This dancing space robot is cool (but illustrates important limitations)

Valkyrie is arguably the most sophisticated humanoid ever built. But even NASA can't make a robot walk like a human.

Who doesn't like a dancing robot? The smarty-pants at NASA know the answer: no one, that's who. This week the space agency released a funky video of its Valkyrie (AKA R5) humanoid getting down like an overly-cautious James Brown.

The video offers a look at some spectacular robotics engineering, but it also serves to underscore one of the biggest limitations of even the most impressive anthropomorphic robots. Namely, the mechanisms underlying the way you walk are incredibly complex. No matter how many servos and joints we give them, robots still can't accurately mimic the human gait.

There's a very specific reason why. Every time you take a step, you're catapulting yourself forward over the upside-down pendulum of your planted leg. In a very literal sense you're allowing yourself to fall and taking it on faith that your swing leg will come through in the nick of time. If it doesn't, you, my friend, are kissing sidewalk.

The miraculous thing about the human gait is that it's an inherently unstable process. It's only thanks to the miracle computer of the human brain, which can simultaneously account for an incredible rush of sensory information and adapt to unforeseen variables with great agility, that we can use that instability for functional locomotion.


But engineers design robotic controls to seek stability. You don't want your robot to be out of control in a real world environment where there are too many variables to quickly account for. Mechanically speaking, that portion of the gait cycle when we allow ourselves to fall is uncontrolled.

Toward the end of the video you can see Valkyrie walking. Compared to the groove-tastic upper body movements it makes throughout the video, which are reasonably close facsimiles of human motion, the steps look cautious and clunky. In a word, they look robotic.

So here's the question: Does it matter whether a humanoid can walk the way we do? The answer is that it probably does. The whole point of designing a robot to look and move like a person is so the system can function in an environment designed for humans. If you send a robot in to shut down a nuclear reactor, the scenario that the Darpa Robotics Challenge is based around, then your machine will need to navigate walkways, doorways, and control rooms and be able to manipulate everything from buttons to valves.

Similarly, if you want to send a robot into space, it's helpful if it can easily ride in a spacecraft designed for a human crew and sit shotgun in a rover to help astronauts explore the surface of Mars. That's why NASA is so interested in a working humanoid. But for humanoids to be useful assistants on space missions, they should be able to navigate terrain as dexterously and quickly as a human can. If your humanoid can't lunge over a crater, its usefulness will be limited.

A couple weeks ago, NASA announced that it was giving one Valkyrie humanoid to MIT and one to Northeastern, along with $250,000 each, to advance the capabilities of the system. Valkyrie had a disappointing showing at the 2013 DRC trials and didn't qualify for the finals, so it makes sense that NASA is looking for some outside help to get its robot, which has cost about $6M to develop so far, closer to mission-ready.

But the hurdles to allowing a robot to move in the real world like a human are enormous. When Valkyrie can dance Gangnam Style, it might be ready for the moonwalk.