Low-Altitude World Models for Unmanned Systems: Foundations, Technical Progress, and Open Challenges

Low-altitude operations require prediction that can inform decisions under changing airspace constraints, disturbances, and interactions among aircraft. This narrative review distinguishes learned aerial world models from autonomous flight systems, airspace representations, and digital twins. It organizes foundational methods, neighboring technical transfer, direct aerial predictive models, and supporting operational evidence through an agent-world-model layer, an airspace-environment layer, and a cross-layer integration layer. The comparative analysis separates offline generation, simulated navigation, hardware-in-the-loop flight, and real-camera control rather than treating these settings as equivalent validation. A targeted update includes recent aerial generation, world-model-based flight, and visual navigation studies. Their results support task-specific predictive and control capabilities, but do not establish joint validation of changing permissions, disturbance-coupled prediction, multi-aircraft coordination, and operational safety. Six related bottlenecks motivate a research agenda linking model design to data, benchmarks, and airspace services. Airspace-aware integration is presented as a proposed direction, not an experimentally validated unified architecture. The scope and limits of the narrative selection and the reliance on recent preprints are made explicit.