EL-QORAYCHY, FATIMA-ZAHRAE (Université de technologie de Belfort Montbéliard (UTBM)), DRIDI, Mahjoub (Université de Belfort Montbéliard), Créput, Jean-Charles (Université de technologie de Belfort Montbéliard (UTBM))

Optimizing Traffic Flow at Unsignalized Intersections Via Centralized Reinforcement Learning for Connected and Autonomous Vehicles

Scheduled for presentation during the Invited Session "S24b-Traffic Control and Connected Autonomous Vehicles: benefits for efficiency, safety and beyond" (TH-EA-T24), Thursday, November 20, 2025, 14:50−14:50, Coolangata 3

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Intersection management is a key challenge in the development of intelligent transportation systems, especially with the rise of Connected and Autonomous Vehicles. Traditional control strategies, such as traffic lights or static priority rules, often fail to adapt to dynamic and dense traffic conditions, resulting in increased delays and suboptimal traffic flow. In this paper, we propose a centralized intersection management framework based on deep reinforcement learning (DRL), where a central agent directly controls the accelerations of vehicles approaching an unsignalized intersection. Leveraging spatial and temporal real-time information on vehicle states and intended trajectories in different traffic flows, the agent learns continuous control policies that ensure optimal navigation in real-world scenarios. We implement the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm to achieve stable and effective learning in a continuous action space. Simulation results using the SUMO traffic simulator demonstrate that our approach significantly outperforms traditional baselines such as fixed-time traffic lights, First-Come-First-Served and DRL-based approaches, such as Soft Actor-Critic (SAC). In high-density scenarios, TD3 reduces average waiting times by up to 87%, improves throughput by 14.5%, and shortens travel times by 28.2%. These results highlight the potential of our approach to enhance urban traffic flow and support future smart transportation systems.