<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="huang-etal-2026-beyond-planning">
    <titleInfo>
        <title>Beyond Planning: Sparse Context Reset for Long-Horizon GUI Agents</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Weiyao</namePart>
        <namePart type="family">Huang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Xinye</namePart>
        <namePart type="family">Wu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Zeming</namePart>
        <namePart type="family">Liu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Xiaoyu</namePart>
        <namePart type="family">Shen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yunpu</namePart>
        <namePart type="family">Ma</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Hung-Yi</namePart>
            <namePart type="family">Lee</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Sunipa</namePart>
            <namePart type="family">Dev</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jessy</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Desmond</namePart>
            <namePart type="family">Elliott</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-459-0</identifier>
    </relatedItem>
    <abstract>Graphical User Interface (GUI) agents powered by vision-language models (VLMs) have emerged as a promising paradigm for computer-use automation, but remain unreliable on long-horizon tasks. Existing planner–executor frameworks address this by shifting more reasoning to the planner, thereby decomposing rather than decoupling long-horizon reasoning. Our failure analysis reveals that many long-horizon failures are not pure capability gaps: from the same failure state, executors recover substantially more often when their accumulated context is reset with a locally executable subtask. To bridge this gap, we propose SCoR, a sparse context reset framework that restores the executor’s short-horizon capability by resetting its accumulated task context. At subtask boundaries, a lightweight Refiner emits a locally executable subtask and resets the Executor’s context; the executor then autonomously executes multiple GUI actions within the reset context. We train the Refiner with turn-level GRPO. In OSWorld, SCoR achieves a success rate of 53.16%, yielding a 25.41% relative improvement over the standalone EvoCUA-8B executor. Our code is available at https://github.com/huia711/SCoR.</abstract>
    <identifier type="citekey">huang-etal-2026-beyond-planning</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.1634/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>37145</start>
            <end>37163</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Beyond Planning: Sparse Context Reset for Long-Horizon GUI Agents
%A Huang, Weiyao
%A Wu, Xinye
%A Liu, Zeming
%A Shen, Xiaoyu
%A Ma, Yunpu
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F huang-etal-2026-beyond-planning
%X Graphical User Interface (GUI) agents powered by vision-language models (VLMs) have emerged as a promising paradigm for computer-use automation, but remain unreliable on long-horizon tasks. Existing planner–executor frameworks address this by shifting more reasoning to the planner, thereby decomposing rather than decoupling long-horizon reasoning. Our failure analysis reveals that many long-horizon failures are not pure capability gaps: from the same failure state, executors recover substantially more often when their accumulated context is reset with a locally executable subtask. To bridge this gap, we propose SCoR, a sparse context reset framework that restores the executor’s short-horizon capability by resetting its accumulated task context. At subtask boundaries, a lightweight Refiner emits a locally executable subtask and resets the Executor’s context; the executor then autonomously executes multiple GUI actions within the reset context. We train the Refiner with turn-level GRPO. In OSWorld, SCoR achieves a success rate of 53.16%, yielding a 25.41% relative improvement over the standalone EvoCUA-8B executor. Our code is available at https://github.com/huia711/SCoR.
%U https://aclanthology.org/2026.emnlp-main.1634/
%P 37145-37163

Markdown (Informal)

[Beyond Planning: Sparse Context Reset for Long-Horizon GUI Agents](https://aclanthology.org/2026.emnlp-main.1634/) (Huang et al., EMNLP 2026)

• Beyond Planning: Sparse Context Reset for Long-Horizon GUI Agents (Huang et al., EMNLP 2026)

ACL

• Weiyao Huang, Xinye Wu, Zeming Liu, Xiaoyu Shen, and Yunpu Ma. 2026. Beyond Planning: Sparse Context Reset for Long-Horizon GUI Agents. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 37145–37163, Budapest, Hungary. Association for Computational Linguistics.