RSU-assisted V2X message fusion via PC5 and Uu with actor–critic modeling for autonomous driving under intersection scenario

Abstract

Autonomous driving systems rely heavily on effective data fusion from Vehicle-to-Everything (V2X) networks, where accurate decisions depend on integrating diverse messages from multiple communication interfaces. However, current single-interface communication methods, either PC5 or Uu, limit the achievable autonomy level due to insufficient reliability and situational awareness. To address these limitations, this paper proposes an efficient RSU-centered Message-level fusion framework tailored for intersection-based autonomous driving. The proposed approach strategically assigns CAM, CPM, and SPATEM to the PC5 interface, while DENM and MAPEM are assigned to the Uu interface. A confidence-weighted fusion algorithm is implemented at the RSU aligns timestamps, filters inconsistent inputs, and resolves conflicts to generate unified situational awareness messages every 100 ms. The onboard decision-making model employs a CNN–GRU enhanced Actor–Critic network to optimize decisions for intelligent lane changing, collision avoidance, and traffic flow management. Simulation outcomes confirm that the proposed dual-interface fusion significantly enhances performance compared to single-interface systems, improving the packet delivery ratio to 0.75 at 300 m and achieving decision accuracy improvements of approximately 14–25% across key use cases. Consequently, our framework meets the criteria for autonomy sub-level L4-C, providing a robust foundation for advanced intersection-based autonomous driving systems.