arXiv 2511.197732025-11-24cs.AIcs.CLcs.CV

在视觉语言模型中实现工具集成推理的规模化代理强化学习

Scaling Agentic Reinforcement Learning for Tool-Integrated Reasoning in VLMs

Virginia Tech（弗吉尼亚理工大学）；Emory University（埃默里大学）；KAUST（阿联酋卡塔尔大学）；Georgia Tech（佐治亚理工学院）；Cisco（思科系统）；TAMU（德克萨斯大学奥斯汀分校）；UT Southwestern Medical Center（德克萨斯西南医学中心）；Eigen AI

Meng Lu, Ran Xu, Yi Fang, Wenxuan Zhang, Yue Yu, Gaurav Srivastava, Yuchen Zhuang, Mohamed Elhoseiny, Charles Fleming, Carl Yang, Zhengzhong Tu, Yang Xie, Guang… 展开作者

Meng Lu, Ran Xu, Yi Fang, Wenxuan Zhang, Yue Yu, Gaurav Srivastava, Yuchen Zhuang, Mohamed Elhoseiny, Charles Fleming, Carl Yang, Zhengzhong Tu, Yang Xie, Guanghua Xiao, Hanrui Wang, Di Jin, Wenqi Shi, Xuan Wang

备注：: 17 pages, 9 figures, work in progress

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出VISTA-Gym，通过多轮轨迹采样和端到端强化学习，提升视觉语言模型的工具集成推理能力，在多个基准测试中取得显著优势。

摘要：

尽管最近的视觉语言模型（VLMs）在图像理解方面表现出色，但它们

英文摘要：

While recent vision-language models (VLMs) demonstrate strong image understanding, their ability to "think with images", i.e., to reason through multi-step visual interactions, remains limited. We introduce VISTA-Gym, a scalable training environment for incentivizing tool-integrated visual reasoning capabilities in VLMs. VISTA-Gym unifies diverse real-world multimodal reasoning tasks (7 tasks from 13 datasets in total) with a standardized interface for visual tools (e.g., grounding, parsing), executable interaction loops, verifiable feedback signals, and efficient trajectory logging, enabling visual agentic reinforcement learning at scale. While recent VLMs exhibit strong text-only reasoning, both proprietary and open-source models still struggle with tool selection, invocation, and coordination. With VISTA-Gym, we train VISTA-R1 to interleave tool-use with agentic reasoning via multi-turn trajectory sampling and end-to-end reinforcement learning. Extensive experiments across 11 public reasoning-intensive VQA benchmarks show that VISTA-R1-8B outperforms state-of-the-art baselines with similar sizes by 9.51%-18.72%, demonstrating VISTA-Gym as an effective training ground to unlock the tool-integrated reasoning capabilities for VLMs.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。