American Scientist

Early on a Saturday morning next March, an unlikely collection of machines will hop, roll, crawl and drive into the rugged country outside Los Angeles. The vehicles will be autonomous, but their human sponsors will have a compelling goal: A cash award of $1 million for the first one to reach Las Vegas within ten hours.

The scene will mark the start of the DARPA Grand Challenge, an off-road race of robot vehicles through the California desert. The contest is the brainchild of the Defense Advanced Research Projects Agency, the central research and development arm of the Department of Defense. In order to reduce casualties and increase combat effectiveness, Congress mandated in 2000 that at least one of every three future Army battle systems be unmanned. The National Research Council recommended that the Army begin by creating autonomous vehicles that can perceive their surroundings and avoid obstacles. Organizers hope this civilian competition will spur some useful innovations.

The precise route of the Grand Challenge will be revealed only two hours before the race starts on March 13, but DARPA promises "extremely rugged, challenging terrain and obstacles" and says the race will go forward in any weather. The autonomous vehicles must find a series of latitude/longitude waypoints along a 250-mile course that includes paved and graded roads, trails and open terrain, ditches, open water, rocks, underpasses and construction.

In order to encourage diverse approaches, DARPA has set no size, weight or propulsion limits on the ground vehicles, though guidelines note that "an extremely heavy vehicle with steel tracks that destroy road pavement, a vehicle that clears a path by setting everything in its way on fire, or a vehicle that digs large holes, are unacceptable." Vehicles cannot attack their competitors and must be able to avoid collisions with noncompeting motor vehicles. They need not obey traffic or driving conventions, but DARPA says it might impose its own speed limit on portions of the route for safety or environmental reasons.

Word of the contest has been spreading steadily. Col. Jose Negron, who is organizing the race for DARPA, notes that a recent competitors' conference in Los Angeles drew more than 400 registrants, ranging from representatives of major corporations and universities to individual enthusiasts in fields such as robotics, off-road racing and computers. Faced with such a challenging course, teams have discussed models ranging from a mechanical centipede to "a hamster in a volleyball" to a leaping automaton known as "the 500-mile-per-hour frog."

Bob La Quey, a former research physicist at the University of California, San Diego, says his team will likely convert a simple dune buggy and focus their energy on sensors and programming. "Our goal is to make this essentially a software project," he says. The team plans to divide the project into modular subsystems and invite open-source software developers around the world to contribute code to be tested on a "mule" near Ensenada, Mexico.

Rhett Creighton is more skeptical. An MIT graduate student who recently won the school's autonomous robotics contest, Creighton plans to assemble a team of up to 10 alumni, graduate students and undergrads.

"Refueling is out of the question for almost any approach," he says. "The only way to win that I know of is to make a robot that fits within their rules but circumvents the hardest design challenges. Right now, it seems like they want something like a jeep to average 30 miles an hour over off-road desert terrain autonomously. That won't be possible for 30 years, simply because your 'bot can't make a mistake at those speeds."

La Quey agrees that the time limit poses a challenge. "The speeds required are high enough so that one mistake and you can lose your vehicle," he says. "Acting on the data from multiple sensors and the best maps possible, one must avoid obstacles and the various vehicle-killing crashes that are all too possible. Integration of all of this and the intelligence needed to make decisions based on this data is going to be very difficult."

The defense agency is still accepting new teams. Applications and design plans are due October 13; the plans will be made available to the public, but intellectual property will remain with the designers. "We certainly hope that someone wins in 2004," says Negron, "but if not, our plan is to do it again at some future date."