 "Photo: Joe Buglewicz")
A white Isuzu sport utility vehicle curves smoothly around the road. It approaches a stalled yellow sedan and manages to turn on its right blinker, swerve to the right and avoid a collision with the defunct car. There is nothing singularly remarkable about the vehicle's driving performance, except that there is no one behind the wheel.
Marvin, the driverless SUV, is one of the 53 vehicles in the nation that will advance to the next stage of the Urban Challenge race sponsored by the Defense Advanced Research Projects Agency.
The vehicle is a collaboration between Austin Robot Technology, an independent Austin-based group of programmers and engineers, and UT computer science professor Peter Stone's class "Autonomous Vehicles: Driving in Traffic."
The team is currently preparing for a site visit on July 17 from officials who will test the SUV's capabilities and determine if it is fit to advance to the National Qualification Event in October. If Marvin fares well there, the team will be able to compete in the Urban Challenge itself in November.
The visit will test the car's performance at a four-way intersection, passing parked cars and following another vehicle.
A match made in technology heaven, Austin Robot Technology supplied the vehicle and control hardware, teaming up with Stone's class in the spring for the more intricate software and artificial intelligence necessary to give the car its behavioral programming.
"Peter's class brought it to another level and took us a huge step forward," said Dave Tuttle, team leader of Austin Robot Technology.
The class is part of the College of Natural Science's Freshman Research Initiative, which funded much of the project's equipment. The program is tailored to give undergraduate students a chance to work with graduate students and faculty through intellectual application and research, said Sarah Simmons, director of the initiative.
"There is nothing like a real-life problem or inquiry-based project to stimulate learning," Simmons said.
Computer science senior Tarun Nimmagadda was inspired to get involved with the challenge after a Stanford University student, who was a member of the 2005 competition's winning team, strongly urged students to participate in what he called "a defining moment for robotics." After Stone announced he would be offering the course, Nimmagadda was even more motivated to join.
Stone heard about the Austin Robot Technology's success with Marvin at the 2005 DARPA Challenge. The vehicle qualified on a national level but did not advance to the Grand Challenge. Austin Robot Technology did not participate in the first Grand Challenge, held in 2004. There were no finishers in the 2004 race, since no team could complete the course in the allotted time.
Instead of conquering rugged desert terrain as before, Marvin will face what millions of Americans deal with every day - traffic.
Sensing and interpreting moving traffic, merging, avoiding collisions, parking and handling stop signs in an urban setting are part of the 60-mile course that the driverless vehicles must finish in less than six hours.
Tuttle said he believes this challenge will be more difficult due to the complexity of the course.
"If you think about the millions of accidents a year, you realize that driving is tough for humans," he said. "Making a vehicle drive on its own by replicating what a human being is doing is a huge technical challenge."
Nimmagadda said the class was fortunate enough to have the Austin Robot Technology's lowlevel controls and hardware in place already, but were presented with software problems that had no pre-existing solutions. With no textbook to reference, the class relied on a mixture of studying what others had done the previous year and using ample time outside the classroom to program the software.
"We had no assigned homework and only met twice a week," Nimmagadda said. "So we really had to motivate ourselves."
The SUV works like a large computer. A mission file, which tells it what to do, is downloaded into the computer system. The vehicle then gathers information about the road ahead through GPS satellites, cameras, lasers and a computer software system.
Patrick Beeson, research associate and member of the UT team, said the real challenge is integrating all of the technology, from the GPS to the robotics, together.
"It is the completion of these large-scale projects that yield further research," Beeson said.
Sponsored by the U.S. Department of Defense's central research organization, the Urban Challenge experimental cars could join military operations as transport vehicles. The department plans to make a third of its transport vehicles autonomous by 2025, Beeson said.
As for civilian life, the future of the autonomous car is a promising reality in the next 10 to 15 years with the continued advancement of technology and funding, Tuttle said.
"As things get more sophisticated and engineers and researchers learn how to make this technology more reliable, we will see it getting more cost-effective for the average consumer," he said.
A glimmer of the potential advancement can be found in modern cars such as the Lexus LS 460, which boasts automatic parking capabilities. But for fully autonomous cars, just one sensor runs a hefty price of $75,000.
The terrain is not the only difference in this year's challenge. DARPA has increased the prize money, promising $2 million to the winner, $1 million to second place and $500,000 to third. The allure is not the prize money, Tuttle said, but the opportunity to work on a project that produces such a rare blend of technology.
"It is really exciting and fun working with such talented people on the 'ultimate real-life robot toy,'" he said.
Be the first to comment on this article!