README.md

Sparsh-skin dataset

For Sparsh-skin the dataset consists of ~4 hours of contact data with different types of household objects, collected via a VR teleoperation using the Meta Quest 3.

There are 14 different objects in the dataset, each containing 10 sequences per object, which are each ~2 mins long. Every object has varied interaction with the object, including sliding, tapping, object reorientation in the hand, and the like.

We provide the sequences used for SSL training in an extracted pickle format. Each sequence in the dataset has the following structure:

.
├── ball
│   ├── 0
│   │   ├── allegro # contains data.pkl with allegro joint states information
│   │   └── xela # contains `data.pkl` and `forces.pkl`; `data.pkl` contains raw data used for training, and `forces.pkl` contains
│   ├── 1
|   |....
|
|-- baseline
│   ├── allegro
│   ├── realsense
│   │   └── color
│   └── xela
|-- urdf