A month after the Grand Challenge robot race, members of the Red Team are still hard pressed to explain why their lead racer, H1ghlander, took almost an hour longer than planned to complete the 132-mile course.

	Defense Advanced Research Projects Agency photo
	H1ghlander, (front) one of two Carnegie Mellon University entries in the Grand Challenge Race in Nevada, drives itself down Beer Bottle Pass, the most difficult part of the 132-mile course.

"Engine trouble" in the modified Hummer H1 is about as specific as they get, at least for now, in explaining why H1ghlander finished in third place, behind the winner, Stanford University's Stanley, and its own stablemate, Sandstorm.

"The bottom line is, we just can't pin it down," said team leader William "Red" Whittaker at a team meeting last week. "We pushed on the gas pedal and it didn't go."

It was strategy -- the decision to program H1ghlander to run at an aggressive pace and Sandstorm at a conservative pace -- that ultimately kept both Red Team entries out of the winner's circle. Both H1ghlander and Sandstorm, a modified Humvee, were capable of beating Stanley's elapsed time of six hours and 54 minutes, Dr. Whittaker said.

The Red Team had programmed H1ghlander to complete the course in 6 hours, 19 minutes. Based on speeds achieved during the qualifying events prior to the race, the Red Team had calculated that no other team was likely to finish the course in much less than seven hours.

But 17 miles after leaving the starting gate, H1ghlander began to fall off its pace. At seven spots, it came to unexplained stops, twice rolling backward on hills. On-board audio recordings indicate the engine repeatedly stalled and, at the finish line, its idle speed was oscillating wildly.

The engine problems have not recurred, however, and commitments to showing the vehicles at shows and events with team sponsors have prevented the team from tearing H1ghlander apart. The vehicle was heavily damaged less than three weeks before the race when it rolled over during testing, but no one knows yet if that contributed to the race day troubles.

Other problems plagued the machine. H1ghlander's gimbal -- an arm that points the vehicle's primary sensor, a laser radar, and isolates it from road shock -- stopped responding at the 57-mile mark, apparently because of fluctuating power generated by the erratic engine.

H1ghlander had an elapsed time of 7 hours, 14 minutes -- 55 minutes slower than programmed and 20 minutes slower than the winner, Stanley.

The Red Team, with the advantage of having two vehicles in the race, decided to program Sandstorm to run at a slow, sure, can't-miss pace, to protect against some miscalculation. It was programmed to finish in 7 hours and 1 minute; it's actual elapsed time was 7 hours, 5 minutes.

Some team members had argued to give both machines their head; others saw merit in the dual-pace strategy.

"At some point," Dr. Whittaker said, "I had the last word."

Stanford didn't sweat strategy. "We just tried to drive at the speed limit," explained team leader Sebastian Thrun, referring to speed limits set for each course segment by the sponsor, the Defense Advanced Research Projects Agency. "We never discussed how fast Carnegie Mellon would go."

But Stanford had only one vehicle. A team with two vehicles might reasonablly choose to run at two different paces, Dr. Thrun said. "It made sense to be careful," he added, noting that all the teams feared a miscalculation might run them off a cliff.

Dr. Whittaker professes no disappointment with the outcome and points with pride not only to two successful finishes, but also to the fact H1ghlander was able to finish at all.

"On its worst damn day," he said, "the thing is really tough."