Saved in:
Bibliographic Details
Main Authors: Tao, Xingjian, Wang, Yiwei, Cai, Yujun, Yang, Zhicheng, Tang, Jing
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2506.15425
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911011977560064
author Tao, Xingjian
Wang, Yiwei
Cai, Yujun
Yang, Zhicheng
Tang, Jing
author_facet Tao, Xingjian
Wang, Yiwei
Cai, Yujun
Yang, Zhicheng
Tang, Jing
contents Multimodal large language models (MLLMs) have enabled GUI agents to interact with operating systems by grounding language into spatial actions. Despite their promising performance, these models frequently exhibit hallucinations-systematic localization errors that compromise reliability. We propose a fine-grained evaluation framework that categorizes model predictions into four distinct types, revealing nuanced failure modes beyond traditional accuracy metrics. To better quantify model uncertainty, we introduce the Peak Sharpness Score (PSS), a metric that evaluates the alignment between semantic continuity and logits distribution in coordinate prediction. Building on this insight, we further propose Context-Aware Cropping, a training-free technique that improves model performance by adaptively refining input context. Extensive experiments demonstrate that our framework and methods provide actionable insights and enhance the interpretability and robustness of GUI agent behavior.
format Preprint
id arxiv_https___arxiv_org_abs_2506_15425
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Understanding GUI Agent Localization Biases through Logit Sharpness
Tao, Xingjian
Wang, Yiwei
Cai, Yujun
Yang, Zhicheng
Tang, Jing
Computation and Language
Multimodal large language models (MLLMs) have enabled GUI agents to interact with operating systems by grounding language into spatial actions. Despite their promising performance, these models frequently exhibit hallucinations-systematic localization errors that compromise reliability. We propose a fine-grained evaluation framework that categorizes model predictions into four distinct types, revealing nuanced failure modes beyond traditional accuracy metrics. To better quantify model uncertainty, we introduce the Peak Sharpness Score (PSS), a metric that evaluates the alignment between semantic continuity and logits distribution in coordinate prediction. Building on this insight, we further propose Context-Aware Cropping, a training-free technique that improves model performance by adaptively refining input context. Extensive experiments demonstrate that our framework and methods provide actionable insights and enhance the interpretability and robustness of GUI agent behavior.
title Understanding GUI Agent Localization Biases through Logit Sharpness
topic Computation and Language
url https://arxiv.org/abs/2506.15425