This story is from the category Artificial Intelligence
Date posted: 09/10/2005
Mojave Desert, Nevada, deep in the Primm Valley. Here stands Buffalo Bill?s Casino, and a car park behind it. This car park is host to something looking a little like a Borg colony. A mish-mash of high tech and big business, university students and produce of garden sheds.
The second annual DARPA Grand Challenge is about to get underway, that military-sponsored desert race using self-driving AI vehicles. Not a human driver in sight.
$2,000,000 us is up for grabs for the winner. A tantalising prize when you consider that last year, nobody made it to the 10-mile mark, let alone the 144-mile finish line.
Twenty-three teams from across the country have made it to the finals, from nearly 200 original applicants and 43 teams that competed in the 8-day semi-finals last week at the California Speedway.
The teams won't know the course they're running until 2 hours before start time. This prevents cheating by pre-programming the robots with details of the course they have to run. The first team taking off, Carnegie Mellon University's H1ghlander, will be given course details at 4 a.m.
CMU is seen as the Microsoft of the race. Team leader Red Whittaker is thought of as the father of robotics, and everyone looksat them, expecting the best of the best. Two vehicles comprise their team, Sandstorm and H1ghlander. H1ghlander will race in pole position as the fastest car there, Sandstorm will begin in third position, as an all-rounder, 20 minutes behind.
Hopefully they will do better than last year?s pittiful
CMU was the reigning champ of the first Grand Challenge, but that wasn't saying much. The modified HUMMV only made it little over 7 miles in a 144-mile race. Still, that was last year. Tech is bound to have improved dramatically since then ? we hope.
The course this year is even longer ? 165 miles. The goal is to finish in less than 10 hours ? or around 16.5 mph. Finishing at all would be rather nice thou. If several finish, the winner will be gauged on how fast it runs the course, according to DARPA officials.
According to DARPA's director Tony Tether, all the robots have improved dramatically since last year's race,. Talking to teams around the parking lot, everyone seems to agree that because they've had more time to prepare since last year, they've been able to improve the software and systems needed to make a driverless vehicle travel relatively well in unforeseen conditions. Tether said they're able to "analyse" situations better so that they can drive more accurately, rather than swerving too far one way to avoid an obstacle only to run into a rock.
Five and a half hours into the challenge, and at about mile 68 of the DARPA Grand Challenge, the halfway point in the military-sponsored robot race, at least five vehicles have driven by without human intervention. Compared to the 7 miles total last year, that is an incredible achievement for just 12 months work in vehicular AI.
"It's exciting to see this compared with last year's race," said Jim Dugan, a representative of Caterpillar, which makes mining engines and is a sponsor of two race teams, CMU and Terra Max.
"It's great for technology to advance that quickly and it shows competition is a good thing," he said.
CMU's two robots, H1ghlander and Sandstorm, are in first and third place, respectively, with Stanford Racing Team's Stanley running second at about noon.
Both cars have passed all course obstacles by noontime, including a darkened tunnel where their GPS technology doesn't work well, bumpy hills and rocky gullies. And that's not including all the desert wind and dust clouds muddying their "sight" on this windy, but sunny day almost 40 miles outside Los Vegas.
At this time alas, 11 entries have failed.
Refreshingly however, the first to fail, Cornell University's "Spider", bit the dirt after 9 miles. When you consider that the BEST robot last year managed 7 miles, that is still impressive. The car hit a barrier when it was driving over a bridge.
Team DAD's "Dad, Are we there yet?" Toyota Tundra truck travelled 26 miles before its sensor stopped spinning. Caltech's "Alice," which replaced last year's "Bob" from the same team, burned out by driving up and hitting a cement wall and nearly taking out a reporter.
Stanford University's Racing Team has accomplished a historic feat, finishing first place in the 2005 DARPA Grand Challenge, 131.5 miles of artificial-intelligence driven car racing..
"We had a great day," said Sebastian Thrun, director of Stanford's artificial intelligence lab and head of the racing team. Stanford's "Stanley," a modified Volkswagen Toureg with sensors and radar mountings, crossed the finish line within eight hours and 14 minutes, under the 10 hour requirement, according to times posted on the DARPA race Web site.
Although an official winner will not be announced until all robots either finish or burn out, the DARPA Grand Challenge "has been conquered," according to a spokesman for the department.
"These are world records," he added.
Ten minutes after Stanley crossed the finish line, Carnegie Mellon University's H1ghlander came through second. H1ghlander completed the course in eight hours and 19 minutes.
CMU's Sandstorm followed minutes later, finishing in roughly eight hours 12 minutes, according to the site. CMU had predicted the odds of finishing the course for both H1ghlander and Sandstorm were about 29 percent and 40 percent, respectively.
The driving times for all the vehicles are not final because they had been paused several times during the race for various reasons. Stanley, for example, was paused at least once for running up too close behind H1ghlander, which it eventually passed.
According to the Stanford team members, they believe they ran the course in about 7 hours.
People in the spectator tent watched on with awe when Stanley drove over and down Beer Bottle Pass, which has 1,000-foot drops and hairpin turns. The packed crowd cheered when the car made it around the first switchback and then began chanting "Stanley, Stanley" as it drove down.
See the full Story via external site: news.com.com
Most recent stories in this category (Artificial Intelligence):
18/04/2013: NASA's Plan to Advance Robotics by Robotically Capturing small Asteroid