November 5, 2007
Junior the robotic car drives in traffic, finishes second in national contest
Junior, the robotic car of the Stanford Racing Team, on Sunday claimed a $1 million prize for finishing second in the DARPA Urban Challenge after successfully navigating through traffic with no human help, relying instead of its own electronic eyes and brain.
The winning team, from Carnegie Mellon University, came in second place in 2005, when Stanford won a race in the desert.
It was clear on Saturday, just after Junior crossed the finish line in Southern California, that team co-leader Sebastian Thrun felt fulfilled. The artificially intelligent Volkswagen Passat had set out on a roughly 60-mile journey, carrying out missions in a simulated cityscape among the traffic of stunt drivers and robot competitors, and had helped make computing history along the way.
"It's a historic event," beamed Thrun, a professor of computer science and of electrical engineering. "It's the first time it has ever happened. The fact that three [eventually six] cars finished is fantastic. The fact that ours is one of them is even better. Do I care about the [money]? A little bit, but this is so much bigger. We are all one big community and we all win."
What Junior accomplished (along with the winning vehicle, Boss, from Carnegie Mellon University, and the other vehicles) on a former Air Force base in Victorville, Calif., was to demonstrate that robots can make decisions about routes and maneuvers, that they can follow traffic laws, and that they can perceive and understand their environment, at least as it pertains to driving.
Throughout the day, the robots drove themselves without any human intervention through a constantly trafficked maze of sinuous routes and traffic circles to complete a series of mock missions. To see the better cars merging into moving traffic, obeying right-of-way at intersections and methodically running errands like soccer moms was to see cars displaying unprecedented capabilities to reason.
What impressed Thrun most was not the robot's race performance, but its grace and its prudence.
"Junior looked competent, elegant and agile, but not too fast, not reckless," Thrun said. "That's the theme we have on our team, which is that speed is not everything. It's really about safety."
The main goal of the team's efforts was not to win prize money, Thrun said, but instead to strive for the day when intelligent cars can make automotive transportation safer, either by assisting human drivers or perhaps giving them the option of giving up the wheel entirely when they don't want to drive or shouldn't drive.
The race was sponsored by the Pentagon's Defense Advanced Research Projects Agency, which seeks to harness autonomous vehicles to keep troops out of harm's way on the dangerous roads of Iraq and future theaters. It was essentially just a moment in the ongoing quest of roboticists to make intelligent machines, but it was a gloriously intense moment for the Stanford team that readily evoked memories of the 2005 DARPA Grand Challenge, a 131-mile race through the Mojave Desert that Stanford won. This race was far more complicated because of the requirements of handling moving traffic and California traffic laws.
The racing team's co-leader, Mike Montemerlo, a senior research engineer in the Computer Science Department, reveled in the successful culmination of the team's yearlong effort to build and tune Junior. The task pooled the efforts of professors, research staff and graduate students with the support of Volkswagen of America's Electronics Research Lab in Palo Alto, Applanix, Google, Intel, MDV-Mohr Davidow Ventures, NXP Semiconductors and Red Bull.
"All of the skills came together," Montemerlo said. "He did great merging. The localization worked great, he kept himself centered in his lane. Parking was perfect."
While Volkswagen's engineers converted a stock Passat wagon into a computer-controllable robot, Montemerlo led the development of Junior's software. The complex set of algorithms allowed Junior to process the data from his many laser, radar and position sensors, analyze the environment and then control the car's throttle, steering, gears and brakes.
According to Peter Truscott, the driver that DARPA designated to shadow Junior around the course, the robot was nearly flawless.
"I think he did a wonderful job," Truscott said. "He didn't really do anything wrong. He was well behaved."
Junior's only sins were a couple of diversions in the same spot of the course onto the wrong side of the road. These flaws were brief and never resulted in a safety hazard. Ultimately, Junior drove a little less than a mile an hour slower than Boss, DARPA officials said.
Not all of the robot contestants were fast or safe. Before the first set of missions was complete, five of the 11 robots that started the competition had been eliminated. Two robots had developed fatal attractions to buildings. Others had become terminally confused. Even among a couple of the remaining six contenders, all of whom eventually finished, there were a few collisions and some other questionable maneuvers.
Montemerlo found it easy to become nonchalant about cars whizzing by without anyone behind the wheel. "At first when a robot drove by we'd all get up and clap, and just over the course of two or three hours, it got to the point where you'd turn around and a robot would go by and you'd say, 'Oh, there goes another robot,'" he said. "It's amazing how quickly you acclimate to this idea of robots driving around in a city."
If the research succeeds in perfecting the technology to the point where it is commercially viable, that may be exactly what everyone will do.
David Orenstein is the communications and public relations manager at the Stanford School of Engineering.