Learning Reliable GUI Agents under Imperfect Priors

Bo Han , Qianyi Wang , Shuai Liu , Xiong Zifan , Changqiao Wu , Yuanfa Li , Pengzhi Gao , Wei Liu , Jian Luan , Heng Qu , Yunpeng Song , Zhongmin Cai

🏛 Institutions: XJTU , Xiaomi
📅 Date: September 30, 2026
📑 Publisher: arXiv
💻 Env: Mobile
🔑 Keywords: imperfect priors exploration data synthesis robustness UI state-transition graph

TLDR

Argues that GUI agents should act correctly under imperfect, drifting priors rather than pursue perfect app knowledge. Structured exploration builds a UI state-transition graph and synthesizes task-trajectory pairs without human annotation, and a noise-aware training strategy injects five types of realistic errors so the agent learns to assess prior reliability before acting.

Open paper Report issue