VLN-Cache: Enabling Token Caching for VLN Models with Visual/Semantic Dynamics Awareness
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913070475902976 |
|---|---|
| author | Zheng, Zihao Mao, Zhihao Zhou, Xingyue Chen, Jiayu Li, Maoliang Sun, Xinhao Zou, Hailong Zhang, Zhaobo Liu, Xuanzhe Cao, Donggang Mei, Hong Chen, Xiang |
| author_facet | Zheng, Zihao Mao, Zhihao Zhou, Xingyue Chen, Jiayu Li, Maoliang Sun, Xinhao Zou, Hailong Zhang, Zhaobo Liu, Xuanzhe Cao, Donggang Mei, Hong Chen, Xiang |
| contents | Vision-and-Language Navigation (VLN) increasingly relies on large vision-language models, but their inference cost conflicts with real-time deployment. Token caching is a promising training-free strategy that avoids redundant computation by reusing stable visual tokens across frames. However, existing methods assume a static camera and fixed semantic focus, assumptions that VLN fundamentally violates. We identify two failure modes: (1) visual dynamics, where viewpoint shift displaces token positions across frames, causing position-wise matching to pair misaligned content; (2) semantic dynamics, where token relevance shifts across task stages as navigation progresses, making cached states stale. We propose VLN-Cache, a visual-dynamic-aware and semantic-dynamic-aware caching framework that introduces view-aligned remapping to recover geometric correspondences and a task-relevance saliency filter to veto reuse at semantic transitions. A layer-adaptive entropy policy further balances the per-layer reuse budget. Experiments on the R2R-CE simulation benchmark show up to 1.52x speedup while maintaining competitive navigation success rates. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_07080 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | VLN-Cache: Enabling Token Caching for VLN Models with Visual/Semantic Dynamics Awareness Zheng, Zihao Mao, Zhihao Zhou, Xingyue Chen, Jiayu Li, Maoliang Sun, Xinhao Zou, Hailong Zhang, Zhaobo Liu, Xuanzhe Cao, Donggang Mei, Hong Chen, Xiang Robotics Machine Learning Vision-and-Language Navigation (VLN) increasingly relies on large vision-language models, but their inference cost conflicts with real-time deployment. Token caching is a promising training-free strategy that avoids redundant computation by reusing stable visual tokens across frames. However, existing methods assume a static camera and fixed semantic focus, assumptions that VLN fundamentally violates. We identify two failure modes: (1) visual dynamics, where viewpoint shift displaces token positions across frames, causing position-wise matching to pair misaligned content; (2) semantic dynamics, where token relevance shifts across task stages as navigation progresses, making cached states stale. We propose VLN-Cache, a visual-dynamic-aware and semantic-dynamic-aware caching framework that introduces view-aligned remapping to recover geometric correspondences and a task-relevance saliency filter to veto reuse at semantic transitions. A layer-adaptive entropy policy further balances the per-layer reuse budget. Experiments on the R2R-CE simulation benchmark show up to 1.52x speedup while maintaining competitive navigation success rates. |
| title | VLN-Cache: Enabling Token Caching for VLN Models with Visual/Semantic Dynamics Awareness |
| topic | Robotics Machine Learning |
| url | https://arxiv.org/abs/2603.07080 |