Coming Soon: Self Driving Cars
Thanks to a supercomputer on a tiny microchip, Eugenio Culurciello says cars are going to drive themselves in the near future.
Culurciello, an associate professor of electrical engineering at Yale Unversity, would know. He helped design the chip that could make it happen. The microchip contains a supercomputer with the ability to recognize and distinguish surrounding objects, such as other cars, people, trees, lampposts and whatever else can be found on the road.
Thanks to a supercomputer on a tiny microchip, Eugenio Culurciello says cars are going to drive themselves in the near future.
Culurciello, an associate professor of electrical engineering at Yale Unversity, would know. He helped design the chip that could make it happen. The microchip contains a supercomputer with the ability to recognize and distinguish surrounding objects, such as other cars, people, trees, lampposts and whatever else can be found on the road.
The supercomputer, which Culurciello and his associates named NeuFlow, was inspired by the mammalian visual system, which imitates a neural network allowing it to interpret the supercomputer's surroundings. Culurciello says the supercomputer is necessary for implementation into a car, because regular computers wouldn't be able to handle the complex algorithims done to distinguish the surrounding objects. He said it differs from the precrash braking systems seen on various car models.
"Those warning systems you see on some cars that break automatically if you're about to crash, those use laser systems. The laser can sense something in the area whereas this is a vision. It actually sees the object. It can actually figure out what things are," Culurciello.
"Those warning systems you see on some cars that break automatically if you're about to crash, those use laser systems. The laser can sense something in the area whereas this is a vision. It actually sees the object. It can actually figure out what things are," Culurciello.
To develop the complex algorithims that made this vision possible, Culurciello turned to Yann LeCun, a silver professor of Computer Science and Neural Science at New York University. To get it to recognize the objects, LeCun and Culurciello's NeuFlow processes tens of megapixel images in real time.
The idea of mass producing these chips, putting them in cars and working with GPS systems, to eventually let cars drive themselves is not far off says Culurciello.
"I think we're very close," he said. "Even though it seems distant, we have the technology to do it. We need the right combination, and that's basically what we've done. It's the right combo of what's out there. We didn't invent anything, we merged things from different disciplines, computer vision, machine learning, computer design and computer architecture, to put the best of everything together."
Still, there is work to be done before the cars of the future become the present.
Culurciello says his team will now have to prove the supercomputer can work in real time. Once the car has tested, the team will analyze data to see where they have made mistakes.
Culurciello says they have not worked with any car companies yet, but would like to. He also said the supercomputer could be used to improve robot navigation into difficult-to-reach locations or provide soldiers with 360-degree synthetic vision.
The idea of mass producing these chips, putting them in cars and working with GPS systems, to eventually let cars drive themselves is not far off says Culurciello.
"I think we're very close," he said. "Even though it seems distant, we have the technology to do it. We need the right combination, and that's basically what we've done. It's the right combo of what's out there. We didn't invent anything, we merged things from different disciplines, computer vision, machine learning, computer design and computer architecture, to put the best of everything together."
Still, there is work to be done before the cars of the future become the present.
Culurciello says his team will now have to prove the supercomputer can work in real time. Once the car has tested, the team will analyze data to see where they have made mistakes.
Culurciello says they have not worked with any car companies yet, but would like to. He also said the supercomputer could be used to improve robot navigation into difficult-to-reach locations or provide soldiers with 360-degree synthetic vision.
envira blog
No comments:
Post a Comment