Guardado en:
Detalles Bibliográficos
Autores principales: Zhang, Shaojie, Fu, Pei, Zhang, Ruoceng, Yang, Jiahui, Du, Anan, Xi, Xiuwen, Wang, Shaokang, Huang, Ying, Qin, Bin, Luo, Zhenbo, Luan, Jian
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2510.27266
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910262255157248
author Zhang, Shaojie
Fu, Pei
Zhang, Ruoceng
Yang, Jiahui
Du, Anan
Xi, Xiuwen
Wang, Shaokang
Huang, Ying
Qin, Bin
Luo, Zhenbo
Luan, Jian
author_facet Zhang, Shaojie
Fu, Pei
Zhang, Ruoceng
Yang, Jiahui
Du, Anan
Xi, Xiuwen
Wang, Shaokang
Huang, Ying
Qin, Bin
Luo, Zhenbo
Luan, Jian
contents Autonomous graphical user interface (GUI) agents rely on accurate GUI grounding, which maps language instructions to on-screen coordinates, to execute user commands. However, current models, whether trained via supervised fine-tuning (SFT) or reinforcement learning (RL), often provide confidence signals that are poorly aligned with actual grounding correctness, leading to overconfident and unreliable predictions. To address this, we propose HyperClick, a novel framework that enhances trustworthy GUI grounding through self-critiqued reinforcement learning (SCRL). HyperClick combines a correctness reward and a confidence alignment reward, training the policy model to output both a click prediction and an explicit confidence estimate. This approach jointly optimizes grounding accuracy and confidence reliability through confidence-based self-assessment. Extensive experiments on challenging benchmarks show that HyperClick maintains strong grounding performance while providing better-aligned confidence estimates. By exposing uncertainty alongside GUI actions, HyperClick supports confidence-based abstention in GUI automation. Code will be released here.
format Preprint
id arxiv_https___arxiv_org_abs_2510_27266
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Enhancing Trustworthy GUI Grounding via Self-Critiqued Reinforcement Learning
Zhang, Shaojie
Fu, Pei
Zhang, Ruoceng
Yang, Jiahui
Du, Anan
Xi, Xiuwen
Wang, Shaokang
Huang, Ying
Qin, Bin
Luo, Zhenbo
Luan, Jian
Computer Vision and Pattern Recognition
Autonomous graphical user interface (GUI) agents rely on accurate GUI grounding, which maps language instructions to on-screen coordinates, to execute user commands. However, current models, whether trained via supervised fine-tuning (SFT) or reinforcement learning (RL), often provide confidence signals that are poorly aligned with actual grounding correctness, leading to overconfident and unreliable predictions. To address this, we propose HyperClick, a novel framework that enhances trustworthy GUI grounding through self-critiqued reinforcement learning (SCRL). HyperClick combines a correctness reward and a confidence alignment reward, training the policy model to output both a click prediction and an explicit confidence estimate. This approach jointly optimizes grounding accuracy and confidence reliability through confidence-based self-assessment. Extensive experiments on challenging benchmarks show that HyperClick maintains strong grounding performance while providing better-aligned confidence estimates. By exposing uncertainty alongside GUI actions, HyperClick supports confidence-based abstention in GUI automation. Code will be released here.
title Enhancing Trustworthy GUI Grounding via Self-Critiqued Reinforcement Learning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2510.27266