Playing video games is also a profession – there are numerous players earning six-figure sums by deploying their skills, but sometimes playing games even serves a particular scientific purpose.
For example, in our case scientists from Darmstadt University in Germany and Intel Labs have decided that aimlessly stealing cars and running over people for no apparent reason in Grand Theft Auto needs to serve a higher purpose. This is because they came up with a neat way of getting data from GTA and using it to teach autonomous cars to better operate on real roads. The researchers argue that data from modern computer games can be “almost as good, or sometimes even better” than real data, while using it is also infinitely cheaper than gathering real-world visual data. The developers cooked a new software which classifies different objects on the roads in the game.
So far 25,000 frames from the computer game have been extracted, with various weather conditions and at different moments of the day. Annotating them only took 49 hours in total – the process would have lasted days with real-world data. The data is then used to have the vehicles learn the difference between humans and other objects throughout the environment, which in turn makes autonomous cars smarter and safer.