Abstract
Unmanned Aerial Vehicles (UAV) are increasingly used in sectors such as smart cities, precision agriculture, disaster response, and last-mile logistics, with Multi-Access Edge Computing (MEC) playing a key role in enhancing their capabilities. In disaster response management, UAV assist in locating survivors, tracking objects, mapping post-disaster areas, and delivering critical supplies to inaccessible regions. However, unstable network conditions in disaster environments pose significant challenges to maintaining reliable video transmission and real-time decision-making. In this paper, we propose a comprehensive orchestration framework that integrates both offline and online strategies to optimize UAV video transmission, multi-UAV networking, and network management. The offline strategy combines policy-based orchestration with batch reinforcement learning (RL) to prepare UAV for deployment by optimizing network settings and video properties. The online strategy leverages reinforcement learning to enable real-time trajectory prediction and adaptive multi-UAV networking, ensuring efficient communication and decision-making during missions. Our experimental results, conducted across various Disaster Response Scenarios (DRS), demonstrate that the DQN-based approach significantly improves network throughput and round-trip time (RTT) compared to traditional methods, e.g. heuristic-based and rule-based, achieving approximately 87% of the Oracle baseline. The proposed framework enhances both the efficiency and adaptability of UAV operations, providing a robust solution for disaster response management.
