Learn How to Act from Your Own Interactions: On-Policy Self-Distillation for GUI Agents
作者: Yan Zhang, Daiqing Wu, Huawen Shen, Liang Li, Gang Cao, Zhi Gong, Wei Dai, Xiaode Zhang, Can Ma, Yu Zhou
分类: cs.AI
发布日期: 2026-09-23
备注: Under Review
💡 一句话要点
提出GUI-SD-v2以解决多轮GUI交互中的自蒸馏问题
🎯 匹配领域: 支柱二:RL算法与架构 (RL & Architecture)
关键词: 图形用户界面 自蒸馏 多轮交互 深度学习 智能代理
📋 核心要点
- 现有的在线自蒸馏方法在多轮GUI代理中面临自教师能力有限和指导不足的挑战。
- 本文提出GUI-SD-v2,通过两阶段训练框架增强特权跟随能力,并选择性蒸馏推理和记忆指导。
- 在AndroidWorld和MobileWorld基准上,GUI-SD-v2在Pass@1和Pass@3成功率上显著优于现有方法。
📝 摘要(中文)
图形用户界面(GUI)代理通过与软件环境的多轮交互来实现复杂用户指令的执行,这要求逐步推理和长期记忆以指导行动和保留任务相关信息。近期的在线自蒸馏(OPSD)方法在GUI基础任务上表现出色,但在多轮GUI代理中扩展现有OPSD方法受到自教师能力和指导不足的限制。本文提出GUI-SD-v2,扩展OPSD至多轮GUI交互,并通过两阶段训练框架解决关键限制。具体而言,GUI-SD-v2通过联合优化有无特权指导的回放来增强特权跟随能力,并通过特权条件自教师选择性蒸馏逐步推理和记忆指导,支持后续交互的行动决策和任务信息保留。在AndroidWorld和MobileWorld两个代表性基准上的广泛实验表明,GUI-SD-v2在Pass@1和Pass@3成功率上均优于现有OPSD基线和最新方法。
🔬 方法详解
问题定义:本文旨在解决现有在线自蒸馏方法在多轮GUI交互中的不足,特别是自教师的特权跟随能力有限和指导不足的问题。
核心思路:提出GUI-SD-v2,通过两阶段训练框架来增强特权跟随能力,并通过特权条件自教师进行选择性蒸馏,以支持后续交互中的推理和记忆。
技术框架:整体架构包括两个主要阶段:第一阶段联合优化有无特权指导的回放,第二阶段通过特权条件自教师进行逐步推理和记忆的蒸馏。
关键创新:最重要的创新在于通过两阶段训练框架有效提升了特权跟随能力,并且选择性蒸馏推理和记忆指导,显著提高了多轮交互的性能。
关键设计:在参数设置上,采用了特权条件自教师的设计,损失函数考虑了推理和记忆的选择性蒸馏,网络结构上优化了回放机制以增强特权跟随能力。
🖼️ 关键图片
📊 实验亮点
在AndroidWorld和MobileWorld基准上,GUI-SD-v2在Pass@1和Pass@3成功率上均显著优于现有的OPSD基线,表现出更强的多轮交互能力,具体提升幅度未知,显示了其在实际应用中的潜力。
🎯 应用场景
该研究的潜在应用领域包括智能助手、自动化测试和人机交互等场景,能够提升GUI代理在复杂任务中的执行效率和准确性。未来,该方法可能推动更智能的交互系统的发展,使得用户与软件的交互更加自然和高效。
📄 摘要(原文)
Graphical User Interface (GUI) agents enable the fulfillment of complex user instructions through multi-turn interactions with software environments, requiring step-wise reasoning and long-horizon memory to guide actions and retain task-relevant information, respectively. Recent on-policy self-distillation (OPSD) methods have achieved strong performance on GUI grounding, a foundational subtask for GUI agents, owing to dense token-level supervision from privilege-conditioned self-teachers. However, extending existing OPSD methods to multi-turn GUI agents is hindered by self-teachers' limited privilege-following ability and insufficient privileged guidance. In this paper, we introduce GUI-SD-v2, the next version of GUI-SD, which extends OPSD from GUI grounding to multi-turn GUI interaction and addresses key limitations through a two-stage training framework. Specifically, GUI-SD-v2 first strengthens privilege following by jointly optimizing rollouts with and without privileged guidance from the same GUI states. Furthermore, it selectively distills step-specific reasoning and memory guidance through a privilege-conditioned self-teacher, supporting action decisions and the retention of task-relevant information for subsequent interactions. Extensive experiments on two representative GUI agent benchmarks, AndroidWorld and MobileWorld, show that GUI-SD-v2 compares favorably with existing OPSD baselines while consistently outperforming the evaluated state-of-the-art methods in both Pass@1 and Pass@3 success rates. Code and training data will be publicly released.