Can RL Improve Generalization of LLM Agents? An Empirical Study

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xi, Zhiheng, Guo, Xin, Liu, Jiaqi, Zhang, Jiazheng, Fan, Yutao, Zhang, Zhihao, Liu, Shichun, Chai, Mingxu, Shi, Xiaowei, Zhai, Yitao, Cai, Xunliang, Gui, Tao, Zhang, Qi, Huang, Xuanjing
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911509060255744
author Xi, Zhiheng
Guo, Xin
Liu, Jiaqi
Zhang, Jiazheng
Fan, Yutao
Zhang, Zhihao
Liu, Shichun
Chai, Mingxu
Shi, Xiaowei
Zhai, Yitao
Cai, Xunliang
Gui, Tao
Zhang, Qi
Huang, Xuanjing
author_facet Xi, Zhiheng
Guo, Xin
Liu, Jiaqi
Zhang, Jiazheng
Fan, Yutao
Zhang, Zhihao
Liu, Shichun
Chai, Mingxu
Shi, Xiaowei
Zhai, Yitao
Cai, Xunliang
Gui, Tao
Zhang, Qi
Huang, Xuanjing
contents Reinforcement fine-tuning (RFT) has shown promise for training LLM agents to perform multi-turn decision-making based on environment feedback. However, most existing evaluations remain largely in-domain: training and testing are conducted in the same environment or even on the same tasks. In real-world deployment, agents may operate in unseen environments with different background knowledge, observation spaces, and action interfaces. To characterize the generalization profile of RFT under such shifts, we conduct a systematic study along three axes: (1) within-environment generalization across task difficulty, (2) cross-environment transfer to unseen environments, and (3) sequential multi-environment training to quantify transfer and forgetting. Our results show that RFT generalizes well across task difficulty within an environment, but exhibits weaker transfer to unseen environments, which correlates with shifts in both semantic priors and observation/action interfaces. In contrast, sequential training yields promising downstream gains with minimal upstream forgetting, and mixture training across environments improves the overall balance. We further provide detailed analyses and deeper insights, and hope our work helps the community develop and deploy generalizable LLM agents.
format Preprint
id arxiv_https___arxiv_org_abs_2603_12011
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Can RL Improve Generalization of LLM Agents? An Empirical Study
Xi, Zhiheng
Guo, Xin
Liu, Jiaqi
Zhang, Jiazheng
Fan, Yutao
Zhang, Zhihao
Liu, Shichun
Chai, Mingxu
Shi, Xiaowei
Zhai, Yitao
Cai, Xunliang
Gui, Tao
Zhang, Qi
Huang, Xuanjing
Artificial Intelligence
Reinforcement fine-tuning (RFT) has shown promise for training LLM agents to perform multi-turn decision-making based on environment feedback. However, most existing evaluations remain largely in-domain: training and testing are conducted in the same environment or even on the same tasks. In real-world deployment, agents may operate in unseen environments with different background knowledge, observation spaces, and action interfaces. To characterize the generalization profile of RFT under such shifts, we conduct a systematic study along three axes: (1) within-environment generalization across task difficulty, (2) cross-environment transfer to unseen environments, and (3) sequential multi-environment training to quantify transfer and forgetting. Our results show that RFT generalizes well across task difficulty within an environment, but exhibits weaker transfer to unseen environments, which correlates with shifts in both semantic priors and observation/action interfaces. In contrast, sequential training yields promising downstream gains with minimal upstream forgetting, and mixture training across environments improves the overall balance. We further provide detailed analyses and deeper insights, and hope our work helps the community develop and deploy generalizable LLM agents.
title Can RL Improve Generalization of LLM Agents? An Empirical Study
topic Artificial Intelligence
url https://arxiv.org/abs/2603.12011