metadata
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- robotics
- manipulation
configs:
- config_name: default
data_files:
- split: train
path: data/**/*.parquet
mydataset (TEST)
Dataset Description
LeRobot v3.0 format dataset collected via the LeRobot Dataset Proxy Server. This dataset contains robot manipulation episodes with synchronized sensor data and video recordings.
Dataset Structure
mydataset/
├── data/
│ └── chunk-000/
│ ├── episode_000000.parquet
│ ├── episode_000001.parquet
│ └── ...
├── videos/
│ └── observation.images.front/
│ └── chunk-000/
│ ├── episode_000000.mp4
│ ├── episode_000001.mp4
│ └── ...
├── meta/
│ ├── info.json
│ ├── episodes.jsonl
│ └── tasks.jsonl
└── README.md
Data Format
Parquet Columns
Each episode parquet file contains the following columns:
| Column | Type | Description |
|---|---|---|
timestamp |
float64 | Timestamp in seconds |
frame_index |
int64 | Frame index within the episode |
episode_index |
int64 | Episode identifier |
action |
list[float] | Robot action vector |
observation.state |
list[float] | Robot joint states |
Video
- Format: MP4
- Camera views:
observation.images.front
Metadata (meta/info.json)
{
"codebase_version": "v3.0",
"robot_type": "so100",
"fps": 30,
"total_episodes": 0,
"total_frames": 0,
"total_tasks": 1,
"total_chunks": 1
}
Collection Details
- Robot: SO-100
- Control Frequency: 30 Hz
- Proxy Server: LeRobot Dataset Proxy (FastAPI + Celery)
- Storage: HuggingFace Hub (LFS for large files)
Usage
With LeRobot
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
dataset = LeRobotDataset("echwang/mydataset")
print(f"Episodes: {dataset.num_episodes}")
print(f"Frames: {dataset.num_frames}")
With surro-cli
# Download
surro dataset download --name mydataset
# Upload a new episode
surro episode upload --dataset mydataset --dir ./episode_001
With HuggingFace Datasets
from datasets import load_dataset
dataset = load_dataset("echwang/mydataset")
License
This dataset is released under the Apache 2.0 License.