In April of last year, we announced that the new Roborace category, based on autonomous racing cars, will use Nvidia hardware. However, the interest of the green giant from Santa Clara is not limited to competitive environments; it also works on solutions compatible with traditional cars. A solid proof is Xavier, a supercomputer that combines the ARM and Volta architectures, presented during the company's keynote at CES 2017.
Autonomous drones are getting closer to starting their commercial adventure, but we cannot say the same about autonomous cars. The media example par excellence is Tesla's Autopilot system (which recently managed to anticipate a car and protect its passengers), but it should be interpreted more as a "driver assistance". Google took its foot off the accelerator in its development, now known as Waymo, a separate company run by Alphabet. And of course, authorities still have doubts about letting cars loose that drive themselves on the streets. Until then, it is necessary to settle for tests or experiments, and at CES 2017 we have already seen some.
https://www.youtube.com/embed/fmVWLr0X1SkXavier in Detail
Let's make room for Xavier, Nvidia's supercomputer that transforms traditional cars into autonomous ones. Eight custom 64-bit ARM cores, 512 cores based on the Volta architecture, 30 trillion operations per second, and a maximum consumption of 30 watts define Xavier's main parameters. How does the computer learn to drive? First, by observing a human being. How good a student is it? If we judge by the video in which it controls a Lincoln car (named BB8), pretty well. Conservative in difficult areas, a little faster on highways, and it gives up control as soon as the driver asks for it by voice.
If Xavier sounds a bit familiar, it's because it is actually the latest variant of the DRIVE PX 2 supercomputer that we should soon see in the prototypes of the Roborace category. To this announcement we must add a kind of co-pilot that supposedly will help us drive better, and a special agreement with the people at Audi, which plans to put on the streets cars with level 4 autonomy (complete, with limitations) by 2020.
