[论文] StopWatch: Goal Grounding for Object-goal Navigation

论文概要

研究领域: CV
作者: Xinhao Liu, Yunsheng Tian, Yuheng Qiu, Pieter Abbeel, Xi Chen, Guanya Shi, Chen Feng
发布时间: 2026-10-16
arXiv: 2610.23841

中文摘要

预训练的视觉-语言模型（VLM）为物体目标导航带来了丰富的世界知识和语义先验，将 VLM 微调为导航策略已成为常见做法。尽管取得了进展，成功率仍然低下，且不清楚这些微调策略缺乏哪种能力。我们的诊断发现目标锚定（goal grounding）是主要失败原因：这些策略探索能力不错，但往往未能注意到目标已经在视野中，或无法估计目标距离。为解决这一失败，我们对目标进行显式锚定。显式锚定显著提升了成功率，在三个场景集合的六个基准上达到了最先进结果。我们还修正了这些基准中的标注缺陷，使未来方法得到更忠实的评估。在清理后的基准上，我们的分析表明使用真值锚定时，策略几乎追平了生成其演示的专家水平，剩余提升空间在感知而非策略学习。我们的工作指出目标锚定是物体目标导航缺失的一环，并为该任务提供了一个简单而有效的解决方案。

原文摘要

Pretrained VLMs bring rich world knowledge and semantic priors to object-goal navigation, and fine-tuning a VLM into a navigation policy has become a common recipe. Despite this progress, success rates remain low, and it is unclear which capability these fine-tuned policies lack. Our diagnosis identifies goal grounding as the dominant failure. These policies explore well, yet often fail to notice that the goal is already in view or to estimate how far away it is. To address this failure, we ground the goal explicitly. Explicit grounding substantially improves success and reaches state-of-the-art results on six benchmarks across three scene collections. We also correct annotation defects in these benchmarks, so that future methods are evaluated more faithfully. Our analysis on the clean ben...

自动采集于 2026-10-20

#论文 #arXiv #CV #小凯