---
license: apache-2.0
authors:
- Mathias
task_categories:
- robotics
tags:
- tsfile
- timeseries
- tabular
- robotics
- lerobot
- so101
- tutorial
modality:
- timeseries
- tabular
pretty_name: SO101 Test13 TsFile
configs:
- config_name: default
  data_files:
  - split: train
    path: data/m3throcks_so101_test13.tsfile
size_categories:
- 10K<n<100K
---

# SO101 Test13 TsFile

Apache TsFile edition of [M3thRocks/so101_test13](https://huggingface.co/datasets/M3thRocks/so101_test13), a LeRobot v2.1 SO101 dataset for grasping a LEGO block and putting it in a bin. The numeric trajectories are stored in one table-model TsFile.

## Source and attribution

- Original author and repository owner: [Mathias (M3thRocks)](https://huggingface.co/M3thRocks). Repository commits are attributed to `des`.
- License: Apache-2.0.
- The source card does not provide a homepage, paper, or completed BibTeX citation.
- Split: train; 45 episodes; 30,519 frame rows; one task; 30 FPS; 45 source Parquet shards.
- Task `0`: `Grasp a lego block and put it in the bin.`

## Data layout

The table is `m3throcks_so101_test13` and contains 30,519 rows across 45 TAG devices. The source Parquet shards total 1,528,259 bytes; the TsFile is 515,055 bytes (33.7% of the source Parquet size).

| Column | TsFile role | Type | Meaning |
|---|---|---|---|
| `Time` | TIME | INT64 milliseconds | `round(timestamp * 1000)`, restarting at zero per episode |
| `episode_index` | TAG | STRING from source INT64 | Source episode identity |
| `task_index` | TAG | STRING from source INT64 | Source task identity |
| `frame_index` | FIELD | INT64 | Frame position within the episode |
| `sample_index` | FIELD | INT64 | Source `index`, renamed for clarity |
| `action_0` ... `action_5` | FIELD | FLOAT | Flattened `action[6]` |
| `observation_state_0` ... `observation_state_5` | FIELD | FLOAT | Flattened `observation.state[6]` |

`timestamp` is not retained as a separate FIELD because it is represented by `Time / 1000` seconds. The vector column names preserve their source prefixes, with dots changed to underscores. No trajectory row, episode, task, action dimension, or state dimension is removed.

## Encoding and compression

- TIME, INT32, and INT64: TS_2DIFF + LZ4
- FLOAT and DOUBLE: GORILLA + LZ4
- BOOLEAN, when present: RLE + LZ4
- TAG values: TsFile table-model device/tag storage

## Videos and alignment

The 90 source AV1 videos are not included here. They remain under [`videos/chunk-000`](https://huggingface.co/datasets/M3thRocks/so101_test13/tree/main/videos/chunk-000) in the original repository:

- [`observation.images.laptop`](https://huggingface.co/datasets/M3thRocks/so101_test13/tree/main/videos/chunk-000/observation.images.laptop): 45 files, 499,994,290 bytes
- [`observation.images.phone`](https://huggingface.co/datasets/M3thRocks/so101_test13/tree/main/videos/chunk-000/observation.images.phone): 45 files, 479,124,167 bytes

The source template is `videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4`. Use `episode_index` and `frame_index` to align each numeric row with both 30 FPS video streams.

## Read example

```python
from tsfile import TsFileReader

reader = TsFileReader("data/m3throcks_so101_test13.tsfile")
with reader.query_table(
    "m3throcks_so101_test13",
    ["episode_index", "task_index", "frame_index", "sample_index", "action_0", "observation_state_0"],
    batch_size=1024,
) as result:
    batch = result.read_arrow_batch()
    print(batch.to_pandas().head())
reader.close()
```