Nvidia is building a driving licence for driverless cars

Vaughn Highfield October 10, 2018

Just like a human driver, driverless cars could need driving licences to be allowed to operate on the road. The trouble is, such systems for testing these driving algorithms don’t exist and being able to test the competency of a computer across every situation would take millions of hours to test.

Nvidia is building a driving licence for driverless cars

Now though, thanks to Nvidia’s Drive Constellation simulation platform, German safety agency Tüv Süd and AVL can work with automakers to create a set of standardised testing measures for autonomous driving systems.

Despite being based in Germany, Tüv Süd is an international certification company that works with automakers and companies all around the world to come up with testing and standards for technological goods. By pairing its expertise with AVL, who work on the testing and simulation of powertrain systems, and Nvidia’s platform, it’s hoped a driving test for autonomous vehicles can be created and rolled out to automakers.

Announced earlier this year as an internal testing project, Nvidia’s Drive Constellation platform essentially traps an autonomous car’s systems entirely within a simulated environment. Utilising two servers in tandem, Nvidia can create a completely simulated world in real-time, changing weather and time conditions and even moving the sun’s position to simulate glare or place human-controlled drivers on the roads to test erratic driving.

This world, and all of its intricacies, is then fed into a second server equipped with an Nvidia Drive AGX Pegasus essentially running the brains of an autonomous vehicle. It’s here, trapped in a virtual world, that the car learns how to drive millions – if not billions – of simualted road learning and training itself on all the situations it would find in the real world without having to put anyone at risk.

By leveraging this hardware-in-the-loop techonology in the automotive space, Nvidia, Tüv Süd and AVL all hope to create measurable testing requirements. Currently, Tüv Süd’s testing covers how an automated system handles common highway traffic situations. One example is how a system handles being cut up by another driver, breaking sufficiently and then accelerating to keep up with traffic flow after.

It’s not designed to be a test to see if a car can avoid crashing, it’s really if it’s capable of driving on roads where other humans and autonomous cars are present. It’s important that it can keep passengers comfortable and drive economically.

In time, the companies hope to expand testing to city driving and really test how cars handle busy intersections and tight city streets.

This is clearly just an early first step, but if it means improved safety and standardised tests, it’s more than likely to accelerate the number of cities and countries comfortable having autonomous driving systems on their roads.