@inproceedings{chen-etal-2026-efficient-agentic,
title = "Efficient Agentic Reinforcement Learning with On-Policy Intrinsic Knowledge Boundary Enhancement",
author = "Chen, Dingwei and
Zong, Zefang and
ma, Zhipeng and
Luo, Leo and
Li, Yang and
Li, Chengming and
Chen, Peng and
Jiang, Jie",
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.1959/",
doi = "10.18653/2026.findings-emnlp.1959",
pages = "41777--41795",
ISBN = "979-8-89176-462-0",
abstract = "Agentic reinforcement learning (RL) has proven effective for training LLM-based agents with external tool-use capabilities. However, we identify that agentic RL training induces increasing redundant tool calls and blurs the model{'}s intrinsic knowledge boundary, where the model fails to distinguish when tools are needed versus when parametric knowledge suffices. Existing solutions based on reward shaping create coarse-grained optimization targets that tend to incentivize indiscriminate tool-call suppression, leading to reward hacking. In this paper, we propose **AKBE** (**A**gentic **K**nowledge **B**oundary **E**nhancement), an on-policy method that dynamically probes the model{'}s intrinsic knowledge boundary through dual-path (with-tool and no-tool) rollouts during training. We define the knowledge boundary as the per-instance determination of whether tools are required and the minimum tool calls necessary. By comparing correctness across paths, AKBE categorizes trajectories and constructs targeted supervisory signals that guide efficient tool-use patterns for each question. These signals are integrated seamlessly into the agentic RL training loop. Experiments on seven QA benchmarks demonstrate that AKBE improves task accuracy by on average $+1.85$ and reduces tool calls by 18{\%} over standard agentic RL, yielding 25{\%} higher tool productivity without any accuracy-efficiency trade-off. Further analysis suggests its plug-and-play compatibility across different RL algorithms and the mechanism of each signal category."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chen-etal-2026-efficient-agentic">
<titleInfo>
<title>Efficient Agentic Reinforcement Learning with On-Policy Intrinsic Knowledge Boundary Enhancement</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dingwei</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zefang</namePart>
<namePart type="family">Zong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhipeng</namePart>
<namePart type="family">ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Leo</namePart>
<namePart type="family">Luo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chengming</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peng</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jie</namePart>
<namePart type="family">Jiang</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>Agentic reinforcement learning (RL) has proven effective for training LLM-based agents with external tool-use capabilities. However, we identify that agentic RL training induces increasing redundant tool calls and blurs the model’s intrinsic knowledge boundary, where the model fails to distinguish when tools are needed versus when parametric knowledge suffices. Existing solutions based on reward shaping create coarse-grained optimization targets that tend to incentivize indiscriminate tool-call suppression, leading to reward hacking. In this paper, we propose **AKBE** (**A**gentic **K**nowledge **B**oundary **E**nhancement), an on-policy method that dynamically probes the model’s intrinsic knowledge boundary through dual-path (with-tool and no-tool) rollouts during training. We define the knowledge boundary as the per-instance determination of whether tools are required and the minimum tool calls necessary. By comparing correctness across paths, AKBE categorizes trajectories and constructs targeted supervisory signals that guide efficient tool-use patterns for each question. These signals are integrated seamlessly into the agentic RL training loop. Experiments on seven QA benchmarks demonstrate that AKBE improves task accuracy by on average +1.85 and reduces tool calls by 18% over standard agentic RL, yielding 25% higher tool productivity without any accuracy-efficiency trade-off. Further analysis suggests its plug-and-play compatibility across different RL algorithms and the mechanism of each signal category.</abstract>
<identifier type="citekey">chen-etal-2026-efficient-agentic</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1959</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1959/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>41777</start>
<end>41795</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Efficient Agentic Reinforcement Learning with On-Policy Intrinsic Knowledge Boundary Enhancement
%A Chen, Dingwei
%A Zong, Zefang
%A ma, Zhipeng
%A Luo, Leo
%A Li, Yang
%A Li, Chengming
%A Chen, Peng
%A Jiang, Jie
%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 chen-etal-2026-efficient-agentic
%X Agentic reinforcement learning (RL) has proven effective for training LLM-based agents with external tool-use capabilities. However, we identify that agentic RL training induces increasing redundant tool calls and blurs the model’s intrinsic knowledge boundary, where the model fails to distinguish when tools are needed versus when parametric knowledge suffices. Existing solutions based on reward shaping create coarse-grained optimization targets that tend to incentivize indiscriminate tool-call suppression, leading to reward hacking. In this paper, we propose **AKBE** (**A**gentic **K**nowledge **B**oundary **E**nhancement), an on-policy method that dynamically probes the model’s intrinsic knowledge boundary through dual-path (with-tool and no-tool) rollouts during training. We define the knowledge boundary as the per-instance determination of whether tools are required and the minimum tool calls necessary. By comparing correctness across paths, AKBE categorizes trajectories and constructs targeted supervisory signals that guide efficient tool-use patterns for each question. These signals are integrated seamlessly into the agentic RL training loop. Experiments on seven QA benchmarks demonstrate that AKBE improves task accuracy by on average +1.85 and reduces tool calls by 18% over standard agentic RL, yielding 25% higher tool productivity without any accuracy-efficiency trade-off. Further analysis suggests its plug-and-play compatibility across different RL algorithms and the mechanism of each signal category.
%R 10.18653/2026.findings-emnlp.1959
%U https://aclanthology.org/2026.findings-emnlp.1959/
%U https://doi.org/10.18653/2026.findings-emnlp.1959
%P 41777-41795
Markdown (Informal)
[Efficient Agentic Reinforcement Learning with On-Policy Intrinsic Knowledge Boundary Enhancement](https://aclanthology.org/2026.findings-emnlp.1959/) (Chen et al., Findings 2026)
ACL
- Dingwei Chen, Zefang Zong, Zhipeng ma, Leo Luo, Yang Li, Chengming Li, Peng Chen, and Jie Jiang. 2026. Efficient Agentic Reinforcement Learning with On-Policy Intrinsic Knowledge Boundary Enhancement. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 41777–41795, Budapest, Hungary. Association for Computational Linguistics.