# GroundBench: A Factorized, Counterfactual Benchmark for Locating VLM Affordance Failures

arXiv: https://arxiv.org/abs/2609.13308 · HTML: https://mlsanity.com/paper/2609.13308/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Embodied and 3D vision-language models, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 16 of its 20 references that are in the map.
- First posted on arXiv 10 Sep 2026.

## Placed in

- 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, 16 of 16)

- [RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control](https://mlsanity.com/paper/2307.15818/) (2023)
- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [Do As I Can, Not As I Say: Grounding Language in Robotic Affordances](https://mlsanity.com/paper/2204.01691/) (2022)
- [PaLM-E: An Embodied Multimodal Language Model](https://mlsanity.com/paper/2303.03378/) (2023)
- [Shortcut Learning in Deep Neural Networks](https://mlsanity.com/paper/2004.07780/) (2020)
- [CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning](https://mlsanity.com/paper/1612.06890/) (2016)
- [Sanity Checks for Saliency Maps](https://mlsanity.com/paper/1810.03292/) (2018)
- [Beyond Accuracy: Behavioral Testing of NLP models with CheckList](https://mlsanity.com/paper/2005.04118/) (2020)
- [MDETR -- Modulated Detection for End-to-End Multi-Modal Understanding](https://mlsanity.com/paper/2104.12763/) (2021)
- [Winoground: Probing Vision and Language Models for Visio-Linguistic Compositionality](https://mlsanity.com/paper/2204.03162/) (2022)
- [Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering](https://mlsanity.com/paper/1712.00377/) (2017)
- [RUBi: Reducing Unimodal Biases in Visual Question Answering](https://mlsanity.com/paper/1906.10169/) (2019)
- [Where2Act: From Pixels to Actions for Articulated 3D Objects](https://mlsanity.com/paper/2101.02692/) (2021)
- [GAPartNet: Cross-Category Domain-Generalizable Object Perception and Manipulation via Generalizable and Actionable Parts](https://mlsanity.com/paper/2211.05272/) (2022)
- [ManipBench: Benchmarking Vision-Language Models for Low-Level Robot Manipulation](https://mlsanity.com/paper/2505.09698/) (2025)
- [Assessing VLM-Driven Semantic-Affordance Inference for Non-Humanoid Robot Morphologies](https://mlsanity.com/paper/2604.19509/) (2026)

## Abstract (arXiv)

A companion evaluation found that naming the target part in a manipulation prompt increased action accuracy by 0.32-0.63 across eight vision-language models, with no model outperforming a constant baseline until the part was named. However, naming the part supplies information that a real system must infer, confounding visual grounding, mechanical reasoning, and category-to-action association. We introduce GroundBench, a diagnostic benchmark that separates these explanations through six branch-and-merge conditions, each adding a controlled information bundle, and a counterfactual re-ask targeting a real alternate part visible in the same image. Across three OpenAI models and 1,068 predictions, supplying the target region without its identity leaves action accuracy at or below the 0.53 majority baseline (0.26, 0.26, and 0.53), although the models largely reproduce the supplied region. Supplying identity without location instead yields 0.74, 0.68, and 0.68. Every above-baseline gain in this curated set occurs where the supplied part category itself determines the action. A no-vision control leaves GPT-5's scores unchanged or improved, providing evidence consistent with substantial category-to-action association. GPT-4o mini declines on one condition, so this interpretation is not universal. Adding joint type and motion axis does not improve accuracy across six model-stratum comparisons. On 74 counterfactual pairs from 32 objects, GPT-5 achieves 0.86 pair-weighted compliance with a 0.07 shortcut rate but fails all observed push-to-lift-vertical cases. GroundBench identifies which supplied information changes affordance behavior and tests whether apparently grounded performance can be reproduced through textual shortcuts.