SA-MAGPPO: Security-Aware Multi-Agent DRL for Edge-Assisted Public Transit Systems in Low-Altitude Intelligent Transportation Environments

Abstract

Public transit systems play an important role in reducing traffic congestion, energy consumption, and greenhouse gas emissions in smart urban transportation environments. However, large-scale transit operations still suffer from inefficient routing, scheduling, and resource management decisions under highly dynamic traffic and passenger demand conditions. Furthermore, existing intelligent transportation approaches often ignore communication unreliability, anomalous transportation observations, computation offloading overhead, and network congestion in low-altitude intelligent transportation systems. To address these challenges, this paper proposes a secure and resilient UAV-assisted edge-enabled transit offloading framework based on Security-Aware Multi-Agent Proximal Policy Optimization (SA-MAGPPO). In the proposed framework, UAVs operate as low-altitude communication assistants that enhance V2I and V2V connectivity. These UAVs provide aerial traffic observations and reduce communication congestion in dense urban transportation networks. The proposed framework jointly optimizes transit routing, charging scheduling, fleet management, and computation offloading decisions between onboard units and edge servers. Unlike conventional MAGPPO, the proposed framework augments the agent state representation with communication reliability and anomaly information and incorporates security-aware policy learning to improve robustness against unreliable traffic observations, communication disruptions, and unstable network conditions. Extensive simulations under multiple operational scenarios demonstrate that the proposed framework consistently outperforms RP, GS, RBH, SPPO, and conventional MAGPPO approaches in terms of energy, operational efficiency, offloading reliability, and robustness against communication anomalies.