Scientists Use Grand Theft Auto Data To Help In Development Of Autonomous Driving Technology
Gathering data can be much faster using the open-world environment of GTAby Kirby Garlitos, on
If you thought that the sole purpose of Grand Theft Auto is to satisfy your carnal urges of running people over, prepare yourselves to be blown away by what a group of scientists from Darmstadt University in Germany and Intel Labs were able to do using the popular video game franchise. According to a video released by YouTube page Visual Inference, these researchers have been using the actual GTA game to import thousands of images and data from the game into computers with the objective of improving the way autonomous driving vehicles work on the road.
Yes, the words “Grand Theft Auto” and “autonomous driving vehicles” were just used in the same paragraph in a manner that could reinvent the way research is done in the development of the technology. According to the research, the scientists were able to gather visual data from Grand Theft Auto using a special software that’s capable of classifying different objects on the roads within the actual game. These data can then be used on actual roads and that the gathering of all the data using GTA is much faster than doing it in the real world.
For a little bit of context, gathering real world data and annotating all of the objects in a particular image could take anywhere from 60 to 90 minutes per image depending on how complex a specific image is. By comparison, using a road image from Grand Theft Auto takes no more than seven seconds on average. That has helped researchers extract over 25,000 images from the game with different weather conditions thrown in and at different times of a specific day. In total, annotation of all these images took 49 hours, or just a speck over two days.
The really interesting and potentially useful part about this exercise is that the annotated data generated from all the images from GTA can be used to help autonomous driving vehicles tell the difference between objects when they’re driving around in the real world.
The discovery isn’t going to help develop autonomous driving technology overnight, but it does point to the possibility of expediting the data-gathering aspect of the development. Of all things, who knew that something like Grand Theft Auto would actually have a role to play in the race to get self-driving cars on the road.
Continue after the jump to read the full story.
You’re not going to look at Grand Theft Auto the same way
There’s something to be said for the state of the development of autonomous driving technology when you have scientists from one of the finest universities in Germany taking data from Grand Theft Auto to help accelerate the steps needed to develop the revolutionary system. I personally didn’t see that one coming, so if the researchers believe that it could really help the development of the tech, then I think it’s something that’s worth considering.
That’s as far as I’ll go, though because I don’t personally know if the annotating data through GTA is going to be useful in the long run. That’s for other researchers, developers, and scientists to find out. If they think that the method is worth pursuing, then I’m not going to go against it. I just think that it’s a little funny that the development of a technology as revolutionary as autonomous driving will rely on a game that’s known for being on the opposite side of driver safety. But who am I to judge, right?
Hopefully, we can learn more from this rather unorthodox method in the future. Who knows, it if turns out being a ground-breaking way to develop the technology, we might end up partially crediting Grand Theft Auto – of all things – for getting self-driving cars on the road sooner than we expected.
Source: Visual Inference