Knowing Where to Look: Interactive Object Search via Spatial-Semantic Fusion
Knowing Where to Look: Interactive Object Search via Spatial-Semantic Fusion
This paper introduces a system that combines semantic intuition from vision-language models with geometric spatial awareness using Bayesian log-odds filtering to create a unified 3D voxel belief field. The system drives action selection for viewing, navigating, or interacting with objects on a bimanual mobile humanoid, demonstrating improved performance over existing geometry-only and language-model-based search methods.
The authors propose a unified 3D voxel belief field that fuses VLM plausibility scores with geometric occupancy evidence to improve interactive object search on a bimanual mobile humanoid, claiming outperformance over geometry-only and LLM-based baselines; however, the abstract does not specify quantitative metrics, test environments, or safety evaluations.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- code: not_reported
- weights: not_reported
- dataset: not_reported
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
nd that it outperforms geometry-only search and LLM-based planning methods. Project website - https://skymanaditya1.github.io/objsearch-website/
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org