Gizmodo’s coverage of the DARPA Robotics Challenge continues with day two at the Homestead Speedway south of Miami. The weekend has arrived and the crowds are here with their kids, cheering on the bots.
Some of our readers have complained that watching this thing over the live feed is kind of like watching paint dry. Another reporter at the Challenge, while watching one of the bots climb a ladder in slow motion yesterday, quipped a bit more poetically that it was like watching a sculpture in motion.
It’s a fair complaint. These machines, although they are competing in timed trials at a race track, are not exactly built for speed.
But the extreme challenge these machines and their operators, builders, and programmers are up against in performing their eight seemingly simple tasks is in sharp focus. That makes this an exciting event in person, even if it’s more like watching golf than an auto race or a ball game: concentration, ingenuity, and perseverance, are all in evidence.
A robot managing to open a door elicits cheers from onlookers, while a stumble provokes groans.
This morning, team MIT put its Atlas robot in a Polaris ATV and actually drove part of the race. This is a feat that many of the teams are not even attempting. The team did it with aplomb — albeit in super slow motion.
The MIT team, sequestered out of sight of their car-driving Atlas robot, had to rely on a deliberately degraded communications link that could only feed them short bursts of data from the bot’s camera and lidar “eyes.”
When the signal came through well enough for the operators to see exactly where the robot was, they had to send a quick command back to the machine to tell it to, for example, turn the wheel 10 degrees to the right and step on the gas. After the command went through, the team next had to wait for the data to come back through the comm link to let them know where the robot had ended up.
This machine-driving-machine challenge is the hardest task among the eight that the robots have to perform, but MIT made it look almost easy. That’s all the more remarkable considering that today’s race is the first time the team has had a chance to actually drive a vehicle with their bot.
Andres Valenzuela, one of the software engineers on Team MIT watching the Atlas drive the Polaris with me and the other spectators, explained that the team had been practicing at home with their Polaris up on blocks.
On top of that, MIT’s practice Polaris was an unmodified, production vehicle, whereas the vehicle on the course has been modified with a speed inhibitor. That keeps the robots from careening out of control before the safety official standing just on the other side of the concrete barricade can mash his hand down on the red emergency stop button. But it also means that the MIT had to adjust its driving technique on the fly, during the competition.
And that’s one of the things that makes the competition exciting. The element of speed in the is race is not with the challenges themselves, but with the extremely short timeframe the teams have had to put together their bots and write their code. It’s amazing what they’ve been able to do in less than 15 months.
The seven teams that are working with Boston Dynamics Atlas robots got a head start and only had to focus on coding, since DARPA provided them with ready-made hardware. But the nine other teams on the course had to not only write software for controlling their robots, but also build the hardware.
The four so-called Track D teams have the biggest challenge of all. Not only did they have to build as well as code, but, unlike the other teams, they also had to do it with no DARPA funding.
Track D team Mojavaton, out of Grand Junction, Colorado also raced in the DARPA Grand Challenge autonomous car races in the Mojave Desert in 2004 and 2005 and made it to the trials in the final Urban Challenge car race in 2007. They’re here now with the lightest, smallest bot on the course. At only 16.8kg, it’s an insect to the NASA Johnson Space Center team, Valkyrie, which weighs 129.7kg.
The Mojavaton folks are in the race for the sheer fun of it as well as the bragging rights to be had by besting one of their million-dollar competitors. As of this morning, they had yet to score a point, but neither had Valkyrie. Just getting into the race is an accomplishment for the Mojavaton team of volunteers, who work on their bot on nights and weekends. Scoring even one point over a major-league team like Valkyrie, with its team of 50 full-time engineers, would be a major victory.
Mojavaton’s Karl Castleton and Skyler Ogden gave me a tour of their bot this morning between challenges.
Image courtesy of DARPA/Team Mojavaton
The spindly machine is made from plastic 3D-printed gears and other parts, an aluminium framework, and off-the-the-shelf electronics that include an Arduino board, motors to power the joints, a borrowed liar unit, and a webcam. The most expensive components on the machine are the motors, which run $US500 each. The entire robot cost 20,000 bucks.
Out of an original 17 teams and their bots that DARPA invited to compete, 16 are on the course. The Chinese team — the Track D Intelligent Pioneer — never did make it to the competition after hitting travel snags.
By the end of this morning, the leading team was SCHAFT — by a wide margin. The team from Tokyo had an unofficial 19 points by the end of this morning. Next-best Team MIT had 13 points. Tartan Rescue, the Carnegie Mellon University team — which blew through the debris-removal task this morning to score the maximum of four points — stood at 12 points.
The competition will finish today. The final results, including who will get to compete for the $US2 million prize purse at next year’s DARPA Robotics Challenge, should be announced soon afterwards. here.
Michael Belfiore is the author of The Department of Mad Scientists: How DARPA Is Remaking Our World, from the Internet to Artificial Limbs.