Jiawen Wang, Haolong Ma, Chengcheng Yang
This study proposes a game-theoretic modeling framework to quantitatively evaluate the safety and operational impacts of external human–machine interfaces (eHMIs) in mixed traffic conditions at unsignalized crosswalks. With the increasing prevalence of autonomous vehicles (AVs), the lack of driver–pedestrian communication introduces greater uncertainty in right-of-way negotiations. Unlike previous studies that focus on fully automated environments, this research addresses the mixed-traffic context in which human-driven vehicles (HDVs), AVs, and AVs equipped with eHMIs (AV-eHMIs) coexist. A unified interaction model is developed to characterize the dynamic interactions between pedestrians and three types of vehicles: HDVs, AVs, and AV-eHMIs. The model incorporates a game-theoretic mechanism to simulate right-of-way negotiations and is calibrated using aerial video data collected in Shanghai. Numerical simulations are conducted under varying traffic demand levels and different penetration rates of AVs and eHMIs, evaluating key performance indicators such as travel delay, collision risk, and congestion risk. The results indicate that while HDVs are associated with relatively lower travel delays, they account for a high share of time spent in risky interactions (94.3%). Conversely, AVs alone are insufficient to significantly reduce this risk due to their limited ability to explicitly communicate intent. However, the introduction of AV-eHMIs reduces the share of time in risky interactions by 77%, though at the expense of increased vehicle delay under high pedestrian volumes. Based on this foundational assessment, the study further proposes an efficiency-oriented interaction strategy. This strategy introduces time-varying efficiency benefit coefficients, enabling AV-eHMIs to proactively manage and temporarily interrupt pedestrian flows when appropriate, thereby achieving a more effective balance between safety and traffic efficiency.