Multi-level signal-vehicle cooperative control to improve safety and efficiency for arterial intersections in...

Cooperative control of traffic signals and connected autonomous vehicles (CAVs) has shown significant promise in improving safety and efficiency at isolated intersections. Nevertheless, extending such strategies to arterial intersections introduces considerable complexity, requiring seamless coordination across multiple intersections and the dynamic control of heterogeneous traffic flows. This paper proposes a Multi-Level Signal-Vehicle Cooperative Control (ML-SVCC) system that integrates multi-agent reinforcement learning (MARL)-based traffic signal control (TSC) and CAV speed advisories to optimize safety and efficiency across the arterial network. In the proposed system, each intersection is modeled as a local agent that uses context-specific traffic conditions as state inputs and performance metrics as the reward function. A global agent, operating within a distributed framework, evaluates the collective performance of all local agents and refines their models to ensure coordinated control. By providing additional rewards, the central agent refines the local agents' models, ensuring coordinated and effective traffic control throughout the network. Additionally, the system features a speed control module embedded in CAVs that adjusts vehicle speeds to align with signal timings, promoting smooth and efficient traffic flow across the arterial network. Simulation results in a real-world arterial setting in Changsha City, China, show that the proposed system reduces traffic conflicts by 42%-54% and delays by 25%-57%, outperforming the traditional TSC, the MARL-based TSC, and the Green Light Optimal Speed Advisory. Furthermore, the system minimizes vehicle stops and the frequency of acceleration and deceleration, demonstrating robust performance as CAV penetration rates increase.