DARPA race pushes robotics forward

ORO GRANDE, Calif.--Carnegie Mellon University and its robotics guru, Red Whittaker, have been vindicated.

On Sunday, CMU's Tartan Racing took home $2 million for first place in DARPA's Urban Challenge--a test of driverless cars on urban streets here at the former George Air Force Base in Southern California's Mojave Desert. By doing so, the team regained its pride after two stinging defeats in 2004 and 2005. And it stole some glory back from 2005's winner, Stanford University, in tackling what was effectively a harder challenge this year. (Stanford claimed the second prize of $1 million this year.)

Apart from a little competitive drama and at least one robot wreck, the DARPA Urban Challenge produced a more important win for robotics this year, one that everyone from Whittaker to Stanford's team leader, Sebastian Thrun, pointed out at the race Saturday. That was simply that the competition seeded the idea in people's minds that self-driving cars are possible. Moreover, proponents say the underlying technology will pave the way for a new generation of cars that will help save lives, either through assisted-driving applications for civilian cars or fully autonomous vehicles for the military.

DARPA Director Tony Tether, who conceived of the races, compared Saturday's event to the Wright brothers' flight at Kitty Hawk in the early years of the 20th century, a small step that eventually led to things like jumbo jets.

"Bot on bot was a new experience (this year), and I saw them pass each other like any person would," Tether said in an interview with CNET News.com. "It was spooky."

Still, there's much work to be done before autonomous vehicles work well, and the race proved how hard it is. Of the 89 original applicants in the competition, officials chose only 35 to compete in a National Qualifying Event, which weeded out another 24 teams. On race day, only six out of the 11 teams completed DARPA's three required missions to cross the finish line, and then only four accomplished the tasks in under the allotted 6 hours.

"This is a fun event, but these cars are not ready for autonomous driving," Stanford's Thrun said during the race. "There's clearly more development needed."

Nearly all of the teams' robots suffered some difficulty. CMU's Tartan Racing, for example, was set to begin the race early Saturday morning, as the first contestant out of the gates. But it was pushed to the No. 10 slot because a nearby Jumbotron was interfering with its Global Positioning System setup so the car couldn't track its own whereabouts on a map. The team couldn't figure out what was wrong with the robot initially.

Later on, race officials stopped Intelligent Vehicle Systems just out of the starting gate because it nearly hit one of the concrete race-chute dividers. It was cut from the race within the first two hours. Then, in what was the lone robot collision of the race, Cornell's robot hit MIT's Land Rover after MIT tried to pass the Cornell car.

One of the most eye-popping traffic violations involved Team Oshkosh's 24,000-plus-pound truck, which rolled over a curb and nearly struck an old Air Force building before officials hit the emergency stop button on the robot.

Videos:

DARPA 2007: Competition heats up
CNET checks out the Grand Challenge scene before the race and as the cars launch.

DARPA 2007: Competition gets interesting
Surprise disqualifiers and some mishaps lead to an interesting race.

DARPA 2007: Finalists emerge
CNET speaks with a stunt car driver and race watchers.

Still, the near-accident was amusing to spectators. "I'm definitely amazed by the lack of the human body being in the car," said James Murrell, who drove 60 miles from Lancaster, Calif., with his girlfriend to see the event.

Ultimately, DARPA judged the winners by how well they obeyed traffic laws and by how fast they finished the course. CMU's Boss finished the course nearly 20 minutes faster than Stanford's robot, Junior.

On the flip side, many more winners emerged from the event than just the obvious. Velodyne, for example, developed a high-powered spinning laser, called a lidar, that was used by 7 of the 11 finalists, including CMU, Stanford and MIT. That technology, which includes 64 data-collecting lasers instead of one, was built specifically for autonomous car navigation and was tested in the 2005 Grand Challenge. Now the company sells the sensor for $75,000, more than the cost of most vehicles in the race.

Team sponsors such as Google also appeared prescient. The search giant backed two teams in the winner's circle, CMU and Stanford.

In contrast to the last two races, 2007's challenge had much more of a circus feel. For example, four horses ridden by Marines were supposed to strut out in front of the cars during the opening ceremony, but the animals got spooked and circled each other for a while. More than one person joked that it was a sign that the animals sensed something was off about letting robotic software take the wheel.

