Driverless cars, unsurprisingly, drive in a computational manner: they rely on pre-analysed routes using complex 3D-scanned maps of the area that enables them to compute the route before setting out on a journey. The problem with this approach is that a driverless car will not navigate very successfully on the fly.
Human drivers, on the other hand, are really good at looking at our environment and matching it up to the sat nav device in front of us. MIT researchers have decided to teach driverless cars to drive like us. They have developed an autonomous control system that studies human steering patterns while learning to read data from video camera feeds and a simple GPS-like map to simulate the information someone would be working with as they drive.
"With our system, you don't need to train on every road beforehand," says MIT researcher behind the project, Alexander Amini. "You can download a new map for the car to navigate through roads it has never seen before.
"If we train an autonomous vehicle to drive in an urban setting such as the streets of Cambridge, the system should also be able to drive smoothly in the woods, even if that is an environment it has never seen before," added study co-author Daniela Rus.
http://news.mit.edu/2019/human-reasoning-ai-driverless-car-navigation-0523
