You probably know that Grand Theft Auto V is not one of the video games that encourage players to stick to posted speed limits. In fact, I’m sure it’s impossible to play this game without committing numerous virtual traffic offenses and other crimes within a few minutes. So why is it used as a training platform for autonomous/self-driving cars?
Of course, the game isn’t limited to heists and random violence. It also features a large, realistic-looking and functioning 3D world – and that’s exactly what research groups like Intel Labs and the University of Darmstadt are interested in.
MITs Technology Review reported that GTA V primarily used to train artificial intelligence to recognize objects – cars, pedestrians, bikes, buildings, roads and other things a vehicle may encounter in the real world.
“The researchers created a software layer that sits between the game and a computer’s hardware, automatically classifying different objects in the road scenes shown in the game. This provides the labels that can then be fed to a machine-learning algorithm, allowing it to recognize cars, pedestrians, and other objects shown, either in the game or on a real street.”
In short, it would take forever if you had to collect this data by driving a car with a camera through a real city and then have people label it. They are immediately available in the game.
The idea is to give software developers a lot more information to work with as they try to teach a self-driving car how to operate in the real world. The goal is (hopefully) not to teach the car how to cause as much chaos and destruction as possible and reach a “five-star wanted level”. That would be much more entertaining, but problematic.
There are some videos of similar software in action on the Internet. The following video shows “DeepDrive,” which uses screenshots to “teach” vehicles.
This “DeepDrive” doesn’t seem to have anything to do with it UC Berkeleys program of the same name to do, which is also dedicated to the study of artificial “perception” for vehicles.
As for the use of GTA V in the studies discussed previously –
“With artificial environments we can effortlessly gather precisely annotated data at a larger scale with a considerable amount of variation in lighting and climate settings,” said Alireza Shafaeia PhD student at the University of British Columbia who specializes in training computers using video games, in conversation with Tech Review. “We showed that this synthetic data is almost as good, or sometimes even better, than using real data for training.”
Still, I have a feeling that a prankster with evil intentions could sneak in and be a bad influence on the cars of the future. But these concerns exist with any type of artificial intelligence.
Related Posts
