Focus area II: Prediction

Predicting the vehicle’s surroundings is a prerequisite for all planning processes in an automated vehicle, as it must navigate in this environment safely and provide passengers with a comfortable ride. This prediction must be carried out in a constantly evolving traffic environment.
The goal of the focus area Prediction in Beyond Validation AI is to predict the future motions of road users and to model, reduce, quantify, and validate the uncertainties associated with these predictions. The focus is on both aleatoric uncertainties resulting from the stochastic behavior of road users and epistemic uncertainties caused by incomplete knowledge, limited data coverage, or model limitations. Strategies for reducing uncertainty by incorporating contextual information and through hybrid architectures that offer physically grounded guarantees are researched. Another aspect is the use of long-term traffic contexts to reduce uncertainty. It is analyzed how long-term observations, information on driving styles, and the prediction of long-term behavioral intentions can improve the quality of environmental prediction. Furthermore, the focus area Prediction also addresses human behavior as a source of uncertainty. To this end, behavioral models are being developed that better capture the cognitive and physical characteristics of human road users. The focus is on both drivers and vulnerable road users, with the goal of improving behavior prediction in mixed traffic and supporting more effective interaction between humans and automated systems.