How do human-driven and autonomous vehicles interact with vulnerable road users at unsignalized intersections? Evidence from a game-theoretic analysis

Understanding the interaction between vulnerable road users (VRUs) and vehicles is essential for improving safety in mixed traffic environments, particularly with the increasing presence of autonomous vehicles (AVs). At unsignalized intersections, VRUs and drivers must make decisions while anticipating each other's actions, leading to inherent uncertainty and interdependence in the interaction process. To address this challenge, this study develops a game-theoretic modeling framework that treats VRU crossing and driver yielding as a strategic interactive decision-making process. Using naturalistic trajectory data extracted from roadside video recordings, which involve both AVs and human-driven vehicles (HDVs), the model incorporates bounded rationality and belief updating within a quantal response equilibrium (QRE) framework. Model estimation results show that the QRE specification outperforms independent logit models, achieving higher log-likelihood values and lower information criteria. The results reveal several key behavioral insights. VRU observation plays a dual role in the decision process, initially reinforcing waiting behavior but facilitating opportunistic crossing under higher perceived conflict risk. Drivers adjust their yielding decisions based on the characteristics of VRUs, including socio-demographic attributes and crossing modes. AVs exhibit lower conflict probabilities than HDVs when interacting with pedestrians and cyclists, but higher conflict probabilities in interactions with scooters, indicating context-dependent differences in risk patterns across vehicle types. Overall, this study provides an empirically grounded framework for analyzing VRU-vehicle interactions in mixed traffic environments and offers new insights into the role of uncertainty, interaction, and heterogeneity in shaping crossing behavior.