UniVLA
- AI model
- Humanoids and robot learning
UniVLA is a vision-language-action model that plans in an embodiment-agnostic action space, learning task-centric latent actions from cross-embodiment videos. It targets generalist robot policies.
Made by OpenDriveLab, Hong Kong SAR China (checked on github.com/OpenDriveLab (opens in a new tab)).
Source on GitHub: github.com/OpenDriveLab/UniVLA (opens in a new tab). Facts read at commit 0ab9e9d (opens in a new tab).
Licenses
- Code
- Apache-2.0 (opens in a new tab)
Read by Robotheca from LICENSE (opens in a new tab) (Apache-2.0) at commit 0ab9e9d (opens in a new tab), on 16 Oct 2026.
- Commercial use
- Yes
What it does
- Plans in a unified, embodiment-agnostic action space as a recipe toward generalist policies.
- Extracts task-centric latent actions from cross-embodiment videos using a VQ-VAE-based latent action model.
- Pretrains on Open X-Embodiment and Ego4D data, then adds lightweight embodiment-specific action decoder heads.
- Provides pretrained checkpoints and fine-tuned models for LIBERO, CALVIN, Room2Room and SimplerEnv evaluations.
- Reports state-of-the-art results on LIBERO and SimplerEnv WidowX tasks, plus real-world experiments on the AgiLex platform.
Written by Robotheca from the project's README (opens in a new tab) at commit 0ab9e9d.
Getting started
- Difficulty
- Advanced
- How to start
- The project's website (opens in a new tab)
Maintenance
- Status
- Maintained: a commit in the last 12 months
- Newest commit
- 19 Nov 2025 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.