揭示视觉工具强化学习的真实效果：主要减少工具副作用，而非真正掌握工具使用。
What Does Vision Tool-Use Reinforcement Learning Really Learn? Disentangling Tool-Induced and Intrinsic Effects for Crop-and-Zoom
- 提出分解框架MED，分离模型内在能力与工具带来的影响。
- 发现性能提升主要来自内在能力进步，工具使用改进有限。
- 适合关注VLM工具使用机制与可靠性研究的读者。
视觉工具使用强化学习（Vision Tool-Use RL）可为视觉语言模型（VLM）赋予如裁剪缩放等视觉操作能力并带来显著性能提升，但尚不清楚这些提升是源于工具使用能力的增强，还是模型内在能力的演化。本文提出MED（Measure--Explain--Diagnose）框架，从粗到细地解耦内在能力变化与工具诱导效应，将工具引起的性能差异分解为增益与损害两项，并探究其演变机制。在两个具有不同工具先验的VLM上，针对裁剪缩放任务，在六个基准测试中进行检查点级分析，结果表明性能提升主要由内在学习驱动，而工具使用强化学习主要作用于减少工具诱发的损害（如更少调用错误、更弱的工具模式干扰），在纠正内在缺陷方面的工具化修正进展有限。总体而言，在所研究的裁剪缩放设置下，当前视觉工具使用强化学习主要学习的是与工具安全共存，而非真正掌握工具使用。
原文摘要 · Abstract (English)
Vision tool-use reinforcement learning (RL) can equip vision language models with visual operators such as crop-and-zoom and achieves strong performance gains, yet it remains unclear whether these gains are driven by improvements in tool use or evolving intrinsic capabilities. We introduce MED (Measure--Explain--Diagnose), a coarse-to-fine framework that disentangles intrinsic capability changes from tool-induced effects, decomposes the tool-induced performance difference into gain and harm terms, and probes the mechanisms driving their evolution. Across checkpoint-level analyses in the crop-and-zoom setting on two VLMs with different tool priors and six benchmarks, we find that improvements are dominated by intrinsic learning, while tool-use RL mainly reduces tool-induced harm (e.g., fewer call-induced errors and weaker tool schema interference) and yields limited progress in tool-based correction of intrinsic failures. Overall, in the crop-and-zoom setting studied here, current vision tool-use RL learns to coexist safely with tools rather than master them.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。