View Single Post
Old 09-24-10, 02:10 PM   #1
News
Registered User
 
Join Date: Jun 2009
Posts: 49,763
Post GTC Closing Keynote Takes GPUs Into the Future with Robot Cars

A furious three days at GTC concluded on a soaring note as Stanford's Sebastian Thrun gave a fascinating, at times comical, look at the evolution of driverless cars.

Taking the stage to the strains of Gary Numan's 'Cars,' Thrun took a packed house through a litany of sorry statistics. Cars are the leading cause of death for young adults. They consume a third of our nation's energy. They are only used for 3 percent of their life. And the highways they drive on are only about 6 percent fully utilized.

The way to do better is with GPU-based technology, which helps drive the computer vision that robotic cars depend on.

Thrun's research began in 2004 when DARPA, the research arm of the U.S. military, sponsored a 150-mile challenge race for robot cars. The initial challenge was nothing short of disastrous, with no car making it more than seven miles. Most failed right out of the gate.

The next year, Thrun entered the challenge with a team of students and a car called Stanley (which now lives at the Smithsonian and was top of the heap in Wired Magazine's 50 best robots of all time). Taking us through Stanley's evolution, Thrun touched on the incredibly complex engineering feat of creating a driverless car. The solution involved wedding computer vision through lasers and cameras with machine learning algorithms so that Stanley could find the road was and drive confidently along it.

Showing some nail-biting footage of the race itself, Thrun included an unforgettable blooper reel of robot cars in confused, hapless encounters with traffic cones, hay bales and makeshift tunnels. One underdog robot motorcycle was a crowd favorite, and it was almost hard not to cheer when Thrun ran video of it finally staying upright.

Profound challenges remain in creating driverless cars, although the sight of a robotic car executing a flawless parallel parking maneuver indicated that the day may not be too far off. Thrun noted that advances in GPU computing could bring the day closer. He closed with a list of what he called '2x challenges' ' among them, making cars twice2x as safe, 2x as efficient, and to live with only half the cars we have today.






More...
News is offline   Reply With Quote