Event

CLOSER: An Active Visual Verification Framework for ObjectNav Stop Determination

1 reports · 1 sources

Pengying Wu et al. propose CLOSER, an evidence-aware active verification framework for object-goal navigation (ObjectNav) that treats stop decisions as closed-loop evidence collection, combining vision-language model (VLM) verification with active perception. On Gibson and Habitat-Matterport 3D, the method improves average success rate by 3.6 points and path-length-weighted success rate by 2.0 points. CLOSER runs locally and can be directly integrated into existing ObjectNav systems without retraining; if the fused judgment remains uncertain, a diffusion policy drives the agent toward the candidate target using only tracking masks and depth information until the judgment becomes unambiguous.