Dataset
RoboSet
Multi-task real robot demonstration data for imitation and offline RL.
Dataset
BC-Z
Google large-scale real robot demos with language-conditioned multi-task manipulation.
Overview
BC-Z is a Google large-scale real robot demonstration dataset with language-conditioned multi-task manipulation. Historically influential for showing that broad language-conditioned imitation can work on real robots at scale.
Access paths have evolved (papers, Kaggle, etc.); verify the current distribution channel and license before planning a project timeline.
Who it is for
Language-conditioned real IL researchers; historical baselines.
Key highlights
- Large multi-task real demos
- Language conditioning pioneer
- Influential BC-at-scale results
When to use
- Language multi-task real IL studies
- Comparisons to modern open sets
When not to use
- Need guaranteed long-term stable hosting without checking
Getting started
- 1Locate current official access
- 2Reproduce a published BC baseline
- 3Compare to Bridge/DROID
Papers & reading
How it compares
Caveats & pitfalls
- Access/license friction possible.
- May be superseded for new projects by DROID/Bridge.
Content reviewed 2026-07-23