Datasets:

lgy0404
/

MemGUI-Bench

license: mit
task_categories:
  - robotics
tags:
  - gui-agent
  - mobile-agent
  - benchmark
  - memory
pretty_name: MemGUI-Bench
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files:
      - split: full
        path: memgui-tasks-all.csv
      - split: mini
        path: memgui-tasks-40.csv

🧠 MemGUI-Bench

A memory-centric benchmark for evaluating Mobile GUI Agents with 128 tasks across 26 applications.

🚀 Usage

from datasets import load_dataset

# Load full benchmark (128 tasks)
dataset = load_dataset("lgy0404/MemGUI-Bench", split="full")

# Load mini subset (40 tasks) for quick testing
dataset_mini = load_dataset("lgy0404/MemGUI-Bench", split="mini")

📊 Dataset Structure

Split	Tasks	Description
`full`	128	Complete benchmark
`mini`	40	Quick testing subset

Fields

Field	Description
`task_identifier`	Unique task ID
`task_description`	Natural language instruction
`task_app`	Required app(s)
`num_apps`	Number of apps involved
`requires_ui_memory`	Whether task requires memory
`task_difficulty`	Difficulty level (1-3)
`golden_steps`	Reference step count

📝 Citation

@misc{liu2026memguibenchbenchmarkingmemorymobile,
  title={MemGUI-Bench: Benchmarking Memory of Mobile GUI Agents in Dynamic Environments},
  author={Guangyi Liu and Pengxiang Zhao and Yaozhen Liang and Qinyi Luo and Shunye Tang and Yuxiang Chai and Weifeng Lin and Han Xiao and WenHao Wang and Siheng Chen and Zhengxi Lu and Gao Wu and Hao Wang and Liang Liu and Yong Liu},
  year={2026},
  eprint={2602.06075},
  archivePrefix={arXiv},
  primaryClass={cs.DC},
  url={https://arxiv.org/abs/2602.06075},
}

📧 Contact

For questions, issues, or collaborations, please contact: guangyiliu@zju.edu.cn