Search papers, labs, and topics across Lattice.
The paper introduces SpatialUncertain, a benchmark to evaluate VLMs' ability to recognize when spatial questions are unanswerable due to occlusion or perspective ambiguity. Experiments on various VLMs reveal a tendency for overconfident answering, with accuracy plummeting to 30% under occlusion and below 10% under perspective ambiguity. Furthermore, VLMs struggle to identify informative viewpoints that could resolve perspective ambiguity, highlighting a critical gap in their spatial reasoning capabilities.
VLMs confidently hallucinate answers to spatial reasoning questions even when visual evidence is occluded or misleading, achieving near-random performance in identifying viewpoints that could resolve the ambiguity.
Spatial reasoning is a fundamental capability for vision-language models (VLMs) deployed in real-world environments. However, visual observations are inherently limited representations of a 3D world: occlusion can render objects invisible, and perspective can make geometric properties misleading. Despite this, existing spatial reasoning benchmarks typically assume that observations are sufficient and reliable, focusing on whether models produce correct answers rather than whether they recognize when a question cannot be answered and what additional observations would be needed. In this work, we challenge this assumption by constructing a controlled evaluation framework, SpatialUncertain, and introducing two types of observation challenges: (1) occlusion, which hides target information, and (2) perspective ambiguity, which produces misleading visual cues. For each configuration, we design spatial questions that are answerable under clean observations but require abstention under the introduced challenges. We further evaluate whether models can identify which additional viewpoints would resolve perspective ambiguity. Our results across a diverse set of frontier open- and closed-source VLMs reveal two consistent failure modes. First, models are prone to overconfident answering, attempting to solve spatial reasoning tasks even when visual evidence is incomplete or misleading, with average accuracy around 30\% under occlusion and below 10\% under perspective ambiguity. Second, even when additional views are available, some models perform near random chance in identifying which would provide reliable evidence. Together, our findings call for moving beyond answer correctness toward evaluating whether models know when to abstain and how to seek reliable evidence.