Underscoring the heft of the event but with a hint of the carnival, a race official said during the ceremony: "You'll see things today that three years ago you'd never have dreamed about."

Still, the drama of the competition was largely between CMU and Stanford.

In 2004, CMU was pegged the favorite in DARPA's first-ever challenge of autonomous driving vehicles, given that the expertise of the university's robotics department and professor Whittaker. But CMU's autonomous car spun its wheels after only 7 miles on the 142-mile desert course, leaving no winner that year.

In 2005, CMU returned to the Grand Challenge more determined than ever with two race vehicles, heavily outfitted and modified Hummers. However, technical problems with the vehicles brought CMU defeat, and Stanford's team led by Thrun--the former protege of Whittaker--claimed the $1 million prize as a first-time entrant in the race.

Stanford also garnered global attention for accomplishing what hadn't been done before: engineering a car to drive itself more than 132 miles in the desert in less than 10 hours. It's rumored that after the race, CMU's team threw darts at a picture of Stanford's robot, Stanley.

This year, Whittaker's team will be remembered for engineering a robot that could master basic traffic rules while driving among other robots.

One race veteran put it like this: "Competition is huge for this event. The spirit of competition focuses everyone to solve the problem at hand."

[SkinnerAZ]

How you solve it is more important

After watching it and listening to how people approached the problem, it's great to see the diversity of solutions. Problem I see is you have MIT doing something SO ridiculous with having a super computer with its own gen-set and air conditioner for the cluster computer (40 cores), getting SPANKED by Stanford running TWO PC's (1-Q6600/box) in Junior. Even though I'm no fan of Stanford, it's absolutely refreshing to see an *elegant* solution doing well in this environment as that is what is really needed. I don't think anyone wants to fill up for both a gen-set and the engine at the pump...

[FOG9]

Pump this up

I can't believe this wasn't "live" on Discovery or PBS or (somebody
wanting innovative programing) ... I watched the video stream and
found it to be compelling, informative entertainment!
With an entry from Germany, I also wondered why there wasn't an
entry from Japan?

[Manhattan2]

Our group is using 4D logic to make real world products

While these games can foster innovation there is already a team of engineers creating products using gpsimaging, location markers and a seeingaid that will revolutionize location based services and navigation.

[compurgator]

Remember: DARPA=Poindexter

Please, please, please tell me that at least some of you remember John Poindexter (he still likes to have himself called, "Admiral," notwithstanding his having fallen so badly from grace with Ollie North, when they both lied before Congress after the Iran-Contra screw-up that was his last tour-de force). DARPA: the spooky little closet program funded by Bush 41 that runs weird programs (in the name of "science") like the use of parasitic worms carrying radio transmitters to monitor the whereabouts of those who eat hot dogs; throwing pianos out of tune so as to make obsessive piano players paralyzed with anxiety; and painting replicas of The Golden Gate Bridge on the side of a mountain near The Bay, causing mariners to run aground and then fall into clinical depression because of their guilt. That's Poindexter. That's DARPA. Go ahead. Participate. Encourage him. He'll be under your bed, soon enough.

[sheldonc]

DARPA....

has been around a lot longer than Dubya. It was started as a
response to Sputnik's launch in 1958.

Sure it's dome a number of odd kinds of things in the past, but it's
created some pretty cool things, like, for instance, the Internet.

Poindexter is a dirty footnote in DARPA's history.

[mzaug]

Urban

The Carnegie Mellon University and others taking part in the DARPA Urban Challenge, signal first attempts at filling a one hundred year old gap in the automobile driving. The current car industry?s computer ?achievements?, such as decapitating, self operating windows, inaccessible doors, the tire pressure checks without regulating it, are no more than promotional gimmicks. None of these contribute to the relief and provide assistance for the, as my cousin Douglas B. a New England gas station operator termed, ?A nut behind the steering wheel?. The poor sop still has to continuously strain eyes to see the road and lurking dangers on and off it, often in hidden and masked by the rain, snow slippery surface, poorly lit by crude headlights. As a hundred years ago, the driver must mentally and manually coordinate and time the movement of his hands, legs and eyes. The only change is in the road congestion, high speeds, and a blossoming road rage.
M. Z. Augustyniak, Architect Nov.6, 07