Guowei-Zou/DMPO-datasets

---
license: mit
tags:
- robotics
- reinforcement-learning
- imitation-learning
- robomimic
- mujoco
- d4rl
language:
- en
task_categories:
- robotics
size_categories:
- 1M<n<10M
---

# DMPO Demonstration Datasets

Pre-processed demonstration datasets for **DMPO: Dispersive MeanFlow Policy Optimization**.

[](http://arxiv.org/abs/2601.20701)
[](https://github.com/Guowei-Zou/dmpo-release)
[](https://guowei-zou.github.io/dmpo-page/)

## Overview

This repository contains pre-processed demonstration data for pre-training DMPO policies. Each dataset includes trajectory data and normalization statistics.

## Dataset Structure

```
gym/
├── hopper-medium-v2/
├── walker2d-medium-v2/
├── ant-medium-expert-v2/
├── Humanoid-medium-v3/
├── kitchen-complete-v0/
├── kitchen-mixed-v0/
└── kitchen-partial-v0/

robomimic/
├── lift-img/
├── can-img/
├── square-img/
└── transport-img/
```

Each task folder contains:
- `train.npz` - Training trajectories
- `normalization.npz` - Observation and action normalization statistics

## Usage

Use the `hf://` prefix in config files to auto-download:

```yaml
train_dataset_path: hf://gym/hopper-medium-v2/train.npz
normalization_path: hf://gym/hopper-medium-v2/normalization.npz
```

## Data Sources

- **Gym tasks**: Derived from [D4RL](https://github.com/Farama-Foundation/D4RL) datasets
- **Robomimic tasks**: Derived from [Robomimic](https://github.com/ARISE-Initiative/robomimic) proficient-human demonstrations

## Citation

```bibtex
@misc{zou2026stepenoughdispersivemeanflow,
title={One Step Is Enough: Dispersive MeanFlow Policy Optimization},
author={Guowei Zou and Haitao Wang and Hejun Wu and Yukun Qian and Yuhang Wang and Weibing Li},
year={2026},
eprint={2601.20701},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2601.20701},
}
```

## License

MIT License