Abstract
Autonomous-vehicle (AV)-dedicated lanes may support AV adoption and improve traffic operations and sustainability. However, most existing studies evaluate their effects under idealized AV behavior, which may not represent real-world operations. This study evaluates the operational, energy, and economic impacts of AV-dedicated lanes using a data-informed microscopic simulation framework. An imitation learning model is trained using trajectories from AV datasets to represent realistic AV car-following behavior, while heterogeneous human-driven vehicle behavior is carefully calibrated from large-scale naturalistic driving datasets. These models are implemented in a map-calibrated simulation of the SR-210E freeway corridor. Results show that increasing AV penetration does not necessarily improve traffic performance. AV-dedicated lanes provide the greatest benefits during the early transition stage, particularly at 6%–12% AV penetration. This study highlights the timely need to reassess their practical deployment as AV services move from hypothetical scenarios into real traffic systems.