arXiv cs.AI· Yang Zhou, Zixuan Huang, Sunzhu Li, Zhuo Yang, Chen Zhang, Shunian Chen, Caijun Yan, Jianyao Xu, Shunyu Liu, Weijie Fu, Peiliang Li, Xiaozhi Chen, Yuxiang Cai·· 4 小时前AI 评分41
SpatialCLI：教会 VLM 先用空间工具推理、再内化摆脱工具
SpatialCLI: Learning to Reason With Spatial Tools, Then Without Them
AI 导读
研究团队提出 SpatialCLI 框架，通过 Call、Learn、Internalize 三阶段让视觉语言模型借助专家视觉模型的空间工具推理，并逐步内化其感知能力，同时发布 516 例的 SpatialCLI-Bench 基准。
正文 · 原文
Authors:Yang Zhou, Zixuan Huang, Sunzhu Li, Zhuo Yang, Chen Zhang, Shunian Chen, Caijun Yan, Jianyao Xu, Shunyu Liu, Weijie Fu, Peiliang Li, Xiaozhi Chen, Yuxiang Cai
Abstract:Vision-language models (VLMs) are increasingly used in embodied agents to interpret visual inputs, reason about spatial relationships, and make task-level decisions based on that reasoning. However, a fundamental capability mismatch remains: general VLMs can reason about the overall task but often miss the visual details that determine success, while specialist vision models can capture those details but cannot translate them into task-level decisions. In this work, we propose SpatialCLI, a framework that teaches VLMs to reason with spatial tools and progressively internalize the specialist perceptual capabilities they provide. SpatialCLI proceeds in three stages: (1) Call exposes specialist vision models as spatial tools to augment the VLM's perception; (2) Learn uses Cold-Start SFT and agentic RL to improve tool use; and (3) Internalize verbalizes successful tool-use trajectories to internalize specialist perceptual capabilities. We further introduce SpatialCLI-Bench, a 516-example benchmark for compositional perception across localization, segmentation, depth, and pose. On MindCube, SpatialCLI raises Qwen3-VL-8B-Instruct from 29.3% to 84.6% with tools, surpassing GPT-5.6 Sol with tools (72.1%), while retaining 73.8% without tools after internalization.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2607.27703 [cs.AI] |
| (or arXiv:2607.27703v3 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.27703 arXiv-issued DOI via DataCite |
Submission history
From: Yang Zhou [view email]
[v1]
Thu, 30 Jul 2026 05:39:01 UTC (5,347 KB)
[v2]
Tue, 4 Aug 2026 12:14:50 UTC (5,346 KB)
[v3]
Mon, 26 Oct 2026 09:20:14 UTC (5,346 KB)
来源：arXiv cs.AI · arxiv.org