From Region Arrival to Instance-Level Grounding in Vision-and-Language Navigation
For VLN researchers, this work addresses a critical but overlooked evaluation gap by providing metrics and a method to improve instance-level grounding, though the approach is incremental.
The paper identifies a 'Last-3-Meter Grounding Gap' in Vision-and-Language Navigation (VLN) evaluation, where agents often succeed in coarse metrics but fail at object-level grounding. It introduces REALM, a plug-and-play module that improves proximity precision and target visibility, achieving consistent gains across four VLN backbones.
Vision-and-Language Navigation (VLN) agents may satisfy conventional success criteria while still failing to establish reliable object-level grounding, because current evaluation protocols mainly reward stopping within a 3-meter radius and largely ignore the agent's final orientation and target visibility. We formalize this limitation as the Last-3-Meter Grounding Gap and introduce three instance-centric metrics to quantify proximity precision, target visibility, and final-view grounding. To mitigate this gap, we propose REALM (Region-to-Entity Alignment for Last-3-Meter Navigation), a plug-and-play, architecture-agnostic refinement module that decouples fine-grained target approaching from long-horizon navigation. REALM uses a visibility-aware stopping strategy to reduce premature termination and improve final viewpoint alignment. We further construct REVERIE-AIM, which provides object-instance-level goals and 180K short-horizon training samples for final-stage target approaching. Extensive evaluations across four diverse VLN backbones show that REALM consistently improves proximity precision and visual grounding success, demonstrating its broad applicability.