DARPA Urban Challenge 2007: George Town Races
by Tim Oren
I’ll start this with a big tip of the hat to DARPA and its director, Dr. Tony Tether, who has one of the world’s best jobs. Not only do they push the bleeding edge and come up with clever ways to engage the research community in their endeavors, but they run well-managed events with a flair for showmanship that belies their status as a government and military agency. As an example of the latter, they had arranged for the Urban Challenge webcast and on-site video to be co-hosted by Jamie Hyneman and Grant Imahara of Myth Busters, the techie crowd’s favorite TV show.
They also have the guts to invite in the world press and the general public while trying something new to the world: Turning multiple autonomous vehicles loose on city streets at the same time, interspersed with human drivers. As Tether said at the start of the program, “If anyone tells you he knows what’s going to happen, he’s lying.”
Since that test could likely take every bit of a short November day, the teams, staff and press assembled for their briefings at a chilly and dark 0600 hours. The day featured robot traffic jams, the world’s first ‘bot vs. ‘bot collision, and the Terramax robot truck’s attempt to take out the old air base PX.
DID has the report – and the pictures…
The Robots of Dawn
It was a day for old rivals to face off again. The Carnegie-Mellon ‘Red Team’ had been narrowly defeated in the 2005 Grand Challenge desert race, and were back as Tartan Racing, with their bot ‘Boss’.
The victors of 2005, Stanford Racing, were also back with ‘Junior’, based on a Volkswagen Passat.
But first, perhaps to remind the assembled technology and automotive junkies that there is a serious mission underlying this effort, there was ceremony. The armed forces’ oldest form of autonomous transportation met the newest.
A fleet of dozens of human-driven Ford Tauruses paraded through the start area and populated the streets of ‘DARPA Town’, in reality an abandoned housing area in the former George Air Force Base.
And then they were off. After the problems caused yesterday by lining up all eleven bots at once, no more than three were staged in the chutes at any one time. ‘Odin’ from Virginia Tech led off the launch.
Followed by all 10 of the other bots, from the smallest, Pennsylvania’s “Little Ben” Prius…
To Oshkosh Truck’s 12-ton Terramax, the monster truck of the ‘bot world and a crowd favorite…
The day’s course for the robots was divided into three missions, each consisting of 6 or 7 subtasks. After each complete mission, the bots would return to the start area to have a new mission file loaded and its sensors cleaned off by the team’s pit crew.
The first mission quickly began to take a toll on the bots. I headed from the start area towards a press zone that overlooked a four-way stop which was going to force the bots to interact with one another. By the time I reached it, the DARPA staff’s radios were already reporting that the vehicles from Team Annieway and IVS, the two most tentative in their movements in yesterday’s practice runs, had already been removed from the course for going too slow and impeding traffic.
At the intersection, Boss soon appeared around a curve, with its chase vehicle behind.
The old housing in the background is in bad shape even for a ghost town. It’s been used for urban warfare training that may have featured live demolition charges and flash-bangs, judging from the many blown out windows and walls. The heavy equipment is waiting to take it down as soon as the DARPA show departs.
Sure enough, the four-way forced the bots to intersect on their courses. Several of them dealt with the traffic correctly, but then the UCF entry froze at one of the stop signs, with its chase car parked close behind. The Cornell bot approached the stopped vehicles from the rear, paused, probably decided they were just obstacles, and pulled out to pass. Then it froze as well.
With one road into the intersection choked, the MIT bot rounded a curve and headed for the opposing stop sign. At the same time, Little Ben rolled up from the right, and the press and DARPA staffers watched in anticipation, with no idea where the bots were supposed to go, or what was about to happen. (Little Ben in the foreground, UCF stuck in the center, Cornell barely visible over Little Ben’s roof, MIT out of view left.)
MIT’s Talos paused, successfully turned right away from the jam (and then seemed to stop for a bit to think about what had happened.) Little Ben had apparently also been tasked with a right turn, which would take it right into the jam of bots and chase cars. It began to roll, turned… (MIT turning, Little Ben beginning to roll).
…and threaded perfectly through the stopped bots and chase cars, and disappeared up the street. A cheer went up from the humans.
Then everything stopped. The race control had called an ‘all pause’, which froze all the bots and the human driven vehicles in their places. Word came down that the Terramax vehicle had appeared to go catatonic for twenty minutes while attempting a parking task, then suddenly accelerated, jumped a curb, and headed straight for the old post exchange building. It was apparently shut down before beginning demolition work, but Terramax’s confident appearence yesterday did not hold up under stress, and its humans arrived and drove it away.
