Nvidia developing an 'autonomous car driver's license' based on its tech
The company is working on standardized "driver's license" testing for self-driving cars based on a new open-source algorithm.
- North American Car, Truck and SUV of the Year (NACTOY) Awards Juror
At its 2019 GPU Technology Conference in San Jose, graphics processor manufacturer Nvidia announced Monday that it will be releasing version 9.0 of its Nvidia Drive platform for autonomous vehicles to developers and, perhaps more interestingly, that it is also working with German standards body TÜV SÜD to develop a sort of driver's license for self-driving cars .
The Nvidia Drive AP2X release 9.0 -- as the new hardware and software platform is more specifically known -- includes a few new features for autonomous vehicle developers to play with. The software will also be able to make decisions at complex intersections, work through congested traffic and automatically change into a faster moving lane without endangering the vehicle's occupants or, presumably, oncoming traffic.
In areas where there is no map data for the autonomous software to use for guidance -- perhaps a private driveway or a newly paved road -- Drive 9.0 can even generate its own map data, piecing together information from cameras, lidar and radar sensors to create what Nvidia calls a Local HD map that can be used for self-driving navigation after a few human-piloted trips.
Most noteworthy is Safety Force Field, a new, open-source collision avoidance algorithm for autonomous vehicles that uses computational predictions to guess where nearby cars, pedestrians and other obstructions will be in the immediate future and then act to shield autonomous vehicles from collision with a combination of emergency braking and intelligent steering around obstructions. Nvidia is making the Safety Force Field algorithm open-source for other developers to use when building their own autonomous platforms.
Nvidia's motivation for making its technology open-source is to encourage standardization of autonomous vehicle software and to make self-driving algorithms easier to validate and certify for safety. To this end, the GPU manufacturer has also announced that it will be working with German certifying and standards body TÜV SÜD to draft a proposed set of testing measures for all automated driving systems.
Much like how much like how today human drivers are tested to be licensed to drive, Nvidia is proposing what is essentially a "driver's license" for self-driving car software that would put every manufacturer's platform through a series of tests to prove their road-worthiness using Nvidia's newly released Drive Constellation simulation software and the Safety Force Field algorithms as the basis.
Nvidia Drive 9.0 will be available to developers and automakers soon, along with a Safety Force Field white paper that will be published to Nvida's developer site, so you can take a look and pick apart their algorithms for yourself.