What it is
Sample embodied dataset for training vision-language-action models, tied to GigaBrain-0.7 and heterogeneous multi-robot data. Sample of heterogeneous embodied video-language-action data for VLA training, drawn from a 37,000+ hour multi-embodiment pretraining regime used by GigaBrain-0.7.
Key results
- Sample dataset on Hugging Face
- Apache-2.0 license
- Modality includes video; language is English
Useful for prototyping instruction-conditioned embodied policy and VLA data pipelines.