EgoHumanoid · Sample G1 Dataset

A small sample dataset collected on the Unitree G1 humanoid platform, released alongside EgoHumanoid: Unlocking In-the-Wild Loco-Manipulation with Robot-Free Egocentric Demonstration for quick experimentation and fine-tuning.

---

Overview

This repository hosts a sample subset intended for:

• Smoke-testing the EgoHumanoid training / inference pipeline
• Fine-tuning a pretrained VLA policy on a small G1 task
• Validating data loaders and the LeRobot-compatible storage format

For the full release and broader documentation, please visit the project page and GitHub repository linked below.

Links

Resource	URL
Project Page	https://opendrivelab.com/EgoHumanoid/
GitHub Repo	https://github.com/OpenDriveLab/EgoHumanoid
Paper (arXiv)	https://arxiv.org/abs/2602.10106

Dataset Structure

EgoHumanoid/
└── example/
    ├── data/
    │   └── chunk-000/
    │       └── episode_*.parquet     # per-episode state / action trajectories
    ├── meta/
    │   ├── info.json                 # schema & dataset metadata
    │   ├── modality.json             # input / output modality definitions
    │   ├── episodes.jsonl            # episode index
    │   ├── episodes_stats.jsonl      # per-episode statistics
    │   └── tasks.jsonl               # language task descriptions
    ├── videos/                       # egocentric RGB recordings
    └── norm_stats.json               # normalization statistics for VLA training

The layout follows the LeRobot v2 convention, so any LeRobot-compatible loader can read it out of the box.

Quick Start

from huggingface_hub import snapshot_download

local_dir = snapshot_download(
    repo_id="SII-JinChen/EgoHumanoid",
    repo_type="dataset",
    allow_patterns="example/*",
)
print(f"Downloaded to: {local_dir}/example")

Or with the CLI:

hf download SII-JinChen/EgoHumanoid --repo-type=dataset --local-dir ./EgoHumanoid

Citation

If you find this dataset useful, please cite the EgoHumanoid paper:

@article{shi2026egohumanoid,
    title={EgoHumanoid: Unlocking In-the-Wild Loco-Manipulation with Robot-Free Egocentric Demonstration},
    author={Shi, Modi and Peng, Shijia and Chen, Jin and Jiang, Haoran and Li, Yinghui and Huang, Di and Luo, Ping and Li, Hongyang and Chen, Li},
    journal={arXiv preprint arXiv:2602.10106},
    year={2026}
}

License

Released under the Apache 2.0 license.