hzxie/DOM

---
license: other
license_name: slab-license
license_link: LICENSE
size_categories:
- 100K<n<1M
task_categories:
- robotics
tags:
- lerobot
- franka
- dynamic
- visual-language-action
- vla
---

# Dynamic Object Manipulation (DOM)

[**Project Page**](https://haozhexie.com/project/dynamic-vla) | [**Paper**](https://huggingface.co/papers/2601.22153) | [**Code**](https://github.com/hzxie/DynamicVLA)

**TL;DR:** DOM is a large-scale dynamic manipulation dataset with 200K episodes, 2,800+ scenes, and 206 objects for training and evaluating VLA models.

## Introduction

The Dynamic Object Manipulation (DOM) benchmark is designed to address the challenges of rapid perception and temporal anticipation in robotics. It includes:
- **200K synthetic episodes** across 2,800+ scenes and 206 objects.
- Support for evaluating VLA models in dynamic scenarios requiring continuous control and closed-loop adaptation.

## Citation

If you find this dataset or the DynamicVLA framework useful for your research, please cite:

```bibtex
@article{xie2026dynamicvla,
title = {DynamicVLA: A Vision-Language-Action Model for
Dynamic Object Manipulation},
author = {Xie, Haozhe and
Wen, Beichen and
Zheng, Jiarui and
Chen, Zhaoxi and
Hong, Fangzhou icon,
Diao, Haiwen and
Liu, Ziwei},
journal = {arXiv preprint arXiv:2601.22153},
year = {2026}
}
```

## Changelog

- [2026/04/26] The dataset is released.
- [2026/01/31] The repo is created.