让GUI智能体更可信：通过新方法减少记忆捷径，提升行动一致性。
Faithful Mobile GUI Agents with Guided Advantage Estimator
- 分两阶段训练：先教智能体不依赖干扰信息，再用引导优势估计强化可信行为。
- 在陷阱任务中成功率从13.88%提升至80.21%，同时保持指令跟随能力。
- 适合关注智能体可解释性与可靠性的研究者，尤其在自动化交互场景。
基于视觉-语言模型的图形用户界面（GUI）智能体虽具备强大交互能力，但常表现出不可信行为，依赖记忆捷径而非屏幕内容或用户指令。为此，我们提出Faithful-Agent——一种以可信性为先的框架，将GUI交互重定义为优先保证证据一致性和内部自洽。该框架采用两阶段流程：(i) 基于信念的监督微调（SFT）阶段，使智能体在证据扰动下学会克制行为；(ii) 反向反馈训练（RFT）阶段，引入基于锚点和方差自适应的优势调节机制——引导优势估计器（GuAE），结合GRPO优化。GuAE有效防止稀疏奖励下低方差轨迹组中的优势崩溃，并辅以思维-动作一致性奖励。实验显示，相比基线，该方法在陷阱任务成功率（Trap SR）上从13.88%提升至80.21%，同时维持强指令遵循性能。
原文摘要 · Abstract (English)
Vision-language model based graphical user interface (GUI) agents have shown strong interaction capabilities. However, they often behave unfaithfully, relying on memorized shortcuts rather than grounding actions in displayed screen evidence or user instructions. To address this, we propose Faithful-Agent, a faithfulness-first framework that reformulates GUI interaction to prioritize evidence groundedness and internal consistency. Faithful-Agent employs a two-stage pipeline: (i) a faithfulness-oriented SFT stage to instill abstainment behaviors under evidence perturbations; (ii) an RFT stage that further amplifies faithfulness by introducing the guided advantage estimator (GuAE), an anchor-based and variance-adaptive advantage tempering mechanism built upon GRPO. GuAE prevents advantage collapse in low-variance rollout groups under sparse GUI rewards, and with a thought-action consistency reward, Faithful-Agent (Stage II) elevates the Trap SR from 13.88\% to 80.21\% relative to the baseline, while preserving robust general instruction-following performance.
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。