README.md

---
pipeline_tag: robotics
library_name: transformers
license: mit
---

This repository contains models for the **VLN-PE Benchmark**, as presented in the paper [Rethinking the Embodied Gap in Vision-and-Language Navigation: A Holistic Study of Physical and Visual Disparities](https://huggingface.co/papers/2507.13019).

VLN-PE introduces a physically realistic Vision-and-Language Navigation platform supporting humanoid, quadruped, and wheeled robots, and systematically evaluates several ego-centric VLN methods in physical robotic settings.

For more details, visit the [project page](https://crystalsixone.github.io/vln_pe.github.io/) or the main [GitHub repository](https://github.com/InternRobotics/InternNav).

## VLN-PE Benchmark
<style type="text/css">
.tg  {border-collapse:collapse;border-spacing:0;}
.tg td{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
  overflow:hidden;padding:10px 5px;word-break:normal;}
.tg th{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
  font-weight:normal;overflow:hidden;padding:10px 5px;word-break:normal;}
.tg .tg-c3ow{border-color:inherit;text-align:center;vertical-align:top}
.tg .tg-0pky{border-color:inherit;text-align:left;vertical-align:top}
.tg .tg-fymr{border-color:inherit;font-weight:bold;text-align:left;vertical-align:top}
</style>
<table class="tg"><thead>
  <tr>
    <th class="tg-c3ow" rowspan="2"><span style="font-weight:bold">Model</span></th>
    <th class="tg-0pky" rowspan="2"><span style="font-weight:bold">Dataset/Benchmark</span></th>
    <th class="tg-c3ow" colspan="7"><span style="font-weight:bold">Val Seen</span></th>
    <th class="tg-c3ow" colspan="7"><span style="font-weight:bold">Val Unseen</span></th>
    <th class="tg-fymr" rowspan="2">Download</th>
  </tr>
  <tr>
    <th class="tg-fymr">TL</th>
    <th class="tg-fymr">NE</th>
    <th class="tg-fymr">FR</th>
    <th class="tg-fymr">StR</th>
    <th class="tg-fymr">OS</th>
    <th class="tg-fymr">SR</th>
    <th class="tg-fymr">SPL</th>
    <th class="tg-fymr">TL</th>
    <th class="tg-fymr">NE</th>
    <th class="tg-fymr">FR</th>
    <th class="tg-fymr">StR</th>
    <th class="tg-fymr">OS</th>
    <th class="tg-fymr">SR</th>
    <th class="tg-fymr">SPL</th>
  </tr></thead>
<tbody>
  <tr>
    <td class="tg-c3ow" colspan="17">Zero-shot transfer evaluation from VLN-CE</td>
  </tr>
  <tr>
    <td class="tg-0pky">Seq2Seq-Full</td>
    <td class="tg-0pky">R2R VLN-PE</td>
    <td class="tg-0pky">7.80</td>
    <td class="tg-0pky">7.62</td>
    <td class="tg-0pky">20.21</td>
    <td class="tg-0pky">3.04</td>
    <td class="tg-0pky">19.3</td>
    <td class="tg-0pky">15.2</td>
    <td class="tg-0pky">12.79</td>
    <td class="tg-0pky">7.73</td>
    <td class="tg-0pky">7.18</td>
    <td class="tg-0pky">18.04</td>
    <td class="tg-0pky">3.04</td>
    <td class="tg-0pky">22.42</td>
    <td class="tg-0pky">16.48</td>
    <td class="tg-0pky">14.11</td>
    <td class="tg-0pky"><a href="https://huggingface.co/InternRobotics/VLN-PE/tree/main/r2r/zero_shot/seq2seq" target="_blank" rel="noopener noreferrer">model</a></td>
  </tr>
  <tr>
    <td class="tg-0pky">CMA-Full</td>
    <td class="tg-0pky">R2R VLN-PE</td>
    <td class="tg-0pky">6.62</td>
    <td class="tg-0pky">7.37</td>
    <td class="tg-0pky">20.06</td>
    <td class="tg-0pky">3.95</td>
    <td class="tg-0pky">18.54</td>
    <td class="tg-0pky">16.11</td>
    <td class="tg-0pky">14.61</td>
    <td class="tg-0pky">6.58</td>
    <td class="tg-0pky">7.09</td>
    <td class="tg-0pky">17.07</td>
    <td class="tg-0pky">3.79</td>
    <td class="tg-0pky">20.86</td>
    <td class="tg-0pky">16.93</td>
    <td class="tg-0pky">15.24</td>
    <td class="tg-0pky"><a href="https://huggingface.co/InternRobotics/VLN-PE/tree/main/r2r/zero_shot/cma" target="_blank" rel="noopener noreferrer">model</a></td>
  </tr>
  <tr>
    <td class="tg-c3ow" colspan="17">Train on VLN-PE</td>
  </tr>
  <tr>
    <td class="tg-0pky">Seq2Seq</td>
    <td class="tg-0pky">R2R VLN-PE</td>
    <td class="tg-0pky">10.61</td>
    <td class="tg-0pky">7.53</td>
    <td class="tg-0pky">27.36</td>
    <td class="tg-0pky">4.26</td>
    <td class="tg-0pky">32.67</td>
