FlySearch: Exploring how vision-language models explore

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Pardyl, Adam, Matuszek, Dominik, Przebieracz, Mateusz, Cygan, Marek, Zieliński, Bartosz, Wołczyk, Maciej
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914104718917632
author Pardyl, Adam
Matuszek, Dominik
Przebieracz, Mateusz
Cygan, Marek
Zieliński, Bartosz
Wołczyk, Maciej
author_facet Pardyl, Adam
Matuszek, Dominik
Przebieracz, Mateusz
Cygan, Marek
Zieliński, Bartosz
Wołczyk, Maciej
contents The real world is messy and unstructured. Uncovering critical information often requires active, goal-driven exploration. It remains to be seen whether Vision-Language Models (VLMs), which recently emerged as a popular zero-shot tool in many difficult tasks, can operate effectively in such conditions. In this paper, we answer this question by introducing FlySearch, a 3D, outdoor, photorealistic environment for searching and navigating to objects in complex scenes. We define three sets of scenarios with varying difficulty and observe that state-of-the-art VLMs cannot reliably solve even the simplest exploration tasks, with the gap to human performance increasing as the tasks get harder. We identify a set of central causes, ranging from vision hallucination, through context misunderstanding, to task planning failures, and we show that some of them can be addressed by finetuning. We publicly release the benchmark, scenarios, and the underlying codebase.
format Preprint
id arxiv_https___arxiv_org_abs_2506_02896
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle FlySearch: Exploring how vision-language models explore
Pardyl, Adam
Matuszek, Dominik
Przebieracz, Mateusz
Cygan, Marek
Zieliński, Bartosz
Wołczyk, Maciej
Computer Vision and Pattern Recognition
Machine Learning
Robotics
The real world is messy and unstructured. Uncovering critical information often requires active, goal-driven exploration. It remains to be seen whether Vision-Language Models (VLMs), which recently emerged as a popular zero-shot tool in many difficult tasks, can operate effectively in such conditions. In this paper, we answer this question by introducing FlySearch, a 3D, outdoor, photorealistic environment for searching and navigating to objects in complex scenes. We define three sets of scenarios with varying difficulty and observe that state-of-the-art VLMs cannot reliably solve even the simplest exploration tasks, with the gap to human performance increasing as the tasks get harder. We identify a set of central causes, ranging from vision hallucination, through context misunderstanding, to task planning failures, and we show that some of them can be addressed by finetuning. We publicly release the benchmark, scenarios, and the underlying codebase.
title FlySearch: Exploring how vision-language models explore
topic Computer Vision and Pattern Recognition
Machine Learning
Robotics
url https://arxiv.org/abs/2506.02896