In SAE Level 3 automated driving, safe vehicle control must be ensured until the driver takes over the driving task. This means that critical situations that require the driver to take over must be detected or predicted early and reliably. With an assumed takeover time of 10 s, a large number of driving decisions and maneuvers may be necessary, especially in very dynamic traffic situations, which must be taken into account by an automated vehicle.

The KISSaF project aims to develop an AI-based situation prediction for the entire scene over a long period of time. By coupling the prediction with the trajectory planning of the ego vehicle, safety is to be increased and automated driving from SAE level 3 is to be made possible.

First, a generic and scalable environment representation will be built during the project, which includes the dynamic and static objects of the scene as well as the infrastructure and the situational context. On this basis, machine learning methods will be used to determine possible scene trajectories that, reflected back to the vehicle, will support its own action planning. The basis for the development and training of the AI algorithms is a specially recorded data set. The results developed for the environment model and for situation prediction are implemented, demonstrated and validated for highway and urban scenarios.

Homepage