    <td class="tg-0pky">19.75</td>
    <td class="tg-0pky">14.68</td>
    <td class="tg-0pky">10.85</td>
    <td class="tg-0pky">7.88</td>
    <td class="tg-0pky">26.8</td>
    <td class="tg-0pky">5.57</td>
    <td class="tg-0pky">28.13</td>
    <td class="tg-0pky">15.14</td>
    <td class="tg-0pky">10.77</td>
    <td class="tg-0pky"><a href="https://huggingface.co/InternRobotics/VLN-PE/tree/main/r2r/fine_tuned/seq2seq" target="_blank" rel="noopener noreferrer">model</a></td>
  </tr>
  <tr>
    <td class="tg-0pky">CMA</td>
    <td class="tg-0pky">R2R VLN-PE</td>
    <td class="tg-0pky">11.13</td>
    <td class="tg-0pky">7.59</td>
    <td class="tg-0pky">23.71</td>
    <td class="tg-0pky">3.19</td>
    <td class="tg-0pky">34.94</td>
    <td class="tg-0pky">21.58</td>
    <td class="tg-0pky">16.1</td>
    <td class="tg-0pky">11.16</td>
    <td class="tg-0pky">7.98</td>
    <td class="tg-0pky">22.64</td>
    <td class="tg-0pky">3.27</td>
    <td class="tg-0pky">33.11</td>
    <td class="tg-0pky">19.15</td>
    <td class="tg-0pky">14.05</td>
    <td class="tg-0pky"><a href="https://huggingface.co/InternRobotics/VLN-PE/tree/main/r2r/fine_tuned/cma" target="_blank" rel="noopener noreferrer">model</a></td>
  </tr>
  <tr>
    <td class="tg-0pky">RDP</td>
    <td class="tg-0pky">R2R VLN-PE</td>
    <td class="tg-0pky">13.26</td>
    <td class="tg-0pky">6.76</td>
    <td class="tg-0pky">27.51</td>
    <td class="tg-0pky">1.82</td>
    <td class="tg-0pky">38.6</td>
    <td class="tg-0pky">25.08</td>
    <td class="tg-0pky">17.07</td>
    <td class="tg-0pky">12.7</td>
    <td class="tg-0pky">6.72</td>
    <td class="tg-0pky">24.57</td>
    <td class="tg-0pky">3.11</td>
    <td class="tg-0pky">36.9</td>
    <td class="tg-0pky">25.24</td>
    <td class="tg-0pky">17.73</td>
    <td class="tg-0pky"><a href="https://huggingface.co/InternRobotics/VLN-PE/tree/main/r2r/fine_tuned/rdp" target="_blank" rel="noopener noreferrer">model</a></td>
  </tr>
  <tr>
    <td class="tg-0pky">Seq2Seq+</td>
    <td class="tg-0pky">R2R VLN-PE</td>
    <td class="tg-0pky">10.22</td>
    <td class="tg-0pky">7.75</td>
    <td class="tg-0pky">33.43</td>
    <td class="tg-0pky">3.19</td>
    <td class="tg-0pky">30.09</td>
    <td class="tg-0pky">16.86</td>
    <td class="tg-0pky">12.54</td>
    <td class="tg-0pky">9.88</td>
    <td class="tg-0pky">7.85</td>
    <td class="tg-0pky">26.27</td>
    <td class="tg-0pky">6.52</td>
    <td class="tg-0pky">28.79</td>
    <td class="tg-0pky">16.56</td>
    <td class="tg-0pky">12.7</td>
    <td class="tg-0pky"><a href="https://huggingface.co/InternRobotics/VLN-PE/tree/main/r2r/fine_tuned/seq2seq_plus" target="_blank" rel="noopener noreferrer">model</a></td>
  </tr>
  <tr>
    <td class="tg-0pky">CMA+</td>
    <td class="tg-0pky">R2R VLN-PE</td>
    <td class="tg-0pky">8.86</td>
    <td class="tg-0pky">7.14</td>
    <td class="tg-0pky">23.56</td>
    <td class="tg-0pky">3.5</td>
    <td class="tg-0pky">36.17</td>
    <td class="tg-0pky">25.84</td>
    <td class="tg-0pky">21.75</td>
    <td class="tg-0pky">8.79</td>
    <td class="tg-0pky">7.26</td>
    <td class="tg-0pky">21.75</td>
    <td class="tg-0pky">3.27</td>
    <td class="tg-0pky">31.4</td>
    <td class="tg-0pky">22.12</td>
    <td class="tg-0pky">18.65</td>
    <td class="tg-0pky"><a href="https://huggingface.co/InternRobotics/VLN-PE/tree/main/r2r/fine_tuned/cma_plus" target="_blank" rel="noopener noreferrer">model</a></td>
  </tr>
</tbody></table>

## Citation
If you find our work helpful, please cite:

```bibtex
@inproceedings{vlnpe,
  title={Rethinking the Embodied Gap in Vision-and-Language Navigation: A Holistic Study of Physical and Visual Disparities},
  author={Wang, Liuyi and Xia, Xinyuan and Zhao, Hui and Wang, Hanqing and Wang, Tai and Chen, Yilun and Liu, Chengju and Chen, Qijun and Pang, Jiangmiao},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  year={2025}
}
@misc{internnav2025,
    title = {{InternNav: InternRobotics'} open platform for building generalized navigation foundation models},
    author = {InternNav Contributors},
    howpublished={\url{https://github.com/InternRobotics/InternNav}},
    year = {2025}
}
```