AI Fusion of Vision and Management for 6G Millimeter-Wave Aerial Networks

This paper aims to enhance beam management for 6G aerial communications by fusing real-time vision data with intelligent beam selection in dynamic urban environments where buildings and mobility-induced blockages frequently obstruct line-of-sight (LoS) connectivity. We propose an artificial intelligence-driven fusion of vision and beam selection (AI-FVBS) framework that integrates YOLOv8-based object detection, reconfigurable intelligent surfaces (RIS), and contextual multiarmed bandit (CMAB) learning to optimize beam selection for aerial platforms such as high-altitude platform stations (HAPS) and uncrewed aerial vehicles (UAVs). Aerial platforms improve 6G wireless network throughput, reduces blockages, and improve LoS transmission. Unlike conventional channel state information (CSI)-based methods, the proposed system leverages real-time visual context to dynamically balance beam exploration and exploitation, enhancing robustness in dense urban environments with frequent blockages. Vision-assisted RIS further extends the communication range when direct visibility is lost. Mounted vision-assisted RIS on buildings enhance signal propagation and quality by intelligently reflecting signals, and assist in beam selection when aerial platforms cannot directly detect mobile IoT systems. Experimental validation using aerial datasets, including DOTA, demonstrates a vehicle detection precision of 89%, and spectral efficiencies of 28 bps/Hz for HAPS and 35 bps/Hz for UAVs. These findings confirm significant gains in throughput and localization accuracy over traditional CSI-dependent schemes, establishing AI-FVBS as an efficient and scalable solution for 6G aerial networks.