Self-propelled cars trained with Grand Theft Auto
Photo: Rockstar Games
Research teams of Intel and the university in the German city of Darmstadt to use the popular game Grand Theft Auto to algorithms for self-propelled cars to improve.
The scientists make use of machine learning to the systems to improve algorithms to improve oneself by analyse large amounts of data.
This kind of systems are already being used, for example, objects to recognize on photos. In these cases, there are huge databases of visual material ‘fed’ to the self-learning system, then getting better at recognizing the objects.
However, it is a lot harder to apply with self-propelled cars; there’s going to be an unworkable amount of time to sit in the collection and tagging of footage of real traffic situations. Also, it is not very convenient to have a self-propelled vehicle during the learning process, each scenario (such as accidents) to continue.
Video: self-Propelled cars learning video game
The researchers also make use of the game Grand Theft Auto, that there are so realistic that it can also be used by the self-learning systems of self-propelled cars. Because the labels directly from the software, can be met, do the images not have to be manually provided annotations.
By the systems to train with the computer images from Grand Theft Auto, it was for the scientists to make the system much better than when it would only work with “real” images.
The scientists are not going to ‘real’ dates do not have to use; the images from the game to serve as addition to the data collected in the real world.