metadata
license: mit
pipeline_tag: robotics
Grounding Vision–Language–Action Models in Scientific Laboratories
Model Description
LabVLA is the first vision–language–action (VLA) model designed specifically for scientific laboratory environments, as introduced in LabVLA: Grounding Vision-Language-Action Models in Scientific Laboratories.
It combines a Qwen3-VL-4B-Instruct vision–language backbone with a DiT flow-matching action expert. The model is trained using a two-stage recipe:
- FAST action token pretraining: Makes the backbone action-aware.
- Flow matching posttraining: Attaches the DiT action expert under knowledge insulation to enable continuous control.
LabVLA addresses the gap in existing policies that are mostly trained on household data, enabling autonomous execution of scientific protocols involving laboratory instruments and transparent liquids.
How to Use
Download
huggingface-cli download zjunlp/LabVLA --local-dir LabVLA
Deployment
Serve the model over the OpenPI msgpack WebSocket protocol:
git clone https://github.com/zjunlp/LabVLA.git
cd LabVLA
bash deployment/deploy.sh
For training, data preparation, and more details, please refer to the GitHub repository.
Citation
@article{ren2026labvla,
title = {LabVLA: Grounding Vision-Language-Action Models in Scientific Laboratories},
author = {Ren, Baochang and Liu, Xinjie and Chen, Xi and Liu, Yanshuo and
Li, Chenxi and Gao, Daqi and Su, Zeqin and Xing, Jintao and
Xue, Zirui and Li, Rui and Zhao, Xiangyu and Qiao, Shuofei and
Pan, Minting and Zuo, Wangmeng and Bai, Lei and Zhou, Dongzhan and
Zhang, Ningyu and Chen, Huajun},
journal = {arXiv preprint arXiv:2606.13578},
year = {2026}
}