FlySearch: Exploring how vision-language models explore

TL;DR

FlySearch introduces a challenging 3D environment to evaluate vision-language models' ability to perform goal-driven exploration, revealing current limitations and potential improvements through finetuning.

Contribution

The paper presents FlySearch, a new benchmark environment for testing VLMs in complex exploration tasks, and analyzes their shortcomings compared to humans.

Findings

VLMs struggle with simple exploration tasks in complex scenes.

Performance gap between VLMs and humans widens with task difficulty.

Finetuning can mitigate some of the identified issues.

Abstract

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.