Datasets:

InternRobotics
/

EgoExplorer-oxt

pretty_name: 'EgoExplorer-oxt: Zarr export of EgoExplorer'
license: cc-by-nc-4.0
language:
  - en
task_categories:
  - robotics
tags:
  - zarr
  - egocentric
  - loco-manipulation
  - tactile
  - hand-pose
  - humanoid
  - imitation-learning
size_categories:
  - 100K<n<1M
viewer: false
extra_gated_heading: Request access to EgoExplorer-oxt
extra_gated_prompt: >-
  **To download EgoExplorer-oxt, please first fill in the access form:
  [https://forms.gle/hPkKRUK7zzHm6BCX6](https://forms.gle/hPkKRUK7zzHm6BCX6)**
  (name, affiliation, verified email, and your Hugging Face username). Requests
  that pass the automatic checks are approved within about a minute and you will
  get a confirmation email. The button below alone does not grant access. Access
  to this repository is separate from the main EgoExplorer repository.
extra_gated_button_content: I have submitted the form

🔐 Gated dataset — request access before downloading

Fill in a short form with your name, affiliation, Google-verified email and Hugging Face username. Requests that pass the automatic check are approved within about a minute, and a confirmation email follows. Access is granted per repository: this form covers EgoExplorer-oxt only; the LeRobot release has its own page and form.

Ten egocentric clips playing side by side, one per task template.

EgoExplorer-oxt

The Zarr (oxt) export of EgoExplorer — one store per task, ready for policy learning

EgoExplorer is a vision–motion–touch observatory of human loco-manipulation: egocentric video, whole-body motion, finger articulation and tactile contact, recorded together on one clock in real laboratories, workshops and homes. See the main dataset page and the paper for the capture rig, the reconstruction pipeline and the full description.

This repository holds the oxt export: the same recordings reorganised as one Zarr v2 store per task, in the flat data/ + meta/episode_ends layout used by diffusion-policy style training code. The stores follow the Open-X-Tactile (OXT) data format — see the OXT website for the full key specification — and are uploaded as one .zarr.zip archive per task, as required by the OXT submission layout. Each store carries egocentric RGB at policy resolution, head-camera and wrist poses, 21-point hand keypoints and joint angles for both hands, Wuji tactile readings over 17 hand regions, and a natural-language sub-task instruction on every frame.

Store	Environment	Episodes	Frames (30 Hz)
`weighing.zarr`	🧪 Laboratory	4	29,978
`extraction.zarr`	🧪 Laboratory	3	35,289
`reaction_setup.zarr`	🧪 Laboratory	3	21,429
`rotary_evaporation.zarr`	🧪 Laboratory	1	3,109
`oven_drying.zarr`	🧪 Laboratory	1	1,095
`column_chromatography.zarr`	🧪 Laboratory	2	13,060
`fold_large_cardboard_box.zarr`	🏭 Industrial workspace	1	6,140
`bookshelf_organization.zarr`	📚 Study	1	3,849
`clothes_folding.zarr`	🛋️ Living room	1	14,374
`sandwich_making.zarr`	🥪 Food-preparation area	1	2,267
Total		18	130,590 (≈ 73 min)

Data format

EgoExplorer-oxt/
└── EgoExplorer/
    ├── weighing.zarr.zip         # unzip → weighing.zarr/
    ├── extraction.zarr.zip
    └── ...

weighing.zarr/                    # after unzipping
├── data/            # one row per 30 Hz frame, all episodes concatenated
│   ├── camera_ego_rgb            (N, 224, 224, 3) uint8
│   ├── camera_ego_pose           (N, 6)  float64
│   ├── {left,right}_wrist_pose   (N, 6)  float64
│   ├── {left,right}_hand_pose    (N, 21, 3) float32
│   ├── {left,right}_hand_joints  (N, 21) float32
│   ├── {left,right}_hand_joints_idx      (N, 21) int64
│   ├── {left,right}_tactile_data_wuji    (N, 17) float32   (+ _contact, _score)
│   ├── {left,right}_tactile_area_wuji    (N, 17) int64     (+ _contact, _score)
│   ├── {left,right}_tactile_sensor_wuji  (N,) str          (+ _contact, _score)
│   ├── {left,right}_tactile_type_wuji    (N,) str          (+ _contact, _score)
│   ├── sub_task_instruction      (N,) str
│   └── timestamps                (N,) float64, seconds, episode-relative
└── meta/
    └── episode_ends              (E,) int64, exclusive end row of each episode

Array	Per-frame shape	Contents
`camera_ego_rgb`	224×224×3	Egocentric RGB from the head camera, resized for policy input
`camera_ego_pose`	6	Head-camera pose: position (m) + 3-D rotation
`*_wrist_pose`	6	Wrist pose: position (m) + 3-D rotation, same world frame
`*_hand_pose`	21×3	21 hand keypoints (wrist first) in the world frame, metres
`*_hand_joints`	21	Finger joint angles (rad); `*_hand_joints_idx` gives the source joint index of each entry
`*_tactile_data_wuji`	17	Normalised Wuji tactile reading per anatomical hand region
`*_tactile_data_wuji_contact`	17	Binary contact per region
`*_tactile_data_wuji_score`	17	Continuous contact score per region
`_tactile_area_wuji`	17	Region IDs for the 17 columns above
`_tactile_sensor_wuji`, `_tactile_type_wuji`	string	Sensor and reading-type tags (`WujiGlove`, `state`, …)
`sub_task_instruction`	string	English description of the step the frame belongs to
`timestamps`	scalar	Episode-relative time (s), 30 Hz

Episode i spans rows episode_ends[i-1]:episode_ends[i] (with episode_ends[-1] = 0). All arrays are Blosc-compressed Zarr v2 with . as the chunk-key separator.

Access

EgoExplorer-oxt is gated. Fill in the access form — full name, affiliation, a Google-verified email and your Hugging Face username. Submissions are checked automatically, and if they pass, access is granted to that Hugging Face account within about a minute; a confirmation email follows. Then download while signed in:

hf auth login
hf download InternRobotics/EgoExplorer-oxt --repo-type dataset --local-dir EgoExplorer-oxt
cd EgoExplorer-oxt/EgoExplorer && for f in *.zarr.zip; do unzip -q "$f"; done

The full collection keeps growing and is being opened up in stages. To be considered for more data, tick “Join the waitlist for the full release” in the same form and we will contact you at your verified email.

Quick start

import zarr  # zarr<3

root = zarr.open("EgoExplorer-oxt/EgoExplorer/weighing.zarr", mode="r")
ends = root["meta/episode_ends"][:]
data = root["data"]

start, end = 0, ends[0]                      # first episode
rgb = data["camera_ego_rgb"][start:end]      # (T, 224, 224, 3) uint8
wrist = data["right_wrist_pose"][start:end]  # (T, 6)
touch = data["right_tactile_data_wuji_contact"][start:end]  # (T, 17)
print(data["sub_task_instruction"][start])

License

EgoExplorer-oxt is released under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license. Commercial use is not permitted.

Citation

If EgoExplorer is useful in your work, please cite:

@misc{zhou2026egoexplorer,
      title={EgoExplorer: A Vision-Motion-Touch Observatory for Human Loco-Manipulation In Situ}, 
      author={Yunsong Zhou and Xuekun Jiang and Haiqi Shen and Hangxu Liu and Chaojian Shi and Yujie Liu and Jingyuan Li and Haoran Yang and Jiafei Cao and Dongzhan Zhou and Dong Wang and Yao Mu and Chunhua Shen and Weinan Zhang},
      year={2026},
      eprint={2610.18872},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2610.18872}, 
}