---
license: apache-2.0
---
# RoboDual

This repo contains the checkpoint for the generalist policy of our paper: \
**Towards Synergistic, Generalized and Efficient Dual-System for Robotic Manipulation**

## Performance
The pre-trained OpenVLA with action chunk size of 8 can achieve around 3.27 average length on CALVIN ABC-D test suite:
|Method| 1    | 2    | 3    | 4    | 5    | Avg.Len. |
|------|------|------|------|------|------|---------|
|RoboDual (Generalist-only)| 91.3 | 77.8 | 62.0 | 52.1 | 43.5 | 3.27    |

## How to use


You can simply modify the ```vla_path``` in [FeintuneConfig](https://github.com/OpenDriveLab/RoboDual/blob/4a31a7a59e2082c67c659c629822b708cb029e91/vla-scripts/train_spacialist_calvin.py#L169) of the specialist training script according to your local path.

## Citation
If you find our code or models useful in your work, please cite [our paper](https://arxiv.org/abs/2410.08001):

```bibtex
@article{bu2024robodual,
  title={Towards Synergistic, Generalized, and Efficient Dual-System for Robotic Manipulation}, 
  author={Qingwen Bu and Hongyang Li and Li Chen and Jisong Cai and Jia Zeng and Heming Cui and Maoqing Yao and Yu Qiao},
  journal={arXiv preprint arXiv:2410.08001},
  year={2024}
}