EUVIMEDEUVIMED entdecken
Crossref

Adaptive Look-Around: Enhancing Single-View VLN via Panoramic Perception and Failure-Memory-Guided Search

Lingfeng Li, Chuhua Huang, Rongjing Zhang, Xue Ren, Yongxing Shen

IEEE Robotics and Automation Letters · 2026-12 · 11 (12) · 13901-13908

DOI: 10.1109/lra.2026.3746085 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Robotics and Automation Letters, 2026-12. Verfasst von Lingfeng Li und 4 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Guizhou Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Zero-shot Vision-and-Language Navigation (VLN) in continuous environments is an important task for embodied agents. Most existing methods rely on panoramic observation, whereas robots in real environments are typically equipped with only a single-view RGB-D camera. This discrepancy in viewpoint results in limited environmental perception and makes the agent prone to repetitive search. To address these issues, we propose AFNav, a single-view zero-shot VLN framework that builds on the ideas of adaptive panoramic perception and failure memory. AFNav employs two core modules that enhance environmental perception and reduce repetitive search. Specifically, the Adaptive Panoramic Perception (APP) monitors navigation degradation to adaptively trigger a panoramic look-around, thereby enhancing environmental perception, while Failure-Memory-Guided Search (FMGS) explicitly constructs a search failure memory and uses it to suppress repetitive search of failed regions. Experimental results show that AFNav outperforms existing single-view zero-shot methods on both the R2R-CE and RxR-CE benchmarks, yielding a 2.0% improvement in SR and a 26.4% reduction in path length on R2R-CE, and a 2.1% improvement in SR and a 15.1% improvement in NDTW on RxR-CE, while also surpassing several panoramic-observation zero-shot methods.

Förderung

  • National Natural Science Foundation of China
  • Natural Science Foundation of Guizhou Province

Zitieren

APA

Li, L., Huang, C., Zhang, R., Ren, X., & Shen, Y. (2026). Adaptive Look-Around: Enhancing Single-View VLN via Panoramic Perception and Failure-Memory-Guided Search. IEEE Robotics and Automation Letters, 11(12), 13901-13908. https://doi.org/10.1109/lra.2026.3746085

Vancouver

Li L, Huang C, Zhang R, Ren X, Shen Y. Adaptive Look-Around: Enhancing Single-View VLN via Panoramic Perception and Failure-Memory-Guided Search. IEEE Robotics and Automation Letters. 2026;11(12):13901-13908. doi:10.1109/lra.2026.3746085

BibTeX
@article{Li2026,
  author = {Li, Lingfeng and Huang, Chuhua and Zhang, Rongjing and Ren, Xue and Shen, Yongxing},
  title = {Adaptive Look-Around: Enhancing Single-View VLN via Panoramic Perception and Failure-Memory-Guided Search},
  journal = {IEEE Robotics and Automation Letters},
  year = {2026},
  volume = {11},
  number = {12},
  pages = {13901--13908},
  doi = {10.1109/lra.2026.3746085}
}
Kostenlos starten →