This week in Victorville, California, 36 teams from around the world are participating in the National Qualifying Event for the 2007 DARPA Urban Challenge, a race between autonomous (that is, driverless) cars. According to DARPA, no more than 20 cars will take part in the final 60-mile race on Saturday, November 3. To determine who goes through, the teams are currently taking part in a 5-day qualifying event, in which their vehicles have to prove themselves in three different test areas designed to assess their abilities to pull safely into moving traffic, avoid static and moving obstacles, and park in designated spots. We got a trackside view of the robot cars testing, and looked on as a number of the teams tweaked, tested, and retested for qualifying.
Junior looked less certain about making the final left-hand turn through a stream of oncoming traffic. On a number of occasions when the traffic had cleared, the car stuttered forward only to apply the brakes again, apparently sensing a phantom obstacle. Due to a series of delays at the same turning--some for nearly a minute at a time--Junior managed to complete just under 11 laps of Test area A. On a positive note, its conservatism ensured that it was one of the few cars not to incur a beep from one of the stunt drivers indicating a hazardous driving maneuver.
Stanford University's autonomous Volkswagen Passat, named Junior, negotiated Test area A without any major incidents. Having activated its turn signal and waited for the traffic (driven by 12 DARPA stunt drivers) to clear, Junior turned swiftly left to merge with traffic. DARPA regulations for the test state that the autonomous cars should stay at least one meter from all obstacles at any one time, making the first turn on Test area A even more challenging.
Team Sting, a collaboration between Georgia Tech and SAIC, suffered probably the biggest setback of the NQE to date as its modified Porsche Cayenne crashed headfirst into a concrete pillar during Saturday testing. The car suffered damage to its front sensor mount. However, the team reported that it was able to completely rebuild and recalibrate its sensors in time to resume competing on Sunday.
Boss relies upon a combination of sensors to enable it to drive autonomously, including five long-range radar modules, eight short-range lidar sensors, a mid-range lidar sensor, and two long-range lidar sensors. Information from all these systems is processed by 10 Intel Core2Duo blades that live in the car's cargo area.
One of the leading contenders for the Urban Challenge is "Boss," a Chevy Tahoe modified by the Tartan Racing team from Carnegie Mellon University. According to the team's leader, William "Red" Whittaker, Boss has undergone thousands of hours of testing in preparation for the Urban Challenge. On Sunday, Boss completed 16 circuits of NQE Test area A, more than any other team to that point.
While its size may have proven to be a hindrance at Test area A, TerraMax does get some advantages from being the biggest vehicle in the Urban Challenge. Cameras mounted at the top of its windshield enable it to see over any cars directly in front of it to get a better idea of the road ahead, and to assist it in passing maneuvers. Other sensors on the autonomous monster truck include a series of lidar (light detection and ranging) detectors and two other vision systems mounted to the front grille for assessing the road surface and detecting traffic at intersections.
Team Oshkosh's four-wheel drive military-supply vehicle, TerraMax, had less success on Test area A. At eight feet wide, it was unable to make the first turn without remaining further than the requisite one meter from the course's lane barrier.
For both the qualifying events and the final itself, teams are given flash drives containing road network data files (RNDF) that show the layout of the various courses, plus specific mission data files (MDF), which are given to the teams just before the start of each event. For the NQE test runs, the DARPA guidelines state that the MDFs must be given to the teams "at least five minutes before the start of the test".
Like Carnegie Mellon, Virginia Tech is a veteran of the DARPA Challenge events, having competed in the previous two. Victor Tango, as its team is called, is entering this year's event with ODIN, a converted Ford Escape hybrid. In its run on the challenging Test Area A, which required cars to make a series of left-hand turns through oncoming traffic, ODIN managed to complete 11 laps, the second-highest number behind CMU's Boss.
MIT is participating in its first DARPA event, but that is not stopping it from bringing out the heavy artillery. Its Land Rover LR3, named Talos II, is bedecked with vertically and horizontally mounted lidar sensors to give the car a 3D impression of its surroundings. These are supplemented by a combination of narrow- and wide-angle cameras that analyze the shape of the road ahead and look for obstacles. Like CMU's Tartan Racing, the MIT team has qualified for government funding for its vehicle.
For some teams, getting the car ready on such short notice can be tough, as the
University of Utah found when trying to get their Chrysler Minivan to leave the chute of NQE test area B. Test area B requires vehicles to navigate a series of winding roads, to avoid disabled vehicles, and to park in a designated spot.
Alice, the California Institute of Technology's autonomous Ford E350, also had trouble at Test area B. After spending a lot of time trying to exit the start chute, Alice found its way to the parking-assessment area, but again struggled to maneuver itself into the designated spot, and the team ran out of time before completing the test.