Go Park Yourself: NVIDIA DRIVE PX Turns Driving’s Biggest Pain Into a Pleasure
Here’s how to make an entrance: Step out of your car. Pull out your phone. Then press a button to tell the car “Go Park Yourself.”
That’s the vision of the future presented in the first public demo Sunday of NVIDIA DRIVE PX by NVIDIA CEO Jen-Hsun Huang, kicking off the International Consumer Electronics Show. (see “NVIDIA Paves Way for Tomorrow’s Cars With NVIDIA DRIVE Automotive Computers”).
“When you’re done with dinner, you can say ‘come back to me’ … and it becomes an auto-valet,” Jen-Hsun says. “That car meanders back out and gets back to the driver.”
The demo is an example of the power of our newly introduced Tegra X1 processor, which packs more than one teraflops of computing power. That’s what the world’s fastest supercomputer — a machine the size of a suburban home — could deliver 15 years ago.
Over the past two decades, NVIDIA has put more than $9 billion into researching visual computing technologies, like computer vision and deep learning. That gives us the unique ability create new products – like NVIDIA DRIVE – where graphics and supercomputing meet.
As a result, NVIDIA DRIVE PX – our new auto-pilot car computer – opens the door to new era for the automotive industry. Powered by two Tegra X1 processors – and a host of advanced algorithms created by NVIDIA’s computer vision engineering team – DRIVE PX gives cars the brains to take over one of driving’s biggest headaches.
To show what DRIVE PX can do, engineers from our computer vision engineering and demo teams created a highly realistic virtual parking garage – using the blueprints from the actual garage at NVIDIA’s headquarters in Santa Clara — and relying on Epic Games’ Unreal Engine 4.
“It’s an incredibly rich and visually complex environment that simulates the way light reflects from cars and obstacles in real time,” says Mark Daly, who leads our demo team. ““It’s a perfect virtual world.”
DRIVE PX soaks up the information from this complex virtual world – simulated on on five PCs with NVIDIA GPUs – via virtual fisheye-lens cameras attached to a simulated car.
“All the car knows about is what it can see via the simulated front, rear and side cameras that are mounted on it,” says Cem Cebenoyan, director of engineering for computer vision devtech at NVIDIA.
NVIDIA PX then uses sophisticated point cloud algorithms to turn those images into a real-time obstacle map as it moves through the garage, controlling the car’s throttle, brake and steering to move safely through the world around it.
The result: NVIDIA PX can search for an open spot in the crowded garage and slip into it with an expert series of turns. Press another button, and NVIDIA PX guides the car out of the space and back to the driver. No scratches or dinged fenders – not even virtual ones.
It’s proof that the once far-out idea of self-driving cars is closer than skeptics may think.
Even Daly – a driving enthusiast who owns a Dodge Viper – embraces the idea. “I am totally cool having my car drive around and find a parking spot,” Daly said.