In the computer world, "debugging" means finding out why a spreadsheet program spits out a wrong number or why a graphics simulation suddenly freezes up.

But out in Nevada, where Carnegie Mellon University's Red Team has been testing its Sandstorm robotic racer this month, debugging means figuring out why the automated Humvee suddenly runs itself into a ditch or pulverizes a fence post, as it did last week.

With only seven days remaining before qualification runs for the Defense Advanced Research Project Agency's $1 million Grand Challenge race, the team is hoping to see fewer flattened fence posts this week and more miles on the odometer.

"At some point, we have to go for miles," said Phillip Koon, the field team leader. "The goal is to get it up to more than 250 miles [in a single trip] ---- preferably more than once."

In the longest run to date, Sandstorm has driven itself 119 miles. That occurred at the former LTV site in Hazelwood. Since arriving on Feb. 3 at the Nevada Automotive Test Center ---- a sprawling desert proving ground outside of Carson City with some 3,000 miles of trails and obstacle courses ---- it has driven no more than 30 miles in a single trip.

Yet, during the March 13 race, Sandstorm and the other robotic competitors will have to drive perhaps 200 miles without human intervention over an as-yet undisclosed course across the Mojave Desert. And the autonomous vehicles will need to travel from Barstow, Calif., to the vicinity of Las Vegas in less than 10 hours.

Autonomous driving at such speeds over a variety of road and off-road environments is unprecedented.

This weekend, the Red Team, headed by famed CMU roboticist William "Red" Whittaker, is making the final decisions about the robot's race configuration. By today, the team planned to have locked-in its race software so that it can concentrate on final testing and preparations.

All 25 teams that DARPA has invited to its qualification, inspection and demonstration event are supposed to report at the California Motor Speedway in Fontana next Sunday night. Late last week, rumors were flying that several teams won't make it that far, though none had officially withdrawn from the field.

"I'd be surprised if more than 20 showed up," said Seth Cabe, leader of Team LogHIC, a small group of enthusiasts in upstate New York that has built its own vehicle from the ground up. His team got its vehicle out on a frozen lake at Walden, N.Y., for its first test only last weekend, but plans to be ready for the qualification trial.

"The race itself is a whole other thing," Cabe added, acknowledging he didn't expect to finish. "All along, most people felt that nobody would win it this year."

"I will be a little disappointed if somebody wins," said Richard Mason, who heads the Golem Group, a California team that is automating a Ford F-150 pickup. Like Cabe, he sees this year's race as a tuneup, and a chance for smaller teams such as his own to gain attention and attract sponsors for next year.

If CMU, the favored team, should win, "That's not a cinematic ending," Mason said. "It would be like Rocky losing to Apollo Creed."

Even Whittaker, Cabe noted, has given his Red Team only about a 40 percent chance of completing the challenge race this year.

As of Friday, Whittaker was still sticking by that estimate.

"I'd say it is an amazingly robust machine," Whittaker said, which thus far has held up under the pounding it takes because of its imperfect, automated driving technique. Systems for isolating its delicate electronics from the rough-and-tumble of desert roads are working better than expected and the road data from its array of laser scanners, radar and stereo vision systems often is remarkable.

More than once, Koon said, drivers of a chase vehicle following in Sandstorm's wake have been blinded by the vehicle's dust cloud and had the unnerving realization that the only way they could navigate was by following the lead of the driverless robot.

Sandstorm's sophisticated perception system, which allows it to see through dust and to navigate both roads and boulder fields, is both its strength and its weakness, Koon said.

It might well be possible to negotiate much of the race course using only the Global Positioning Satellite coordinates that DARPA will use to outline the course, he said.

But to win the race ---- to complete the course in 10 hours and go faster than competitors ---- will require a perception system that works and is reliable.

"We believe perception is a race winner; you've got to have it," Koon said. "I keep thinking we're close to the point of saying, 'Wow, it works.' " Thus far, it hasn't demonstrated reliability over long stretches. "But we're getting close."

One key, innovative component developed by Bryon Smith, Yu Kato and other team members is a gimbal-mounted laser scanner that enables the vehicle to look in the direction it is turning, much as a human driver can.

The gimbal also can compensate for the jittering of the vehicle, providing a rock-steady view.

"It works fine for a while," Koon said. "Then it stops."

Whittaker said the laser scanner would be the primary sensor used for navigating, and that the ability to point it will be part of the race system.

But team members were pondering whether they could do without the stabilization feature, thus simplifying its operation.

Whittaker said the shock isolation of the electronic box on which the gimbal is mounted has worked so well that the additional stabilization offered by the gimbal might not be needed.

As of Thursday, the Red Team had spent $382,000 on its entry. When Whittaker factors in the value of equipment and services provided by sponsors such as Boeing, Intel and SAIC, and the labor provided by students and volunteers, he estimates that the equivalent of $2.7 million has been invested thus far.

"I believe we can get there and we can do it," said Koon of the Red Team's chances of winning the race. "We just need to stick with it."