RDT-1B
Largest open diffusion foundation model for bimanual manipulation, pre-trained on 1M+ multi-robot episodes with a unified physically-interpretable action space, then fine-tuned on 6k self-collected bimanual episodes.
- 74%
- 34K
- 30 Hz
- 1.0M
Benchmarks
Numbers as reported by the authors on the named evaluation suite.
| Evaluation | Suite | Score |
|---|---|---|
Unseen-object dexterity | RDT bimanual eval | 74% |
vs Octo / OpenVLA | 7 bimanual tasks | 56% |
One-shot new skill | RDT eval | 61% |
Training
Diffusion transformer with unified action space
1.2B
48x A100 80GB
9.6K accelerator-hours
13K h
1.0M
Scalable diffusion transformer with QK-norm and MLP decoder; pre-train on heterogeneous robots then fine-tune bimanual.
One-shot learning of new skills after fine-tuning; unified action space avoids per-robot heads.
Linked datasets, robots & papers
Resolving related models, datasets and papers…