Yesterday afternoon, DARPA held a briefing to discuss the forthcoming DARPA Robotics Challenge Finals. It's been about six months since the DRC Trials were held in Miami, so we've been expecting an update, and DARPA certainly delivered.
Program manager Gill Pratt spent over an hour explaining what we have to look forward to in Southern California (yes, the Finals will be held in California!) next June (yes, the Finals are not happening this year, as DARPA decided to give teams some extra time) and we've got all the highlights for you.
Rather than just communicate everything in giant paragraphs of doom, let's instead go with a nice, friendly bulleted list of how things went down in the briefing. All the quotes come from Gill Pratt.
WHEN AND WHERE
- The DRC Finals will be held at Fairplex in Pomona, Calif., on June 5th and 6th, 2015. This is about six months later than they'd originally planned, but teams will get additional funding to make up for it.
- Why push the Finals back? The DRC Trials went better than expected, and DARPA is changing the scope of the Finals to reflect that, by increasing the difficulty and complexity of the competition. "This will require quite a bit of innovation from the teams. We're raising the bar. In order to let [the teams] get there, we'll give them more time and more funding."
TEAMS AND PRIZE
- 11 teams are prequalified for the DRC Finals based on their performance in the DRC Trials.
- Google-owned Team SCHAFT, which finished first in the Trials, has elected to withdraw from the DRC Finals to "focus on the development of their first commercial product." (Early this year, DARPA had announced that SCHAFT was going to stay in the competition, but apparently something made Google change its mind about that decision.)
- Governments from the European Union, Japan, and South Korea have decided to sponsor teams from their countries to join the DRC Finals (as self-funded teams). DARPA said more details will be available in coming weeks.
- The prize is still $2 million to the winning team. Scoring will emphasize first task completion, and second speed. Each team will get two chances to complete the course, and DARPA will score the best run.
- Self-funded teams can still qualify for the DRC Finals. They'll have to send in a video of a robot performing basic tasks, among other things. DARPA is expecting about 24 teams, but they're prepared to accommodate more.
- The specific DRC Final tasks have not yet been decided on; DARPA is getting advice from teams and partners on what they should be.
- The tasks will definitely not be discrete like in the Trials: they'll be strung together into a sequential disaster scenario. Three or four robots will be running through scenarios at once. The tasks will likely be arranged from easiest to most difficult, with the exception of the driving task: it'll come first, but teams will have the option to instead skip the driving (and the points that come with it) and instead walk their robot to the beginning of the second task.
- One of the eight tasks will be a surprise (yay!), and teams won't know anything about it until they show up at the finals. There will likely be a different surprise task on the second day.
- DARPA wants the entire task sequence to be completed in less than an hour (!), and speed will be weighted more heavily in scoring. DARPA expects robots to move "at least four times faster than they did in the Trials, and perhaps much more than that."
ROBOTS AND INFRASTRUCTURE
- Robots will not be connected to offboard power, wired communications, or even safety harnesses (as they were on every task in the DRC Trials). They'll have to run on batteries (or other power source) for long enough to complete the entire task sequence, and if they get stuck or fall over, they'll have to be able to get themselves on their feet again.
- Communications will be degraded even more than before to better mimic real disasters, including fully intermittent communications. "We're going to make the comm link a whole lot worse. For a substantial fraction of the time available to the teams, the comm link will be blacked out. It will be turned off. It will occasionally bounce back, and then the latency will be long enough that you can't teleop during that time."
- The DRC Finals will involve substantially more autonomy than the DRC Trials did. DARPA is going to make sure of this through aggressive time limits and poor communications, which will make it necessary to allow the robots to operate to some extent autonomously: any team that waits for good communications to execute tasks based on low-level planning will rapidly run out of time. Higher level commands will be necessary.
- DARPA thinks that cloud robotics is going to be big, and they're making it a significant component of the DRC. The context of cloud robotics is going to be much different in the DRC than it is in most commercial applications, and DARPA sees that as a benefit, since it'll encourage things like figuring out how to make cloud robotics work in environments where connectivity is intermittent.
- Although teams will have degraded communication with their robots, their communication with the Internet will not have any limitations, and they'll be able to use as much computing power and as many other experts as they want, as a sort of a crowdsourced, cloud-based asset.
- Most of the teams will probably field hardware that is mostly the same as what they had in the DRC Trials. During the finals, the focus is going to be on better software, not hardware.
- ATLAS is going to get some upgrades: the biggest one is an onboard battery system to allow the robot to operate untethered, which is somehow going to be weight neutral.
- The robot will also get new arms that are stronger with a better range of motion to help it get up on its own if (when) it falls over, and more robustness to help keep it from accidentally smashing itself it to bits.
The big news here (besides all the other big news) is that SCHAFT dropped out. It's certainly a disappointment, but not a total surprise, although we were betting that they'd be in it until the end. From the sound of things, it wasn't an easy decision for them to make, but Gill Pratt isn't sure that they'd have repeated the blow-out performance that they had during the DRC Trials anyway:
"I think the Finals are going to be both more difficult and also about things that are somewhat different than what the SCHAFT team excelled at. In my estimation, the SCHAFT team had extremely good hardware, they had very high-torque joints which allowed their robot to be much more stable than others. But, the way that the Trials were set up, it was still possible for human beings to control the robots at a relatively low level. In the Finals, the control will be at a higher level. It'll be at a task level: 'open the door.' 'Go up the stairs.' 'Move the debris out of the way.' That's what we expect the level of the commands to be, and we expect the robots to have to have a significantly higher capability in autonomy, in software, for perception and for planning, and for execution control in order to accomplish that without a human being helping. And we actually believe that amongst the competitors that we have right now, we have some extremely strong groups who can do that. So we think that we are in fact going to get some of the best technology in this field. We're moving away from having the competition primarily about whether the hardware is good enough, to now having a competition of, is the supervised autonomy good enough, and is the human-robot interface good enough.
We are actually very happy that Team SCHAFT is going to commercialize the technology that they developed in the DRC and that we helped with, with the investment from DARPA, that we're actually very proud of. I do not know how they came to decide not to continue on in the Finals, but I think it was a difficult decision: they told me they wanted to do both. They wanted to both continue in the contest, and they wanted to also focus on the development of their first commercial product, but it's difficult to do two things at once, and so ultimately they had to decide to focus on the development of their first product. That's what they told me, and we didn't try to get them to change their mind. We're very happy for them."
SCHAFT or no SCHAFT, we're looking forward to a lot more excitement in 2015, and we'll be right there in Pomona to bring you all of it.
Evan Ackerman is a senior editor at IEEE Spectrum. Since 2007, he has written over 6,000 articles on robotics and technology. He has a degree in Martian geology and is excellent at playing bagpipes.