University of California Institute of Transportation Studies

As more automated vehicles (AVs) gradually appear on our roads, they must be able to safely interact with human drivers as well as existing infrastructure designed with human drivers in mind. Current car-following computer models—which determine how AVs adjust their speed and position relative to other vehicles—often struggle to replicate human drivingpatterns. This deficiency could lead to unpredictable AV behavior, potentially increasing crash risks, disrupting traffic flow, and creating problems at traffic lights and intersections designed for human drivers. If AVs brake much earlier or later than humans, drivers may be caught in ‘dilemma zones’ — unable to safely stop or proceed through the intersection. To address these challenges, we conducted a comprehensive analysis of existing car-following models and developed a novel multi-phase projection-based model that ensures safety while exhibiting human-like driving characteristics. This model is called ‘projection-based’ because it anticipates future braking scenarios rather than just reacting to current conditions.