A social-aware game-theoretic decision-making model for autonomous vehicle considering interaction styles

Autonomous vehicles encounter critical decision-making challenges in mixed traffic environments, particularly at unsignalized intersections, where the uncertainty of human driver intentions and multi-agent interactions lead to complex and potentially risky scenarios. To address these challenges, this paper proposes a social-aware game-theoretic decision-making model for left-turn scenarios at unsignalized intersections. Initially, an interaction parameter is introduced to estimate the social interaction styles of agents, based on an analysis of driving data extracted from a naturalistic driving dataset. Subsequently, a level-k model is employed to frame the decision-making problem, and through real-time estimation of the social interaction styles of other participants, the model accounts for multi-modal social interaction styles. The game-theoretic model is solved using Monte Carlo Tree Search method. Simulation results demonstrate that the proposed framework enables autonomous vehicles to accurately estimate human drivers’ social interaction styles and generate human-like, interactive decisions in unsignalized intersection scenarios. Compared to baseline methods, the approach improves safety and interpretability while adapting to diverse driving behaviors.