UI-Evol: Automatic Knowledge Evolving for Computer Use Agents

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Ziyun, Liu, Xinyi, Zhang, Xiaoyi, Wang, Jun, Chen, Gang, Lu, Yan
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911245209174016
author Zhang, Ziyun
Liu, Xinyi
Zhang, Xiaoyi
Wang, Jun
Chen, Gang
Lu, Yan
author_facet Zhang, Ziyun
Liu, Xinyi
Zhang, Xiaoyi
Wang, Jun
Chen, Gang
Lu, Yan
contents External knowledge has played a crucial role in the recent development of computer use agents. We identify a critical knowledge-execution gap: retrieved knowledge often fails to translate into effective real-world task execution. Our analysis shows even 90% correct knowledge yields only 41% execution success rate. To bridge this gap, we propose UI-Evol, a plug-and-play module for autonomous GUI knowledge evolution. UI-Evol consists of two stages: a Retrace Stage that extracts faithful objective action sequences from actual agent-environment interactions, and a Critique Stage that refines existing knowledge by comparing these sequences against external references. We conduct comprehensive experiments on the OSWorld benchmark with the state-of-the-art Agent S2. Our results demonstrate that UI-Evol not only significantly boosts task performance but also addresses a previously overlooked issue of high behavioral standard deviation in computer use agents, leading to superior performance on computer use tasks and substantially improved agent reliability.
format Preprint
id arxiv_https___arxiv_org_abs_2505_21964
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle UI-Evol: Automatic Knowledge Evolving for Computer Use Agents
Zhang, Ziyun
Liu, Xinyi
Zhang, Xiaoyi
Wang, Jun
Chen, Gang
Lu, Yan
Human-Computer Interaction
Computation and Language
External knowledge has played a crucial role in the recent development of computer use agents. We identify a critical knowledge-execution gap: retrieved knowledge often fails to translate into effective real-world task execution. Our analysis shows even 90% correct knowledge yields only 41% execution success rate. To bridge this gap, we propose UI-Evol, a plug-and-play module for autonomous GUI knowledge evolution. UI-Evol consists of two stages: a Retrace Stage that extracts faithful objective action sequences from actual agent-environment interactions, and a Critique Stage that refines existing knowledge by comparing these sequences against external references. We conduct comprehensive experiments on the OSWorld benchmark with the state-of-the-art Agent S2. Our results demonstrate that UI-Evol not only significantly boosts task performance but also addresses a previously overlooked issue of high behavioral standard deviation in computer use agents, leading to superior performance on computer use tasks and substantially improved agent reliability.
title UI-Evol: Automatic Knowledge Evolving for Computer Use Agents
topic Human-Computer Interaction
Computation and Language
url https://arxiv.org/abs/2505.21964