#### NOTE
This is a **Hugging Face dataset**. For large datasets, ensure `huggingface_hub>=1.1.3` to avoid rate limits. Learn more in the Hugging Face integration docs.
# Dataset Card for Xense Earbud Case Insertion (834 episodes)
This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 834 samples. Each sample is one teleoperated episode.
## Installation
```bash
pip install -U fiftyone
```
## Usage
### Load from the Hub
```python
import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub
dataset = load_from_hub("Voxel51/xense-earbud-case-834ep")
session = fo.launch_app(dataset)
```
### Download a snapshot and load it locally
```python
import fiftyone as fo
from huggingface_hub import snapshot_download
snapshot_download(
repo_id="Voxel51/xense-earbud-case-834ep",
repo_type="dataset",
local_dir="xense-earbud-case-834ep",
)
dataset = fo.Dataset.from_dir(
dataset_dir="xense-earbud-case-834ep",
dataset_type=fo.types.LeRobotDataset,
name="xense_earbud_case",
)
session = fo.launch_app(dataset)
```
## Dataset Details
### Dataset Description
834 teleoperated episodes (1,213,518 frames, 30 fps) on robot type `bi_flexiv_rizon4_rt`. The task is “Pick up each earbud case from the left stands, insert the matching earbuds, close the lid, and place the case in the box.” Each episode has three RGB camera streams, four tactile video streams (left and right sensor on each side), and 20-D TCP state and action. The source is a LeRobot v3.0 repo recorded with the Xense LeRobot fork.
- **Shared by:** `Xense` on Hugging Face
- **Language(s):** en
- **License:** apache-2.0
### Dataset Sources
- **Repository:** https://huggingface.co/datasets/Xense/earbud_case_insertion_teleop_0515 (code: https://github.com/Vertax42/lerobot-xense)
## Dataset Structure
**Topology.** `media_type: multimodal`, 834 samples, one per episode, 30 fps, 968-2,342 frames (32.3-78.1 s), 1,213,518 frames total. Samples hold a `media_reference` to the per-frame parquet data and 7 video streams.
| Field | FiftyOne type | Description |
|-------------------|---------------------------|-------------------------------------------------------|
| `media_reference` | `LeRobotEpisodeReference` | Pointers into `data/` parquet and the 7 video streams |
| `episode_index` | `IntField` | Episode index (0-833) |
| `task` | `StringField` | The single task instruction |
| `tasks` | `ListField(StringField)` | All task strings of the episode |
| `length` | `IntField` | Frames in the episode |
| `duration` | `FloatField` | Episode duration in seconds |
| `robot_type` | `StringField` | `bi_flexiv_rizon4_rt` |
| `fps` | `FloatField` | 30 |
**Data behind `media_reference`** (per-frame, from `meta/info.json`):
| Feature | Shape | Description |
|-----------------------------------------------------------------------------------------------------------|-----------|--------------------------------------------------------------------------------------------------------------|
| `observation.state` | (20,) | Bimanual TCP: `left_tcp.{x,y,z,r1..r6}`, `right_tcp.{x,y,z,r1..r6}`, `left_gripper.pos`, `right_gripper.pos` |
| `action` | (20,) | Same 20 names and layout as `observation.state` |
| `observation.images.head`, `left_wrist`, `right_wrist` | 480x640x3 | RGB cameras (H.264 yuv420p) |
| `observation.images.left_tactile_left`, `left_tactile_right`, `right_tactile_left`, `right_tactile_right` | 400x700x3 | Four tactile video streams (H.264 yuv420p) |
**Label types and why.** The importer produces episode-level samples with a `media_reference`; numeric streams (state, action) are read through the viewer and are not stored as FiftyOne label fields. The source has a single task, so `task` is constant. No custom fields were added, so the FiftyOne dataset matches what the LeRobot export contains.
**`dataset.info`.** `lerobot` entry with `format`, `format_major`, `episode_count`, `imported_episode_count`, `skipped_episodes` (empty).
**Parsing decisions.**
- All episodes imported from the full repo (6.9 GB). No source file was modified, and no transcoding or metadata repair was needed.
- All 7 video streams were imported; no modality was excluded. All streams are standard H.264 yuv420p.
- The source does not state the tactile sensor model or the physical meaning of the tactile images.
## Dataset Creation
### Source Data
#### Data Collection and Processing
The source card states only that it was created using LeRobot. The citation names the Xense LeRobot fork, `LeRobot-Xense: LeRobot with Xense Tactile Robotics Support`.
### Annotations
#### Annotation process
A single task string is stored in `meta/tasks.parquet`.
## Citation
**BibTeX:**
```bibtex
@misc{vertax2026lerobotxense,
author = {vertax42 and Xense Robotics Team},
title = {LeRobot-Xense: LeRobot with Xense Tactile Robotics Support},
howpublished = {\url{https://github.com/Vertax42/lerobot-xense}},
year = {2026}
}
```