DatasetFeatured
RH20T
Robotic manipulation dataset with rich contact and multi-modal sensing for contact-rich skills.
110k+ sequences (paper)ResearchBest for pretrainBest for finetuneSingle-armUpdated 2026-07-28
DatasetFeatured
DROID (full release notes)
Distributed robot interaction dataset across many scenes and skills — in-the-wild co-training staple.
Why we recommend it
Largest open in-the-wild robot interaction set
Overview
DROID (full release notes) is a curated dataset on Embodied AI Hub.
Distributed robot interaction dataset across many scenes and skills — in-the-wild co-training staple.
Why it matters: Largest open in-the-wild robot interaction set
Always verify install pins, licenses, and hardware requirements on the official site before production use.
Who it is for
Builders mapping open embodied stacks; researchers comparing SOTA baselines.
Key highlights
- Distributed robot interaction dataset across many scenes and skills — in-the-wild co-training staple.
- Type: dataset; tags: 数据集, 真机, 野外, 开源
- Hub recommended
When to use
- You need this category of open resource in your stack map
- You are shortlisting baselines before deep evaluation
When not to use
- License or hardware constraints block adoption
- You only need a closed commercial stack with vendor SLA
Getting started
- 1Open the official URL and read the README / model card.
- 2Check license and citation requirements.
- 3Run the smallest official example before scaling.
- 4Cross-link related hub resources for a full pipeline.
Papers & reading
Caveats & pitfalls
- Hub cards are curated summaries — verify upstream docs.
- Stars and release names drift; treat metadata as approximate.
Content reviewed 2026-07-28