Room-Mediated Co-occurrence for Zero-Shot Object-Centric Semantic Navigation via Frontier Scoring

2026-07-28

AI Summary

This work addresses the limitation of existing zero-shot semantic navigation methods, which rely solely on direct semantic similarity between objects and fail to capture spatial co-occurrence patterns prevalent in real-world environments. To overcome this, the paper proposes a training-free, open-vocabulary navigation framework that introduces rooms as semantic intermediaries. Specifically, it leverages CLIP to map both target instructions and observed scene objects into room probability vectors (RPVs). Co-occurrence scores are derived from the overlap between RPV distributions, and these scores are propagated across the environment using the fast marching method with adaptive decay based on geodesic distance, yielding a semantic value map that guides frontier-point prioritization. Evaluated on the HM3D validation set, the approach improves success rate (SR) by 3% and success weighted by path length (SPL) by 1.3% over an image-level baseline, while offering strong interpretability and open-vocabulary generalization.

Abstract

Zero-shot ObjectNav methods increasingly use vision-language priors, but direct object-object similarity in the latent space is often a weak proxy for spatial co-occurrence. We present an analytical, training-free semantic navigation pipeline that mediates object relationships through a compact room lexicon. Each object label is mapped to a CLIP-derived Room Probability Vector (RPV), and object-target co-occurrence is computed from RPV distribution overlap. These scores are projected onto a value map using geodesic flood-fill propagation (Fast Marching Method), with adaptive signal decay, and are used to rank frontiers by semantic score for navigation. Together, these components form an integrated, training-free, object-centric pipeline for open-vocabulary zero-shot navigation. Results show that our object-centric approach improves Success Rate (SR) and Success by weighted inverse Path Length (SPL) by a relative 3% and 1.3%, respectively, compared to image-holistic baselines on the HM3D dataset validation split, while preserving interpretability and open-vocabulary flexibility. Code is available at: uts-ri.github.io/RPV-SemNav.

Problem

• Zero-Shot Object Navigation
• Spatial Co-occurrence
• Vision-Language Priors
• Object-Centric Navigation
• Semantic Navigation

Innovation

• Room-Mediated Co-occurrence
• Zero-Shot Semantic Navigation
• Room Probability Vector (RPV)
• Frontier Scoring
• Training-Free Navigation