Posted: 10/10/2005 2:56:20 PM EDT
Robot vehicles conquer US desert race
Three modified driverless vehicles have crossed the finish line and entered the history books after traversing 210 kilometres of desert terrain, guided only by laser sensors and on-board computers.
The race was designed to spur the development of driverless vehicles that one day could carry water, fuel and other supplies for the US military in war zones.
No winner has been declared yet for the more than $2.5 million in prize money.
The Defence Advanced Research Projects Agency (DARPA), which organised the race, says it is waiting for final race times from the three driverless vehicles that finished and two others still on the course.
"We have a winner, we just don't know who it is," DARPA director Tony Tether said.
Last year, in the inaugural race sponsored by DARPA, called the Grand Challenge, every machine failed within sight of the starting line. The Pentagon decided to double the prize money and hold the event again this year.
Twenty-three modified Humvees, SUVs, and dune buggies were sent into the mountains and valleys in the Nevada desert to navigate man-made obstacles, tunnels and a dry lake bed just after sunrise on Saturday.
One broke down at the starting line.
A blue Volkswagen, 'Stanley', built by a Stanford University team, overtook an automated Humvee, 'H1ghlander', built by Carnegie University students at 164 kilometres and arrived at the finish line first.
Shortly afterward, 'Highlander' and another Humvee, 'Sandstorm', also built by the Carnegie Mellon team, finished the race.
The rugged, twisting Mojave desert course was chosen because of its similarity to terrain where the US military is currently most active - Iraq and the Middle East.
At one point, the vehicles had to climb through a steep valley that organisers said was "reminiscent of a mountain pass in Afghanistan".
DARPA designed a much more difficult course this year, saying at least a third of the contestants would be able to cross the finish line given the level of technology demonstrated in qualification events earlier this week.
But many of the vehicles simply stopped running on the course. One stopped after a tire went flat and another hit a bridge. Still left on the course was a huge six-wheeled truck called 'TerraMax' and a modified SUV called 'GrayBot'.
Using global positioning satellites and inertial navigation, the vehicles were programmed to follow a pre-defined course disclosed only hours before the race.
Radar, lasers and cameras mounted on the vehicles guided on-board computers that steered the vehicles around obstacles.
"This is the first step in the evolution of truly automated vehicles," said Sebastian Thrun, leader of the Stanford University team, adding he was confident Stanley had clinched the prize.
He said Stanley's technology could be used in the near future to assist drivers by detecting potential accidents.
"It's a no-brainer that 50 to 60 years from now, cars will drive themselves," Mr Thrun said.
Both Mr Thrun and Mr Tether compared Saturday's race to the first controlled flight by the Wright brothers in 1903, calling the race a "historic achievement".
By hosting the event, the US military is aiming to comply with a congressional mandate for a third of US military vehicles to be unmanned by 2015.
But DARPA's Mr Tether said that could be accomplished earlier, based on the technology shown so far in this year's race.
"It's closer than we think," Mr Tether said. "We could see convoys deployed in as early as five years."
In a Grueling Desert Race, a Winner, but Not a Driver
(from New York Times)
By JOHN MARKOFF
Published: October 9, 2005
PRIMM, Nev., Oct. 8 - Stanley, a robotic vehicle designed by a Stanford University team, appeared to earn its creators a $2 million prize on Saturday by being the fastest finisher on a 132-mile course through the Nevada desert.
The race, called the Grand Challenge, was a Pentagon project meant to promote the development of technologies for 21st-century automated warfare. The car was not immediately declared the winner because officials were doing final calculations, but race times on the event's Web site indicated that it had come in several minutes ahead of two entries from Carnegie Mellon University.
Driverless Vehicles Race Through Mojave Desert
Darpa Grand Challenge Web Site The Stanford scientists who led the 18-month effort to build Stanley said they saw their victory as a significant leap forward in the field of artificial intelligence, a discipline that has long suffered from big promises that did not pan out.
"This is for people who say, 'Cars can't drive themselves,' " said Sebastian Thrun, the director of the Stanford Artificial Intelligence Laboratory and co-leader of the Stanford team. "These are the same people who said the Wright brothers wouldn't fly."
The three finishers were the survivors from a starting group of 23 teams fielded by alliances of computer, automotive and aerospace firms, university researchers and others.
