FlySearch: Exploring how vision-language models explore
This addresses the limitation of VLMs for practical applications like robotics or search in unstructured environments, though it is incremental as it identifies issues and proposes finetuning solutions.
The paper tackles the problem of whether Vision-Language Models (VLMs) can perform active exploration in messy, real-world conditions, finding that state-of-the-art VLMs fail to reliably solve even simple tasks, with performance gaps up to increasing difficulty compared to humans.
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.