@inproceedings{ma-etal-2026-ddcg,
title = "{DDCG}: Decoupled Dual-Critic Guidance for Embodied Agents",
author = "Ma, Shaojin and
Zhang, Min and
Tang, Hongyao and
Wang, Ruitao and
Hao, Jianye and
Zheng, Yan",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.445/",
pages = "9339--9360",
ISBN = "979-8-89176-462-0",
abstract = "Large Language Models (LLMs) have enabled embodied agents to perform high-level planning, yet grounding language-based plans into the physical world remains challenging. While feedback-based closed-loop planning is the dominant paradigm, existing methods typically rely on a single scalar feedback signal, which conflates physically infeasible actions that violate hard constraints with strategically sub-optimal but valid choices. We identify this structural limitation as Signal Confounding, which leads to ambiguous feedback and unreliable plan correction. To address this issue, we propose Decoupled Dual-Critic Guidance (DDCG), a framework that explicitly disentangles feedback into two complementary signals: a Feasibility Critic ($C_F$) that enforces physical and logical executability constraints, and a Quality Critic ($C_Q$) that evaluates the strategic value of feasible actions. This decoupled guidance enables precise error attribution during planning without requiring parameter updates to the LLM planner. We provide a theoretical interpretation of DDCG as critic-regularized planning with hard feasibility constraints. Experiments on two embodied planning benchmarks, VirtualHome and ScienceWorld, show that DDCG consistently improves plan executability and task success, particularly in long-horizon settings, highlighting the importance of structured feedback for reliable LLM-based planning."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ma-etal-2026-ddcg">
<titleInfo>
<title>DDCG: Decoupled Dual-Critic Guidance for Embodied Agents</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shaojin</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Min</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hongyao</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruitao</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianye</namePart>
<namePart type="family">Hao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yan</namePart>
<namePart type="family">Zheng</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Large Language Models (LLMs) have enabled embodied agents to perform high-level planning, yet grounding language-based plans into the physical world remains challenging. While feedback-based closed-loop planning is the dominant paradigm, existing methods typically rely on a single scalar feedback signal, which conflates physically infeasible actions that violate hard constraints with strategically sub-optimal but valid choices. We identify this structural limitation as Signal Confounding, which leads to ambiguous feedback and unreliable plan correction. To address this issue, we propose Decoupled Dual-Critic Guidance (DDCG), a framework that explicitly disentangles feedback into two complementary signals: a Feasibility Critic (C_F) that enforces physical and logical executability constraints, and a Quality Critic (C_Q) that evaluates the strategic value of feasible actions. This decoupled guidance enables precise error attribution during planning without requiring parameter updates to the LLM planner. We provide a theoretical interpretation of DDCG as critic-regularized planning with hard feasibility constraints. Experiments on two embodied planning benchmarks, VirtualHome and ScienceWorld, show that DDCG consistently improves plan executability and task success, particularly in long-horizon settings, highlighting the importance of structured feedback for reliable LLM-based planning.</abstract>
<identifier type="citekey">ma-etal-2026-ddcg</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.445/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>9339</start>
<end>9360</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T DDCG: Decoupled Dual-Critic Guidance for Embodied Agents
%A Ma, Shaojin
%A Zhang, Min
%A Tang, Hongyao
%A Wang, Ruitao
%A Hao, Jianye
%A Zheng, Yan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F ma-etal-2026-ddcg
%X Large Language Models (LLMs) have enabled embodied agents to perform high-level planning, yet grounding language-based plans into the physical world remains challenging. While feedback-based closed-loop planning is the dominant paradigm, existing methods typically rely on a single scalar feedback signal, which conflates physically infeasible actions that violate hard constraints with strategically sub-optimal but valid choices. We identify this structural limitation as Signal Confounding, which leads to ambiguous feedback and unreliable plan correction. To address this issue, we propose Decoupled Dual-Critic Guidance (DDCG), a framework that explicitly disentangles feedback into two complementary signals: a Feasibility Critic (C_F) that enforces physical and logical executability constraints, and a Quality Critic (C_Q) that evaluates the strategic value of feasible actions. This decoupled guidance enables precise error attribution during planning without requiring parameter updates to the LLM planner. We provide a theoretical interpretation of DDCG as critic-regularized planning with hard feasibility constraints. Experiments on two embodied planning benchmarks, VirtualHome and ScienceWorld, show that DDCG consistently improves plan executability and task success, particularly in long-horizon settings, highlighting the importance of structured feedback for reliable LLM-based planning.
%U https://aclanthology.org/2026.findings-emnlp.445/
%P 9339-9360
Markdown (Informal)
[DDCG: Decoupled Dual-Critic Guidance for Embodied Agents](https://aclanthology.org/2026.findings-emnlp.445/) (Ma et al., Findings 2026)
ACL
- Shaojin Ma, Min Zhang, Hongyao Tang, Ruitao Wang, Jianye Hao, and Yan Zheng. 2026. DDCG: Decoupled Dual-Critic Guidance for Embodied Agents. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 9339–9360, Budapest, Hungary. Association for Computational Linguistics.