There are problems with the way humans drive that could be solved by self-driving cars, and even though Uber is looking to bring more autonomous vehicles to the road very soon, the technology hasn’t been perfected yet, and self-driving cars still have plenty to learn. Somebody’s got to educate them, and it looks like that somebody will be Grand Theft Auto V, the same game in which you can try to jump a car into a cargo plane.
Researchers at Intel Labs and Germany’s Darmstadt University are less interested in the carnage the game offers, however, and are focusing more on the gigantic and realistic 3D world filled with real-life objects like cars, people, bicycles, and other things you’d encounter on your daily commute (via Jalopnik). So with that in mind, it makes sense that the game is being used to train self-driving cars to recognize objects, which is a lot more practical and safe than these vehicles doing this learning out in the real world.
For example, below is a video of a car being driven in-game by artificial intelligence, which learned using what is called a neural network. As you can see, it stays in its lane and avoids other objects on the road, which seems like a good start to us:
If you’re not up to speed on what a neural network is or what it does, this video on a computer beating a Super Mario World level explains it well. Essentially, the computer runs a bunch of simulations and learns through trial and error what inputs it needs to execute in order to achieve the desired outcome. This means that once self-driving cars are on the road, they shouldn’t be running down pedestrians on the sidewalk and jumping over buildings to avoid the police, so that’s good.
Featured image: Rockstar Games