Finding a spot, parking legally and then leaving the lot without a fender bender is a task Boss will need to perform this fall in the Urban Challenge, a 60-mile competition for autonomous vehicles sponsored by the Defense Advanced Research Projects Agency (DARPA). The Tartan Racing team made great progress on developing this skill, as well as learning how to cope with stop-and-go traffic and to pass stopped vehicles, during a testing stint at General Motors' Desert Proving Ground in Mesa, Ariz.
"Improvement of robot driving doesn't equate to the mastery needed for Urban Challenge success," said William "Red" Whittaker, team leader and the Fredkin Research Professor at Carnegie Mellon's Robotics Institute. "Sensing, computing and software capabilities are already leaps beyond those of prior robot races, but so is the Urban Challenge's degree of difficulty.
"It is not yet clear whether any machine can succeed, and that uncertainty is some of the appeal of the competition," he added. "Many automotive, academic, defense and international contenders are new to the competition and they raise the bar above anything yet known in robotics."
The team left Arizona March 26 bound for Pittsburgh, with planned stops along the way in Peoria, Ill., and Detroit to visit three of the team's sponsors - General Motors, Caterpillar and Continental AG. Boss is due back in Pittsburgh by April 6.
The Tartan Racing team, which includes Carnegie Mellon faculty, students and staff, began modifying two 2007 Chevy Tahoes for autonomous operation last fall in preparation for the Nov. 3 Urban Challenge.
Robots are expected to compete simultaneously in a rally race format on the same course. "Nothing like that has ever been attempted," Whittaker said. The urban course is likely to have several segments with intermediate pit stops, but details are not yet known. The best overall time for the day takes a $2 million prize. DARPA will announce details about the format, location and semifinalists Aug. 10.
By early this year, the Boss vehicles were capable of completing basic missions, including negotiating four-way stops, at a former industrial site in Pittsburgh. To escape winter weather, the team temporarily relocated to Arizona in late January. "Having snow on the ground, as was the case in Pittsburgh, made it difficult to develop new driving skills," said Chris Urmson, the team's director of technology.
The team is developing a long list of skills, including long-range perception, predicting the behavior of other vehicles, and seeing berms and lane markings. Parking lot skills were a major emphasis in Arizona. An inexperienced human driver might welcome the freedom of movement in an uncrowded parking lot, but that high degree of freedom is itself a challenge for autonomous vehicles.
"If you're driving down the street, the vehicle knows it has to go in the direction of the street," Urmson explained. "But in a parking lot, there's a lot more freedom and, therefore, a lot more decisions that the vehicle must make. GPS can tell Boss where to park, but figuring out how to get there so that it is properly aligned with the parking space requires a great deal of planning."
When parking, Boss must make a number of decisions, often involving forward and reverse motions, and execute some tight turns to position itself correctly. Thus far, Boss has contended with parking lots cluttered with obstacles, but will need to improve its ability to navigate around moving obstacles. "We don't want the vehicle being surprised by a shopping cart," Urmson added.
Tartan Racing sponsors include General Motors, Caterpillar, Continental AG, Intel, Google, Applanix, TeleAtlas, Vector, Ibeo, Mobileye, CarSim, CleanPower Resources, MA/COM, NetApp Vector CANtech and Hewlett Packard. DARPA has also selected Tartan Racing as one of 11 teams that will receive up to $1 million in federal funds for technology development.
For additional information, visit Tartan Racing on the Web at www.tartanracing.org
For additional information regarding the Urban Challenge, visit www.darpa.mil/grandchallenge