Reproduce “Structure over Reasoning: Unified Zero-Shot Navigation with a Small Vision-Language Model” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zeqi Li, Xunyi Zhao, Zerui Li, Sihao Lin, Qi Wu, ShiJie Li
The authors present an indoor navigation agent that uses Qwen3-VL-4B for visual judgments, while a map and trail check select waypoints and verify stopping. It accumulates directional scores into a belief and accepts stop votes only when supported by the trail. At a 30-step budget, fine-grained mean success is 36.20%, versus 27.13% for the strongest baseline; coarse-grained advantage remains uncertain.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.