arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.05044cs.AI
收藏

WebFactory:自动化将基础语言智能压缩为 grounded Web agents

WebFactory: Automated Compression of Foundational Language Intelligence into Grounded Web Agents

Fudan University(复旦大学);IMean AI;The Chinese University of Hong Kong(香港中文大学);Tsinghua University(清华大学)

Sicheng Fan, Qingyun Shi, Shengze Xu, Shengbo Cai, Tieyong Zeng, Li Ling, Yanyi Shang, Dehan Kong

AI总结:

WebFactory 提出了一种自动化闭环强化学习流程,将大语言模型的潜在知识高效压缩为可操作的 agent 行为,实现了数据效率和泛化能力的提升。

摘要:

当前训练 GUI agent 的范式受到对 unsafe、非可复现的 live web 交互或 costly、稀缺的人工数据和环境的依赖。我们主张这种对数据量的关注忽略了更关键的因素:将大语言模型(LLM)的潜在知识压缩为可操作 agent 行为的效率。我们引入 WebFactory,一种全新的、完全自动化的闭环强化学习流程,系统性地将 LLM 编码的互联网智能压缩为高效、 grounded 的动作。我们的流程包括可扩展的环境合成、知识感知的任务生成、 LLM 驱动的轨迹收集、分解奖励的 RL 训练以及系统性的 agent 评估。令人印象深刻的是,我们的 agent 表现出卓越的数据效率和泛化能力。在仅使用 WebFactory 中 10 个网站的合成数据训练后,其性能可与在更大环境集合中训练的相同量级的人工标注数据的 GUI agent 相媲美。这种优越的性能在我们的内部离线和在线迁移基准中保持一致,其中我们的 agent 还显著优于基础模型。我们进一步提供了关于不同 LLM 基础的 'embodiment potential' 的关键洞察,为模型评估提供了一个新的轴线。这项工作提出了一种可扩展且成本效益高的范式,将被动的互联网知识转化为活跃的、 grounded 的智能,标志着迈向通用交互 agent 的关键一步。

英文摘要:

Current paradigms for training GUI agents are fundamentally limited by a reliance on either unsafe, non-reproducible live web interactions or costly, scarce human-crafted data and environments. We argue this focus on data volume overlooks a more critical factor: the efficiency of compressing a large language model's (LLM) latent knowledge into actionable agent behavior. We introduce WebFactory, a novel, fully automated closed-loop reinforcement learning pipeline for GUI agents, systematically compressing LLM-encoded internet intelligence into efficient, grounded actions. Our pipeline features a process of scalable environment synthesis, knowledge-aware task generation, LLM-powered trajectory collection, decomposed reward RL training, and systematic agent evaluation. Remarkably, our agent demonstrates exceptional data efficiency and generalization. Trained on synthetic data from only 10 websites within WebFactory, it achieves performance comparable to GUI agents trained on the same amount of human-annotated data from a much larger set of environments. This superior performance is consistent across our internal offline and online transfer benchmarks, where our agent also significantly outperforms the base foundation model. We further provide critical insights into the "embodiment potential" of different LLM foundations, offering a new axis for model evaluation. This work presents a scalable and cost-effective paradigm for transforming passive internet knowledge into active, grounded intelligence, marking a critical step towards general-purpose interactive agents.

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。

↑