Grand Theft Auto is teaching driverless cars how to drive better

Grand Theft Auto has historically been a video game allowing the player to run riot and cause all kinds of mayhem, both in cars and on foot.

But now, some rather intelligent people from Darmstadt University in Germany and Intel Labs have developed a method of using the video game to teach driverless cars how to drive better!

A method of sourcing visual data from the video game could then be fed to computers to teach vehicles how to drive autonomously. Machine learning is enabling computers to pick up new skills much more quickly, but this requires large amounts of information to work effectively, which can often be hard to gather.

It takes thousands of hours of “real-world” visual data in many different scenarios to teach cars how to drive on their own, but the researchers have said that the information from games is easy to collect and can be “almost as good, or sometimes even better than real data.

This is where the ultra-realistic environment in the latest instalment of Grand Theft Auto comes in to play. The scientists created a software capable of classifying different objects on roads within the game.

This generated data, taken from a range of different scenarios and weather conditions, can then be used to teach self-driving cars the difference between pedestrians and other objects when they are navigating the roads in the real world.

Watch the video below for how the researchers set about gathering their data from Grand Theft Auto; it’s rather impressive!

Why not check out our car selling app as well whilst you’re at it? It might not be able to teach self-driving cars to drive better, but it can help you get more money when it comes to selling your car, and that’s pretty neat as well!



Photo courtesy of BagoGames on Flickr, under Creative Commons