect.

blog comments powered by Disqus