LeRobot Community Datasets v3
LeRobot Community Datasets v3 is a large-scale community-contributed robotics dataset for vision-language-action learning, featuring 791 datasets across 46 robot types from contributors worldwide. Released in December 2025 under Apache 2.0 for commercial use, it enables cross-embodiment pretraining for generalist robot policies through the unified LeRobot v2.1 data format. The dataset aggregates contributions primarily focused on tabletop manipulation using the SO-100 and SO-101 low-cost robot arms — the dominant platform in the open-source robotics community. With 791 sub-datasets it is the largest community-curated robotics dataset collection available, enabling researchers to train across more embodiment configurations than any other single dataset. The Apache 2.0 license permits unrestricted commercial use.
| Year | 2025 |
|---|---|
| Embodiments | SO-100 arm, SO-101 arm, Franka Panda, various community robots |
| Modalities | rgb, proprioception, language |
| Task categories | manipulation, pick-and-place, bimanual, dexterous-manipulation |
| Data format | parquet, mp4, lerobot |
| License | Apache 2.0 |
| Access | open — commercial use permitted |
| Maintainer | Hugging Face, LeRobot Community |
| Origin country | US |
What is it?
LeRobot Community Datasets v3 is a large-scale community-contributed robotics dataset for vision-language-action learning, featuring 791 datasets across 46 robot types from contributors worldwide. Released in December 2025 under Apache 2.0, it aggregates contributions in the unified LeRobot v2.1 format primarily focused on tabletop manipulation using SO-100 and SO-101 low-cost robot arms. With 791 sub-datasets it is the largest community-curated robotics dataset collection, enabling cross-embodiment pretraining across more robot types than any other single dataset.
Who is it for?
Researchers building generalised manipulation policies using the Hugging Face LeRobot library. The community contribution model means coverage expands continuously as new contributors submit datasets — making it a living collection rather than a fixed snapshot. Particularly valuable for teams working with affordable robot hardware (SO-100, SO-101) who want to leverage community data rather than collecting their own.
Key specifications
- Sub-datasets: 791 from community contributors
- Robot types: 46 distinct embodiments
- Primary platform: SO-100 and SO-101 arms (low-cost open-source)
- Format: LeRobot v2.1 (Parquet, MP4)
- License: Apache 2.0 — commercial use permitted
How it compares
LeRobot Community v3 covers the most distinct robot types (46) of any single dataset collection. Open X-Embodiment covers 22 embodiments but with higher average data quality per embodiment. LeRobot Community v3's breadth comes at the cost of quality variability — community contributions range from expert to novice demonstrations. The SO-100/SO-101 focus reflects the growing open-source robot hardware community.
Limitations and access notes
Quality varies significantly across community contributions — no standardised data quality floor. Apache 2.0 permits unrestricted commercial use.
Linked professions
Frequently asked questions
What is the SO-100 robot?
The SO-100 is a low-cost open-source robot arm designed for robot learning research, costing approximately $100-200 to build. It is the dominant platform in the open-source robotics community, enabling researchers and hobbyists without access to expensive arms like Franka Panda to contribute to robot learning datasets.
Can LeRobot Community Datasets v3 be used commercially?
Yes. Apache 2.0 licensed, permitting unrestricted commercial use for all 791 sub-datasets in the collection.
How does the community contribution model work?
Contributors record demonstrations using any supported robot, format data in LeRobot v2.1 standard, and submit via Hugging Face. Hugging Face maintains quality guidelines but does not guarantee expert-level demonstrations. The collection grows continuously — v3 supersedes v1 and v2 with more datasets and robot types.
How do I access LeRobot Community Datasets v3?
Available on Hugging Face at huggingface.co/datasets/lerobot/community_v3. No registration required. The LeRobot library provides direct loading utilities.
What is the LeRobot v2.1 format?
LeRobot v2.1 is Hugging Face's standardised format for robot demonstration datasets, storing observations, actions, and metadata in Parquet files alongside MP4 video. It is the dominant community standard for robot learning datasets and is compatible with the LeRobot training library.