Building Intelligent UAVs for Non-Terrestrial Networks: Large Language Models, Vision and Reasoning on-board

Unmanned Aerial Vehicles (UAVs) are revolutionizing Non-Terrestrial Networks (NTNs) by serving as intelligent, connected platforms capable of real-time situational awareness and advanced analytics. This paper presents a cost-effective quadcopter system that integrates 5G connectivity, edge computing, and Artificial Intelligence (AI) to address critical challenges in NTN applications. Equipped with a 360-degree camera, on-board high-performance computing, and Large Language Models (LLMs), the system enables seamless object detection, situational analysis, and immersive operator interaction through Virtual Reality (VR) interfaces. Field tests demonstrate the system’s ability to process visual data with minimal latency while maintaining reliable communication over 5G networks. Integrating LLMs further enhances operational efficiency by providing contextual insights and filtering acquired data for actionable decision-making. Applications such as disaster response, infrastructure inspection, and environmental monitoring highlight the platform’s versatility in challenging environments. A video demonstration of the platform is available at https://youtu.be/F1fDpXw-kBg, illustrating its real-world performance.