How Mobile World Model Guides GUI Agents?

Weikai Xu , Kun Huang , Yunren Feng , Jiaxing Li , Yuhan Chen , Yuxuan Liu , Zhizheng Jiang , Heng Qu , Pengzhi Gao , Wei Liu , Jian Luan , Xiaolin Hu , Bo An

🏛 Institutions: NTU , MiLM Plus , Xiaomi , Renmin University of China , Wuhan University , Xiamen University
📅 Date: May 11, 2026
📑 Publisher: arXiv
💻 Env: Mobile
🔑 Keywords: world model model-based planning data synthesis test-time guidance

TLDR

This study compares delta-text, full-text, diffusion-image, and renderable-code mobile world models and evaluates how their generated rollouts guide GUI agents. Renderable code gives strong in-distribution fidelity and training supervision, while text feedback is more robust for online out-of-distribution execution; generated trajectories also improve end-to-end training, but do not preserve the original data distribution.

Open paper Report issue