Research map / Vision-language and multimodal / Robot action and navigation
COSMO: Combination of Selective Memorization for Low-cost Vision-and-Language Navigation
Siqi Zhang, Yanyuan Qiao, Qunbo Wang, Zike Yan, Qi Wu, Zhihua Wei, Jing Liu
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 |
|---|---|---|---|
| add RSS instead of vanilla Mamba layer (ablation) | COSMO · REVERIE val unseen | SR/SPL: SR decrease of 3.61%, SPL decreased by 3.89% without RSS | Improved |
| add RSS instead of Bi-Mamba layer (ablation) | COSMO · REVERIE val unseen | SPL: 1.16% lower in SPL | Improved |
| add CS3 instead of Bi-Mamba over concatenated tokens (ablation) | COSMO · REVERIE val unseen | SR/SPL: decrease of 3.86% in SR and decline of 4.53% in SPL without CS3 | Improved |
| use Bi-Mamba instead of Mamba in single-stream VLN (ablation) | single-stream Mamba · REVERIE val unseen | SR: 32.25 -> 35.61 | Improved |
| use COSMO instead of vanilla single-stream Mamba (ablation) | DUET · REVERIE val unseen | SR: 32.25 -> 50.81 | Improved |
| use SSM before Transformer instead of Transformer-only (Trans+Trans) (ablation) | COSMO · REVERIE val unseen | SR/SPL: 49.47 -> 50.81 SR; 31.10 -> 35.83 SPL | Improved |
| SSM+SSM instead of SSM+Trans (ablation) | COSMO · REVERIE val unseen | SR: 50.81 -> 41.92 | Hurt |
| order SSM before Trans rather than Trans+SSM (ablation) | COSMO · REVERIE val unseen | SPL: 31.38 -> 35.83 | Improved |
| COSMO node encoder vs DUET-mini (ablation) | COSMO · REVERIE val unseen | SR: 46.61 -> 49.47 | Improved |
| add COSMO to DUET on long REVERIE instructions (>30 words) | DUET · REVERIE val unseen long instr. | SR: 37.46 -> 44.88 | Improved |
| add COSMO to BEVBert on long REVERIE instructions | BEVBert · REVERIE val unseen long instr. | SR: 38.87 -> 45.94 | Improved |
| add COSMO to DUET on R2R-CE test | DUET · R2R-CE test unseen | SR/SPL: 42 -> 47 SR; 36 -> 40 SPL | 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.
- DUET — DUET: Cross-Modal Semantic Grounding for Contrastive Zero-Shot Learning (AAAI 2023) · better on 12 of 13 shared results
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
- Towards Long-Horizon Vision-Language Navigation: Platform, Benchmark and Method (CVPR 2025 · similarity 0.89)
- Cross-Modal Map Learning for Vision and Language Navigation (CVPR 2022 · similarity 0.88)
- Multimodal Transformer with Variable-Length Memory for Vision-and-Language Navigation (ECCV 2022 · similarity 0.88)
- ImagineNav: Prompting Vision-Language Models as Embodied Navigator through Scene Imagination (ICLR 2025 · similarity 0.87)
- Run, Ruminate, and Regulate: A Dual-process Thinking System for Vision-and-Language Navigation (AAAI 2026 · similarity 0.87)
- OVER-NAV: Elevating Iterative Vision-and-Language Navigation with Open-Vocabulary Detection and StructurEd Representation (CVPR 2024 · similarity 0.86)
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.
