Nvidia Drive PX 2: A Supercomputer for Self-Driving Cars

Nvidia's Drive PX 2 artificial intelligence platform for self-driving cars can process the inputs of 12 video cameras, plus lidar, radar and ultrasonic sensors.


Nvidia at CES 2016 unveiled its Drive PX 2, a new artificial intelligence (AI) platform for self-driving cars. The PX 2 is the sequel to the Drive PX in-car computer that debuted in 2015 and currently is used by more than 50 automotive companies.

Nvidia says Drive PX 2 processes up to 24 trillion deep leaning operations per second, which is 10 times the performance of the Drive PX.

The PX 2 has the processing power of 150 Macbook Pros, the company says. It can process the inputs of 12 video cameras, plus lidar, radar and ultrasonic sensors. It takes all that information to accurately determine where the car is relative to its environment and calculate the safest driving route.

Volvo will be the first automaker to use the Drive PX 2. The car maker will use the deep-learning computer in 100 Volvo XC90 SUVs beginning next year as part of its “Drive Me” autonomous-car pilot program.

“Our vision is that no one should be killed or seriously injured in a new Volvo by the year 2020,” Marcus Rothoff, director of the Autonomous Driving Program at Volvo said. “Nvidia’s high-performance and responsive automotive platform is an important step towards our vision and perfect for our autonomous drive program and the Drive Me project.”

Nvidia also announced Drivenet, deep learning network that can identify five different classes of objects, including pedestrians and motorcyclists, and it learns over time.




About the Author

Steve Crowe · Steve Crowe is managing editor of Robotics Trends. Steve has been writing about technology since 2008. He lives in Belchertown, MA with his wife and daughter.
Contact Steve Crowe: scrowe@ehpub.com  ·  View More by Steve Crowe.




Comments

Totally_Lost · January 8, 2016 · 1:05 pm

Way awesome ... but the reality is the algorithms need to be compiled to silicon (C to gates), and the algorithms need to be memory access optimized (using massive distributed memory) to increase both scan rates and effective memory bandwidth, as a front end to a more traditional architecture navigation system. Tools like the Drive PX 2 certainly provide platforms for development, for early adopter sales.

Including this technology in volume for actual everyday cars provides the NRE costs to justify a dedicated ASIC solutions, that both lower cost for wide spread adoption, and greatly improve performance to enhance safety, to allow multiple redundant implementations in the vehicle.


Totally_Lost · January 8, 2016 at 1:05 pm

Way awesome ... but the reality is the algorithms need to be compiled to silicon (C to gates), and the algorithms need to be memory access optimized (using massive distributed memory) to increase both scan rates and effective memory bandwidth, as a front end to a more traditional architecture navigation system. Tools like the Drive PX 2 certainly provide platforms for development, for early adopter sales.

Including this technology in volume for actual everyday cars provides the NRE costs to justify a dedicated ASIC solutions, that both lower cost for wide spread adoption, and greatly improve performance to enhance safety, to allow multiple redundant implementations in the vehicle.


Log in to leave a Comment


in the Future Tech Hub

Editors’ Picks

Autonomous Snake-like Robot to Support Search-and-Rescue
Worcester Polytechnic Institute is creating autonomous snake-like robots that can navigate through...

Love Writing About Robotics and AI? Robotics Trends is Hiring!
Robotics Trends and sister site Robotics Business Review are growing and adding...

WiBotic PowerPad Wirelessly Charges Drones
WiBotic’s PowerPad wirelessly charges everything from large industrial drones to smaller...

Meet Jing Xiao: WPI’s New Director of Robotics
In January 2018, Jing Xiao will become the new director of the Robotics...