Towards Large-Model-Guided UAV Navigation in Complex Environments

Unmanned aerial vehicles have been widely used in scenarios such as urban inspection, disaster response, and exploration in complex environments. However, achieving autonomous navigation in unknown environments remains challenging. In recent years, large models have shown strong capabilities in visual understanding and language reasoning, which provide new possibilities for intelligent navigation. In this paper, we explore the use of large models in UAV navigation and design a large-model-guided navigation framework. The framework takes multi-view visual observations, natural language instructions, and UAV state information as input, and processes them in a unified manner using a large model to generate navigation decisions. This work focuses on a relatively simple modeling approach and aims to study the feasibility of applying large models to UAV navigation tasks. Experimental results in a simulation environment show that the proposed method can utilize multimodal information to produce reasonable navigation behaviors. The results demonstrate the potential of large models for UAV navigation and provide a basic reference for future work.