So there you have it, it’s possible to learn good habits from playing Grand Theft Auto.
Advertisement
It takes thousands of hours of real-world visual data in various scenarios to teach cars how to navigate on their own, but the researchers say information from games is easy to collect and can be “almost as good, or sometimes even better” than real data.
Grand Theft Auto’s speeding, drive-bys and frequent crashes might not seem like an ideal model to base driverless cars on, but that hasn’t stopped several groups from using the game in their research.
By using a technique known as machine learning, they’re enabling computers to quickly comprehend and process the world around them, like identifying faces and recognizing speech.
One team from Intel Labs and Darmstadt University in Germany created a software layer between the game and a computer’s hardware that automatically classifies all objects in the game’s road scenes. This is especially problematic for real-world tasks like automated driving.
The game is an efficient way to teach algorithms, as recording and labelling real-life imagery by humans would be extremely laborious and virtually impossible. It’s also impractical to go through every possible scenario in real life, like crashing a vehicle into a brick wall at a high speed.
Advertisement
“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 Shafaei, a student and the University of British Columbia. And for those anxious about potential issues that might arise from using data from a video game, the scientists say that data from games like GTA can be just as good, if not better than data from real life.
