README.md

---
task_categories:
  - robotics
license: cc-by-sa-4.0
tags:
  - robotics
  - lerobot
  - so101
  - so-101
  - vision-language-action
  - imitation-learning
  - lerobot-dataset
---

# SO-101 Early Collection (superseded)

The project's **first, miscellaneous recordings**, made before the team settled on four
fixed tasks, a consistent recording protocol, and systematic prompt variation.

> [!WARNING]
> **This dataset is superseded and not recommended for training.** It is retained for
> provenance and to document the project's history. The only model trained on it —
> [SmolVLA V1 Misc](https://huggingface.co/Project-IRA/TPSoSe2026_SmolVLA_LeRobot_SO101_Finetuning_V1_Misc_Dataset) —
> does not work.

Part of **[Project-IRA](https://huggingface.co/Project-IRA)** — Interactive Robotic Arm.
Code: https://github.com/Project-IRA/interactive-robotic-arm

## Use these instead

| Dataset | Episodes | Content |
|---|---|---|
| [**Full Merged Final V1**](https://huggingface.co/datasets/Project-IRA/TPSoSe2026_Dataset_Full_Merged_Final_LeRobot_SO101_V1) | **930** | The actual training set — all four tasks, 93 prompts |
| [Lego](https://huggingface.co/datasets/Project-IRA/TPSoSe2026_Dataset_Lego_LeRobot_SO101) | 460 | Colour sorting |
| [Desk Cleanup](https://huggingface.co/datasets/Project-IRA/TPSoSe2026_Dataset_Desk_Cleanup_LeRobot_SO101) | 350 | Desk clearing |
| [Dice Throw](https://huggingface.co/datasets/Project-IRA/TPSoSe2026_Dataset_Dice_Throw_LeRobot_SO101) | 120 | Dice cup |
| [Fetch Ball](https://huggingface.co/datasets/Project-IRA/TPSoSe2026_Dataset_Fetch_Ball_LeRobot_SO101) | 100 | Ball handover |

## Why it was superseded

The early recordings mixed several exploratory task ideas that were later dropped —
including putting a ball on a plate, putting a box on a marked X, waving, and placing a
chess piece on a Rubik's cube. Episode counts per task were small and inconsistent, and
prompts were not varied systematically.

Fine-tuning on this data produced a model that was useless at essentially everything.
That negative result is what motivated the structured protocol used for every later
dataset: four fixed tasks, exactly 10 episodes per prompt, deliberate prompt variation
every 10 episodes, and documented object-placement schemes.

## Robot setup

| | |
|---|---|
| Robot | SO-101 follower arm (6-DOF), `robot_type: so_follower` |
| Teleoperation | SO-101 leader arm |
| Control frequency | 30 fps |
| State / action space | 6-dim: `shoulder_pan.pos`, `shoulder_lift.pos`, `elbow_flex.pos`, `wrist_flex.pos`, `wrist_roll.pos`, `gripper.pos` |
| Camera `observation.images.desk_view` | 800x600, h264 (recording) |
| Camera `observation.images.wrist_left` | 640x480, h264 (recording) |

> **Inference note:** both cameras are run at **640x480 during inference**, not at their
> recording resolutions, to reduce the payload sent to the inference server.

## Licensing

Released under **CC BY-SA 4.0**. This is a share-alike licence: you may use, share and
adapt this dataset, including commercially, provided you give attribution and release any
derivative dataset under the same licence. It cannot be taken closed-source.

Recorded with [LeRobot](https://github.com/huggingface/lerobot) (Apache-2.0).

## Citation

```bibtex
@misc{project_ira_2026,
  title        = {Project-IRA: Interactive Robotic Arm},
  author       = {Baten, Cleo and Keppler, Bela and Sapper, Jonas},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/Project-IRA}},
  note         = {Code: \url{https://github.com/Project-IRA/interactive-robotic-arm}}
}
```