LabVLA
- AI model
- Robot arms and industry
LabVLA is a vision-language-action model that turns a Qwen3-VL backbone into a real-time robot controller for scientific laboratory experiments.
Made by ZJUNLP, China (checked on github.com/zjunlp (opens in a new tab)).
Source on GitHub: github.com/zjunlp/LabVLA (opens in a new tab). Facts read at commit 1b941bd (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 1b941bd (opens in a new tab), on 16 Oct 2026.
- Commercial use
- Yes
What it does
- Built on a Qwen3-VL-4B-Instruct backbone with a DiT flow-matching action expert that generates 50-step continuous action chunks.
- Trained in stages: FAST action-token pre-training, flow-matching post-training with knowledge insulation, then task fine-tuning.
- Provides a single training entrypoint for all stages, launched through Accelerate and DeepSpeed ZeRO-2 in bf16.
- Includes data tools to scan, clean, and compute normalization statistics for robot datasets, plus dataset schemas for new data.
- Offers a WebSocket inference server for connecting a robot or simulator client to run rollouts.
Written by Robotheca from the project's README (opens in a new tab) at commit 1b941bd.
Getting started
- Difficulty
- Advanced
- How to start
- The project's website (opens in a new tab)
Maintenance
- Status
- Active: a commit in the last 90 days
- Newest commit
- 9 Sept 2026 on main
- Latest release
- None published on GitHub
Checked on GitHub on 16 Oct 2026.
Links
Robots that can run 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.