An MIT engineer made a pit stop on campus yesterday to divulge the details of his team’s performance in the “Urban Challenge” – a government contest in which engineers and scientists developed self-driving vehicles capable of maneuvering through city traffic.
Since 2004, the Defense Advanced Research Projects Agency has sponsored two competitions to foster innovation and create a “competent autonomous automobile”. This most recent competition, the 2007 DARPA Urban Challenge, instead sought to produce an unmanned vehicle capable of safely navigating urban traffic, crossing intersections, and coping with surprise situations.
Seth Teller, a professor of computer science and electrical engineering at MIT, explained the planning and development process of his team’s self-driving car to an audience of about 40 people in the Engineering Sciences building yesterday.
According to Teller, his team’s autonomous vehicle, Talos – a modified Land Rover LR3 with self-navigation capabilities – won fourth place in the competition. Teller said Talos was only one of six vehicles that actually completed the 60-mile race, although there were 89 contestants.
Talos, Teller said, was designed to contain a large quantity of processing power in order to prevent setbacks.
“We really wanted a lot of resources…. [We] threw in a lot of sensors and CPU power in order to not get stuck during the contest,” Teller said.
To ensure accurate data was received by internal processors, Teller said Talos utilized a multitude of sensors throughout the race – as many as two to three sensors at the same time – ranging from regular visual cameras to 360º sensors. The various sensors allowed the car to produce a 3D dataset in order to perceive lane markers, road intersections and various obstacles in real time.
According to Teller, Talos got into two collisions during the competition. A competing vehicle from Cornell was partially responsible for one of the collisions.
“We tried to pass, but [Talos] turned back into the lane too quickly,” Teller said. “Cornell [University] hits the gas and hits us.”
Teller said that during the competition, Talos decided to bypass a roadblock by driving up on the curb, which, while effective and commonly used by human drivers, did not sit too well with the DARPA judges.
“We got docked some points for that,” Teller said.
The development of autonomous vehicles such as Talos, Teller said, has the potential to vastly improve the current state of road transportation.
“[With human drivers,] a lot of productivity is lost just keeping themselves alive from home to work,” Teller said. “If, in the long run, cars can drive themselves and if cars can synchronize, [we] can get rid of intersections.”
However, the age of robotic taxi drivers is still quite far on the horizon. Teller and his team are continuing to study what caused some of the failures during the competition, such as the hallucination of curbs and the confusion caused by shadows on the road.
Teller said he intends to supplement his efforts in the automobile industry with research in other fields of robotic transportation, such as creating a voice-activated motorized wheelchair and an autonomous forklift.