Analyzing human-driven vehicles’ following behavior in response to autonomous vehicles under different traffic conditions

The car-following behavior of human-driven vehicles (HDVs) in traffic flows mixed with autonomous vehicles (AVs) is highly complex, shaped by macroscopic traffic conditions, microscopic vehicle interactions, and drivers' subjective perceptions and driving habits. Analyzing how HDVs follow AVs in such context is critical for enhancing the efficiency of mixed traffic systems. This study employed field experiments and questionnaire surveys to analyze car-following behavior under the combined influence of traffic conditions and leading vehicle types. A machine learning-based car-following model was developed, incorporating both internal (driver-related) and external factors. Simulations confirmed the feasibility of the developed model for mixed traffic flows. Furthermore, strategies to enhance the efficiency of mixed traffic were proposed, with a focus on AV penetration rates and drivers’ trust in AVs. The results indicate that traffic conditions and leading vehicle types jointly influence drivers' AV-following behavior. The efficiency of mixed traffic first decreases and then increases as AV penetration rises, with critical thresholds of 70 % and 60 % observed under uncongested and congested conditions, respectively. Additionally, higher driver trust in AVs leads to greater traffic efficiency in mixed platoons. The developed model—integrating both internal and external (traffic condition/vehicle type-related) factors—achieved fitting accuracy exceeding 93.27 % for speed and 89.14 % for acceleration, outperforming both machine learning models that neglect internal factors and the Intelligent Driver Model (IDM). The findings present a reproducible car-following simulation model for mixed traffic across diverse conditions, alongside management strategies to boost efficiency and reduce congestion. Collectively, this work offers theoretical and model support for simulating and optimizing car-following behavior in mixed traffic scenarios.