A style-aware belief-based game model for automated-vehicle discretionary lane-changing decisions in mixed traffic

Discretionary lane changing in mixed traffic requires automated vehicles to account for uncertain responses from surrounding human-driven vehicles, particularly the target-lane rear vehicle. This study proposes SADP-QRE, a vehicle-level style-aware belief-based game model for automated-vehicle lane-changing decisions, integrating style-conditioned behavioral learning, online opponent-type belief updating, TTC-based risk assessment, and quantal response equilibrium. Naturalistic NGSIM trajectories are used to learn style-conditioned lane-changing and rear-vehicle response tendencies. During decision making, observed rear-vehicle motion updates the belief over latent opponent types, and the belief-aware payoff matrix generates probabilistic lane-changing decisions. The model is validated on held-out NGSIM LC/LK samples and evaluated in SUMO against expected-utility and leader–follower benchmark models. The trajectory-level validation yields ROC-AUC values of 0.764–0.782 and PR-AUC values of 0.713–0.749. In simulation, SADP-QRE reduces mean time loss by 3.71%–7.78% in a straight-freeway scenario and by 11.32%–22.24% in a merge–weave scenario, without reducing throughput.