This is how artificial intelligence is learning to drive your car

Researchers at Nvidia are figuring out the nuts and bolts of how artificial intelligence present in autonomous driving systems processes data in order to learn how to drive.

What they’ve established so far is that AI systems take an approach so obvious and simple, it escapes most of us lame mortals completely: it maps all of the intricacies of human driving and just takes out the mistakes. Duh.

In a nutshell, software drives like us, only better.

Although almost every big automotive and tech company these days is developing some kind of autonomous vehicle, according to the MIT Technology Review, the intricacies of how artificial intelligence actually handles and understands information is largely unknown.

Tech industry leader Nvidia – recognised mainly for their graphic cards and innovative chip-sets – has been working for years on their own autonomous driving system. Their research is one of the most robust in the industry, and they have already signed head-turning partnerships with Tesla and Bosch.

In their research paper titled Explaining How A Deep Neural Network Trained With End-to-End Learning Steers A Car Nvidia scientists describe their project: “As part of a complete software stack for autonomous driving, NVIDIA has created a neural-network-based system, known as PilotNet, which outputs steering angles given images of the road ahead.”

PilotNet is trained using road images and data provided by a human-driven car. According to the paper, learning from observing human drivers “eliminates the need for human engineers to anticipate what is important in an image and foresee all the necessary rules for safe driving.”

The software learns from its human counterparts to identify the multiple features and objects encountered in the driving experience like lane markings, streetlights, bushes and, of course, other cars.

Road tests show PilotNet indeed drives better than most of us puny humans, regardless of weather or road conditions.

“What’s revolutionary about this is that we never directly told the network to care about these things,” said Urs Muller, Nvidia’s chief architect for self-driving cars. “I can’t explain everything I need the car to do, but I can show it, and now it can show me what it learned.”