Commercial License Available

The Most Precise
Hand Manipulation
Dataset

Built for robotics AI. Available off-the-shelf or custom-built to your exact specification.

View Datasets Request Custom Data
MANO pose · 21 joints · 3D / 2D
[ Replace with actual video ]
0
Tasks Available
0
Joints Tracked Per Hand
8K
Video Resolution
Commercial
License
Differentiation

Why Our Data?

EgoDex by Apple is research-only. We offer the same quality — commercially licensed, and built to order.

EgoDex (Apple)
Our Dataset
License
Research only
Commercial ✓
Egg Cracking
Not included
Included ✓
Laundry Folding
Basic only
Full sequences ✓
Chopstick Use
Not included
Included ✓
Custom Tasks
Not available
On demand ✓
Failure + Recovery
Not included
Included ✓
Custom Order

We Build What You Need

Don't settle for open-source datasets limited to research use. Tell us your target task — we'll capture, process, and deliver exactly what your model needs.

01

Define Your Task

Describe the manipulation task your robot needs to learn. Any bimanual or single-hand action.

02

We Capture

8K smartphone + Intel RealSense D435i. Egocentric view, synchronized RGB-D, multiple performers.

03

Extract & Label

MANO pose via WiLoR/HaMeR, 21-joint 3D/2D, task phase labeling, success/failure flags.

04

Deliver

HDF5 + LeRobot v3 format. HuggingFace Hub compatible. Commercial license included.

Start a Custom Order
Available Now

Available Datasets

TASK 01
[ Replace with video / image ]
Bimanual Deformable Contact-rich

Egg Cracking

Bimanual contact-rich manipulation with a deformable, fragile object. Not present in EgoDex.

4 phases · Failure sequences included
Request Sample
TASK 02
[ Replace with video / image ]
Bimanual Deformable High Demand

Laundry Folding

Full folding sequences with deformable fabric. More complete than EgoDex's basic version.

4 phases · Failure sequences included
Request Sample
TASK 03
[ Replace with video / image ]
Single Hand Extreme Dexterity

Chopstick Use

Fine-grained finger coordination with Asian cultural specificity. Not present in EgoDex.

4 phases · Failure sequences included
Request Sample
Technical

Data Specification

◎

Pose Annotation

  • MANO pose params (48D per hand)
  • 21 joints 3D per hand
  • 21 joints 2D per hand
  • WiLoR / HaMeR extracted
◎

Capture Setup

  • RGB: 8K smartphone (egocentric)
  • Depth: Intel RealSense D435i
  • Chest-mounted, first-person view
  • RGB-D synchronized
◎

Labels & Format

  • Task phase labels (4 stages)
  • Language instruction per episode
  • Success / failure flag
  • HDF5 + LeRobot v3
◎

Compatibility

  • HuggingFace Hub ready
  • LeRobot v3 format
  • Commercial license
  • Custom format on request
episode_0042.hdf5 
episode_0042.hdf5
├── obs/rgb_egocentric      (T, 1080, 1920, 3)   uint8
├── obs/left_hand/
│   ├── mano_pose           (T, 48)               rotation matrix
│   ├── mano_shape          (10,)                 per sequence
│   ├── joints_3d           (T, 21, 3)            camera frame
│   └── joints_2d           (T, 21, 2)            normalized
├── obs/right_hand/         (same structure)
└── attrs:
    task_name · language_instruction · fps · success
Get in Touch

Ready to Build Your Dataset?

Whether you need an existing dataset or a fully custom capture — let's talk.

Send a Request Request Sample Data
contact@handdata.ai