Open X-Embodiment / RT-X
Pooled dataset from 33 labs and 22 robot types; RT-1-X beat single-lab models by 50% on average.
Open dataset with 1M+ episodes, 500+ skills and 150,000 tasks, plus RT-1-X and RT-2-X trained across embodiments. RT-1-X averaged 63% vs 41% for per-lab baselines; RT-2-X roughly tripled emergent-skill success (about 75% vs 25%).
- Date
- Tuesday, 3 October 2023
- Lab
- Google DeepMind
- Kind
- model
- Access
- research preview
Figures
| Measure | Value | Measured by |
|---|---|---|
| RT-1-X vs original (5 labs) | 63% vs 41% mean success; 50% relative improvement | company |
| Data pooled | 33 institutions, 22 robot types 1M+ episodes | company |
Dataset was an academic-industry collaboration, not DeepMind alone.
Sources
This record was checked against its sources on 6 October 2026. How we check