Saved in:
Bibliographic Details
Main Authors: Lin, Sihao, Li, Zerui, Zhao, Xunyi, Zhou, Gengze, Wang, Liuyi, Wei, Rong, Tang, Rui, Li, Juncheng, Wang, Hanqing, Pang, Jiangmiao, Hengel, Anton van den, Liu, Jiajun, Wu, Qi
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.19021
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909973300117504
author Lin, Sihao
Li, Zerui
Zhao, Xunyi
Zhou, Gengze
Wang, Liuyi
Wei, Rong
Tang, Rui
Li, Juncheng
Wang, Hanqing
Pang, Jiangmiao
Hengel, Anton van den
Liu, Jiajun
Wu, Qi
author_facet Lin, Sihao
Li, Zerui
Zhao, Xunyi
Zhou, Gengze
Wang, Liuyi
Wei, Rong
Tang, Rui
Li, Juncheng
Wang, Hanqing
Pang, Jiangmiao
Hengel, Anton van den
Liu, Jiajun
Wu, Qi
contents Despite remarkable progress in Vision-Language Navigation (VLN), existing benchmarks remain confined to fixed, small-scale datasets with naive physical simulation. These shortcomings limit the insight that the benchmarks provide into sim-to-real generalization, and create a significant research gap. Furthermore, task fragmentation prevents unified/shared progress in the area, while limited data scales fail to meet the demands of modern LLM-based pretraining. To overcome these limitations, we introduce VLNVerse: a new large-scale, extensible benchmark designed for Versatile, Embodied, Realistic Simulation, and Evaluation. VLNVerse redefines VLN as a scalable, full-stack embodied AI problem. Its Versatile nature unifies previously fragmented tasks into a single framework and provides an extensible toolkit for researchers. Its Embodied design moves beyond intangible and teleporting "ghost" agents that support full-kinematics in a Realistic Simulation powered by a robust physics engine. We leverage the scale and diversity of VLNVerse to conduct a comprehensive Evaluation of existing methods, from classic models to MLLM-based agents. We also propose a novel unified multi-task model capable of addressing all tasks within the benchmark. VLNVerse aims to narrow the gap between simulated navigation and real-world generalization, providing the community with a vital tool to boost research towards scalable, general-purpose embodied locomotion agents.
format Preprint
id arxiv_https___arxiv_org_abs_2512_19021
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle VLNVerse: A Benchmark for Vision-Language Navigation with Versatile, Embodied, Realistic Simulation and Evaluation
Lin, Sihao
Li, Zerui
Zhao, Xunyi
Zhou, Gengze
Wang, Liuyi
Wei, Rong
Tang, Rui
Li, Juncheng
Wang, Hanqing
Pang, Jiangmiao
Hengel, Anton van den
Liu, Jiajun
Wu, Qi
Computer Vision and Pattern Recognition
Robotics
Despite remarkable progress in Vision-Language Navigation (VLN), existing benchmarks remain confined to fixed, small-scale datasets with naive physical simulation. These shortcomings limit the insight that the benchmarks provide into sim-to-real generalization, and create a significant research gap. Furthermore, task fragmentation prevents unified/shared progress in the area, while limited data scales fail to meet the demands of modern LLM-based pretraining. To overcome these limitations, we introduce VLNVerse: a new large-scale, extensible benchmark designed for Versatile, Embodied, Realistic Simulation, and Evaluation. VLNVerse redefines VLN as a scalable, full-stack embodied AI problem. Its Versatile nature unifies previously fragmented tasks into a single framework and provides an extensible toolkit for researchers. Its Embodied design moves beyond intangible and teleporting "ghost" agents that support full-kinematics in a Realistic Simulation powered by a robust physics engine. We leverage the scale and diversity of VLNVerse to conduct a comprehensive Evaluation of existing methods, from classic models to MLLM-based agents. We also propose a novel unified multi-task model capable of addressing all tasks within the benchmark. VLNVerse aims to narrow the gap between simulated navigation and real-world generalization, providing the community with a vital tool to boost research towards scalable, general-purpose embodied locomotion agents.
title VLNVerse: A Benchmark for Vision-Language Navigation with Versatile, Embodied, Realistic Simulation and Evaluation
topic Computer Vision and Pattern Recognition
Robotics
url https://arxiv.org/abs/2512.19021