KI-basierte Crasherkennung für das Sichere Automatisierte Fahren (KICSAFe)

The use of artificial intelligence (AI) methods offers great potential for quickly classifying complex situations and deriving recommendations for action. This makes AI methods highly interesting for use in the field of automated driving as well as in vehicle safety. However, a major challenge here lies in validation, traceability and trust in AI, as wrong decisions must be avoided at all costs. This is why vehicle manufacturer BMW, system supplier Continental and the C-ISAFE research institute at Ingolstadt University of Applied Sciences are conducting research in the new joint research project "AI-based crash detection for safe automated driving (KICSAFe)" into the possibility of predicting unavoidable accidents using AI and triggering safety systems before a collision occurs. The project, with total costs of €1.2 million, is scheduled to run for three years and is being funded by the Bavarian State Ministry of Economic Affairs, Regional Development and Energy.

The prediction of accidents is based on the precise and reliable detection and prediction of road users and other objects in the vehicle environment. Camera-based sensor systems in particular enable cost-effective, complete environment detection and interpretation. For robust detection of unavoidable collisions, algorithms must be developed that enable prediction of the accident constellation and, from this, estimation of the accident consequences, regardless of the object properties of the collision partner. Pole-like objects such as trees, street lamps or bollards must be recognized just as reliably as other cars or trucks. In KICSAFe, the focus is on the particularly critical accident scenarios of pole impacts and truck underride collisions, which are also difficult for sensors to detect. Methods are being investigated for crash detection as well as for determining the situation and unavoidability, which enable the traceability of system decisions. Together, BMW, Continental and the C-ISAFE research institute want to use their expertise and the available test options to bring safe automated driving one step closer through the use of AI in safety-critical functions.