There Is Now A Virtual Driving School For AI Vehicles
Scientists have developed a simulation program designed to teach autonomous cars how to handle real obstacles
Scientists from the Computer Vision Center in Barcelona have developed Synthia (Synthetic collection of Imagery and Annotations of urban scenario), a virtual city simulation that trains autonomous cars to recognize obstacles, signs, and other situations the software will encounter while driving on city roads.
Synthia’s breakthrough has to do with automatic annotation of images which need to be registered by driving AIs in order to be trained in recognizing them and reacting accordingly. Until now, this had to be done by hand, pixel by pixel, resulting in an expensive, painstakingly slow process that didn’t provide a complete solution to the problem.
The software functions like a videogame, complete with a fully designed city and changing seasons, and a virtual car that drives around capturing images that can be interpreted as real pictures of different situations. The data collected can then be transferred to neural networks—the software behind self-driving cars—in order to train AI vehicles.
Synthia looks like a promising step toward a future where driverless cars are commonplace.