MIT finishes fourth in DARPA challenge for robotic vehicles

MIT's self-driving Land Rover takes to the road. <a onclick="MM_openBrWindow('urban-truck-enlarged.html','','width=509, height=583')">
<span onmouseover="this.className='cursorChange';">Open image gallery</span>
<noscript> <a href="urban-truck-enlarged.html">
<em>(no JavaScript)</em>

There was one fender-bender, and a few cars had to be physically removed from competition when they lost control. But in the end this year's robotic vehicle race in California was a great success, with six driverless cars making it through a daunting drive in the equivalent of downtown traffic.

MIT's automated Land Rover, packed with computers and electronic sensors, finished fourth in the DARPA Grand Challenge, a road race for robotic vehicles. The MIT vehicle, competing for the first time, was one of only six to complete the challenging 55-mile course, out of 89 original applicants and 35 that were selected to compete in the finals.

The first-, second- and third-place cars, from Carnegie-Mellon University, Stanford, and Virginia Tech, respectively, were built by teams that had finished in the top ten in the 2005 race.

"My heartfelt congratulations to the top three teams," said MIT team leader John Leonard, professor of mechanical engineering. "They were all repeat performers, and there are some things you can only learn by doing it. But I'd like to think we brought in some fresh approaches."

There were two overall challenges the vehicles had to contend with, Leonard said: Navigation, which emphasizes locating the roads, lanes and intersections, and traffic avoidance, to deal with other moving vehicles. For safety reasons, Leonard said, DARPA chose to emphasize the traffic issues, whereas the MIT vehicle's greatest strength lies in its navigational abilities.

MIT's Land Rover, caled Talos, carried "about twice as many" computers and sensors as the other vehicles, Leonard said, including 10 quad-core computers, along with 15 radar systems, six cameras and a dozen lasers.

Only 11 vehicles survived several qualifying rounds to make it to the final contest, which involved driving through a traffic-filled urban environment complete with moving traffic, stop signs and traffic lights. The racecourse was in an abandoned military base, which had been set up to resemble a busy downtown area.

Except for emergency cutoff switches held by the race judges, all control of the vehicles was completely in the hands of their onboard computers once the race began. Despite the challenges of driving through traffic with totally autonomous systems, there were only two collisions in the race. In the first, a German-built car sustained serious damage when it crashed into MIT's entry, and it was eliminated from the race.

In the second, which Leonard called "almost humorous," the MIT car barely made contact with one from Cornell. The other car had stopped after getting confused in a traffic circle. The MIT car tried to pass it, but just then the other car started up again, and the two "nudged up against" each other and "their noses just touched," said one of the MIT team's co-leaders, professor Jonathan How of aeronautics and astronautics. (The other co-leaders were Seth Teller, a professor in electrical engineering and computer science, and Professor David Barrett [MIT Ph.D. 1996] of Olin College.) The encounter, which caused no damage, was deemed a "no-fault accident" by the race organizers.

DARPA (the U.S. Defense Advanced Research Projects Agency) has organized the robotic vehicle races as a way of spurring the development of technology that could provide an important new capability in future battles, just as automated pilotless aircraft have already begun to play an important role in warfare.

But some of the innovations could also find their way into ordinary passenger cars, which is why all the major car companies have been involved in sponsoring teams. "They see that this technology, if it can be channeled toward improving safety, is going to have a huge impact," How said. "It could revolutionize the way people drive."

A version of this article appeared in MIT Tech Talk on November 7, 2007 (download PDF).

Topics: Aeronautical and astronautical engineering, Computer science and technology, Innovation and Entrepreneurship (I&E), Mechanical engineering, Artificial intelligence, Security studies and military, Contests and academic competitions

Back to the top