In the wake of the Uber accident, Nvidia is turning to simulations to test autonomous cars

© Nvidia
© Nvidia

Nvidia announces the DRIVE Constellation system for testing autonomous driving tech in a simulated setting

Share

Nvidia is shifting its focus towards testing self-driving vehicle systems in simulated settings with the announcement of its DRIVE Constellation system at its GPU Technology Conference (GTC) in Silicon Valley this week.

The announcement comes at an important time for the self-driving vehicle industry following the news earlier this month that an Uber self-driving car had struck and killed a woman police identified as Elaine Herzberg as she crossed a street in Arizona with her bicycle.

The accident sent ripples through the industry and caused Nvidia to suspend its fleet of self-driving cars, which it had been testing on public roads in US states such as California.

Speaking about the accident during a press Q&A session this week, Nvidia CEO and founder Jensen Huang was keen to distance his company from the incident, while also positioning Nvidia's new Constellation testing system as a safer alternative.

"First of all, Uber does not use Nvidia's DRIVE technology," Huang said. "Uber develops their own sensors and driving technology so that's the first thing. Secondly, we're looking forward to see what we can learn and this is obviously an important moment."

Read next: Nvidia supercharges deep learning at GTC 2018

The CEO also confirmed that the development of Nvidia DRIVE - the vendor's own AI-powered platform for designing, deploying and now testing self-driving vehicle systems - is still in progress and the company decided to put a stop to testing due to good engineering practice.

Now, during GTC, the company announced DRIVE Constellation, a cloud computing platform which runs across two servers and allows developers to test autonomous vehicle systems within a photorealistic simulated version of the real world.

Its first server runs on Nvidia DRIVE Sim Software to simulate everything from autonomous vehicle sensors like cameras and radar, whilst the second server includes a DRIVE Pegasus AI car computer which runs the full vehicle software stack.

"One of the beautiful things in virtual reality is we can create whatever scenario we want," Huang added, as it allows developers to potentially run simulations of adverse weather or the glare of the sun.

DRIVE Constellation will be made available to early access partners in the third quarter of 2018.

Nvidia has been busy touting its VR credentials during GTC this week, including the announcement of its most powerful graphics card to date: the Quadro GV100 graphics card and RTX ray tracing technologies, with the aim of achieving the elusive goal of delivering intensive ray tracing imaging in real time.

Read next: How Nvidia is pursuing the 'holy grail' of real time photorealistic VR

"Recommended For You"

Self-driving cars could save more than 21,700 lives, $450B a year University building simulated city to test driverless cars