Research on potential conflict resolution for CAV at unsignalized intersections in human-machine mixed driving environments

At unsignalized intersections, mixed traffic scenarios involve Connected and Autonomous Vehicles (CAVs) and Human-Driven Vehicles (HDVs) safety deficiencies and poor interactivity. To address this, this paper proposes a hierarchical conflict resolution framework integrating machine learning predictions: at the upper layer, potential conflict CAV alliances coordinately make decisions through cooperative game; at the lower layer, CAV alliances act as leaders engaging in Stackelberg game with HDVs. By designing a three-dimensional utility function encompassing safety, efficiency and comfort, the framework unifies individual optimization with system-wide optimization. Simulation results based on the INTERACTION Dataset demonstrate that this method significantly enhances traffic safety and travel efficiency compared to baseline strategies. It exhibits good computational feasibility and scalability potential, offering new insights for intelligent driving decision-making in mixed traffic environments.