Student-programmed Autonomous Car Advances to Next Stage of “Urban Challenge” Race

May 21, 2007

AUSTIN, Texas—Marvin, an "autonomous" Isuzu SUV programmed with the help of computer science undergraduates from The University of Texas at Austin, is one of 53 vehicles selected to advance to the next stage of Defense Advanced Research Projects Agency's (DARPA) 2007 "Urban Challenge" race, the agency announced on May 11.

video icon VIDEO: Watch a video demonstration of Marvin, the “autonomous” Isuzu SUV, submitted to DARPA by Austin Robot Technology and The University of Texas at Austin for the “Urban Challenge” race:

Marvin is a collaboration between Austin Robot Technology, an independent Austin-based group of programmers and engineers, and the students of computer sciences professor Peter Stone's CS378 class, "Autonomous Vehicles: Driving in Traffic."

Stone's students are writing much of the artificial intelligence software that the vehicle uses to learn to drive itself.

"We are involved in making history by participating in the race that is going to be heralded as the big-bang of serious field robotics," says Tarun Nimmagadda, a computer sciences major in Stone's class, which is part of the College of Natural Sciences' Freshman Research Initiative.

The vehicle must prove its mettle at a site visit by DARPA representatives this summer by successfully navigating a test course that includes a four-way intersection and moving traffic.

If chosen, it will advance with 29 other teams' vehicles to the National Qualification Event in October. From there, it would advance to the Urban Challenge itself, which will be held on November 3.

This year's race is the third such race that DARPA has sponsored, and it is by far the most difficult. In the 2005 Grand Challenge, which only five out of eighteen finalists even completed, vehicles had to traverse a 150-mile dirt road through the desert, avoiding stationary obstacles in the process.

In 2007, by contrast, vehicles will confront a vastly more complex Urban Challenge that demands not just that the vehicles drive within a typically complex urban landscape—replete with potholes, intersections and improperly parked cars—but that they do so in a dynamic environment, obeying traffic laws in the process.

"This year's challenge is more interesting from an artificial intelligence perspective," says Stone. "You have to deal with a lot more uncertainties and unknowns. There will be other vehicles, possible route re-plannings, and much sparser waypoints. You get a map of the roads and where you need to get to, but it's more about the car planning for itself where it's going to go."

In recognition of the success of the previous two races, DARPA has upped the ante for 2007, promising $2 million, $1 million and $500,000 awards to the top three teams that finish the course in under six hours.

The objective of challenges like this one, says Stone, is to push the field forward toward the day—which he sees coming in the next few decades—when fully autonomous vehicles are safe to deploy on the public roads. The point of the course, however, is simply to expose his students to the excitement and experience of research.

"These could be the students who end up making a mark on the world in the future," he says, "and I want this to be the kind of course that can be the inspiration that really sets it off."

For more information contact: Lee Clippard, media relations, College of Natural Sciences, 512-232-0675.