metadata
license: mit
language:
- en
size_categories:
- 1K<n<10K
🏰 BesiegeField Human Dataset for ColdStart
📎 Links
- Project Page: https://besiegefield.github.io/
- GitHub: https://github.com/Godheritage/BesiegeField
- arXiv: https://arxiv.org/abs/2510.14980
📌 Overview
This dataset collects human-made Besiege machines and reformats them for the LLM Cold-Start stage.
Processing details are described in Paper §F.1.
🗂️ Data Fields
trainable_dataset: Dataset root.ID: Unique id for each data.xxx.json: The tree format of machine.construction_target.txt: A brief description of the functionality of the machine, generated by gpt-4o.machine_type.txt: If this machine is a land-using machine, generated by gpt-4o.
📊 Data Volume
- Total samples: 6 469
🤝 Acknowledgement & Citation
Special thanks to the Besiege community for the amazing creations!
If you use this dataset, please cite:
@article{zhang2025besiegefield,
title={Agentic Design of Compositional Machines},
author={Zhang, Wenqian and Liu, Weiyang and Liu, Zhen},
journal={arXiv preprint arXiv:2510.14980},
year={2025}
}