Research map / Vision-language and multimodal / Robot action and navigation
Learning from Unlabeled 3D Environments for Vision-and-Language Navigation
Shizhe Chen, Pierre-Louis Guhur, Makarand Tapaswi, Cordelia Schmid, Ivan Laptev
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pretrain DUET on HM3D-AutoVLN | DUET · REVERIE val unseen | SPL: 33.73 -> 40.85 | Improved |
| pretrain DUET on HM3D-AutoVLN | DUET · SOON val unseen | SPL: 22.58 -> 30.69 | Improved |
| add HM3D-AutoVLN pretraining over REVERIE-only training (ablation) | DUET · REVERIE val unseen | SPL: 30.79 -> 39.36 | Improved |
| fine-tune jointly on HM3D-AutoVLN and REVERIE rather than REVERIE only (ablation) | DUET · REVERIE val unseen | SR: 50.87 -> 55.89 | Improved |
| add speaker data from ground-truth annotations on seen buildings (ablation) | DUET · REVERIE val unseen | SPL: 30.79 -> 33.73 | Improved |
| increase HM3D environments from 200 to 900 (ablation) | DUET · REVERIE val unseen | SPL: 37.65 -> 40.85 | Improved |
| use 900 environments instead of 400 at fixed 99,224 instructions (ablation) | DUET · REVERIE val unseen | SPL: 36.79 -> 38.35 | Improved |
| raise object distance limit d_o from 2 m to 3 m (ablation) | DUET · REVERIE val unseen | SPL: 40.85 -> 40.38 | Hurt |
| remove object distance limit (d_o infinite) (ablation) | DUET · REVERIE val unseen | SPL: 40.85 -> 38.03 | Hurt |
| use GPT-2 speaker instead of sentence-template instructions (ablation) | DUET · REVERIE val unseen | SR: 52.20 -> 55.89 | Improved |
| use GPT-2 speaker instead of LSTM speaker (ablation) | DUET · REVERIE val unseen | SR: 49.59 -> 55.89 | Improved |
| merge 2D predictions with cross-view consistency in 3D (ablation) | Mask2Former pseudo labels · 300 manually labeled boxes | label accuracy (%): 37.4 -> 58.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Airbert — Airbert: In-Domain Pretraining for Vision-and-Language Navigation (ICCV 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- D3D-VLP: Dynamic 3D Vision-Language-Planning Model for Embodied Grounding and Navigation (CVPR 2026 · similarity 0.90)
- Volumetric Environment Representation for Vision-Language Navigation (CVPR 2024 · similarity 0.90)
- Human-Aware Vision-and-Language Navigation: Bridging Simulation to Reality with Dynamic Human Interactions (NeurIPS 2024 · similarity 0.90)
- Dynam3D: Dynamic Layered 3D Tokens Empower VLM for Vision-and-Language Navigation (NeurIPS 2025 · similarity 0.90)
- RoomTour3D: Geometry-Aware Video-Instruction Tuning for Embodied Navigation (CVPR 2025 · similarity 0.89)
- LangMap: A Human-Verified Benchmark for Hierarchical Open-Vocabulary Goal Navigation (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
