Stanford University engineers are attributing last month's $2-million Grand Challenge victory to a form of software-based "experience" that enabled their driverless vehicle to stay on the road and avoid obstacles.
The school's vehicle, known as Stanley, finished the 132-mile race through the Mohave Desert in six hours and 53 minutes without any human intervention, thus enabling Stanford to claim the $2 million prize set up by the U.S. Defense Department's Defense Advanced Research Projects Agency (DARPA).
Stanford engineers say that the difference-maker for Stanley was its use of a "machine learning" process, in which the vehicle itself pored over mounds of previous driving data to look for patterns that would give it a better understanding of obstacles.
"Using machine learning, we reduced Stanley's error rate from about 12 percent in the beginning to about 1 in 50,000 by race time," says David Stavens, a Stanford Ph.D. student who worked on Stanley. "We gave Stanley a framework to approach the problems he had to tackle."
To accomplish that, Stanford engineers say that they wrote special machine learning code, which represented a substantial portion of the 100,000-plus lines of code incorporated in the vehicle. The machine learning code provided Stanley with the ability to examine old data, particularly sensor data, which served as a basis for its decisions during the October 8th race.
"It's all about the big picture," Stavens says. "Using machine learning, Stanley was able to look at the big picture more effectively than a human could have."
Stanley processed the data on seven Pentium M-based computer blades, which were housed in a rugged rack-mount chassis made by Intel Corp. The ruggedized chassis, originally designed for telecommunication applications, enabled the blades to stand up to the shock and vibration of the off-road desert race.
Still, Stanford engineers say that the software was the key to the victory. "The real challenge was to get the intelligence right, and the intelligence was in the software," Stavens says.
Although DARPA has declared the Grand Challenge finished, Stanford says that it will continue to research driverless vehicles.
"The project is just beginning," Stavens says. "We feel that, ultimately, you'll be able to go to the dealer and buy a car that drives itself."
Using machine learing, "Stanley" figured out how to stay on the road and avoid obstacles.