Nvidia Shows Autonomous Vehicle Driving on Unmarked Roads at Night

Now, the Silicon Valley-based company is pushing strongly to take over the automotive industry, and after securing some lucrative contracts with the likes of BMW or Volkswagen or the autonomous racing series RoboRace, it looks like it might succeed in achieving its goals. Aside from the big German names, Nvidia has also managed to secure itself the Tesla account after Elon Musk's company went through a bad breakup with Israeli firm MobilEye.

If Nvidia was initially supplying the hardware for the infotainment system and not much else, the advent of the autonomous technology meant that a lot more onboard processing power was needed, and when carmakers started asking for it, Nvidia was ready to deliver.

A recent video posted on YouTube by Nvidia showcases the capabilities of a self-driving car to adapt to new road conditions. Using the deep learning technique (also employed by George Hotz's Comma.ai startup and other companies as well), the vehicle was able to clear a tricky section filled with traffic cones that, at some point, even took the car off the actual paved road.

Nvidia pushed what was asked of the vehicle by making it tackle blind corners on unmarked, really narrow country roads as well as drive on then at night. The short text accompanying the video says that "all it takes is about twenty example runs driven by humans at different times of the day. Learning to drive in these complex environments demonstrates new capabilities of deep neural networks."

The clip also points out the universality of the vehicle's learning ability. The car we can see was trained exclusively in California, but the footage shown is shot in New Jersey, proving it doesn't just memorize the environment mechanically, but actually takes decisions based on what its sensors can detect.