Grand Theft Auto V Is Helping Teach Driverless Car AI
Driverless vehicles are coming, and we'll be relying on them to transport us around in the next decade if everything goes to plan. But before that can happen, the artificial intelligence controlling the cars needs to prove itself to be completely reliable. How do you do that? With lots of learning!
Car AI needs to drive a lot. That may seem obvious, but the AI needs to experience any and all driving situations so as to learn and demonstrate the right reaction. It's very difficult to experience all situations on real roads, so simulation is key. And it turns out the best simulation is a game we've been enjoying for a few years now: Grand Theft Auto V.
If you look past the violence, GTA V is about as complete a city and road simulation as you can get. As detailed on Bloomberg, entering the world of GTA V gives an AI access to 262 different types of car, 14 weather conditions, every road situation you can think of and more than 1,000 different types of pedestrian to deal with. Driving around the city of Los Santos day-after-day would certainly prove to be a good learning experience.
Princeton Autonomous Vehicle Engineering (PAVE) realized this, with PAVE advisor and Princeton University professor Alain Kornhauser referring to GTA V as, "the richest virtual environment that we could extract data from." Last year, Intel Labs and Darmstadt University of Technology went so far as to pull visual data from the game, and now algorithms are being developed using GTA V data to help improve driverless car AI.
So when we do all start trusting an AI to safely drive us to a destination, remember it was in part due to the tireless population of Los Santos going about their daily business and acting as a teaching tool. And don't forget, this kind of sumlation isn't limited for use by the big car manufacturers. In January, the DeepDrive project opened this up to everyone.