South Korean Team Takes $2 Million Top Prize At DARPA Robotics Challenge

DARPA wants to inspire the creation of robots capable of disaster recovery. But there’s a long way to go and a lot of work still to do.

Members of Team KAIST jumped up and down and grabbed each other in an emphatic embrace as their robot, known as DRC-Hubo, placed its second foot atop the platform. The official raised his arm in the air.


Mark it: eight points. In a dominating time of just a hair over 44 minutes. Although several leading teams still had yet to take the course, the DARPA Robotics Challenge (DRC) looked like it was all but over.

And it was. By the end of the day Saturday, the second and last full day of the finals of a three-year-long search for a champion, Team KAIST, from the Korea Advanced Institute of Science and Technology, had claimed the $2 million top prize and permanent rights to brag that it had bested worthy competitors from august institutions like Carnegie Mellon University, the Massachusetts Institute of Technology, NASA’s Jet Propulsion Laboratory, and others across the world.

DARPA, the federal Defense Advanced Research Projects Agency, launched the competition after robots sent to help tackle the disaster at the Fukushima nuclear facility in Japan in the wake of that country’s horrific 2011 earthquake and tsunami were unable to alleviate mounting pressure inside the plant. The explosions that followed left the nearby area a radioactive “no-man’s land,” according to Gill Pratt, the DRC program manager, which “had a profound effect on many of us at DARPA.”


Rescue Robots

With the DRC, DARPA hopes to inspire the development of robots capable of operating “in spaces that are designed for humans, use tools that are built for humans, and can be controlled with only minimal training by people who are not robotics experts,” according to an agency document. “Achieving these goals would mean that robots could be rapidly put to use assisting in human-led response to future emergencies.”

Dozens of teams had participated in earlier rounds of the competition, but the best 24 gathered at the Fairplex, a massive event space east of Los Angeles, to vie for the crown this past weekend. In the end, in front of several thousand people, including Google CEO Larry Page and Tesla and SpaceX CEO Elon Musk, Team KAIST blew away the field.

Each team’s robot got two shots at the course, attempting to score points by driving a utility vehicle down a dirt track, exiting the vehicle, opening and going through a door, closing a valve, cutting through a piece of drywall, tackling a surprise task, crossing a field of rubble, and then climbing a short set of stairs. Each task was worth a point, for a total maximum of eight. Winning meant having the most points, or the best time in case of a tie.


Finishing was a daunting task, as evidenced by the fact that by the end, just three teams collected all eight points. Among those, Team KAIST finished the course in 44 minutes, 28 seconds. Running Man, from Team IHMC Robotics, came in second and won $1 million with a time of 50 minutes, 26 seconds, and Carnegie Mellon University’s CHIMP grabbed third place, and $500,000, finishing in 55:15.

All About Software And Control Systems

To the casual observer, it would have been hard to tell some of the robots apart. Seven teams used Boston Dynamics’ Atlas robot, while KAIST and the team from the University of Nevada, Las Vegas used the Hubo platform–developed by the Koreans themselves. In truth, the DRC was almost certainly more of a software and control systems competition than a hardware challenge.

“Winning first place in this challenge doesn’t mean [ours is] the best robot in the world,” Jun Ho Oh, the leader of Team KAIST, told Fast Company. “It just means it got the best points, no more, no less….The thing is the balance. Some robots have strong power, but less mobility. Some robots have very smart [movement], but less vision, and the body’s too big. And some…have very good autonomy, but [are] over-autonomized. So, [winning was about] good balance between supervisory and autonomy.”


That’s for sure, said Scott Lavalley, one of the Atlas design leads at Boston Dynamics.

“They’re all the same,” Lavalley said of the many teams competing in the DRC with an Atlas robot. “They can modify it with sensors and computing. It’s all controls, all software [which differentiates the teams]. We deliver the hardware, and then it’s up to them to write software and navigate it through the course.”

Of course, many teams believed their approach put them in the best position to take the DRC crown. For example, Carnegie Mellon University National Robotics Engineering Center’s custom-built CHIMP finished the first day as the only robot with eight points. Afterwards, the team’s director, Herman Herman, explained to Fast Company that what was required to succeed was a whole lot of experience and a design tailored for the problems at hand.


“What you need is a robust [robot], so it can recover if it makes a mistake,” Herman said. “It’s supposed to do a rescue mission. It cannot need to have to be rescued itself.”

Falling Down

Watching the competition, that quickly became very clear. One after another, robots stumbled and fell, often in very dramatic fashion, and usually eliciting a loud groan from the audience. A video posted of all the falls was viewed more than 128,000 times. It was funny, but made the point that building and designing robots that can successfully navigate a course like that of the DRC is extremely hard. And doing it without humans needing to come in to rescue the rescue robots is a daunting challenge.

Although it scored eight points on Saturday, CHIMP didn’t avoid falling. But Herman argued that one of the things that set the robot apart was that its software and hardware design enabled it to get back up without help–the only robot at the DRC to do so.


KAIST’s Ho Oh made the same point after the competition, with the benefit, of course, of having won without his robot falling down during its dominating Saturday performance.

“The real disaster was falling down,” Ho Oh said. “This is about ‘disaster recovery.’ That means that all the robots should be free from disaster or they have to recover from [falling] without having any help from outside.”

Otherwise, he argued, a recovery robot working in a contaminated environment, or one too dangerous for people to enter, will never be able to help anyone.


What Was Learned

Although the DRC was nominally a competition aimed at advancing the state of the art of using robots for disaster recovery, there were some on hand who felt that something much more important had been achieved.

Among hard-core roboticists, there is a sense that the public, spurred by media hype, is scared of what robots mean for the future. Are they coming for our jobs? Are they coming to take over? Few storylines in popular media, with perhaps Disney’s Big Hero 6 as a recent exception, are about robots helping people.

Yet, said DARPA’s Pratt, speaking of the robots that competed in the DRC, “These are not small machines, and you would assume [that] the many thousands of spectators would have been filled with fear and anxiety. In the news [it’s all about how] they’re going to kill us, and we should all run away.”


But, Pratt added, the opposite proved to be true.

“They cheered,” he said, “despite the fact that the actual event was one robot [at a time] going through eight very simple tasks and taking nearly an hour to do it. It’s ridiculously slow….There is a new discovery, I think, that we made here: That there’s some incredible untapped affinity between people and robots that I think we saw for the first time today.”

Tony Stentz, the CHIMP team leader, agreed.


“I think the future of robotics is one where robots collaborate with humans,” Stentz said. “For that to be successful, it’s important that the two build bonds.”

In the end, Pratt spoke eloquently about where things stand in the hunt for robots that can be counted on in situations like the Fukushima disaster. There’s a very long way to go, he said, despite the many impressive DRC performances.

“Have we finished the work that’s to be done in robotics,” Pratt asked, rhetorically. “No. Are we in the middle of the work that’s to be done? No. Are we perhaps at the beginning of the work that’s to be done? Yes….We’ve shown the world just a little bit, a tiny bit, of what’s possible.”


About the author

Daniel Terdiman is a San Francisco-based technology journalist with nearly 20 years of experience. A veteran of CNET and VentureBeat, Daniel has also written for Wired, The New York Times, Time, and many other publications