Guangyi Liu (刘广义)

Guangyi Liu (刘广义)

Ph.D. candidate at ZJU

About me

I am a Ph.D. candidate in the Department of Control Science and Engineering at Zhejiang University (ZJU), where I have been since 2023, under the supervision of Prof. Yong Liu, Prof. Jiangning Zhang, and Dr. Liang Liu at the ZJU APRIL Lab. My research explores the foundations of universal digital agents: intelligent systems that can understand digital environments, adapt to new tasks, and assist people across apps, devices, and software workflows.

Citations: Google Scholar citations First-author citations First-Author Stars: First-author project stars

🔥 News

📝 Selected Work

Preprint MobileForge main performance

MobileForge: Annotation-Free Adaptation for Mobile GUI Agents with Hierarchical Feedback-Guided Policy Optimization

Guangyi Liu, Pengxiang Zhao, Gao Wu, Yiwen Yin, Mading Li,
  • Identifies two bottlenecks in annotation-free mobile GUI adaptation and proposes MobileGym to unify target-app exploration, curriculum mining, rollout execution, and hierarchical evaluation.
  • Proposes HiFPO to turn multi-attempt feedback and corrective hints into hint-contextualized step-level GRPO updates, yielding ForgeOwl-8B as the strongest open-data mobile GUI agent in our evaluation.
GitHub Repo stars Media coverage: 机器之心 (Synced China)
Preprint MemGUI-Agent ConAct framework

MemGUI-Agent: An End-to-End Long-Horizon Mobile GUI Agent with Proactive Context Management

Guangyi Liu, Gao Wu, Congxiao Liu, Pengxiang Zhao, Liang Liu,
  • Introduces MemGUI-Agent, an end-to-end long-horizon mobile GUI agent built on ConAct, which unifies history folding, UI memory, and self-describing step outputs within one policy.
  • Constructs MemGUI-3K with 2,956 full ConAct-annotated trajectories and trains MemGUI-8B-SFT, achieving the best open-data 8B performance on MemGUI-Bench and generalizing to out-of-distribution MobileWorld.
GitHub Repo stars Media coverage: 量子位 (QbitAI)
ACM MM 2026 MemGUI-Bench overview

MemGUI-Bench: Benchmarking Memory of Mobile GUI Agents in Dynamic Environments

Guangyi Liu, Pengxiang Zhao, Yaozhen Liang, Qinyi Luo, Shunye Tang,
  • Establishes a systematic memory taxonomy and a 128-task, 26-app benchmark where 89.8% of tasks test cross-temporal and cross-spatial memory retention under pass@1/pass@k protocols.
  • Introduces MemGUI-Eval with Progressive Scrutiny across 3 stages and 7 hierarchical metrics, evaluating 11 agents to reveal memory deficits, baselines, and 7 failure modes.
GitHub Repo stars Google Scholar citations
ACL 2026 LearnAct teaser

LearnAct: Few-Shot Mobile GUI Agent with a Unified Demonstration Benchmark

Guangyi Liu, Pengxiang Zhao, Liang Liu, Zhiming Chen, Yuxiang Chai,
ACL 2026
  • Introduces LearnGUI with 2,252 offline tasks and 101 online tasks for demonstration-based learning.
  • Designs DemoParser, KnowSeeker, and ActExecutor to extract, retrieve, and execute with human demonstration knowledge.
GitHub Repo stars Google Scholar citations Media coverage: 量子位 (QbitAI)
TMLR Phone GUI agent survey taxonomy

LLM-Powered GUI Agents in Phone Automation: Surveying Progress and Prospects

Guangyi Liu, Pengxiang Zhao, Yaozhen Liang, Liang Liu, Yaxuan Guo,
TMLR
  • Provides a systematic map of LLM-powered mobile GUI agents, covering frameworks, models, data, evaluation, and open challenges.
  • Clarifies how phone automation is moving from static scripts toward perception-reasoning-action agents.
GitHub Repo stars Google Scholar citations Media coverage: 机器之心 (Synced China)

Selected Collaborations

📖 Education

💻 Experiences

🏆 Honors & Awards

  • Silver Reviewer, ICML 2026.
  • Excellent Graduate Student of Zhejiang University, 2024.
  • Excellent Student Cadre of Zhejiang University, 2024.
  • Jiangsu Provincial Outstanding Graduate, 2023.
  • Jiangsu Provincial Outstanding Student Cadre, 2022.
  • National Scholarship, 2020 and 2021.

🎓 Academic Service

  • Reviewer (Conferences): ICLR, NeurIPS, ICML, ECCV, ACM MM, EMNLP.
  • Reviewer (Journals): TMLR.