AI-Powered Traffic Management Systems for Kigali’s Urban Mobility

Abstract

Kigali’s rapid growth has intensified congestion and degraded air quality along major corridors. This study designed and validated an AI-powered traffic management framework that fused multi-source real-time data—CCTV/YOLO counts, IoT detectors, GPS travel times, and near-road air-quality sensors—to optimize signal control in line with the City Development Strategy (CDS) smart-mobility goals. A Long Short-Term Memory (LSTM) model generated reliable short-term flow forecasts (R² = 0.91; RMSE = 8.6 vehicles/interval), which were used to train a Deep Q-Network (DQN) controller to adapt phase splits, cycles, and offsets across complex junctions. Over a six-month evaluation on priority corridors (e.g., CBD–Remera, Nyabugogo–Kacyiru), the AI system reduced average control delay by ~29%, increased intersection throughput by ~36%, and lowered corridor travel time by ~30%. Environmental co-benefits were observed, with fuel use declining by ~26% and near-road CO₂ and PM₂․₅ concentrations decreasing by ~25–30% during peak periods. These gains persisted across rainy conditions and demand variability and were confirmed by paired statistical tests and robustness checks. The results demonstrate deployment-ready potential: corridor-level coordination, incident-aware operating playbooks, and deep bus-priority integration (including headway stabilization) can be operationalized within the city’s control center. We outline a phased scale-up path to network coordination and propose aligning the controller with electric-bus expansion and charging strategies to maximize mobility and air-quality benefits. While coverage and sensor limitations remain, the evidence indicates that data-driven, adaptive control can materially advance Kigali’s goals for a smart, green, and resilient urban transport system.