Ego-Tactile Manipulation: Egocentric Video with 160-Channel Two-Hand Tactile Glove Pressure and Touch-Grounded Action Labels (OpenGraph Labs)

Provider
OpenGraph Labs
License
CC-BY-4.0 (source)
Size
612 files
Format
PNG, MP4, Parquet (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/OpenGraphLabs-Research/ego-tactile-manipulation)
Access
Free and public. No account with the data provider is needed.
Category
Robotics
Source
https://huggingface.co/datasets/OpenGraphLabs-Research/ego-tactile-manipulation

About this dataset

A sample of synchronized egocentric video and hand touch data for learning human manipulation, from OpenGraph Labs' physical-AI data pipeline. A person wore a head camera (1920x1080 at 30 fps, intrinsics included) and OGLO tactile gloves on both hands, each with 5 fingers x 4x4 taxels (80 pressure channels per hand, 160 total, raw at 90 Hz and aligned at 30 Hz), plus a 9-channel head IMU and 6-channel IMUs on both wrists. The set has 72 episodes, 1.28 hours and 138,629 frames of tabletop tasks such as pouring, capping, cutting, gripping and arranging. Action boundaries come from the physical contact and grip-force signal rather than human judgment, and concise English labels for action, sub-action, anchor, grasp, hand and object are attached within each span. Packaged as LeRobot v3.0 (meta/, per-frame Parquet in data/, video in videos/), with byte-exact raw tactile and IMU in extra/ and one Rerun .rrd recording per episode for visualization; about 6 GB (measured 2026-11-21). Load with lerobot's LeRobotDataset or pyarrow. Released under CC-BY-4.0, including commercial use with attribution.

How to load this dataset in Python

from datasets import load_dataset
ds = load_dataset("OpenGraphLabs-Research/ego-tactile-manipulation")
print(ds)

Data structure

File types in sampled files: PNG, MP4, Parquet

Files

Example file: data/chunk-000/file-000.parquet

Analyze this dataset without downloading it

On Scigantic, Ego-Tactile Manipulation: Egocentric Video with 160-Channel Two-Hand Tactile Glove Pressure and Touch-Grounded Action Labels (OpenGraph Labs) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Robotics datasets