Most people will recognize Nvidia as a company that makes graphics cards for computers. But the company is also heavily involved in the auto industry, making computer chips with powerful processing capabilities that control everything from infotainment to self-driving systems.
It turns out Nvidia is also developing its own self-driving system, and judging from this demonstration video it looks like Nvidia’s systems could be one of the most advanced. The key is machine learning (also referred to as deep learning), a form of artificial intelligence where computers “learn” how to perform actions on their own, i.e. without the need to be explicitly programmed.
Nvidia says its system didn’t require any programming for object detection, mapping, path planning or control components. Instead, the system learns on its own to create all the necessary requirements simply by observing human drivers. All it takes is about 20 trial runs of a certain situation driven by humans at different times of the day, Nvidia says.
The result is a self-driving car that can successfully navigate a construction site, a road that is overgrown with grass and bushes, and even dirt roads with no markings whatsoever. (Audi recently demonstrated how machine learning could be used to create a self-parking system.)
Importantly, the system can apply what it learns in one situation in other, similar situations. And over time it becomes more and more capable. For example, the demonstration video also shows a prototype initially trained on California roads later successfully navigating roads in New Jersey.