README.md

---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- robotics
- tsfile
- timeseries
- format:tsfile
pretty_name: Random50 v3 (TsFile)
size_categories:
- 10K<n<100K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/random50_v3_train.tsfile
modality:
- tabular
- timeseries
---

# Random50 v3 (TsFile)

This dataset is an Apache TsFile conversion of the Hugging Face dataset
[`iiyudana/random50_v3`](https://huggingface.co/datasets/iiyudana/random50_v3).
The source dataset was created with
[LeRobot](https://github.com/huggingface/lerobot) and contains ALOHA robot
demonstrations for transferring a cube between arms.

Modalities: Time-series. The original repository also contains synchronized
top-camera videos; videos are not included in this converted repository.

## Source Dataset

- Original dataset: [`iiyudana/random50_v3`](https://huggingface.co/datasets/iiyudana/random50_v3)
- License: `apache-2.0`
- LeRobot codebase version: `v2.1`
- Robot type: `aloha`
- Task: `Pick up the cube with the right arm and transfer it to the left arm.`
- Split: `train` (`0:50`)
- Scale: `50` episodes, `20,000` frames, `1` task
- Sampling rate: `50` fps
- Source videos: `50` MP4 files from `observation.images.top`
- Source data layout: `data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet`
- Source video layout: `videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4`

## Converted File

- TsFile: `data/random50_v3_train.tsfile`
- TsFile table: `random50_v3_train`
- Converted rows: `20,000`
- Episodes: `50`
- Time precision: milliseconds
- TAG columns: `episode_index`, `task_index`
- File size: `2,195,882` bytes

All source episodes in the train split are merged into one TsFile. The source
`episode_index` and `task_index` columns are retained as TAG columns, allowing
queries to select an episode without creating synthetic tag aliases.

## Schema

`Time` is computed as `round(timestamp * 1000)` in milliseconds and restarts in
each episode. At 50 fps, consecutive source frames are approximately 20 ms
apart. The source `timestamp` column is dropped because it is redundant with
`Time / 1000` seconds. No source rows are dropped.

TAG columns:

- `episode_index`
- `task_index`

FIELD columns:

- `frame_index`
- `sample_index` (renamed from source `index`)
- `observation_state_0` through `observation_state_13` (FLOAT)
- `action_0` through `action_13` (FLOAT)

The 14 elements in both `observation.state` and `action` use this source order:

1. `left_waist`
2. `left_shoulder`
3. `left_elbow`
4. `left_forearm_roll`
5. `left_wrist_angle`
6. `left_wrist_rotate`
7. `left_gripper`
8. `right_waist`
9. `right_shoulder`
10. `right_elbow`
11. `right_forearm_roll`
12. `right_wrist_angle`
13. `right_wrist_rotate`
14. `right_gripper`

Vector names preserve the full source feature name: `.` is replaced with `_`
and the element index is appended.

## Video Policy

The source feature `observation.images.top` is not converted or uploaded. It is
480 x 640 RGB AV1 video at 50 fps. Use the original dataset for the synchronized
videos:
[`iiyudana/random50_v3/videos`](https://huggingface.co/datasets/iiyudana/random50_v3/tree/main/videos).

The numeric rows retain `episode_index`, `frame_index`, `task_index`, and
`sample_index`, preserving their alignment with the original per-episode video.

## Metadata

The source `meta/` files are mirrored in this repository. `meta/info.json` is
updated so `data_path` points to `data/random50_v3_train.tsfile`. Its
`tsfile_conversion` object records the actual table name, source episode-file
count, converted TsFile count, Time formula, TAG columns, row count, flattened
features, renamed and dropped fields, and frame/video alignment. The converted
`total_videos` value is `0`; the original count of `50` is preserved as
`tsfile_conversion.source_video_count`.

## Validation

The converted file was validated with the project pipeline and read back with
the TsFile Python SDK:

- staged Parquet rows: `20,000`
- TsFile metadata rows: `20,000`
- TsFile query rows: `20,000`
- duplicate `(episode_index, task_index, Time)` rows: `0`
- TsFile size: `2,195,882` bytes

## Usage

```python
from tsfile import TsFileReader

path = "data/random50_v3_train.tsfile"
with TsFileReader(path) as reader:
    schemas = reader.get_all_table_schemas()
    table = schemas["random50_v3_train"]
    print([(column.get_column_name(), column.get_category())
           for column in table.get_columns()])
```