---
license: apache-2.0
task_categories:
- robotics
tags:
- tsfile
- timeseries
- time-series
- LeRobot
- so101
- robotics
pretty_name: so101_test (TsFile)
size_categories:
- 10K<n<100K
configs:
- config_name: default
  data_files: data/*.tsfile
---

# so101_test (TsFile)

Apache TsFile version of [`enpeicv/so101_test`](https://huggingface.co/datasets/enpeicv/so101_test).

## Overview

A single-arm robot manipulation dataset recorded with an SO-101 arm (`so101`,
6 DOF). The task in every episode is to grasp a Lego block and put it in the bin.
It was created with [LeRobot](https://github.com/huggingface/lerobot) (codebase
version v2.1).

- **Scale:** 50 episodes, 22,378 frames total.
- **Sampling rate:** 30 fps.
- **Robot:** `so101` — single SO-101 arm, 6 DOF (5 joints + gripper).
- **Task:** a single task, `task_index = 0` — "Grasp a lego block and put it in the bin".

## Schema (TsFile structure)

- **Time** (INT64, milliseconds) — frame timestamp, `round(timestamp * 1000)`.
  It restarts from 0 at the beginning of each episode (~33 ms between frames at 30 fps).
- **episode_index** (TAG) — which of the 50 episodes the row belongs to.
- **task_index** (TAG) — the task identifier (always `0` here).
  Both tags form the TsFile device dimension; query one episode with
  `WHERE episode_index=0`.
- **frame_index** (FIELD, INT64) — frame number within the episode.
- **sample_index** (FIELD, INT64) — the original global `index` column.
- **action_0 .. action_5** (FIELD, FLOAT) — the commanded target positions, 6
  values: shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper.
- **observation_state_0 .. observation_state_5** (FIELD, FLOAT) — the measured
  joint/gripper positions, same 6-element layout as `action`.

The source `timestamp` column is dropped (it equals `Time ÷ 1000` seconds). Vector
columns are flattened single-precision FLOAT (`.` → `_`, element index appended,
the `observation.` prefix stripped).

## Usage

Read the `.tsfile` files with the Apache TsFile Java or Python SDK.

## Notes

- Camera video streams are **not** included in this repository. The original
  dataset has 2 camera views (`observation.images.laptop`,
  `observation.images.phone`, both 640×480) under its `videos/` directory — see the
  original dataset for them.
- `meta/` from the source is mirrored alongside the data. Aside from the redundant
  `timestamp` column noted above, no columns or rows are dropped.

## Source & license

- Original dataset: https://huggingface.co/datasets/enpeicv/so101_test
- Author / publisher: enpeicv
- Paper: not declared by the original dataset.
- License: apache-2.0 (from the original dataset card).