Robot Car with Duke Radar Races Across Desert
A modified, driverless Humvee using a radar system developed by Duke students finished second by 11 minutes Oct. 8 in a demanding seven-hour, 131.6-mile Nevada desert race sponsored by the Defense Department to pave the way for autonomous military vehicles for future warfare.
The 1986 robot truck called Sandstorm beat its stable mate, a 1999 Hummer named H1ghlander, by nine minutes. Both vehicles were developed by the “Red Team” put together by Carnegie Mellon University’s Robotics Institute.
First place in the race against the clock went to a Volkswagen Touareg that was turned into a robot racer named Stanley by Stanford University’s School of Engineering. It started out second, six minutes behind H1ghlander, but overtook the frontrunner 102 miles into the race. Stanford won the $2 million prize put up by the Defense Advanced Research Projects Agency (DARPA).
Official times announced Oct. 9 had Stanford completing the course in 6 hours 53 minutes 58 seconds; Sandstorm finishing in 7 hours 4 minutes 50 seconds; and H1ghlander finishing third in 7 hours 14 minutes.
The other two vehicles to finish were Gray Team’s GrayBot, a Ford Escape sponsored by the Gray Insurance Co., Metairie, La., and a converted military truck caller TerraMax, built by Oshkosh Truck Corp. GrayBot finished in 7 hours 30 minutes. TerraMax placed fifth, completing the course Oct. 9, after a night-time layover.
In all, 23 vehicles started the race across the dusty, winding course in the Mojave Desert near Primm, Nev. Only five finished, but that was far better than the initial DARPA race, called the Grand Challenge, last year when no vehicles were able to finish.
“These vehicles haven’t just achieved world records, they’ve made history,” said DARPA director Tony Tether, as the five finishers lined up at the awards ceremony.
Shortly after the first three robots completed the course, Tether compared the feat to the first successful airplane flight a century ago. “When the Wright Brothers flew their little plane, they proved it could be done,” he said. “And just as aviation ‘took off’ after those achievements, so will the very exciting and promising robotics technologies displayed here today.”
All five finishers successfully negotiated the toughest portion of the course, a winding mountain road called Beer Mountain Pass. At one point, the narrow dirt road had a 200-foot drop on one side, and a rocky mountain wall on the other.
Carnegie Mellon’s Sandstorm used Global Position Systems and inertial measurements to determine position and looked ahead with laser sensors, cameras and the radar. The robot used computers to calculate the course, set pace and react to contingencies. H1ghlander was set up similarly, except it did not use radar.
William “Red” Whittaker, robotics professor at Carnegie Mellon and leader of the robot vehicles project, said the radar supplied by Duke worked very well on Sandstorm, helping guide it past obstacles including tank traps deliberately placed along the course.
The students at Duke’s Pratt School of Engineering, members of the Duke Robotics Club, turned a commercial radar system into short and long-range sensors integrated into the “brains” of the robot vehicle. The students were assisted by engineers from the Boeing Co., and funded by Science Applications International Corp. (SAIC).
Two members of the Red Team graduated from Duke in May. They are Josh Johnston and Jason Ziglar, who both worked during the past year on the radar system while at Pratt. Both received medals along with the other members of the Red Team and the teams of the other finishers.
"The real advantage of our radar system is what is being called 'horizon sensing’,” Johnston said before the race. "We are able to look out as far as 60 meters -- almost double the distance of any other sensor system on the vehicles. Without radar, the vehicles can't detect obstacles soon enough to avoid them at high speed. With radar, we've safely navigated at 35 miles per hour."
The radar can detect obstacles as small as 1 foot tall and doesn't get confused by multiple objects, developers said. Additionally, radar is well suited for the desert race because it penetrates smoke and dust, which also are common to the battlefield environment.
Whittaker said the third place finish by H1ghlander was a surprise because it had been programmed to run 40 minutes faster. He said the day after the run that it had not yet been determined what crippled it.
“It’s navigation was hot-on, but it didn’t have the steam,” the Carnegie Mellon team said in its web site race log.
Sandstorm, other the other hand, had been programmed to run a slower, more conservative race, and it did, Whittaker said.
The Carnegie Mellon group is an alliance of individuals, non-profit institutions and corporations. In addition to Boeing, Duke and SAIC, sponsors included Caterpillar, Intel, Am General, TTTech, Applanix, HD Systems, KVH, Snap On, Chip Ganassi Racing, Google, CM Labs, HMH Magazine and Wired Magazine.
To get to the Grand Challenge, the contestants had to pass qualifying runs, which were held a few days earlier at the California Speedway in Fontana, Calif. Forty-three vehicles were selected by DARPA to compete in the qualifying runs and the 23 finalists were selected. H1ghlander was judged the qualifying winner, Stanley was second and Sandstorm third. That determined their starting times in the Grand Challenge.