Sistema de semáforos inteligentes para la regulación de tránsito vehicular

Abstract

En movilidad urbana, los sistemas de semáforos inteligentes ajustaron fases en tiempo real para reducir demoras y mejorar la confiabilidad del viaje. Este trabajo implementó un sistema de semáforos inteligentes basado en aprendizaje por refuerzo multiagente usando el algoritmo multi-agent proximal policy optimization (MAPPO). Se construyeron cuatro escenarios representativos de la Ciudad de Guatemala, parametrizados con criterios de ingeniería de tránsito. La política se optimizó con una función de recompensa cuyo objetivo era minimizar acumulación de retraso. Los resultados mostraron convergencia estable y mejoras significativas: el tiempo de viaje promedio descendió en 73.25 %, el tiempo de espera promedio se redujo en 72.52 % y la velocidad promedio aumentó 13.57 %.

In urban mobility, Intelligent Traffic Signal Systems adjust phases in real time to reduce delays and improve travel-time reliability. This work implements an Intelligent Traffic Signal System based on Multi-Agent Reinforcement Learning using the MAPPO algorithm. Four representative scenarios of Guatemala City (Bulevar Los Próceres, Avenida Elena–Anillo Periférico, Calzada Atanasio Tzul, and Avenida Reforma) were constructed and parameterized with traffic-engineering criteria. The policy was optimized with a reward function aimed at minimizing the accumulation of delay in the city’s traffic. In the results, the trained models exhibit stable convergence and improvements over current vehicular coordination setups. Specifically, averaging across the four environments, average travel time decreases by 73.25 %, average waiting time decreases by 72.52 %, and average speed increases by 13.57 %. In conclusion, the Intelligent Traffic Signal System designed with MAPPO achieves strong performance across the evaluated environments because it captures corridor dependencies and mitigates multi-agent non-stationarity, prioritizing the dissipation of queues over peak speeds. This translates into longer green-wave progressions and fewer blockages, suggesting deployment in corridors with sufficient signal density and traffic demand suitable for Intelligent Traffic Signal Systems in Guatemala City.