Abot-M0-MetaData

# Abot-M0-MetaData

`Abot-M0-MetaData` is the metadata package used in **Abot-M0**.

This release provides the `meta` files used by the pretraining datasets for **LeRobot v2** data pipelines, focusing on dataset organization, indexing, statistics, and training configuration alignment, without redistributing full large-scale raw data.

Released dataset metadata in this package includes the following open-source datasets:
- `OXE`
- `OXE-AugE`
- `RoboCoin`
- `Agibot-Beta`
- `InternData`
- `Galaxea`

👋 For more information, you can refer to the following links:
- Github: [ABot-M0](https://github.com/amap-cvlab/ABot-Manipulation)
- arXiv Link: [ABot-M0 (arXiv:2602.11236)](https://arxiv.org/abs/2602.11236)
- Project Page: [ABot-M0 Website](https://amap-cvlab.github.io/ABot-Manipulation/)

---

## 1. Dataset Positioning

This repository primarily contains the **`meta` folder for Abot-M0 pretraining datasets (LeRobot v2 format)**, which is used for:

- Sample indexing and split management
- Statistics and normalization-related configuration alignment
- Metadata loading in training and evaluation pipelines

> Note: This repository distributes metadata only, not the complete raw multimodal dataset.

---

## 2. How to Use (LeRobot v2)

Place the `meta` directory from this dataset into the corresponding location under your LeRobot v2 data root, and keep the same folder structure as your local data files.

Suggested workflow:

1. Download `Abot-M0-MetaData`
2. Align the local directory hierarchy
3. Before running visualization/training scripts, verify that the `meta` path is correctly readable

If you are using the official ABot-Manipulation codebase, please follow its dataset preparation and training instructions first:

- <https://github.com/amap-cvlab/ABot-Manipulation>

---

## 3. Citation
If this dataset is useful for your research or engineering work, please consider citing ABot-Manipulation:

```bibtex
@article{yang2026abot,
title={ABot-M0: VLA Foundation Model for Robotic Manipulation with Action Manifold Learning},
author={Yang, Yandan and Zeng, Shuang and Lin, Tong and Chang, Xinyuan and Qi, Dekang and Xiao, Junjin and Liu, Haoyun and Chen, Ronghan and Chen, Yuzhi and Huo, Dongjie and others},
journal={arXiv preprint arXiv:2602.11236},
year={2026}
}
```

# Abot-M0元数据包（Abot-M0-MetaData）

`Abot-M0-MetaData`是**Abot-M0**所使用的元数据包。

本发布包提供了适配**LeRobot v2**数据流水线的预训练数据集所用的`meta`文件，聚焦于数据集组织、索引构建、统计信息对齐以及训练配置适配，且不会分发完整的大规模原始数据集。

本数据包中包含的已发布数据集元信息覆盖以下开源数据集：
- `OXE`
- `OXE-AugE`
- `RoboCoin`
- `Agibot-Beta`
- `InternData`
- `Galaxea`

👋 更多信息可参考以下链接：
- GitHub：[ABot-M0](https://github.com/amap-cvlab/ABot-Manipulation)
- arXiv链接：[ABot-M0（arXiv:2602.11236）](https://arxiv.org/abs/2602.11236)
- 项目主页：[ABot-M0官网](https://amap-cvlab.github.io/ABot-Manipulation/)

---

## 1. 数据集定位
本仓库主要包含适配**LeRobot v2格式**的Abot-M0预训练数据集的`meta`文件夹，其用途如下：
- 样本索引与数据集划分管理
- 统计信息与归一化相关的配置对齐
- 训练与评估流水线中的元数据加载

> 注意：本仓库仅分发元数据，不提供完整的多模态原始数据集。

---

## 2. 使用方法（LeRobot v2）
将本数据包中的`meta`目录放置到本地LeRobot v2数据根目录下的对应路径，并保持与本地数据文件一致的文件夹层级结构。

推荐操作流程：
1. 下载`Abot-M0-MetaData`数据包
2. 对齐本地目录层级结构
3. 在运行可视化或训练脚本前，确认`meta`路径可正常读取

若您使用官方ABot-Manipulation代码库，请先遵循其数据集准备与训练说明：
- <https://github.com/amap-cvlab/ABot-Manipulation>

---

## 3. 引用说明
若本数据集对您的研究或工程工作有所帮助，请引用ABot-Manipulation相关论文：
bibtex
@article{yang2026abot,
title={ABot-M0: VLA Foundation Model for Robotic Manipulation with Action Manifold Learning},
author={Yang, Yandan and Zeng, Shuang and Lin, Tong and Chang, Xinyuan and Qi, Dekang and Xiao, Junjin and Liu, Haoyun and Chen, Ronghan and Chen, Yuzhi and Huo, Dongjie and others},
journal={arXiv preprint arXiv:2602.11236},
year={2026}
}