---
pretty_name: Dexterious EE (TsFile)
tags:
- tsfile
- timeseries
- robotics
- lerobot
- r1_humanoid
- dexterous
- modality:video
- format:tsfile
modality:
- tabular
- timeseries
configs:
- config_name: default
  data_files:
  - split: train
    path: data/dexterious_ee.tsfile
size_categories:
- 10K<n<100K
---

# Dexterious EE (TsFile)

This dataset is an Apache TsFile conversion of
[`Beable/dexterious_ee`](https://huggingface.co/datasets/Beable/dexterious_ee).
It stores the numeric LeRobot frame data as one TsFile table while preserving
the source metadata. The source camera videos are not mirrored in this converted
repository; they remain in the original dataset under
[`videos/`](https://huggingface.co/datasets/Beable/dexterious_ee/tree/main/videos).

## Source Dataset

- Original dataset: [`Beable/dexterious_ee`](https://huggingface.co/datasets/Beable/dexterious_ee)
- Source commit: `88f573a60a320c6b49e7435fb3fc2fc6b9256086`
- Author/owner: `Beable`
- License: not declared in the source dataset card at conversion time
- LeRobot codebase version: `v3.0`
- Robot type: `r1_humanoid`
- Task: `robot_arm_control`
- Source split: `train: 0:40`
- Scale: 40 episodes, 27,237 frames, 1 task
- Sampling rate: 10 fps
- Source video stream: `observation.images.rgb`, 360x640x3, H.264, 10 fps

## TsFile Data

- TsFile path: `data/dexterious_ee.tsfile`
- Table name: `dexterious_ee`
- Rows: 27,237
- Metadata: `meta/` is mirrored from the source, with `meta/info.json` updated
  to describe the converted TsFile schema and video policy.

## Schema

| Column | TsFile role | Type | Notes |
|---|---:|---|---|
| `Time` | TIME | INT64 | Milliseconds from `round(timestamp * 1000)` |
| `episode_index` | TAG | INT64 | Source episode identifier |
| `task_index` | TAG | INT64 | Source task identifier |
| `frame_index` | FIELD | INT64 | Source frame index, retained |
| `sample_index` | FIELD | INT64 | Source `index`, renamed to avoid ambiguity |
| `observation_state_0` ... `observation_state_16` | FIELD | FLOAT | Flattened source `observation.state` vector |
| `action_0` ... `action_16` | FIELD | FLOAT | Flattened source `action` vector |

The 17-element vector order for both `observation.state` and `action` is:
`x`, `y`, `z`, `roll`, `pitch`, `yaw`, `thumb_joint_roll_r`,
`index_proximal_joint_r`, `middle_proximal_joint_r`,
`ring_proximal_joint_r`, `little_proximal_joint_r`,
`thumb_proximal_joint_r`, `index_proximal_joint_r_1`,
`middle_proximal_joint_r_1`, `ring_proximal_joint_r_1`,
`little_proximal_joint_r_1`, `thumb_proximal_joint_r_1`.

## Conversion Notes

- `Time` is synthesized from source `timestamp` in integer milliseconds.
- Source `timestamp` is dropped as a separate field because it is redundant with
  `Time / 1000` seconds.
- Source `index` is renamed to `sample_index`.
- Vector columns are flattened by preserving the full source name, replacing `.`
  with `_`, and appending the element index.
- Source video feature `observation.images.rgb` and files under `videos/` are
  omitted from this converted repository. Use the original dataset for video
  playback and frame alignment.
- No numeric rows were intentionally dropped during conversion.

## Validation

- `pipeline.py validate` passed for `data/dexterious_ee.tsfile`.
- TsFile metadata row count: 27,237.
- TsFile query readback rows: 27,237.
- Staged Parquet rows: 27,237.

## Minimal Read Example

```python
from tsfile import TsFileReader

path = "data/dexterious_ee.tsfile"
table = "dexterious_ee"
fields = ["frame_index", "sample_index", "action_0", "observation_state_0"]

with TsFileReader(path) as reader:
    with reader.query_table(table, fields) as result:
        for row in result:
            print(row)
            break
```