Assessing the crash avoidance potential of cut-in crashes

Abstract

Road traffic crashes remain a major safety concern. To reduce their occurrence, automated driving functions (ADFs) are being developed—but these systems must be thoroughly validated. Virtual simulations have been used to assess their performance, often comparing results between different systems without establishing a benchmark for what is theoretically achievable. This study focuses on cut-in crashes, a relatively common and challenging scenario for automated driving systems. It aims to assess the proportion of such crashes that could theoretically be avoided through braking alone and to compare this benchmark with the performance of two reference driver models and an ADF. This was achieved by developing an idealized model that reacts earlier and brakes harder than realistically possible, ensuring that no other model should be capable of outperforming it. This ideal model as well as the reference driver models and the ADF were then applied in virtual counterfactual simulations to estimate the proportion of crashes they could avoid. The cut-in scenarios simulated were categorized as either frontal or non-frontal cut-ins. The study was able to establish an upper limit for the non-frontal cut-in crashes but not for the frontal ones, as the ideal model avoided all frontal collisions. The two reference driver models avoided 38.5% and 81.8% of the frontal crashes, respectively, illustrating a large discrepancy. It remains unclear whether this reflects differences in the modeled driver’s behavior or limitations in how well the models represent real human drivers.