jt-2026/so101-breakfast

SO-101 Breakfast Table-Setting Dataset

Teleoperated demonstration dataset for training a π₀.₅ VLA policy on a real SO-101 manipulator arm.

Task

Breakfast table-setting — a multi-step, long-horizon manipulation task:

"First put the block onto the plate, move the plate to the center of the table, place the spoon on the right side of the plate, and place the cup on the left side of the plate."

This requires four sequential object rearrangements in a single language-conditioned rollout.

Recording setup

Item	Detail
Robot	SO-101 follower arm (6-DoF)
Teleoperation	JoyCon-based leader–follower
Cameras	1× global front scene camera + 1× side wrist camera
FPS	30
Action space	float32[6] — joint positions (shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper)
State space	float32[6] — joint positions (same as action)

Dataset stats

Field	Value
Format	LeRobot v2.1 (parquet + mp4 video streams)
Total episodes	49
Total frames	75,682
Tasks	1 (breakfast table-setting)
Size	~790 MB

Structure

├── data/chunk-000/
│   ├── episode_000000.parquet
│   ├── episode_000001.parquet
│   └── ...
├── videos/chunk-000/
│   ├── observation.images.front/
│   │   ├── episode_000000.mp4
│   │   └── ...
│   └── observation.images.side/
│       ├── episode_000000.mp4
│       └── ...
├── meta/
│   ├── info.json
│   ├── stats.json
│   ├── episodes.jsonl
│   └── tasks.jsonl
└── task_index.json

Loading with LeRobot

from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

ds = LeRobotDataset("jt-2026/so101-breakfast")
print(ds[0])  # {'observation.images.front': ..., 'observation.state': ..., 'action': ...}

Fine-tuning with openpi

git clone https://github.com/ljt228/pi05-so101-finetune.git
cd pi05-so101-finetune
uv sync

# Place dataset under ~/.cache/huggingface/lerobot/local/record-breakfast_49src_v21/
# Then train:
uv run scripts/train.py pi05_so101_lora_finetune --exp so101_lora_v1

Citation

@misc{so101_breakfast_2025,
  title={SO-101 Breakfast Table-Setting Dataset},
  author={jt-228},
  year={2025},
  howpublished={\url{https://huggingface.co/datasets/jt-2026/so101-breakfast}}
}