让智能体在真实桌面环境中先搜索验证再训练，提升长程操作成功率。
ENVS: Environment-Native Verified Search for Long-Horizon GUI Agents
- 在真实系统中实时搜索并验证操作序列，构建高质量训练数据。
- 在300个任务上达29.0的通过率，比在线强化学习节省约30%算力。
- 能应对干扰并恢复，适合需要鲁棒性的自动化桌面应用。
随着多模态智能体从界面理解迈向真实软件控制，如何在真实桌面环境中发现成功操作轨迹成为关键挑战。GUI任务需一系列精确的鼠标键盘操作，但反馈稀疏、延迟且通过虚拟机回滚成本高。本文提出环境原生验证搜索（ENVS），一种训练时的搜索与过滤流程：在真实操作系统虚拟机中分支探索不同行为动作，验证成功路径，并基于全局均衡的步级监督进行策略优化。为评估在现实桌面干扰下的鲁棒性，我们引入动态基准OSWorld-Noisy，保留原始任务的同时测试智能体在实时扰动下是否能重新聚焦、忽略、等待或恢复。在300任务的OSWorld数据集上，ENVS在原版评测中达30.3 pass@8，噪声环境下仍保持29.0，优于同类在线强化学习方法，同时将计算量从184–192降至138–153 GPU小时；仅使用30%搜索数据时仍达27.0 pass@8，超越基线模型。在含噪环境中训练还更好保留了视觉推理能力，如在OSWorld-G Refusal上达16.7（对比1.9）和BLINK Functional Correspondence上达26.2（对比23.1）。
原文摘要 · Abstract (English)
As multimodal agents move from interface understanding to real software control, successful trajectory discovery in live desktop environments becomes a key challenge. GUI tasks require long-horizon sequences of precise mouse and keyboard actions, while feedback is sparse, delayed, and costly to obtain through VM rollouts. We propose Environment-Native Verified Search (ENVS), a training-time search-and-filter pipeline that uses the environment to construct verified supervision before policy optimization: it branches over behaviorally distinct GUI actions in live OSWorld VMs, verifies successful leaves, and trains from globally balanced step-level supervision. To evaluate robustness under realistic desktop interruptions, we also introduce OSWorld-Noisy, a dynamic benchmark for recoverable desktop interruptions that preserves the original tasks while testing whether agents can refocus, dismiss, wait, or recover under live perturbations. On the 300-task OSWorld pool, ENVS reaches 30.3 pass@8 on original evaluations and 29.0 on OSWorld-Noisy, outperforming matched ARPO-style online RL while reducing compute from 184-192 to 138-153 GPU-hours; even with only 30% of its search data, ENVS reaches 27.0 pass@8, exceeding ARPO from the base model. Training from noisy environments also better preserves visual-reasoning abilities on auxiliary benchmarks, including OSWorld-G Refusal (16.7 vs. 1.9) and BLINK Functional Correspondence (26.2 vs. 23.1).
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。