# Predicting When to Trust Vision-Language Models for Spatial Reasoning

arXiv: https://arxiv.org/abs/2601.11644 · HTML: https://mlsanity.com/paper/2601.11644/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Spatial and embodied reasoning, one of 17 directions in LLM reasoning.
- Builds on 11 of its 12 references that are in the map.
- First posted on arXiv 14 Jan 2026.

## Placed in

- LLM reasoning / [Spatial and embodied reasoning](https://mlsanity.com/lr/spatial-and-embodied-reasoning/)
- Multimodal LLMs / vision-language models / [Embodied and 3D vision-language models](https://mlsanity.com/mllm/embodied-and-3d-vision-language-models/)

## Builds on (most cited first, 11 of 11)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://mlsanity.com/paper/2301.12597/) (2023)
- [XGBoost: A Scalable Tree Boosting System](https://mlsanity.com/paper/1603.02754/) (2016)
- [Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection](https://mlsanity.com/paper/2303.05499/) (2023)
- [Language Models (Mostly) Know What They Know](https://mlsanity.com/paper/2207.05221/) (2022)
- [Measuring and Narrowing the Compositionality Gap in Language Models](https://mlsanity.com/paper/2210.03350/) (2022)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models](https://mlsanity.com/paper/2303.08896/) (2023)
- [Visual Spatial Reasoning](https://mlsanity.com/paper/2205.00363/) (2022)
- [What's "up" with vision-language models? Investigating their struggle with spatial reasoning](https://mlsanity.com/paper/2310.19785/) (2023)
- [Reinforcing Spatial Reasoning in Vision-Language Models with Interwoven Thinking and Visual Drawing](https://mlsanity.com/paper/2506.09965/) (2025)

## Built on by (most cited first, 5 of 5)

- [Don't Blink: Evidence Collapse during Multimodal Reasoning](https://mlsanity.com/paper/2604.04207/) (2026)
- [MATEX: Multi-scale Attention and Text-guided Explainability of Medical Vision-Language Models](https://mlsanity.com/paper/2601.11666/) (2026)
- [When Depth Is Better Told Than Shown: Depth-Ordinal Prompting for Vision-Language Spatial Reasoning](https://mlsanity.com/paper/2607.11173/) (2026)
- [Symmetry-Aware Likelihood-Orbit Aggregation for Selective Left-Right Claim Verification](https://mlsanity.com/paper/2609.17004/) (2026)
- [RECAP: Relation Evidence Calibration for Detecting Spatial Relation Hallucinations in Vision-Language Models](https://mlsanity.com/paper/2609.26093/) (2026)

## Abstract (arXiv)

Vision-Language Models (VLMs) demonstrate impressive capabilities across multimodal tasks, yet exhibit systematic spatial reasoning failures, achieving only 49% (CLIP) to 54% (BLIP-2) accuracy on basic directional relationships. For safe deployment in robotics and autonomous systems, we need to predict when to trust VLM spatial predictions rather than accepting all outputs. We propose a vision-based confidence estimation framework that validates VLM predictions through independent geometric verification using object detection. Unlike text-based approaches relying on self-assessment, our method fuses four signals via gradient boosting: geometric alignment between VLM claims and coordinates, spatial ambiguity from overlap, detection quality, and VLM internal uncertainty. We achieve 0.674 AUROC on BLIP-2 (34.0% improvement over text-based baselines) and 0.583 AUROC on CLIP (16.1% improvement), generalizing across generative and classification architectures. Our framework enables selective prediction: at 60% target accuracy, we achieve 61.9% coverage versus 27.6% baseline (2.2x improvement) on BLIP-2. Feature analysis reveals vision-based signals contribute 87.4% of model importance versus 12.7% from VLM confidence, validating that external geometric verification outperforms self-assessment. We demonstrate reliable scene graph construction where confidence-based pruning improves precision from 52.1% to 78.3% while retaining 68.2% of edges.