Graph neural networks for real-time optimization of autonomous urban transit systems

This paper proposes an AI-driven framework for optimizing Autonomous Public Transport Systems (APTS) in urban environments using a Spatio-temporal Graph Neural Network (GNN). Unlike traditional or heuristic methods, the proposed model leverages real-time traffic data, dynamic passenger demand, and energy constraints to achieve efficient and equitable routing of autonomous vehicle fleets. The urban transport network is modeled as a dynamic graph, enabling adaptive decision-making that accounts for spatial and temporal dependencies. Simulations using real-world urban topologies and synthetic traffic flows demonstrate that the GNN-based system outperforms rule-based and DQN baselines, reducing average passenger wait times by 56%, lowering energy consumption by 26%, and enhancing service equity across diverse urban zones. These findings highlight the transformative potential of AI-integrated APTS for sustainable and inclusive urban mobility, with significant implications for smart city planning and intelligent transport policy.