Assessment of traffic safety interventions using virtual randomized controlled trials: potential of connected and automated driving including V2X for collision reduction at urban intersections

Safely traveling through urban traffic, for example performing a right turn at an intersection, is a complex task for human drivers. Multiple stimuli such as leading vehicles, cyclists or spontaneously crossing pedestrians compete for attention. Limited field of view and sampling capabilities of humans make a parsimonious and sometime error-prone selection of most relevant information necessary, thus contributing to the occurrence of collisions. Especially with regard to vulnerable road users, such incidents still pose a high risk to life and health. A potential solution to compensate for human limitations are intelligent driver assistance functions up to connected and automated driving, which enable a continuous sampling of the surrounding environment. However, given the characteristics of current urban infrastructure, these systems will be confronted with frequent physical obstructions, e.g. due to construction, delivery vehicles, parked cars, vegetation or bus stops. These sensor occlusions likely lead to ambiguous situations for decision algorithms because of an only partially observable traffic environment. In this regard, external sensors delivering information from a different point of view via V2X could support the system's decision-making confidence. Since the implementation of such technology can consume considerable resources, especially if installed at every urban intersection, an estimation of the specific benefit (given the characteristics of intersections) is reasonable to support decision makers. The present article demonstrates a methodological approach to quantitative assessment of safety efficacy of new technologies by the technique of virtual randomized controlled trials. Thereby, the main focus in the scope of this document is to discuss requirements and the general process of this technique. It will be discussed how the statistical power is maximized and which sub-processes need to be modeled in order to come up with valid predictions. First simulation results will be presented and discussed with respect to their validity based on the implemented sub-process models.