Adaptive Traffic-Signal Control Using Reinforcement Learning for Congestion and Emission Reduction: An Illustrative Urban Mobility Study

This article presents a simulated empirical study in transportation engineering, urban planning, and artificial intelligence. The experiment covered 180 simulated operating hours and approximately 1.4 million vehicle movements across all scenarios. A simulated 4-by-4 urban road network with sixteen signalized intersections was modeled under morning peak, evening peak, off-peak, and incident conditions. Fixed-time, actuated, and reinforcement-learning signal controllers were compared. Vehicle arrivals included private cars, buses, two-wheelers, and delivery vehicles. Outcomes included average delay, queue length, stops, travel time, bus delay, and a speed- profile-based emissions index. Reinforcement-learning control reduced simulated peak-hour average delay by 23.8% relative to fixed timing and by 12.1% relative to actuated control. The emissions index fell by 9.6% because fewer stop-start cycles occurred. Performance weakened during the untrained incident scenario, showing the importance of robustness and safe fallback timing plans. All numerical values and outcomes are synthetic and are used only to demonstrate a plausible journal-style research design. No real participants, companies, cities, laboratories, utilities, or field sites were studied. The article combines a distinct analytical structure with topic-specific metrics, visuals, and limitations so that the five papers in the issue remain independent in subject and presentation.