SPAN-Nav: Generalized Spatial Awareness for Versatile Embodied Navigation

SPAN-Nav is an end-to-end foundation model that enhances embodied navigation with universal 3D spatial awareness using RGB video streams. It extracts spatial priors across diverse scenes through occupancy prediction and uses a single token to encapsulate coarse-grained cues for navigation. The model achieves state-of-the-art performance across three benchmarks and validates robust generalization in real-world scenarios.