← All notes

Summary

This paper provides datasets in standardized formats and models so the field can train generalist X-robot policies, adapted to new robots, tasks, and environments. The authors assemble a dataset from 22 different robots collected through a collaboration between 21 institutions, demonstrating 527 skills (160266 tasks). Later they also describe the repository as pooling 60 existing robot datasets and containing 1M+ real robot trajectories spanning 22 robot embodiments. A high-capacity model trained on this data, which they call RT-X (with RT-1-X and RT-2-X variants), shows positive transfer across robots. More details can be found on the project website, robotics-transformer-x.github.io. [S1]

Claims

• id: C1 claim: 'The dataset comes from 22 robots collected across 21 institutions.' source: S1 evidence: "22 different robots collected through a collaboration between 21 institutions"
• id: C2 claim: 'The authors report 527 skills (160266 tasks) in the assembled dataset.' source: S1 evidence: "demonstrating 527 skills (160266 tasks)"
• id: C3 claim: 'The project website is given as robotics-transformer-x.github.io.' source: S1 evidence: "The project website is robotics-transformer-x.github.io"

Verification

• Code available: Paper: the project website https://robotics-transformer-x.github.io/, which returned HTTP 200 on 2026-11-11. Models and data were not downloaded or run.
• Reproduced: not reproduced (2026-11-11). Code not run.
• Related work: Live notes ao-evals-multimodal-agents.
• Limits: Training mixtures used in the reported experiments use fewer embodiments than the full repository, and the discussion section lists open problems in cross-robot transfer and evaluation.
• License: CC BY 4.0 on the abs page (2026-11-11; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 13 Oct 2023 (v1); last revised 14 May 2025 (v9). Sources: [S1] https://arxiv.org/abs/2310.08864, v9 (14 May 2025). checked_on: 2026-11-11.

When to apply

Use when asking what the Open X-Embodiment dataset contains and how the RT-X models were trained across many robot platforms.

Facts

• id: F1 claim: 'The assembled dataset comes from 22 robots and 21 institutions, with 527 skills (160266 tasks).' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F2 claim: 'The authors also describe the repository as pooling 60 existing datasets and more than 1M real robot trajectories across 22 embodiments.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F3 claim: 'RT-X is the name they give the high-capacity model trained on this data; the project site is robotics-transformer-x.github.io.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0

Sources

• [S1] arXiv 2310.08864, https://arxiv.org/abs/2310.08864. License: CC BY 4.0. v9 (14 May 2025). Accessed 2026-11-11.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.

@misc{iil_note_31jrp6fk,
  author = {{'iinventedlight (AI-assisted)'}},
  title = {{Open X-Embodiment: Robotic Learning Datasets and RT-X Models}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-11},
  url = {https://iinventedlight.com/library/notes/2310-08864-open-x-embodiment/},
  note = {Permanent ID iil:note:31jrp6fk. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2310-08864-open-x-embodiment.md.