Sensor-Fusion Driven Deep Reinforcement Learning for Dynamic Traffic Signal Optimization and Emergency Routing

Traditional fixed-time traffic signal controllers fail to accommodate the high-variance nature of dynamic urban traffic, leading to severe intersection congestion and critical delays for emergency responders. This paper presents VisionX, a sensor-fusion-driven Deep Reinforcement Learning (DRL) architecture designed for real-time traffic signal optimization and automated emergency vehicle routing. We model the traffic intersection as a Markov Decision Process (MDP) and employ a Deep Q-Network (DQN) with experience replay to dynamically calculate optimal phase durations based on live queue lengths. These queue metrics are extracted via edge-deployed YOLO object detection. Evaluated using the Simulation of Urban Mobility (SUMO) platform under the Krauss car-following model, our DRL agent achieved a 40% reduction in total cumulative wait time compared to standard 30-second fixed-cycle controllers. Furthermore, the integration of a bidirectional Node.js and React synchronization bridge demonstrates the system's viability for seamless Edge-to-Cloud monitoring and manual override preemption for emergency infrastructure.