2026/02/10 by Chuan-Chi Lai, Chi Jai Choy
#cs.NI
Unmanned Aerial Vehicles (UAVs) provide pivotal on-demand wireless coverage for mission-critical 6G Air-Ground Integrated Networks (AGINs). However, traditional Deep Reinforcement Learning (DRL) orchestration struggles with multi-agent non-stationarity and balancing Energy Efficiency (EE) with service equity. To address these challenges, we propose ORCHID (Orchestration of Resilient Coverage via Hybrid Intelligent Deployment), a stability-enhanced two-stage learning framework. First, ORCHID utilizes Ground Base Station (GBS)-aware topology partitioning to mitigate the exploration cold-start problem. Second, a Reset-and-Finetune (R&F) mechanism within the Multi-Agent Proximal Policy Optimization (MAPPO) architecture enhances learning stability by synchronizing learning-rate decay with optimizer resetting, thereby reducing gradient variance and mitigating policy degradation. Furthermore, by formulating the resource allocation problem as an Egalitarian Bargaining Game (EBG), our theoretical analysis provides new insights into the relationship between fairness and energy efficiency. Specifically, the proposed Max-Min Fairness (MMF) design provides a theoretical explanation for the emergence of a more dispersed and load-balanced UAV topology, while experimental results further demonstrate that this spatial organization improves system energy efficiency compared with conventional Proportional Fairness (PF) schemes. Moreover, ORCHID deliberately sacrifices opportunistic throughput peaks in favor of more stable long-term service performance, resulting in consistently lower performance variance while maintaining a higher minimum service level and substantially improving service fairness. Extensive experimental results demonstrate robust topology adaptation, stable policy convergence, and consistent performance gains over representative state-of-the-art baselines.