README.md

---
license: mit
---
# Aux-Think: Exploring Reasoning Strategies for Data-Efficient Vision-Language Navigation

<div align="center" class="authors">
    <a href="https://scholar.google.com/citations?user=IYLvsCQAAAAJ&hl" target="_blank">Shuo Wang</a>,
    <a href="https://yongcaiwang.github.io/" target="_blank">Yongcai Wang</a>,
    <a>Wanting Li</a>,
    <a href="https://scholar.google.com/citations?user=TkwComsAAAAJ&hl=en" target="_blank">Xudong Cai</a>, <br>
    <a>Yucheng Wang</a>,
    <a>Maiyue Chen</a>,
    <a>Kaihui Wang</a>,
    <a href="https://scholar.google.com/citations?user=HQfc8TEAAAAJ&hl=en" target="_blank">Zhizhong Su</a>,
    <a>Deying Li</a>,
    <a href="https://zhaoxinf.github.io/" target="_blank">Zhaoxin Fan</a>
</div>


<div align="center" style="line-height: 3;">
  <a href="https://horizonrobotics.github.io/robot_lab/aux-think" target="_blank" style="margin: 2px;">
    <img alt="Homepage" src="https://img.shields.io/badge/Homepage-green" style="display: inline-block; vertical-align: middle;"/>
  </a>
  <a href="https://arxiv.org/abs/2505.11886" target="_blank" style="margin: 2px;">
    <img alt="Paper" src="https://img.shields.io/badge/Paper-Arxiv-red" style="display: inline-block; vertical-align: middle;"/>
  </a>
</div>

## Introduction
Aux-Think internalizes Chain-of-Thought (CoT) only during training, enabling efficient Vision-Language Navigation without explicit reasoning at inference, and achieving strong performance with minimal data.

![](https://horizonrobotics.github.io/robot_lab/aux-think/stats/x3.png)


## Citation

```bibtex
@article{wang2025think,
  title={Aux-Think: Exploring Reasoning Strategies for Data-Efficient Vision-Language Navigation},
  author={Wang, Shuo and Wang, Yongcai and Li, Wanting and Cai, Xudong and Wang, Yucheng and Chen, Maiyue and Wang, Kaihui and Su, Zhizhong and Li, Deying and Fan, Zhaoxin},
  journal={arXiv preprint arXiv:2505.11886},
  year={2025}
}
```