• Lightweight, Edge-Friendly Adaptive Traffic Control: The proposed algorithm is a lightweight, real-time adaptive traffic signal control system designed to run efficiently on edge devices like Raspberry Pi. It leverages real-time traffic data from map services, removing the need for complex infrastructure while allowing quick deployment in dynamic urban environments. • MAB-Driven Traffic Signal Intelligence: At its core, it uses a Multi-Armed Bandit (MAB) model to adjust green signal timings. Unlike heavier deep learning models, MAB offers a simple, fast, and adaptive method that learns from traffic patterns on the go, making it ideal for real-time decision-making with minimal computational load. • Digitally Driven, Effortlessly Scalable, and Infrastructure-Light: By relying entirely on map-based traffic data, the system avoids the need for physical sensors or cameras. This sensorless design not only reduces installation and maintenance costs but also improves scalability, especially in large, complex road networks like those found in countries such as India. • Sustainability Through Smarter Traffic Flow: Optimizing: Signal timings reduce vehicle idle time and improves intersection throughput, which in turn cuts fuel consumption and carbon emissions. The proposed system supports sustainable urban mobility while also enhancing economic efficiency through smoother traffic movement. Handling urban traffic efficiently is a critical challenge for current cities, demanding innovative solutions to tackle mounting congestion. This paper proposes an edge device-oriented, lightweight, sensorless traffic control algorithm that utilizes real-time data from mapping services to dynamically optimize green signal timings. By eliminating the need for physical sensors, the proposed approach is not only cost-effective but also easier to implement, scale, and maintain. The proposed algorithm operates in two stages: it first computes real-time congestion intensity by comparing real-time estimated arrivals with free-flow estimated arrivals to refine green times. In the next stage, a multi-armed bandit approach using reinforcement learning is employed to iteratively optimize these timings further, ensuring adaptive and effective traffic management. Tested with real-world data from Kaloor City, Ernakulam, India, and validated through simulations in the SUMO traffic environment, the proposed algorithm demonstrated exceptional performance in optimizing traffic flow and intersection efficiency, significantly surpassing established benchmark methodologies based on deep learning, swarm intelligence, and evolutionary algorithms. Specifically, the proposed algorithm achieved a remarkable 33.86% increase in intersection throughput, a 35.60% reduction in vehicular density, and a 35.30% decrease in travel time. Beyond throughput optimization, the algorithm demonstrated superior efficacy in mitigating vehicle queues and minimizing CO₂ emissions, achieving frequent reductions to levels below 100,000 mg/s, with significantly lower computational complexity. A hardware prototype of this system has been successfully developed and validated using an edge device. These findings support the practicality of deploying the proposed algorithm on edge infrastructure for sustainable, real-time urban traffic optimization.
K et al. (Sun,) studied this question.