sparsh-skin-dataset

# 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:

```bash
.
├── 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
```

# Sparsh-skin 数据集

本数据集包含约4小时的各类家用物品接触数据，通过基于Meta Quest 3的VR遥操作技术采集得到。

数据集中共包含14种不同的家用物品，每种物品对应10条采集序列，单条序列时长约2分钟。每条序列涵盖丰富的物品交互动作，包括滑动、轻触、手持物品重定向等各类操作。

我们提供了用于自监督学习（SSL）训练的序列，其存储格式为提取后的pickle格式。数据集内每条序列的目录结构如下：
bash
.
├── ball
│ ├── 0
│ │ ├── allegro # 包含存储allegro关节状态信息的data.pkl文件
│ │ └── xela # 包含`data.pkl`与`forces.pkl`文件；其中`data.pkl`为训练所用的原始数据，`forces.pkl`则存储
│ ├── 1
| |....
|
|-- baseline
│ ├── allegro
│ ├── 英特尔实感（RealSense）
│ │ └── color
│ └── xela
|-- urdf