Unmanned vehicles compete in Pentagon's desert challenge

LOS ANGELES — Wanted by the Pentagon: a muscular, outdoorsy specimen. Must be intelligent and, above all, self-driven.

When 20 robotic vehicles face off next month in a rugged race across the Nevada desert, the winning machine (if any crosses the finish line) will blend the latest technological bling and the most smarts.

The military sponsors the race to speed the development of unmanned vehicles for combat. The project had an inauspicious start: Last year's inaugural contest ended soon after it began when the robots careened off course or abruptly stalled. One even got tangled in barbed wire.

Fast-forward 18 months, and double the prize to $2 million. Newcomers have joined a handful of last year's teams to form a motley mix of garage tinkerers, academics and corporations.

Teams have beefed up their vehicles' artificial intelligence through improved computer algorithms that will help them avoid such pitfalls as ditches and boulders strewn across the roughly 150-mile-long course. To get there, the robots must compete in a semifinal showdown that starts Wednesday.

Entrants include several converted SUVs, souped-up passenger sedans, a modified all-terrain vehicle, a behemoth military truck and even a motorcycle.

This year's race shows signs of being extremely competitive. Some vehicles have logged hundreds of self-guided miles in the Southwestern desert during summer practice runs. Several even tested on last year's course, which spanned the Mojave Desert between Barstow, Calif., and Primm, Nev.

The ramped-up preparation reflects the higher stakes for the so-called Grand Challenge. While sweetening the purse, organizers promise that the course will be tougher.

Vehicles will have to drive on dirt and gravel, maneuver mountain switchbacks, squeeze through choke points and avoid man-made and natural obstacles.

The sponsor of the Grand Challenge is the research arm of the Pentagon known as the Defense Advanced Research Projects Agency (DARPA).

On Wednesday, 40 teams and three alternates will compete for a spot in the Oct. 8 race during the semifinals at the California Speedway in Fontana.

The vehicles must negotiate a two-mile stretch of the track using on-board computers, global positioning satellites, and various lasers and radar. The top 20 performers will advance to the final starting line.

The course for the finals will not be revealed until two hours before the start time, when DARPA will give contestants a CD with GPS coordinates that chart the route.

Last year's semifinals were disappointing. Only seven entrants completed a flat, 1.4-mile obstacle course. Even so, organizers let 15 vehicles compete in the finals.

One of the favorites again this year is the Red Team from Carnegie Mellon University in Pittsburgh, led by robotics professor William "Red" Whittaker.

During last year's finals, Carnegie Mellon's converted Humvee, nicknamed Sandstorm, traveled the farthest — all of 7 ½ miles — before breaking down. This year, the university entered two robots — an improved Sandstorm and a converted Hummer named H1ghlander.

Among the newcomers: the Stanford Racing Team, whose modified Volkswagen Touareg, Stanley, recently drove 200 miles in the Arizona desert without interruption or human help.

Team leader Sebastian Thrun, a computer-science professor at Stanford University, declined to reveal how long the journey took. "That's our best-kept secret," he said.

The autonomous robotic vehicles use "drive-by-wire" technology, in which on-board computers control steering, braking and other movement. As a result, many of the mechanical links to the engine are absent.

The vehicles have sensors that pinpoint their location and see whether obstacles lie ahead. The sensors feed data to computers that, with the help of a 3-D camera, let vehicles distinguish a boulder from tumbleweed and calculate whether a chasm is too deep to cross.

Participants generally agree that the sturdier the vehicle, the better it can handle curves and maneuver rocky terrain. But the secret weapon, many say, is each robot's computer brain. It must have all the right algorithms and programming to gather information, plot its path and change course to avoid danger.

Anthony Tether, DARPA director, said this year's selection process was rigorous. Nearly 200 participants from 37 states and three foreign countries applied, double that of last year.

If no one wins this year, the stakes will be raised again, to $4 million.