CM3-VSL: Cooperative Multi-goal Multi-stage Multi-agent VSL Traffic Control

Variable speed limit (VSL) Systems play a crucial role in proactively optimizing traffic control. As a matter of fact, many countries have deployed VSL Systems to improve road safety and resolve traffic breakdown. Most of smart VSL strategies are deployed to optimize traffic flow within a single road segment only, while real-world scenarios often involve complex bottleneck situations arising from multiple ramps. In response, we introduce a novel Cooperative Multi-goal Multi-stage Multi-agent VSL (CM3-VSL) framework where a diverse set of VSL agents collaboratively work towards both individualized local goals and shared global objectives, addressing the complexities of real-world traffic scenarios. The VSL agents are trained using micro-simulations on a real-world Moroccan highway network. Employing a cooperative strategy, each VSL agent pursues both individual and collective goals. Evaluation against a baseline no-VSL scenario and a single-agent multi-objective Reinforcement Learning VSL demonstrates that CM3-VSL achieves superior performance, contributing to advancements in intelligent traffic control systems.