The competition was organized by the Defense Advanced Research Projects Agency, or Darpa, and was intended to tap into the talents of researchers and innovators who might not otherwise be found by the nation's military technology firms. Darpa gave birth to the Internet's predecessor, the Arpanet, along with the Predator drone and the stealth fighter.
Eleven of the competitors were repeat entrants, but the start of this year's race was much different than that of the first Grand Challenge, held in March 2004. During that race, a number of vehicles failed to leave the starting gate, and within four hours the rest of the 15 entrants were stalled and broken.
On Saturday, as darkness fell, two cars were still out on the course, and competition officials were considering suspending the race until Sunday.
The race was run over dusty unpaved roads, mountain passes and flat lake beds north of the gambling town of Primm on the California border. At 6:40 a.m., the first three robots entered a starting gate in front of about 2,000 spectators: H1ghlander and Sandstorm, a Hummer and a Humvee from Carnegie Mellon's Red Team, and Stanford's Stanley, a modified Volkswagen Touareg.
"It's time to see what has transpired over the last year and a half," said Anthony J. Tether, Darpa's director.
Just minutes after sunrise, as three helicopters hovered overhead, the robots set off at five-minute intervals, traveling slowly past a long stretch of bleachers and then turning south and accelerating smartly, leaving clouds of dust in their wake. The robots made constant siren or horn sounds for safety reasons, and each was closely followed by a Darpa pickup truck.
Soon the vehicles were traveling at speeds above 30 miles an hour. Shortly after 8 a.m., Spider, the entry of a team from Cornell, hit a bridge at the 21-mile mark and was out of action. An hour later, four other vehicles were immobilized on the course.
The contest grew tense at the halfway point as five leaders stayed close together, averaging slightly more than 20 miles an hour. There were a number of stalls and a crash in front of an audience viewing area when Alice, created by the California Institute of Technology, plowed into a concrete barrier and then attempted to drive over it.
At the 102-mile mark, Stanley took the lead, passing one of the Carnegie Mellon Hummers. The robots ran through Beer Bottle Pass near the end of the course, skirting a cliff with a 100-foot drop. The course also included several tunnels where the vehicles were cut off from satellite navigation signals.
The robots drove without human intervention and were equipped with a wireless "kill switch" that would allow a chase vehicle to stop them in an emergency. This year Darpa took special care to protect against cheating, conducting wireless surveillance on the course and watching for rogue radio signals.
The team behind Stanley, which averaged more than 17 miles an hour, also included representatives of Volkswagen, Intel and other companies.
"The Grand Challenge has been conquered," Dr. Tether said.
Darpa set up the competition in response to a Congressional mandate that the Pentagon develop technology to make one-third of the military's land vehicles autonomous by 2015.
The race has forced developers to refine the software at the heart of the vehicles' navigation systems. The hardware used by many of the vehicles, including radar, vision systems, laser trackers and fast microprocessors, has remained largely unchanged over the past 18 months. In contrast, there has been tremendous refinement of the software, which employs artificial intelligence techniques, several developers said.
In addition to supporters for the various teams, this year's race attracted robotics hobbyists and several Silicon Valley celebrities. Larry Page, a founder of Google, was in the crowd, and Stephen Wozniak, the co-founder of Apple Computer, raced around the pit area on a Segway scooter in a bicycle helmet.
The race's official Web site, which tracked the progress of the vehicles, also drew a large audience, receiving more than 12 million hits in eight hours.
While the military uses contemplated by Darpa were hypothetical for most of the entrants this year, several of the vehicles were developed by teams closely involved with military contractors.
The most imposing entrant was TerraMax, which was a modified version of Oshkosh Truck's Medium Tactical Vehicle Replacement. The company has manufactured 7,000 of the machines for the Marines and the Navy, and as many as 1,500 have been deployed to Iraq.
"We think that this control system has a real place in the military for the trucks we produce," said Gary W. Schmiedel, vice president for advanced products engineering at Oshkosh. The challenge, he said, is protecting American soldiers who are driving the roads in Iraq: "If we can't eliminate the threat, we can certainly reduce it."
Mr. Thrun, of the Stanford team, said advances in the field of self-driving vehicles would start to come more quickly. "Extrapolate two, three or four years out, and then let your imagination play," he said.
They came a long way in just a year.
I agree. The Army's FCS p'rolly looks a lot better right now.
Yea just a last year ago most were predicting 5-10 years before a vehicle could complete the challenge.