Latest in Carnegiemellonuniversity

Image credit:

The bots that bucked the humanoid trend at DARPA's challenge


Humanoids are supposed to be best suited for a world that's engineered for humans. They can climb stairs, open doors and drive cars. At DARPA Robotics Challenge, most of the participating humanoids succeeded in completing those tasks, but when it came to walking, they were clumsier than the species they were designed to emulate. Getting a machine to put one foot in front of the other has been a priority for roboticists for decades. Bipedal locomotion would presumably make it easier for robots to navigate a man-made world and perhaps make them more relatable. But the movement requires sophisticated control software and advanced AI technology that isn't quite ready yet.

DARPA's challenge was never about the fastest robot or about the one with the most tricks up its metal sleeves. It was about finding a mobile machine that could aid, or even replace, a first responder in the event of a disaster. Entering the competition with a biped wasn't a requisite, but they were the popular choice. All but three of the 23 competing robots were humanoids. Out on the obstacle course though, non-humanoid robots looked sturdier than their bipedal competitors.

Carnegie Mellon University's four-limbed CHIMP was one of the strongest contenders. It was designed to roll on the tracks installed on its limbs, which gave him an edge and flexibility over its competitors. The bright red robot picked itself up after a fall on day one and bounced back from a driving accident on day two, when it rammed its vehicle into a wall. It eventually grabbed the third spot in the overall challenge.

NASA's Jet Propulsion Laboratory brought RoboSimian (aka King Louie), a robot with four identical limbs that could be used as arms or legs depending on the task. It was adept at getting out of a vehicle, which is one of the more challenging 3D environments for a four-limbed machine, but its sluggish pace held it back. It came in fifth in the race.

The third non-humanoid, Team NimBro's Momaro (short for Mobile Manipulation Robot), beat RoboSimian on time. It had an anthropomorphic upper body and a mobile base with four legs that ended in steerable wheels. "With the bipeds, stability is an issue," said Sven Behnke, Momaro's team leader and professor at University of Bonn in Germany. "We have seen some of the bipeds falling. But if you have four legs, it's easier to navigate the terrain." During its final run, the peculiar robot almost tipped over a couple of times. But the operator stabilized the machine and it soon wheeled out of the rubble that stood in its way.

"I'm not using a legged robot because I think it's the best for disaster response," said Russ Tedrake, professor at MIT's Computer Science and Artificial Intelligence Lab. "But it's hard for the right reasons. The problem of balancing a humanoid robot is helping us [figure out] how to make manufacturing robots or the ones that go into your body and help a surgeon." Even though his own team led an Atlas humanoid in the race, he thought legs were a liability in the race. "Some people argue that humanoid robots are uniquely capable of operating in a human made environment," he said. "I'm not sure I totally buy that. I think in many situations a wheeled robot would do very well."

In the end, Team KAIST's DRC-Hubo, a humanoid equipped with wheels, beat 22 robots to win the challenge. It switched between walking and rolling to traverse the terrain.

The challenge brought some of the world's best robots out to play. But most of the machines struggled through mundane tasks and stumbled their way to the finish line. "All these robots are unsuitable for disaster response. They're meant for a challenge that has been inspired by a disaster," says Behnke. "It's a technical challenge with some requirements that are much lower than real disaster response robots would need to be [prepared for]."

Irrespective of their shapes, these robots aren't ready to be deployed for search and rescue missions just yet. But, despite their sluggishness and face-forward falls, they're not going to be cooped in a lab for too long either. "We're only an iteration or two away from RoboSimian being a commercial item," said Brett Kennedy, principal investigator at NASA's JPL. "We've already licensed the technology and we hope to find it in the field very soon."

From around the web

ear iconeye icontext file