These capabilities are particularly critical at higher frequency bands, where communication becomes highly sensitive to blockage and environmental dynamics. Conventional beam management methods, based on exhaustive beam sweeping and reactive measurements, suffer from high latency and overhead while lacking predictive capability.

To address this, we develop an AI-driven beam tracking framework using the NVIDIA Aerial Omniverse Digital Twin. A digital replica of the Hanyang University campus in Seoul, South Korea, was constructed using an OpenStreet 3D map, and realistic mobility scenarios were generated for training and evaluation. By combining digital-twin-based environment awareness, AI-based positioning, and partial beam sweeping, the proposed framework predicts the best beam with reduced latency and signaling overhead. Experimental results show that partial beam sweeping achieves beam tracking performance comparable to full beam sweeping while using significantly fewer resources, and it also provides higher real-time data rate performance.