arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2606.17871cs.AI
收藏

StepGuard: 通过单步校准保护网页导航

StepGuard: Guarding Web Navigation via Single-Step Calibration

School of Software Engineering, Xi’an Jiaotong University(西安交通大学软件工程学院);School of Computer Science and Information Engineering, Hefei University of Technology(合肥工业大学计算机与信息工程学院);Xiamen University(厦门大学);Zhejiang Lab(之江实验室);CSIRO(澳大利亚联邦科学与工业研究组织)

Zhihao Cui, Yuchen Zhang, Xiyang Sun, Yaxiong Wang, Li Zhu, Jinpeng Hu, Liu Liu, Mengjia Li, Yujiao Wu

AI总结:

针对网页导航中单步脆弱性问题,提出StepGuard框架,通过动态双策略优化(DDPO)解决奖励冲突,并利用置信度引导的自适应导航反射(CANR)校准单步误差,显著提升导航与答案准确率。

摘要:

网页导航要求智能体遵循自然语言目标,与网页交互并生成准确答案。尽管近期进展利用了视觉-语言模型和强化学习,现有方法仍因奖励错位和错误传播而存在单步脆弱性。为解决奖励纠缠,我们设计了动态双策略优化(DDPO),在探索的导航优先模式与问答的答案优先模式之间动态切换,以缓解奖励冲突。为校准单步误差,我们提出置信度引导的自适应导航反射(CANR),该机制估计每步置信度,仅在必要时触发反思,并使用对比奖励鼓励自我修正以校准单步不准确性。以上述组件为核心,我们最终开发了StepGuard,一种通过单步校准保护网页导航的新框架。实验表明,我们的方法显著提升了导航与答案准确率,在标准网页导航基准上取得了新的最佳性能。

英文摘要:

Web navigation requires agents to follow natural language goals, interact with web pages, and produce accurate answers. While recent advances leverage vision-language models and reinforcement learning, existing methods still suffer from single-step fragility due to reward misalignment and error propagation. To tackle the reward entanglement, we design Dynamic Dual-Policy Optimization (DDPO), which dynamically switches between a navigation-first mode for exploration and an answer-first mode for question-answering to mitigate reward conflict. To calibrate the single-step error, we propose Confidence-Guided Adaptive Navigation Reflection (CANR), a mechanism that estimates per-step confidence, triggers reflection only when necessary, and uses contrastive rewards to encourage self-correction to calibrate the single-step inaccuracy. With the above as the main components, we finally develop our StepGuard, a new framework of Guarding Web Navigation via Single-Step Calibration. Experiments demonstrate that our approach significantly improves navigation and answer accuracy, setting new state-of-the-art performance on standard web navigation benchmarks.

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。

↑