arXiv cs.AI· Shreya Gummadi, Mateus V. Gasparino, Gianluca Capezzuto, Marcelo Becker, Girish Chowdhary·· 2 天前AI 评分30
ZeST：基于 VLM 的未知环境零样本可通行性导航
ZeST: A VLM-based Zero-Shot Traversability Navigation for Unknown Environments
AI 导读
ZeST 将 VLM 的重复输出视为随机测量并融合为不确定性感知后验，实现零样本可通行性预测，避免为采集数据将机器人置于危险环境。在室内受控与户外非结构化环境实验中，ZeST 导航成功率达 90-100%，而所有基线均不超过 60%，且在查询延迟高达 3.6s 时仍保持完全成功。
正文 · 原文
Abstract:Advances in robotics and autonomous navigation systems hinge on the ability to predict terrain traversability accurately. Traditional methods for generating datasets to train these prediction models often involve putting robots into potentially hazardous environments, posing risks to equipment and safety. To solve this problem, we present ZeST, a novel approach that treats repeated VLM outputs as stochastic measurements and fuses them into an uncertainty-aware posterior. Our approach not only performs zero-shot traversability and mitigates the risks associated with real-world data collection but also accelerates the development of advanced navigation systems, offering a cost-effective and scalable solution. To support our findings, we present navigation results, in both controlled indoor and unstructured outdoor environments. As shown in the experiments, ZeST provides safer navigation with a 90-100% success rate while no baseline exceeds 60%, and keeps complete success under query latencies up to 3.6s.
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2508.19131 [cs.RO] |
| (or arXiv:2508.19131v4 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2508.19131 arXiv-issued DOI via DataCite |
Submission history
From: Shreya Gummadi [view email]
[v1]
Tue, 26 Aug 2025 15:30:19 UTC (8,161 KB)
[v2]
Fri, 17 Oct 2025 19:27:35 UTC (8,161 KB)
[v3]
Sun, 4 Oct 2026 15:01:28 UTC (8,204 KB)
[v4]
Thu, 15 Oct 2026 17:24:07 UTC (4,257 KB)
来源：arXiv cs.AI · arxiv.org