Self-driving cars can learn the rules of the road by studying virtual traffic on videogames such as Grand Theft Auto V, according to a new study.
A team led by University of Michigan at Ann Arbor professor Matthew Johnson-Roberson trained an algorithm solely using Grand Theft Auto V and tested it against an algorithm trained on real-world images of traffic. Johnson-Roberson says the algorithm trained on Grand Theft Auto V performed just as well at spotting cars in a pre-labeled dataset.
He notes the video-game version needed about 100 times more training images to reach the same standard, but given that 500,000 images can be generated from the game overnight, he says that is not a problem.
AI is being trained on images from similar locations, at similar times of day, under similar weather conditions, and then tested under similar conditions, notes German Ros at the Autonomous University of Barcelona in Spain. However, he says it is difficult to determine whether computers are recognizing cars or just memorizing a particular dataset.
Ros points out videogames can present driverless car systems with a variety of vehicles and road conditions.
From New Scientist
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA