CompassNav: Steering from Path Imitation to Decision Understanding in Navigation

LinFeng Li ^1,2 Jian Zhao ² Yuan Xie ¹ Xin Tan ¹ Xuelong Li ² 

¹ East China Normal University; ² The Institute of Artificial Intelligence (TeleAI), China Telecom 

>

The top panel contrasts our End-to-End Goal Navigation paradigm with traditional approaches. Unlike Vision-Language Navigation (VLN), which relies on dense, step-by-step instructions, and complex Modular Navigation pipelines, CompassNav directly maps a high-level goal (e.g., "find the plant") to an action through integrated spatial logical reasoning.

The bottom panel details our core contribution-how to stimulate model reasoning ability: a paradigm shift from "Path Imitation" to "Decision Understanding." While traditional methods train agents to replicate a single expert trajectory and penalize any deviation, our agent learns to evaluate the relative quality of all feasible paths at each decision point. This approach cultivates a true "internal compass," enabling the agent to make more intelligent and flexible decisions in unseen environments.

TODO

• Release Compass-Data-22k
• Release CompassNav-7B
• Release CompassNav training code
• Release CompassNav Object Goal Nav/Instance Image-Goal Nav test code