Publication

Every year, more than 5000 pedestrians and 2000 cyclists die on European roads. These vulnerable road users (VRUs) are especially at risk when interacting with cars. Intelligent safety systems (ISSs), designed to mitigate or avoid crashes between cars and VRUs, first entered the market a few years ago, and still need to be improved to be effective. Understanding how drivers interact with VRUs is crucial to improving the development and the evaluation of ISSs. Today, however, there is a lack of knowledge about driver behaviour in interactions with VRUs. To address this deficiency and contribute to realising the full potential of ISSs, this thesis has multiple objectives: 1) to investigate and describe the driver response process when a VRU crosses the driver path, 2) to devise models that can predict the driver response process, 3) to inform Euro NCAP with new knowledge about driver interactions with crossing VRUs that may guide the development of their test scenarios, and 4) to develop a framework for ISS evaluation through counterfactual simulation and analyse the impact of the chosen driver model on the simulation outcome. The thesis results show that the moment when a VRU becomes visible to the driver has the largest influence on the driver’s braking response process in driver-VRU interactions. Data gathered in driving simulators and on a test track were used to devise different predictive models: one model for the pedestrian crossing scenario, and three for the cyclist crossing scenario. The model for the pedestrian crossing scenario can estimate the moments at which key components of the driver response process (e.g. gas pedal fully released and brake onset) happen. For the cyclist crossing scenario, the first model predicts the brake onset time and the second predicts the experienced discomfort score given the cyclist appearance time. The third predicts the continuous deflection signal of the brake pedal based on the interaction of two visually-derived cues (looming and projected post-encroachment time). These models could be used to improve the design and evaluation of ISSs. From the models, appropriate warning or intervention times that are not a nuisance to the drivers could be adopted by the ISSs, therefore maximizing driver acceptance. Additionally, the models could be used in counterfactual simulations to evaluate ISS safety benefits. In fact, it was shown that driver models are a critical part of these simulations, further demonstrating the need for the development of more realistic driver models. The knowledge provided by this thesis may also guide Euro NCAP towards an improved ISS test protocol by providing information about scenarios that have not yet been evaluated.