RT-X
Cross-embodiment robotics transformers trained on the Open X-Embodiment dataset — one model family spanning 22 robot types.
About
RT-X is what happened when Google DeepMind pooled real-robot data from 33 academic labs into a single training run. The Open X-Embodiment dataset behind it covers 22 robot types, more than a million trajectories, and over 150,000 tasks, all in one unified format. The headline models were RT-1-X and RT-2-X: the same RT-1/RT-2 architectures retrained on the pooled cross-embodiment mixture. DeepMind reported roughly a 50% lift for RT-1-X and a 3x lift for RT-2-X on evaluation tasks. Those are the lab's own numbers and I found no independent replication. The dataset and an RT-1-X TensorFlow checkpoint were released publicly, which made RT-X the starting point for most of the open VLA work that followed, OpenVLA included. The release paper appeared at ICRA 2024. The practical catch: only the RT-1-X checkpoint is public — the RT-2-X weights were never released — and the reported gains mostly show on tasks close to the pooled data distribution.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Paper (arXiv 2310.08864, ICRA 2024): RT-1-X ~50% improvement, RT-2-X ~3x improvement over RT-1/RT-2 (author-reported)
- No independent benchmark replication found
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (TIER verified, metrics-v1). Evidence: ICRA 2024 paper with a 33-lab collaboration; largest open real-robot dataset at release (1M+ trajectories, 22 robot types); public RT-1-X checkpoint on GitHub. Strengths: open data and weights, heavily cited, seeded the open VLA wave including OpenVLA. Docked for: (1) the 3x/50% gains are author-reported with no independent replication found; (2) only the RT-1-X checkpoint is public — RT-2-X weights were never released. License: Apache-2.0 (open_x_embodiment repo).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 8.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · DeepMind + 33 labs paper arXiv 2310.08864 (Oct 2023), ICRA 2024; 1M+ trajectories, 22 robot types, 150k+ tasks
- · Public RT-1-X TensorFlow checkpoint on GitHub; Open X-Embodiment dataset in RLDS format
- · RT-1-X ~50% lift, RT-2-X ~3x lift (author-reported); seeded the open VLA wave including OpenVLA