DARPA Urban Challenge 2007Saturday, 3rd November 2007, the former George Air Force Base in California, US.
Eleven teams gathered for the DARPA 2007 Urban Challenge, the third of the DARPA (U.S. Defense Advanced Research Projects Agency) artificial intelligence challenges, intended to boost development of driverless, self-navigating vehicles.
The previous two challenges were grand challenges: the effort was on getting a vehicle to drive itself along a hundred and sixty miles of deserted desert road.
In the first grand challenge, in 2004, the best robot managed 11 miles.
This third challenge raised the stakes much higher: driving on simulated urban streets, interacting with fifty human-controlled vehicles, with the winning car being capable of qualifiying for a California driver's license.
The closed grounds of the base, opened for spectators at 6am local time, and the cars cleared preparation by 7:30, in place, for a start, a few minutes after 8.
Competition was divided into three 'missions', each of which had multiple sub-missions, or stages to complete. The fastest car to complete all stages was not the important marker. Instead, the safest car to complete, and the most accurate, were the points judged upon.
35 robots vied for a place in the urban challenge, and entered the pretrials, which ended on Wednesday the 31st of October 2007. Of these 35 robots, only eleven were eligable to compete.
Of these eleven robots, by the time the first 'mission' was over, five robots were out of the competition.
Once driving started for each car, an on-board klaxon sounded, followed by a siren which continued until the car was out on the roads. This was to remind everyone in the vicinity that the vehicle moving was completely self-driven. There was a human in the loop, watching from a secure control centre inside one of the buildings. Each car had one operator at a computer terminal. However, the operator's only control was an emergency stop. Beyond using this stop, they could only watch. The car itself, made all decisions.
The cars had a list of goals, and co-ordinates to be at for each goal, along with on-board GPS, and a road map of the base. Beyond that, they were on their own. All real-time decision making was down to the robot's AI.
There were surprisingly few collisions, with the MIT entry clipping the walls a few times, and one nasty side smash, caused by a human controlled pace car incorrectly estimating the distance to clear a building.
Some of the obstacles included clouds of dust raised by the robot cars, the pace cars, or the other vehicles on the roads. These clouds were a hazard as they obscured the laser sensors and viewfinders. Blinded, the vehicles could not navigate until the dust cleared, or was wiped away.
In one incident, the car 'boss' stalled out in something resembling panic, as it had a near miss with a human-controlled car. The car was moving crazily as part of a test. It swung across a T junction Boss was slowly moving out of, cutting across Boss's path, as happens every day between humans on the road. Boss was almost frozen in fear, as far as its actions showed - it sat there, half in the junction and half out, unmoving for several minutes. Apparently waiting, in case any other cars came along. Eventually, however, it crept timidly out and resumed its journey, slowly picking up speed.
Interestingly, the largest entrant, team Oshkosh's truck, amassive vehicle packed with sensor equipment, did not do anything like as well as the smaller, nimbler, car based robots, which had had more 'training' and 'experience' driving themselves before the race, but had vastly reduced sensor arrays in comparison. In fact, the truck veered off the road and drove into a shop front during mission one.
Speed wise. Stanford's Junior (pictured top), the winner of 2005's grand challenge, completed the urban challenge first, and Boss, Tartan Racing's entry, completed it second, just 30 seconds later. Times alone were not enough to win. However, it was immensely gratifying to see the robot cars complete this challenge.
Odin, of Virginia Tech, came in third, a good 10 minutes later.
The cars were vying not just for prestige, but for substantial prize money.
Ben Franklin Racing team, took fourth place with Little Ben. Fifth place went to MIT's Talos, who overtook Skynet in the last straight.
Cornell University's entry. Skynet, came in last, after a last minute reboot. They crossed at 4:37 pm local time, 83 full minutes before the final time limit.
Speed of completing the course was not the winning factor. All the data generated was examined overnight, looking for point deductions, safety violations, errors of machine judgement, and violations of California state driving law.
The final results of the 2007 DARPA urban challenge are:
1st Place, and $2,000,000 - Tartan Racing, Pittsburgh, with Boss
2nd Place, and $1,000,000 - Stanford Racing Team, Stanford, with Junior
3rd Place, and $500,000 - Victor Tango, Blacksburg, with Odin