Adaptive Urban Traffic Control Multimodal with AI and Visible Light Communication (VLC)

This work addresses multimodal adaptive urban traffic control by integrating artificial intelligence with Visible Light Communication (VLC), leveraging public LED lighting infrastructure as a dual-purpose system for illumination and data transmission. This approach enables vehicles, public transport, and other road users to receive critical real-time information on traffic signal states and road conditions, reducing dependency on congested radio frequency networks. The proposed framework explicitly targets multimodal traffic environments, incorporating heterogeneous agents such as private vehicles, buses, cyclists, and pedestrians. Using Multi-Agent Reinforcement Learning (MARL), particularly Deep Q-Learning and Multi-Agent Proximal Policy Optimization, each intersection operates as an intelligent agent. These agents continuously monitor queue lengths, modal priorities, and accumulated waiting times, collaboratively optimizing signal control policies to minimize overall network delay while ensuring equitable treatment across transport modes.A key contribution is the introduction of a dynamic bus lane management strategy, which adapts lane allocation in real time to prioritize public transport flow without significantly degrading the performance of other modes. Additionally, the system integrates pedestrian-aware perception mechanisms, promoting safer crossings and balanced decision-making in complex urban scenarios. Overall, the proposed architecture advances toward a holistic, multimodal traffic management paradigm, improving efficiency, sustainability, and safety in next-generation smart cities.