Edge AI for Urban Traffic Signal Optimization: A City Council Case Study

Thesis Abstract

The rapid urbanization of metropolitan areas has amplified the inefficiencies in traditional traffic signal control, resulting in increased congestion, elevated emissions, and suboptimal incident response times; conventional fixed-time and actuated control approaches fail to adapt to real-time traffic dynamics and edge conditions. This study addresses the problem by evaluating edge artificial intelligence as a scalable solution for urban traffic signal optimization within a mid-sized city governed by a unified City Council. The aim is to develop and validate an edge-enabled traffic signal control framework that dynamically optimizes signal timing using locally collected sensor data to reduce total travel time, stops per kilometer, and emissions while improving emergency vehicle accessibility. Specific objectives include (1) designing an edge AI architecture that processes streamed data from 200+ intersections in real-time, (2) evaluating the performance of learning-based control policies against baseline OPC/SCATS-like systems under peak and off-peak conditions, (3) estimating the impact on travel time reliability and access for ambulance routes, and (4) assessing scalability, energy consumption, and privacy considerations of edge deployment. The study tests the following hypotheses H1 edge-based control significantly reduces average travel speed variance and stop counts compared with fixed-time and actuated controls; H2 adaptive edge policies yield measurable improvements in emergency route clearance times; H3 task offloading to edge devices does not compromise data privacy or system stability under high traffic loads. A mixed-methods research design combines quantitative simulations and field validation with qualitative stakeholder insights. The population comprises the city’s 7,000 active traffic signal events and 2,500,000 annual vehicle trips, with a stratified sample of 250 intersections representing arterial, collector, and local streets. Data collection integrates (i) high-resolution loop sensor, camera-derived vehicle counts, and GPS probe data, (ii) infrastructure telemetry on signal timing plans, and (iii) stakeholder interviews with traffic engineers and emergency services personnel. Instruments include a data fusion pipeline, a microservice-based edge runtime, a standardized signal-optimization benchmark, and a semi-structured interview guide. Validity and reliability are established through cross-validation with historical DNS-based metrics, test-retest reliability for timing decisions, and triangulation of sensor data. The data analysis plan employs a combination of time-series modeling, reinforcement learning (deep Q-learning and proximal policy optimization - PPO) for edge control policies, and comparative statistical analyses (paired t-tests, ANOVA, and non-parametric equivalents) to evaluate performance relative to baseline systems. A concurrent analytic framework processes real-time data at the edge, with cloud-based validation for longitudinal evaluation. Model specification includes an objective function minimizing weighted travel time, stops, and emissions, subject to safety constraints and urban planning policies. Ethical considerations address data privacy, system resilience, and potential bias in sensor data. Expected findings indicate that the edge AI framework achieves a statistically significant reduction in average travel time (10–15%), travel time variance (12–20%), and number of stops (15–25%) during peak hours compared with baseline controllers, with further improvements in emergency vehicle clearance times (8–12%). The edge approach demonstrates robust performance under sensor downtime and varying traffic patterns, aided by transfer learning and federated updates across intersections. Energy consumption on edge devices remains within hardware duty-cycle budgets, and privacy risks are mitigated through on-device processing and differential privacy techniques when ad hoc data sharing is necessary. The study contributes to knowledge by empirically demonstrating the viability of edge AI for municipal traffic optimization, clarifying practical deployment considerations, and offering a replicable framework for other cities with analogous regulatory and infrastructural contexts. The theoretical contribution aligns with the situated learning of adaptive control in urban environments and extends existing traffic theory by integrating edge-native reinforcement learning with real-time, privacy-preserving data fusion. The main conclusion posits that edge AI-enabled traffic signal optimization yields measurable efficiency gains, improves emergency response performance, and scales within a city’s governance and infrastructural constraints. Recommendations include (1) phased deployment prioritizing arterial corridors, (2) standardization of edge hardware and software interfaces across intersections, (3) a governance model for federated learning to maintain data privacy and system stability, and (4) ongoing evaluation protocols incorporating stochastic traffic modeling to sustain service levels under disruptive events.