F1: A Vision Language Action Model BridgingUnderstanding and Generation to Actions
🚀 Key Innovations
- 🧠 Predictive Inverse Dynamics: Visual foresight generation for planning-based control
- 🏗️ Mixture-of-Transformer: Three specialized experts (Understanding, Generation, Action)
- 📈 Three-Stage Training: Progressive alignment, pretraining, and adaptation
🤖 Real-World Robot Experiments
📊 Performance Summary
| Task | Platform | F1 | π0 | Improvement |
|---|---|---|---|---|
| Multi-task | Genie-1 | 82.2% | 65.2% | +17.0% |
| Adaptation | Franka | 66.7% | 53.3% | +13.4% |
| Long-horizon | ARX LIFT II | 40.0% | 0.0% | +40.0% |
| Dynamic Env | ARX LIFT II | 66.7% | 33.3% | +33.4% |
Usage
Please refer to our official repo F1-VLA.
📚 Citation
If you find our work helpful, please cite:
@article{f1_vla_2025,
title={F1: A Vision Language Action Model Bridging Understanding and Generation to Actions},
author={Qi Lv and Weijie Kong and Hao Li and Jia Zeng and Zherui Qiu and Delin Qu and Haoming Song and Qizhi Chen and Xiang Deng and Jiangmiao Pang},
journal={Conference/Journal Name},
year={2025},
url={https://arxiv.org/abs/2509.06951}
}
License
This work is under the cc-by-nc-sa-4.0.
Acknowledgements
This repository is based on Lerobot, Any4lerobot, and VAR.