Saved in:
Bibliographic Details
Main Authors: Zhang, Siwei, Xiong, Yun, Chen, Xi, Jia, Zi'an, Huang, Renhong, Xu, Jiarong, Zhang, Jiawei
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.03078
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918367612370944
author Zhang, Siwei
Xiong, Yun
Chen, Xi
Jia, Zi'an
Huang, Renhong
Xu, Jiarong
Zhang, Jiawei
author_facet Zhang, Siwei
Xiong, Yun
Chen, Xi
Jia, Zi'an
Huang, Renhong
Xu, Jiarong
Zhang, Jiawei
contents Agentic Reinforcement Learning (Agentic RL) has shown remarkable potential in large language model-based (LLM) agents. These works can empower LLM agents to tackle complex tasks via multi-step, tool-integrated reasoning. However, an inherent limitation of existing Agentic RL methods is their reliance on a pure on-policy paradigm for exploration, restricting exploration to the agent's self-generated outputs and preventing the discovery of new reasoning perspectives for further improvement. While recent efforts incorporate auxiliary off-policy signals to enhance exploration, they typically utilize full off-policy trajectories for trajectory-level policy estimation, overlooking the necessity for the fine-grained, step-level exploratory dynamics within agentic rollout. In this paper, we revisit exploration in Agentic RL and propose Retrieval-Augmented Policy Optimization (RAPO), a novel RL framework that introduces retrieval to explicitly expand exploration during training. To achieve this, we decompose the Agentic RL training process into two phases: (i) Hybrid-policy Agentic Rollout, and (ii) Retrieval-aware Policy Optimization. Specifically, we propose a Hybrid-policy Agentic Rollout strategy, which allows the agents to continuously reason over the retrieved off-policy step-level traces. It dynamically extends the reasoning receptive field of agents, enabling broader exploration conditioned on external behaviors. Subsequently, we introduce the Retrieval-aware Policy Optimization mechanism, which calibrates the policy gradient estimation with retrieval reward and importance shaping, stabilizing training and prioritizing retrieval-illuminating exploration. Extensive experiments show that RAPO achieves an +5.0% average gain on fourteen datasets across three agentic reasoning tasks, while delivering 1.2x faster training efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2603_03078
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle RAPO: Expanding Exploration for LLM Agents via Retrieval-Augmented Policy Optimization
Zhang, Siwei
Xiong, Yun
Chen, Xi
Jia, Zi'an
Huang, Renhong
Xu, Jiarong
Zhang, Jiawei
Artificial Intelligence
Agentic Reinforcement Learning (Agentic RL) has shown remarkable potential in large language model-based (LLM) agents. These works can empower LLM agents to tackle complex tasks via multi-step, tool-integrated reasoning. However, an inherent limitation of existing Agentic RL methods is their reliance on a pure on-policy paradigm for exploration, restricting exploration to the agent's self-generated outputs and preventing the discovery of new reasoning perspectives for further improvement. While recent efforts incorporate auxiliary off-policy signals to enhance exploration, they typically utilize full off-policy trajectories for trajectory-level policy estimation, overlooking the necessity for the fine-grained, step-level exploratory dynamics within agentic rollout. In this paper, we revisit exploration in Agentic RL and propose Retrieval-Augmented Policy Optimization (RAPO), a novel RL framework that introduces retrieval to explicitly expand exploration during training. To achieve this, we decompose the Agentic RL training process into two phases: (i) Hybrid-policy Agentic Rollout, and (ii) Retrieval-aware Policy Optimization. Specifically, we propose a Hybrid-policy Agentic Rollout strategy, which allows the agents to continuously reason over the retrieved off-policy step-level traces. It dynamically extends the reasoning receptive field of agents, enabling broader exploration conditioned on external behaviors. Subsequently, we introduce the Retrieval-aware Policy Optimization mechanism, which calibrates the policy gradient estimation with retrieval reward and importance shaping, stabilizing training and prioritizing retrieval-illuminating exploration. Extensive experiments show that RAPO achieves an +5.0% average gain on fourteen datasets across three agentic reasoning tasks, while delivering 1.2x faster training efficiency.
title RAPO: Expanding Exploration for LLM Agents via Retrieval-Augmented Policy Optimization
topic Artificial Intelligence
url https://arxiv.org/abs/2603.03078