The on-board computer needs more than just information on the dimensions of the vehicle, signage and traffic management. It also needs reliable data about the probable behavior of other road users. Researchers at the Uni Ulm have reported that they have improved this movement forecasting with an AI procedure. Previously, this only worked for about one second, but thanks to the researchers in Ulm, correct predictions can be made for three seconds. To do this, artificial images of the vehicle’s environment are generated. A neurological network is then trained to derive probability-based assessments of future movements from these images. The scientists from Ulm have already won a competition in the USA with this procedure.

The research is casting a light on the huge challenges that still exist in autonomous driving. The ADAC reported that five gigabytes of data were processed in a car per minute as a basis for decisions on automatic driving actions during a VW fleet test. And that’s just on a test route: in the future, these vehicles will not only have to master test routes, but all possible traffic scenarios in the entire world, too.