The equivalent of 10 high-powered desktop computers run all the scanners and lasers on board the Massachusetts Institute of Technology SUV that took part in the DARPA (Defense Advanced Research Projects Agency) Grand Challenge.

The Providence Journal / Steve Szydlowski

CAMBRIDGE, Mass. It is one thing to program a vehicle to drive itself along a road while navigating around stationary objects. It is quite another to program it to navigate around moving traffic, whether merging onto a highway or deciding when to proceed at a four-way intersection.

Such intuitive driving involves inferring intention, or what another vehicle intends to do based on position and velocity, which is very difficult to program, according to Jonathan How, Professor at the Massachusetts Institute of Technology’s Department of Aeronautics and Astronautics and Planning and Control Team Leader of MIT’s entry.

But that was the challenge posed by the Pentagon’s Defense Advanced Research Projects Agency, which just completed the finals of its DARPA Urban Challenge at the decommissioned George Air Force Base, Victorville, Calif., in the Mojave Desert.

Carnegie-Mellon University’s Chevrolet Tahoe “Boss” won the $2 million top prize, followed by Stanford’s VW Passat “Junior” ($1 million) and Virginia Tech’s Ford Escape Hybrid “Odin” ($500,000).

Just missing the cut was MIT whose Land Rover LR3 “Talos II” came in fourth and out of the money. But How said just to finish was a great accomplishment.

“I’m really, really happy we got into the race — the qualifications were so stringent — and that we finished the race, although we would have loved to have won some money,” he said in a recent interview.

Indeed. Of the original 82 teams, 36 were invited to go to California to qualify. Eleven did so, but only six completed the course, with Cornell and the University of Pennsylvania rounding out the finishers.

“There was nobody on board,” How said. “It was fully autonomous, no human interaction at all.”

Spearheading Team MIT with How were John Leonard from Mechanical Engineering, Seth Teller from Electrical Engineering and Computer Science and David Barrett from Franklin W. Olin College. Sponsors included Draper Laboratory, Ford-MIT Alliance, Olin College as well as a number of MIT departments in addition to $1 million from DARPA itself.

This was the third in a series of challenges that DARPA has organized in an effort to meet the mandate of the National Defense Authorization Act for 2001, which states in part: “It shall be a goal of the Armed Forces to achieve the fielding of unmanned, remotely controlled technology such that … by 2015, one-third of the operational ground combat vehicles are unmanned.”

In short, the Pentagon wants to keep as many personnel out of harm’s way as possible and having supply trucks, as an obvious example, travel autonomously would cut down on driver casualties.

At the same time, the technology has potential peacetime applications such as helping the elderly or visually handicapped to drive and controlling the spacing between vehicles to create more efficient traffic flow on highways.

Certainly, the technology has come a long way since the first Grand Challenge was held in 2004. None of the vehicles finished the route along a simple road with static obstructions; Carnegie-Mellon’s vehicle won by managing to travel about 7.5 miles. The 2005 race was similar but this time more than 20 vehicles traveled farther than 7.5 miles.

However, the dynamic quality of this year’s Grand Challenge posed a new set of problems for the teams — in addition to the requirement that the 60-mile course be completed in six hours.

MIT’s How said the initial task was to program the vehicle with basic driving instructions. “We basically encoded the California Department of Motor Vehicles’ Driver handbook,” he said, adding that to qualify for the race there were a number of driver’s tests and “it was a little more complicated than parallel parking on a hill.”

Once the vehicle was programmed to drive safely, How said the teams were told the basic route. “We were told where the road was but (the vehicle) also had to infer the route through sensors,” he said.

Talos II is swathed in laser and radar sensors to identify objects around the car as well as inferring the route by identifying painted lines and curbs. The data, which includes a million data points a second from a rotating $75,000 laser set atop the vehicle, is processed by the equivalent of 10 high-end desk top computers set in the rear of the vehicle.

“Here’s the beast where everything happens,” he said, opening up the rear to reveal a mass of wires and computerized technology.”

The rich combination of long range, medium range and short range data is mixed with GPS and topographical mapping technologies to achieve a very detailed localized picture of what is right around the car and where the car is situated on the route.

“It’s vision processing,” he said of the challenge to develop “a consistent view of what’s out there and how you find your way around the world.”

As well as following the route, he said the vehicles had to navigate construction sites and parking lots with large open areas, as well as identifying such objects as buildings, trees, barrels, cones and other cars.

Finally, with the radar sensors determining the velocity of objects, the vehicles had to navigate the dynamic aspect of the course in the form of “a combination of human-driven cars and robot-driven cars.”

How said a blockage in the road posed a relatively simple problem given the basic task of following the best route from A to B. “You go around (the blockage) to get to the goal from the opposite direction,” he said.

“(The moving cars made it) “a much, much more complicated environment,” he said. “Inferring intent is extraordinarily difficult thing to do.”

The vehicles had to be able to be able to make decisions that involved merging with other moving cars or deciding on how to proceed at a four-way intersection.

He said human drivers coming up to four-way intersection proceed in the order at which they arrive. Human drivers are “patient and polite,” he said.

However, he added that there is a limit to patience.

“There is a fine line between polite and being confused,” he said. “At one intersection the other robot car did not move so we did not move. A human driver drove up and seeing that we were going nowhere, drove off. When a third robot came up, we drove off.”

“It was not just about the rules, but the intention of the rules,” he said. “How well do you understand the world?”

How said MIT had acquired the vehicle from Land Rover last November and after having heavy duty racks attached to carry the sensors the team of some 15-to-20 professors and students worked on integrating the technology and developing the software. By May it made its first autonomous trip.

“It sounds like a clichÉ, but we were really not in this for the money,” he said, noting that the project had gotten MIT into Unmanned Ground Vehicle (UGV) research.

He said the team was constantly upgrading this or that piece of technology, at one point replacing the tires, at another replacing the generator. He said it ran into trouble toward the end of testing at a military base in California due to the recent wildfires.

“There was no danger, but a lot of smoke and ashes got into the generator system,” he said. “It really messed up our last week of testing.”

For more information, check out:

www.darpa.mil/ grandchallenge/

www.darpa.mil/ grandchallenge/Teams/ TeamMIT.asp

pelsworth@projo.com