DORAEMON is a cognitive framework developed to overcome the limitations of zero-shot autonomous navigation based on the Visual Language Model (VLM). DORAEMON consists of Ventral and Dorsal Streams, which mimic human navigational abilities. It integrates hierarchical semantic-spatial fusion, topological maps, RAG-VLM, and Policy-VLM to address spatiotemporal discontinuities, unstructured memory representations, and insufficient task understanding. Furthermore, Nav-Assurance ensures navigational safety and efficiency. DORAEMON achieves state-of-the-art performance on the HM3D, MP3D, and GOAT datasets, and introduces a new evaluation metric, AORI, to better assess navigational intelligence.
