---
license: apache-2.0
pipeline_tag: robotics
library_name: transformers
---
# Mantis

> This is the official checkpoint of **Mantis: A Versatile Vision-Language-Action Model
with Disentangled Visual Foresight**

- **Paper:** https://arxiv.org/pdf/2511.16175
- **Code:** https://github.com/zhijie-group/Mantis

### 🔥 Highlights
- **Disentangled Visual Foresight** augments action learning without overburdening the backbone.
- **Progressive Training** preserves the understanding capabilities of the backbone.
- **Adaptive Temporal Ensemble** reduces inference cost while maintaining stable control.

### How to use
This is the Mantis model trained on the [LIBERO](https://huggingface.co/datasets/Yysrc/mantis_libero_lerobot/tree/main) object dataset. For detailed usage please refer to [our repository](https://github.com/zhijie-group/Mantis).

### 📝 Citation
If you find our code or models useful in your work, please cite [our paper](https://arxiv.org/pdf/2511.16175):
```
@article{yang2025mantis,
  title={Mantis: A Versatile Vision-Language-Action Model with Disentangled Visual Foresight},
  author={Yang, Yi and Li, Xueqi and Chen, Yiyang and Song, Jin and Wang, Yihan and Xiao, Zipeng and Su, Jiadi and Qiaoben, You and Liu, Pengfei and Deng, Zhijie},
  journal={arXiv preprint arXiv:2511.16175},
  year={2025}
}
```