نسخة أولية وصول مفتوح
Seek Before You Move: Evidence Seeking for Progress Grounding in Vision-Language Navigation
Vision-Language Navigation (VLN) requires agents to continuously ground task progress from long-horizon instructions and partial egocentric observations. Existing VLM-based navigation agents typically reason only over available observations and may remain confident even when task-relevant evidence is missing. For examp …