Nvidia is Teaching Audi's Autonomous Cars How to Drive

And the two companies plan to have advanced self-driving cars on the road by 2020.

Nvidia

A lack of lanes, varying road surfaces, and unexpected construction detours—those are the type of difficult autonomous driving scenarios that Nvidia's neural network is designed to handle. The semiconductor company demonstrated the capability of its Drive PX 2 processing unit in an autonomous Audi Q7 on a short closed-loop track at the 2017 Consumer Electronics Show.

As it navigated around cones, Nvidia showed how the car can make decisions on the fly as they abruptly put up construction signs and created detours.

"The point is that this isn't programmed to follow a specific course," said Danny Shapiro, Nvidia's senior director of automotive. "It's sensing in real time the environment and making adjustments."

This exercise demonstrated Nvidia's deep neural network capabilities, which came about as a result of 18 months of internal research and development, but just months of work Audi. "No coding was involved," said Shapiro.

But at the same time, the car is only as smart as its been taught. The self-driving vehicle would struggle, for example if it encountered road signs that it has never seen before, or familar ones written a different language. That said, it's certainly not outside the realm of possibility that, one day, Nvidia's smart cars will be able to learn on the fly.

"Getting it to understand natural language is possible, it just requires different training and AI in the car," Shapiro explained.

And while the highlight of the joint program was Nvidia and Audi revealing their plan to put the "world's most advanced AI car" on the street by the year 2020, you won't have to wait that long to see the results of their self-driving partnership in public. The 2018 Audi A8, due out this year, will be capable of Level 3 automated driving in part thanks to Nvidia hardware and software.

You can watch Nvidia's Drive PX2 self-driving program at work—albeit in a Lincoln—in the video below.