Self-Driving Cars Learning A Thing Or Two From GTA 5

Since there aren’t enough hours in a day to cover all the necessary real-world miles that autonomous cars need in order to learn to drive themselves, developers are also relying on Grand Theft Auto V.

And no, it’s not because of Trevor’s antics or Michael’s self-righteous attitude or Franklin’s swagger. It’s simply because this video game has been chosen as a suitable simulation platform by researchers and engineers, looking to teach cars how to respond to different scenarios.

“Just relying on data from the roads is not practical,” said Davide Bacchet, in charge of the simulation effort for Nio, a startup looking to launch an autonomous electric car in the US by the year 2020.

“With simulation, you can run the same scenario over and over again for infinite times, then test it again.”

While it may seem strange at first, ultra-realistic video games such as GTA 5 are able to generate data that’s very close to what AI-powered vehicles can extract on real roads.

According to Autonews, researchers are now deriving algorithms from GTA 5 software that’s been tweaked for use in the fast-expanding self-driving sector. In fact, GTA 5 offers approximately 262 types of vehicles, over 1,000 different unpredictable pedestrians and animals, 14 weather conditions and countless bridges, traffic signals, intersections and quite a few tunnels as well.

It may not be a perfect substitute for the real thing, but the city of Los Santos still makes for the “richest virtual environment that we could extract data from,” stated Alain Kornhauser, a Princeton University professor who advises the Princeton Autonomous Vehicle Engineering team.