CrossFormer
- AI model
- Humanoids and robot learning
CrossFormer is a transformer-based robot policy trained on about 900K trajectories across 20 robot embodiments, for controlling many different robot types from one model.
- Stale
Made by Robotic AI & Learning Lab Berkeley.
Source on GitHub: github.com/rail-berkeley/crossformer (opens in a new tab). Facts read at commit 4a56b64 (opens in a new tab).
Licenses
- Code
- MIT (opens in a new tab)
Read by Robotheca from LICENSE (opens in a new tab) (MIT) at commit 4a56b64 (opens in a new tab), on 16 Oct 2026.
- Weights
- MIT (opens in a new tab)
Read on huggingface.co/rail-berkeley/crossformer (opens in a new tab).
- Commercial use
- Yes
Labels
- StaleNo commit in 12 months: the newest is from 26 Aug 2024.
What it does
- Controls single and dual arm manipulators, wheeled robots, quadcopters, and quadrupeds with one pretrained policy.
- Can be instructed through language commands or goal images.
- Uses a modular attention structure so it can be finetuned to new sensors, action spaces, and morphologies with a small dataset.
- Pretrained checkpoint is a 130M parameter model, available on Hugging Face and loadable from Python.
- Includes scripts for pretraining, finetuning on RLDS-format datasets, and serving the model to remote robots.
- Built on the Octo codebase.
Written by Robotheca from the project's README (opens in a new tab) at commit 4a56b64.
Getting started
- Difficulty
- Advanced
- How to start
- The project's README (opens in a new tab)
Maintenance
- Status
- Stale: no commit in 12 months
- Newest commit
- 26 Aug 2024 on main
- Latest release
- None published on GitHub
Checked on GitHub on 16 Oct 2026.
Links
Robots on Robotheca that use it
No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.
Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.