Pauses struck several more times during the first mission, and after too many inappropriate movements or blockages were caused, the bots from both UCF and Team CarOLO were removed from the course as well. By then the faster bots were returning to the start for their new mission assignments. But with five down and six left, the rate of attrition might end the race early.
Cranking It Up
With the weaker bots out of the way, and with confidence growing in their own performance, several teams appeared to be turning up the maximum speed of their vehicles during their second missions. The number of problematic interactions also dropped dramatically, and there were no further disqualifications during the second mission.
Here MIT’s Talos gets a bit of body lean rounding the traffic circle.
By the start of the third mission CMU, Stanford and VT were building up a lead. Cornell and Little Ben had never turned up the speed, and MIT’s increase seemed to be getting its bot into troubles. I overheard one CMU team member on the radio saying something to the effect of ‘let them hit us for aggressiveness, just turn it up.’ The top three were now running close to the posted limit of 25 (which doesn’t seem like much until you’ve had a driverless car doing 25 zip past you ten feet away – on the other side of a K rail, fortunately). Stanford’s Junior is pursued by Virginia Tech’s Odin down the back straightaway of the traffic circle, with Tauruses chasing.
Then one of the slower bots, Cornell, pulled onto the traffic circle and stalled (due to a bug in a throttle controller, a team member told me later). MIT’s Talos approached from the rear, regarded the stopped bot and chase car, and pulled out to pass the obstacles. As it cut back into the right lane, the Cornell vehicle suddenly moved forward, and the first bot-to-bot contact of the event occurred as bumpers met. This could have been disastrous, as the Cornell team had mounted many of their sensors inside the bumper of their Tahoe. Fortunately, neither vehicle suffered critical damage and they were backed off and restarted.
The scuff on the bumper below is one half of the world’s first bot-on-bot body shop work:
While Junior had been paused to allow the accident to be cleared, the other bots continued to run, and from that point on none of the three leaders appeared to have substantial delays in completing their tasks. Watching them round the circle to start the next part of the mission started to seem almost routine, and soon the announcer said that all three were on their final laps, and the viewing stands refilled. A cheer went up as Stanford’s Junior appeared on the circle first, turned into the finish chute, and took the checkered flag, wielded by Dr. Tether. Team Tartan’s Boss was not far behind.
In a remarkable achievement for the world’s first attempt at such an event, 6 of 11 bots that started finished the entire course. Stanford’s Junior crossed the finish line first, followed by CMU’s Boss and Virginia Tech’s Odin. ‘Little Ben’, MIT, and Cornell also finished the course, but well out of the running.
That does NOT mean that Stanford is the winner, as the finish times need to be adjusted for staggered starts and stoppage time on the course. Also the total times will be further adjusted for traffic violation ‘tickets’ issued for moving and other violations on the course.
Tartan Racing / Carnegie-Mellon University team has taken first place in the 2007 DARPA Urban Challenge with its ‘Boss’ autonomous vehicle. Stanford’s ‘Junior’ placed second, a reversal of the results from the 2005 desert Grand Challenge. Virginia Tech’s Odin came third, and MIT came in fourth.
In the end, the event was in fact decided on adjusted time, as none of the leading teams committed any major violations. DARPA had a recon aircraft orbiting the site for the whole day, so the officials were able to replay incidents reported from the field to determine if they were in fact unsafe.
The length of the course averaged 55 miles, but varied by team since the routes weren’t explicitly provided, only the target waypoints. Team Tartan’s average winning speed was about 14 miles per hour, with Stanford about 13 mph.
Can’t wait for the next one.
Tim Oren has been a venture capitalist, research manager, and CEO, and remains an ongoing, unabashed tech enthusiast. He attended Urban Challenge 2007 as a DID correspondent. When not writing articles for DID, he can be found online at his blog Due Diligence.
Tim Oren’s thoughts about the wider implications and trends suggested by this event can be found in his follow-on article, “Urban Challenge 2007: A Tech Exec’s Reflections.”
- Stars and Stripes – They call him the Crusher. “The robotics project, the culmination of almost four years of work by the Defense Advanced Research Projects Agency (DARPA), is designed to show military planners what unmanned, unsupervised machines can potentially handle on the battlefield. The seven-ton Crusher is programmed to get itself from point A to point B without any instructions other than some basic GPS coordinates…”