| title | SuperMap |
|---|---|
| subtitle | A Spatio-Temporal SLAM System for Visual-Language Navigation |
| layout | page |
| show_sidebar | false |
| hide_footer | false |
| hero_height | is-large |
| hero_image | /img/place_holder_01.png |
<div class="is-size-5 publication-authors" style="margin-top: 1.5rem;">
<span class="author-block">Carnegie Mellon University — AirLab</span>
</div>
<div class="publication-links">
<span class="link-block">
<a href="https://github.com/gfchen01/semantic_mapping" class="external-link button is-normal is-rounded is-dark" target="_blank">
<span class="icon"><i class="fab fa-github"></i></span>
<span>Code</span>
</a>
</span>
<span class="link-block">
<a href="#" class="external-link button is-normal is-rounded is-dark">
<span class="icon"><i class="fas fa-file-pdf"></i></span>
<span>Paper</span>
</a>
</span>
<span class="link-block">
<a href="#bibtex" class="external-link button is-normal is-rounded is-dark">
<span class="icon"><i class="fas fa-quote-left"></i></span>
<span>Citation</span>
</a>
</span>
</div>
</div>
</div>
SuperMap
Robotic navigation in human environments requires a spatio-temporal semantic representation that can reconcile open-vocabulary perception with long-term environmental changes. While foundation models provide strong zero-shot recognition, their predictions are intermittent and view-dependent, and naively integrating them into mapping pipelines leads to identity drift and stale semantics over time.
We present SuperMap, a 4D spatio-temporal mapping framework for language-guided navigation that integrates high-frequency geometric SLAM with asynchronous open-vocabulary perception. Our core contribution is a consistency-driven mapping engine that combines 3D-aware instance association and re-activation with a principled existence-and-label confidence update to maintain stable object identities and prune outdated map content under occlusions and scene changes.
SuperMap produces a queryable 4D scene-graph representation that interfaces naturally with Vision-Language Models by supporting compositional queries over object semantics, relations, and history. We demonstrate SuperMap on benchmarks and real robots, including dynamic scenes with appearance/disappearance and relocation, and provide ablations and runtime analysis. We will release the full system as open-source to provide the community with a deployable baseline for open-vocabulary spatio-temporal mapping.
An online robotic system that builds a persistent, queryable open-vocabulary 4D scene memory suitable for downstream language-conditioned tasks — running fully onboard in real time.
An online pipeline that integrates 2D–3D association, validation, and change-aware updates to maintain instance consistency under occlusions, partial observations, label variability, and scene change.
A 4D scene graph that incorporates spatial and temporal information for each object, equipping robots with instance-level reasoning — e.g., locating moved objects, recalling past scenes.
Three-Layer Pipeline
1
2
Per-frame open-vocabulary detections (GroundingDINO + SAM2) are associated to existing 3D map objects via a hybrid 2D–3D tracker. A probabilistic geometric consistency update and Bayesian semantic fusion maintain stable object identities across long time horizons under occlusions and scene change.
3
The object map is abstracted into a scene graph G = (V, Es, Et) with spatial edges (geometric predicates: on, beside, under) and temporal edges (object trajectory history). The graph is serialized as structured text for compositional VLM queries over object semantics, spatial relations, and history.
<div class="section-card" style="margin-bottom: 2rem;">
<div class="section-badge">Class-level Segmentation — ScanNet</div>
<p style="margin: 1.5rem 0 1rem;">SuperMap achieves competitive accuracy against state-of-the-art object-level mapping methods while running fully online.</p>
<table class="results-table">
<thead><tr><th>Method</th><th>Approach</th><th>mIoU (%)</th><th>f-mIoU (%)</th><th>Acc (%)</th></tr></thead>
<tbody>
<tr><td>ConceptGraphs</td><td>object-level</td><td>21.62</td><td>24.32</td><td>31.05</td></tr>
<tr><td>HOV-SG</td><td>object-level</td><td>26.79</td><td>36.05</td><td>35.17</td></tr>
<tr class="ours"><td>SuperMap (Ours)</td><td>object-level</td><td>27.42</td><td>43.50</td><td>55.48</td></tr>
</tbody>
</table>
</div>
<div class="section-card" style="margin-bottom: 2rem;">
<div class="section-badge">Instance-level Segmentation — ScanNet (mAP<sub>50</sub>)</div>
<p style="margin: 1.5rem 0 1rem;">SuperMap significantly outperforms prior scene-graph methods on instance-level detection.</p>
<table class="results-table">
<thead><tr><th>Method</th><th>Chair</th><th>Window</th><th>Refrigerator</th><th>Sofa</th><th>Door</th></tr></thead>
<tbody>
<tr><td>HOV-SG</td><td>4.58</td><td>0.00</td><td>0.00</td><td>30.00</td><td>9.70</td></tr>
<tr><td>ConceptGraphs</td><td>0.00</td><td>0.00</td><td>0.00</td><td>0.00</td><td>0.00</td></tr>
<tr class="ours"><td>SuperMap (Ours)</td><td>63.76</td><td>42.20</td><td>62.50</td><td>33.35</td><td>10.00</td></tr>
</tbody>
</table>
</div>
<div class="section-card">
<div class="section-badge">Spatio-Temporal Change Detection Recall</div>
<p style="margin: 1.5rem 0 1rem;">SuperMap achieves perfect recall on appearance events and strong recall on disappearance events, significantly outperforming prior methods.</p>
<table class="results-table">
<thead>
<tr><th>Method</th><th>Appeared (Bucket)</th><th>Appeared (Cart)</th><th>Appeared (Sign)</th><th>Disappeared (Plant)</th><th>Disappeared (Trash)</th><th>Disappeared (Chair)</th></tr>
</thead>
<tbody>
<tr><td>Khronos</td><td>—</td><td>—</td><td>—</td><td>—</td><td>—</td><td>—</td></tr>
<tr><td>DualMap</td><td>0.000</td><td>0.000</td><td>0.000</td><td>0.310</td><td>0.000</td><td>0.000</td></tr>
<tr class="ours"><td>SuperMap (Ours)</td><td>1.000</td><td>0.262</td><td>0.583</td><td>0.755</td><td>0.434</td><td>1.000</td></tr>
</tbody>
</table>
</div>
</div>
</div>