Fraunhofer Institute for High-Speed Dynamics, Ernst-Mach-Institut, EMIOne third of road fatalities are pedestrians. To improve their safety, Fraunhofer EMI is developing traffic simulations that integrate detailed pedestrian models for the first time.
Statistics from the European Commission report 7,665 fatal road accidents in urban areas in the EU in 2021. 31 % of the victims were pedestrians who collided with motorized vehicles. By 2023, the number of fatal accidents in urban areas had risen to 7,807, with the proportion of pedestrians increasing to 33 %. At the same time, the growing variety of modes of transport and increasing infrastructure congestion are making road safety more difficult. Autonomous vehicles are intended to better protect particularly vulnerable road users such as pedestrians and minimize human error in critical situations. In order to develop such systems, extensive test data from around 2.1 billion kilometers is required to ensure that each relevant traffic situation occurs at least once with a probability of 50 %. This problem is tackled using synthetic data from simulation environments. Until now, the focus of microscopic traffic simulations has been on motorized road users. However, a realistic representation of pedestrians is crucial, especially in cities. Fraunhofer EMI is closing this gap and has integrated improved modeling of pedestrians into the traffic simulation as a first step. In doing so, it is using its experience from the development of agent-based simulations for crowds at major events. These simulations already take into account the interaction between individuals and their reactions to obstacles, whether stationary or moving. They also model pedestrians with individual characteristics, such as goals, needs and the willingness to take risks.
Finally, optimization algorithms developed at EMI can be used for integrated behavioural models, which ensure an optimal choice of model parameter values based on a statistical comparison with real data. In addition to the classic rule-based simulation algorithms, EMI is also researching the use of AI methods. In particular, the question of whether decision-making processes, such as crossing roads, can be realistically predicted using reinforcement learning algorithms is being investigated.
