EcoDriveAI: Real-Time Smart Traffic Signal Control Using V2I and Deep Reinforcement Learning

Traffic jams in the modern urbanisation of the world have since become a daily ritual, especially with the influx of rural to urban migration and the increase in ownership of private vehicles. Conventional traffic signal systems that rely on fixed or semi-adaptive timing plans fail to respond dynamically to changing traffic patterns or critical events such as the arrival of emergency vehicles. In response to these limitations, we introduce EcoDriveAI, a real-time, edgedeployed intelligent traffic signal control framework based on Vehicle-to-Infrastructure (V2I) communication and Deep Reinforcement Learning (DRL). We have EcoDriveAI, a Deep Q-Network (DQN) agent trained in the SUMO environment to minimize vehicle delays, fuel consumption, and $\text{CO}_{2}$ emissions. It also gives clearance to emergency vehicles. Decentralisation based on a small edge layer based on Raspberry Pi Zero 2 W allows the architecture, however, to work without relying on server centres, as it has a V2I communication interface, internal emergency response logic, etc. The reward function is environmentally friendly, with regard to throughput, idling penalties, fuel efficiency, and emergency response performance. The detection of emergency vehicles is done via valid V2X (Vehicle-to-Everything) messages, triggering the signal preemption procedure where emergency vehicle(s) can pass freely with minimal disturbance to flow. Simulations show that from average delay to pollution and emergency reaction times compared to a conventional traffic control systems EcoDriveAI comes out well ahead. Its modular and scalable architecture ensures interoperability with existing infrastructure, facilitating widespread adoption of seamless integration into smart city applications. EcoDriveAI offers a seamless, low-latency, and green-powered solution for smart urban mobility management.