DARPA Robotics Challenge Finals: What We Learned on Day 1

Photo: Evan Ackerman/IEEE Spectrum
Look ma, one hand!

We’ve been looking forward to the DARPA Robotics Challenge Finals for years. The DRC Trials in December of 2013 showed us how much potential disaster robots had, and yesterday and today, those robots are competing against each other to prove their utility in hardware and software in a mock disaster scenario.

Hopefully, you’ve been following our posts and social media coverage. If not, a good place to get started is this post about the robots and this post about the course, and we’re tweeting like mad from @AutomatonBlog and @BotJunkie. DARPA is streaming everything they can, so you should tune in to catch a little bit of the action.

All kinds of amazing things happened yesterday, so in pictures and commentary, just before we start with Day 2 of the Finals, we’re going to take you through everything we learned on Day 1.

On Thursday, the day before the Finals started, each team got to do to a trial run. The trial runs went really well for most of them: several teams ended up with a full 8 points. As of Friday morning, many teams seemed to be pretty confident that everything would go just as well. Having said that, most teams were very conservative on Friday, because nobody wanted to smash their robot on the first day of the competition. Teams will have another run today to see if they can improve their scores and times. In fact, if the top teams all score 8 points, it will come down to which team completed the course the fastest. So today, speed is more important than ever.

All the Tasks Are Doable, but Some Are Tricky

We knew going in that egress from the Polaris vehicle was going to be hard, but plenty of teams pulled it off successfully. The surprise task turned out to be easy, too: we say this because teams would skip the wall task to do the surprise task first, and then come back to the wall task if they had time. Wall task, which requires grasping and operating a drill, was tough, and the stairs task was tough as well, to the point that many teams decided to not do the stairs on Day 1 for fear of falling over and not getting another run on Day 2. The other task that turned out to be tough? Door. Not because opening the door was hard (robots only had to pull down the handle and the door would swing open), but because some of the robots were too big. These robots had to pass through the narrow doorway sideways or by tucking in their arms as close as possible to their bodies, and as they attempted to do that, some of the them fell over. We heard more than one person in the audience joking that, if you want to stop the robot uprising, just close your door. 

Lots of Robots Fell Over and Didn’t Break

Lots of robots fell over yesterday. Some of these robots are big, complicated, and not intended to be super robust against falls. But what was a good surprise was that most robots that fell didn’t get damaged significantly. Most of the time, teams called a reset (which carries a 10-minute penalty), and got their robot up and running after that. Will the falls today be even more spectacular? Probably. Teams want to both score points and get a good time, and that means they’ll be taking more chances today.

ATLAS Teams Were Especially Terrified of Falling

As far as we know, none of the ATLAS teams tried letting their robots fall in testing. It’s easy to understand why: the robots aren’t really designed to fall, and you don’t want to risk your super complex and expensive robot if you don’t have to. Boston Dynamics recommended that if at all possible, teams should try to get their robots to fall backwards and not forwards, but if you could control where your robot was going to fall, you’d just have it not fall, right? The good (and slightly surprising) news is that ATLAS robots are actually pretty resilient. Most ATLAS teams were surprised too, not just us. Everybody with an ATLAS relaxed a little bit when the very first robot to move, HKU, faceplanted with their ATLAS on literally the very first step, and then reset and walked successfully right afterwards. This isn’t to say that ATLAS robots are immune to breaking, but the expectation was that a fall would be catastrophic for the robot more often than not, so it was great to see that despite a bunch of missteps, as far as we know no ATLAS robots are out of commission.

Getting Up Is Hard, But We’re Not Yet Sure How Hard


Trying to get up after a fall is probably the most difficult maneuver in the DRC Finals, because the robots aren’t designed for it, and teams are reluctant to practice it. It’s also dangerous, because attempting to get up from a fall can just result in another fall. It’s probably for this reason that most teams chose to do a reset on Friday and take the 10-minute penalty; since tomorrow is the last day of the competition, we’re hoping to see more robots trying to get themselves back up off the ground. They may not be successful, of course, but we’re hoping that they do at least try.

Maybe Walking Robots Aren’t the Best Idea

It rapidly became clear that at this point, robots that aren’t bipedal walkers have a significant advantage. We’re not saying that a biped can’t do as well as one of the robots that can (say) roll on wheels when it needs to, but right now, biped walking is still really hard, and rolling is really easy: better stability for doing tasks, risk of falls much lower, and faster movement on flat ground. In fact, we’re a bit surprised that so many teams adopted conventional humanoid designs, with two legs that need to support a relatively big torso and arms. Sure, researchers have been working on such robots for decades, but we were expecting to see some more bold, unconventional designs. Well, maybe at the DRC 2?

CHIMP Is Awesome

Speaking of unconventional designs, CHIMP is really awesome. We have a whole post on how awesome CHIMP was on Friday, and you should go read it if you haven’t yet. It’s one of the most inspiring displays of robotics that we’ve ever seen. Ever.

NIMBRO Is the Surprise Team to Beat

The second place team is NIMBRO, from the University of Bonn in Germany. Their four legged, two armed, wheeled robot with a head that continuously spins around was a little bit janky looking, if we’re honest, but it turned out to be insanely good. The robot drove, did egress, moved on to door, valve, surprise, and got to the rubble in just 17 minutes. After the rubble, they drove back to the wall task, and ended with 7 points in just 34 minutes. Wow.

Expect IHMC, MIT, WPI-CMU, KAIST, and RoboSimian to Also Do Well

Despite some falls, both IHMC and MIT scored well, completing 7 and 6 tasks respectively. WPI-CMU didn’t fall, and completed 7 tasks (including the stair climb) with just 25 seconds left on the clock. KAIST also completed 7 tasks, one minute faster than WPI. RoboSimian completed 7 tasks in 48 minutes, but decided not to attempt the stairs.

Here’s a video of RoboSimian’s egress technique, which Gill Pratt told us was one of his favorite moments of the competition:

Today Will Be Even Better

Like we said at the top of this article, teams were intentionally conservative on Friday in order to preserve their hardware. Today is the last day of the competition, so there’s nothing to lose, and we’re expecting a lot more aggression: robots moving faster, taking shortcuts where possible, and getting themselves up off the ground if they fall. And they will fall, even more than on Friday. Today is going to be the real test of speed and endurance, and we're expecting greatness.

[ DARPA Robotics Challenge Finals ]



IEEE Spectrum’s award-winning robotics blog, featuring news, articles, and videos on robots, humanoids, drones, automation, artificial intelligence, and more.
Contact us:  e.guizzo@ieee.org

Erico Guizzo
Senior Writer
Evan Ackerman

Newsletter Sign Up

Sign up for the Automaton newsletter and get biweekly updates about robotics, automation, and AI, all delivered directly to your inbox.