File size: 8,970 Bytes
1afa827 19e05fc d39ca20 19e05fc d39ca20 19e05fc d39ca20 19e05fc d39ca20 19e05fc d39ca20 19e05fc d39ca20 19e05fc d39ca20 19e05fc d39ca20 19e05fc |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 |
---
pipeline_tag: robotics
license: mit
---
# Rethinking the Embodied Gap in Vision-and-Language Navigation: A Holistic Study of Physical and Visual Disparities
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).
* **Project Page:** https://crystalsixone.github.io/vln_pe.github.io/
* **Code Repository:** https://github.com/InternRobotics/InternNav
## Abstract
Recent Vision-and-Language Navigation (VLN) advancements are promising, but their idealized assumptions about robot movement and control fail to reflect physically embodied deployment challenges. To bridge this gap, we introduce VLN-PE, a physically realistic VLN platform supporting humanoid, quadruped, and wheeled robots. For the first time, we systematically evaluate several ego-centric VLN methods in physical robotic settings across different technical pipelines, including classification models for single-step discrete action prediction, a diffusion model for dense waypoint prediction, and a train-free, map-based large language model (LLM) integrated with path planning. Our results reveal significant performance degradation due to limited robot observation space, environmental lighting variations, and physical challenges like collisions and falls. This also exposes locomotion constraints for legged robots in complex environments. VLN-PE is highly extensible, allowing seamless integration of new scenes beyond MP3D, thereby enabling more comprehensive VLN evaluation. Despite the weak generalization of current models in physical deployment, VLN-PE provides a new pathway for improving cross-embodiment's overall adaptability. We hope our findings and tools inspire the community to rethink VLN limitations and advance robust, practical VLN models. The code is available at this https URL .
## 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> |