Multi-agent Decision-Making Framework Based on Value Decomposition for Connected Automated Vehicles at Highway On-Ramps

Recognition of the necessity of connected and automated vehicles (CAVs) in transportation systems is gaining momentum. CAVs can improve the transportation network efficiency and safety by sharing information and cooperative control. This article addresses the problem of coordinating CAVs at highway on-ramps to achieve smooth traffic flow. We develop a multi-agent reinforcement learning (MARL) method based on value decomposition and centralized control to coordinate CAVs. The simulation results show that the proposed collaborative decision-making framework can effectively coordinate dynamic traffic flows and improve the metrics by more than 10% compared to the baseline methods under high traffic demand scenarios.