VLN-NF: Feasibility-Aware Vision-and-Language Navigation with False-Premise Instructions

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Su, Hung-Ting, Wang, Ting-Jun, Yeh, Jia-Fong, Sun, Min, Hsu, Winston H.
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910145502511104
author Su, Hung-Ting
Wang, Ting-Jun
Yeh, Jia-Fong
Sun, Min
Hsu, Winston H.
author_facet Su, Hung-Ting
Wang, Ting-Jun
Yeh, Jia-Fong
Sun, Min
Hsu, Winston H.
contents Conventional Vision-and-Language Navigation (VLN) benchmarks assume instructions are feasible and the referenced target exists, leaving agents ill-equipped to handle false-premise goals. We introduce VLN-NF, a benchmark with false-premise instructions where the target is absent from the specified room and agents must navigate, gather evidence through in-room exploration, and explicitly output NOT-FOUND. VLN-NF is constructed via a scalable pipeline that rewrites VLN instructions using an LLM and verifies target absence with a VLM, producing plausible yet factually incorrect goals. We further propose REV-SPL to jointly evaluate room reaching, exploration coverage, and decision correctness. To address this challenge, we present ROAM, a two-stage hybrid that combines supervised room-level navigation with LLM/VLM-driven in-room exploration guided by a free-space clearance prior. ROAM achieves the best REV-SPL among compared methods, while baselines often under-explore and terminate prematurely under unreliable instructions. VLN-NF project page can be found at https://vln-nf.github.io/.
format Preprint
id arxiv_https___arxiv_org_abs_2604_10533
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle VLN-NF: Feasibility-Aware Vision-and-Language Navigation with False-Premise Instructions
Su, Hung-Ting
Wang, Ting-Jun
Yeh, Jia-Fong
Sun, Min
Hsu, Winston H.
Robotics
Computation and Language
Computer Vision and Pattern Recognition
Conventional Vision-and-Language Navigation (VLN) benchmarks assume instructions are feasible and the referenced target exists, leaving agents ill-equipped to handle false-premise goals. We introduce VLN-NF, a benchmark with false-premise instructions where the target is absent from the specified room and agents must navigate, gather evidence through in-room exploration, and explicitly output NOT-FOUND. VLN-NF is constructed via a scalable pipeline that rewrites VLN instructions using an LLM and verifies target absence with a VLM, producing plausible yet factually incorrect goals. We further propose REV-SPL to jointly evaluate room reaching, exploration coverage, and decision correctness. To address this challenge, we present ROAM, a two-stage hybrid that combines supervised room-level navigation with LLM/VLM-driven in-room exploration guided by a free-space clearance prior. ROAM achieves the best REV-SPL among compared methods, while baselines often under-explore and terminate prematurely under unreliable instructions. VLN-NF project page can be found at https://vln-nf.github.io/.
title VLN-NF: Feasibility-Aware Vision-and-Language Navigation with False-Premise Instructions
topic Robotics
Computation and Language
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.10533