DARPA Robotics Challenge Trials: What You Should (and Shouldn't) Expect to See

We're super excited for the DARPA Robotics Challenge Trials that run tomorrow and Saturday, and we hope you are too. Before we get all wrapped up in it, though, it's important to understand what's realistic to expect from the robots during the competition, to prevent yourself from (let's be honest here) experiencing some level of (potentially profound) disappointment.

The context in which you should think about the DRC Trials is similar to the first DARPA Grand Challenge for autonomous vehicles that took place back in 2004. DARPA announced the event in 2002, giving 15 teams less than two years to develop fully autonomous vehicles that could navigate a rugged Mojave desert course of 150 miles. On 13 March 2004, the day of the event, two teams withdrew immediately, and one vehicle managed to flip itself over at the starting line. Of the rest, only four teams made it past the 5 mile mark, before breaking or crashing. The team that did best that day was from Carnegie Mellon, as DARPA's final data describe:

"Vehicle 22 Red Team (Carnegie Mellon): At mile 7.4, on switchbacks in a mountainous section, vehicle went off course, got caught on a berm and rubber on the front wheels caught fire, which was quickly extinguished. Vehicle was command-disabled."

Yep. A little more than 7 miles, or less than 5 percent of the entire course, was the distance record for the race. Not surprisingly, the media reaction to the event was was quite unenthusiastic. DARPA, however, was undaunted:

"One of the best ways to motivate engineers is to tell them that there's something that can't be done. And what you saw today was people taking on that challenge and saying, Nah, it's not impossible, I'm gonna try," said Tom Strat, deputy program manager of the DARPA Grand Challenge. "Even though nobody got more than about 5 percent of the way through the course, this has made these engineers even more determined."

Just one year later, most of those determined engineers went back to the Mojave with better hardware, better software, and more experience. The second Grand Challenge in 2005 was as much of a success as the first Grand Challenge wasn't, with all but one team making it farther than the original 7.4 mile mark, and four teams completing the entire 132 mile course, completely autonomously. Later, the DARPA Urban Challenge, held in 2007, was also a success.

 


 

The whole point of these challenges that DARPA puts together is to (first) discover what the current state of the technology is, and then (second) help boost that technology into a usable form. So really, the way to think about these initial challenges (the 2004 Grand Challenge as well as the 2013 DRC Trials) are just that: initial challenges, as opposed to final competitions. This is why we've got the DRC Trials this week, and the DRC Finals won't take place for another year. These trials are more of a "calibration" event, as DARPA Program Manager Gill Pratt explained to IEEE Spectrum when we asked him about the science-fictiony public expectation that's built up around robots as of late:

"I think part of the good that can come out of the trials is that we'll actually help calibrate the public to what the reality is in this field. Part of the difficulty with science fiction is that if there's no counterexample of science fact, people can get the idea that these [robots] aren't very hard to build. So, besides calibrating ourselves to what the state of the art is, I think a lot of the good that we can do here is to calibrate the public."

So with that in mind, what can we expect for the DRC Trials? Generally, DARPA suggests that you expect "robots with roughly the mobility and dexterity of a one-year-old child who can barely walk, falls down frequently, and can’t execute complex manipulation tasks." Specifically, this is what we're assuming we'll be seeing:

  • Most of the time, nothing will be happening. The robots are going to spend a lot of time staring at things and not moving, as their operators interpret data and do path planning. For spectators, this is going to be super duper boring.
  • When things do happen, they will happen very, very slowly. Each robot is allocated up to 30 minutes to complete each task, and DARPA expects that many (if not most) teams will take advantage of most of that time.
  • Autonomy will be limited. In fact, none of the robots being developed for the DRC will act completely autonomously. We'll be seeing a lot of teleoperation, and possible some assistive autonomy, like a human instructing a robot to pick something up and the robot deciding how best to do that. But even the assistive autonomy may not really kick in until next year.
  • Robots will screw up. Expect robots to miss grasps and drop things. Frequently.
  • Robots will fall down. Also expect robots to miss foot placements or lose their balance. Frequently. Most of the robots will be tethered for power, and most tasks will include a safety cable to prevent a total face plant, but there will likely be a lot of stumbling that may require human intervention to rectify.
  • Robots may do nothing at all. These robots are all very, very mechanically complex, and very complex robots just break a lot of the time. Failures are hard to predict, but our guess is that more than one team may not be ready to compete for every required event, especially towards the end of the second day, when they may already have been repaired multiple times.

We're definitely not trying to be pessimistic here, and we're also not trying to minimize the importance of the DRC Trials and the amazing effort by all the teams, nor the robotic wonderment that we'll be seeing this weekend. Rather, we want you to go into this knowing what to expect so that you can be amazed when a robot does something that may not necessarily seem amazing. Like, if a robot is able to get in and out of a vehicle by itself, that's amazing. Or if a robot is able to locate and turn a valve without being aggressively micromanaged by its human controllers, that's amazing. Or, if a robot is able to walk over rough terrain without falling over, that's amazing, too. A robot that can do any of these things in (say) 15 minutes instead of 30 is even more amazing.

These examples that we're giving are, to some extent, being pulled out of pretty much nowhere. As we talked about before, we're not yet calibrated to what the current state of the art in robotics is, and I'm guessing that even the teams themselves aren't entirely sure how things are going to go during the competition either. And there's certainly a chance that we'll be wrong about all this, and that the robots will totally wreck (figuratively wreck) the entire course, forcing DARPA to make it harder for next year.

No matter what happens over the next two days, we'll be experiencing the future of robotics. It may not happen tomorrow, but we're already looking forward to a massive increase in ability by the DRC Finals, and within a few years for sure there will be humanoid robots (whether these robots or their descendants) ready to assist in disaster relief and in many other tasks. In the same way that DARPA's autonomous vehicle challenges inspired real-world pre-consumer technologies like Google's autonomous cars, our hope is that five or 10 years from now (or sooner), we'll be looking back at the DRC Trials and Finals as a major step towards making robots an integral part of our lives.

[ DARPA Robotics Challenge ]

Advertisement

Automaton

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

Editor
Erico Guizzo
New York, N.Y.
Senior Writer
Evan Ackerman
Berkeley, Calif.
 
Contributor
Jason Falconer
Canada
Contributor
Angelica Lim
Tokyo, Japan
 

Newsletter Sign Up

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

Advertisement