After months of planning, building, coding, and troubleshooting, Team MIT arrived with its autonomous car at the finals of the Defense Advanced Research Projects Agency Urban Challenge on Saturday, Nov. 5 as one of 11 teams out of the original field of 89. The car — pushed to its limits — finished the race alongside only five of its competitors.
MIT finished the race in fourth place behind Carnegie Mellon University, Stanford University, and Virginia Tech. Unlike MIT, the top three schools had competed in previous DARPA autonomous challenges.
Because of the difficulty of the competition, “there was a lot of uncertainty as to how we were going to do,” said Jonathan P. How PhD ’93, professor of aeronautics and astronautics and one of the team leaders.
“Our system design relied primarily on perception to find the road, instead of placing a high reliance on GPS and a priori maps,” said John J. Leonard, professor of mechanical and ocean Engineering and another team leader. This approach meant that the car needed to be more cautious in certain situations.
MIT’s car performed well over all, moving smoothly and quickly through the course, stopping at stop signs and maintaining an appropriate speed throughout. Its weakest point, How said, was probably driving downhill. “Driving down a big dirt hill was not a part of our testing strategy.”
“A bigger testing team would have helped free up the developers to work more on the software,” said team member Gaston A. Fiore G. “The biggest challenge for the car in the future is getting a testing site nearby that we can use.”
During the course of the challenge, MIT’s vehicle was involved in two accidents. In both incidents, DARPA did not fault MIT. In the first, a German team was disqualified after crashing into MIT’s entry. In the second, MIT and Carnegie Mellon bumped fenders in a “no-fault” accident.
Despite all of the perception gear on board the car, How said that it was necessary to do “resilient planning in order for the car to continue moving forward as well as it could, even in the face of uncertainty.” The resilient planning allowed the car to recognize when it needed to turn around and find another route to the goal. The fact that MIT’s car finished the race proves that the car was able to accomplish this part of the task.
When asked what the best part of the competition was, How said, “It’s quite humorous at times to watch the robots run around racing each other. It was so similar to what humans would do.”
The plans for the car and the software that it uses are not yet clear, though there are many military and civilian applications for an autonomous ground vehicle. DARPA has been holding Grand Challenges for automated automobiles in order to meet a Congressional mandate from 2001 that requires one-third of operational ground combat vehicles to be unmanned by 2015. On the civilian side, the Ford-MIT alliance has been looking at increasing car safety through automation of some aspects of driving.
The car will probably be used as a platform for future research, rather than placed in a museum. Fiore, for example, said he will base his thesis on the motion planning system of the vehicle. Other graduate students from the team will also incorporate the work into their master’s theses.
“Overall, it was an amazing experience for all on our team, and we look forward for the chance to compete again in a future challenge,” Leonard said.