CVC’s Research group ADAS developed a virtual simulation to accelerate and cheapen machine learning process in autonomous cars.
ADAS recently developed a video game-like computer simulation called SYNTHIA that aims to accelerate and improve the way artificial intelligences learn to understand the city and its features such as vehicles, buildings, trees, roads and more importantly pedestrians.
Many worldwide companies including Google and Tesla are racing to launch the first autonomous driving cars in the market. However, in the current technology, teaching the artificial intelligence to recognize objects is very long and costly process. Project’s main researcher Germán Ros indicates that “This is great leap for the autonomous vehicles to teach them identify what happens around.  These vehicles use artificial intelligence to simulate the functioning of human neural connections. Our simulator, SYNTHIA , has a qualitative advantage in this process. SYNTHIA is presented as a new technology that makes the use of a virtual simulator for generating artificial intelligence easily and without human intervention. So far, information has been given by hand by different operators which is a very expensive process, because millions of images are needed. For example Tesla has 2.000 operators for this process. But our simulation removes this barrier in an easier and cheaper way”
The research have been published in CVC under the title of “The SYNTHIA Dataset: A Large Collection of Synthetic Images for Semantic Segmentation of Urban Scenes” and will be presented in Las Vegas in the International Conference on Computer Vision and Pattern Recognition.
Click to read the spanish news on Lavanguardia