Saved in:
Bibliographic Details
Main Authors: He, Shuo, Feng, Lang, Wei, Qi, Cheng, Xin, Feng, Lei, An, Bo
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.22817
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910033812389888
author He, Shuo
Feng, Lang
Wei, Qi
Cheng, Xin
Feng, Lei
An, Bo
author_facet He, Shuo
Feng, Lang
Wei, Qi
Cheng, Xin
Feng, Lei
An, Bo
contents Group-based reinforcement learning (RL), such as GRPO, has advanced the capabilities of large language models on long-horizon agentic tasks. To enable more fine-grained policy updates, recent research has increasingly shifted toward stepwise group-based policy optimization, which treats each step in a rollout trajectory independently while using a memory module to retain historical context. However, we find a key issue in estimating stepwise relative advantages, namely context inconsistency, where steps within the same group may differ in their historical contexts. Empirically, we reveal that this issue can lead to severely biased advantage estimation, thereby degrading policy optimization significantly. To address the issue, in this paper, we propose Hierarchy-of-Groups Policy Optimization (HGPO) for long-horizon agentic tasks. Specifically, within a group of rollout trajectories, HGPO assigns each step to multiple hierarchical groups according to the consistency of historical contexts. Then, for each step, HGPO computes distinct advantages within each group and aggregates them with an adaptive weighting scheme. In this way, HGPO can achieve a favorable bias-variance trade-off in stepwise advantage estimation, without extra models or rollouts. Evaluations on two challenging agentic tasks, ALFWorld and WebShop with Qwen2.5-1.5B-Instruct and Qwen2.5-7B-Instruct, show that HGPO significantly outperforms existing agentic RL methods under the same computational constraints. Code is available at https://github.com/langfengQ/verl-agent/tree/master/recipe/hgpo.
format Preprint
id arxiv_https___arxiv_org_abs_2602_22817
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Hierarchy-of-Groups Policy Optimization for Long-Horizon Agentic Tasks
He, Shuo
Feng, Lang
Wei, Qi
Cheng, Xin
Feng, Lei
An, Bo
Machine Learning
Artificial Intelligence
Group-based reinforcement learning (RL), such as GRPO, has advanced the capabilities of large language models on long-horizon agentic tasks. To enable more fine-grained policy updates, recent research has increasingly shifted toward stepwise group-based policy optimization, which treats each step in a rollout trajectory independently while using a memory module to retain historical context. However, we find a key issue in estimating stepwise relative advantages, namely context inconsistency, where steps within the same group may differ in their historical contexts. Empirically, we reveal that this issue can lead to severely biased advantage estimation, thereby degrading policy optimization significantly. To address the issue, in this paper, we propose Hierarchy-of-Groups Policy Optimization (HGPO) for long-horizon agentic tasks. Specifically, within a group of rollout trajectories, HGPO assigns each step to multiple hierarchical groups according to the consistency of historical contexts. Then, for each step, HGPO computes distinct advantages within each group and aggregates them with an adaptive weighting scheme. In this way, HGPO can achieve a favorable bias-variance trade-off in stepwise advantage estimation, without extra models or rollouts. Evaluations on two challenging agentic tasks, ALFWorld and WebShop with Qwen2.5-1.5B-Instruct and Qwen2.5-7B-Instruct, show that HGPO significantly outperforms existing agentic RL methods under the same computational constraints. Code is available at https://github.com/langfengQ/verl-agent/tree/master/recipe/hgpo.
title Hierarchy-of-Groups Policy Optimization for Long-Horizon Agentic Tasks
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2602.22817