GAN-MADRL-Driven UAV-Assisted Vehicular Edge Computing for Resilient and Smart Transportation Systems in 6G Networks

Unmanned aerial vehicles (UAVs) integrated with vehicular edge computing (VEC) in 6G-enabled Intelligent Transportation Systems (ITS) offer a resilient platform to enhance urban mobility, road safety, and public transport efficiency. Acting as agile edge nodes, UAVs enable low-latency communication, distributed intelligence, and adaptive coverage in dynamic and infrastructure-limited environments. Nevertheless, challenges such as UAV energy constraints, intermittent connectivity, and heterogeneous data distributions persist. This paper presents a generative adversarial network (GAN)-enhanced Multi-Agent Deep Reinforcement Learning (MADRL) framework for UAVassisted VEC in 6G-enabled ITS. The proposed framework aims to minimize UAV energy consumption and network latency while ensuring reliable connectivity and accurate traffic prediction. Specifically, it jointly addresses traffic prediction, UAV path planning, and computational offloading (TPCO), formulated as a Markov Decision Process (MDP). UAVs, vehicles, and edge servers operate as decentralized agents that learn collaboratively to optimize latency, energy consumption, and connectivity. GAN modules enrich sparse and heterogeneous datasets, thereby improving training robustness and convergence. The simulation results indicate that the proposed GAN-MADRL framework achieves about 30% better traffic-prediction accuracy, 35% higher energy efficiency, and 40% lower latency compared to CDL, FL, PPO, and Greedy baselines. Keywords: Generative Adversarial Networks, Vehicular Edge Computing, Intelligent Transportation Systems, UAV trajectory optimization, Traffic prediction