智能体
arxiv.org ·
学习终端智能体的可泛化行为
该 arXiv 论文研究强化学习如何提升终端智能体，提出“智能体组合泛化”假设：RL 主要塑造高层决策行为，用于组合和路由预训练与 SFT 阶段已获得的低层技能，而非从头学习领域技能，因此决定哪些行为被强化的验证器质量比单纯增加环境数量与多样性更重要。基于此提出 River 训练配方，通过过滤低质量环境并以过程级行为正则增强结果奖励。作者称其 RL 训练的 8B 模型在四个终端智能体基准上优于所评估的开源 RL 8B 模型，且仅用不到 30% 的 TMax 训练环境，在 2B 至 27B 模型上于 Terminal-Bench-Lite 与 Terminal-Bench-v2.1 分别平均提升 RL 收益 106% 与 30%。
AI推荐理由 arXiv 论文（v3，2026-09-24 修订），作者与摘要明确，属可验证的一手研究。核心新信息是提出 Agentic Compositional Generalization 假设：RL 主要塑造高层决策行为、组合并路由预训练与 SFT 阶段获得的低层技能，而非从零学习领域技能；据此提出 River 训练配方，通过过滤低质量环境并用过程级行为正则增强结果奖励来提升奖励质量。摘要称在四个 terminal-agent 基准上，其 RL 训练的 8B 模型在受评估的开源 RL 8B 模型中表现最佳，且仅用不到 30% 的 TMax 训练环境，在 2B–27B 模型上于 Terminal-Bench-Lite 和 Terminal-Bench-v2.1 分别平均提升 RL 收益 106% 和 30%。局限：结论基于作者自述实验，未提供独立复现，基准与训练环境细节需查原文；对终端 Agent 训练与 RL 奖励设计有参考价值。
78
正文
Computer Science > Machine Learning
arXiv:2608.22631 (cs)
[Submitted on 23 Aug 2026 (v1), last revised 24 Sep 2026 (this version, v3)]
Title:Learning Generalizable Behaviors for Terminal Agents
Authors:Yihang Yao, Bo Pang, Xuan Phi Nguyen, Ding Zhao, Shafiq Joty, Semih Yavuz
View a PDF of the paper titled Learning Generalizable Behaviors for Terminal Agents, by Yihang Yao and 5 other authors
View PDF
HTML (experimental)
Abstract:Terminal agents are a compelling application of large language models (LLMs), with the potential to integrate deeply into users' daily workflows. Reinforcement learning (RL) is a key technique for improving their capabilities, making scalable training environments a central challenge. Since public real-user interaction data are scarce, synthetic environments provide a practical alternative, but often suffer from domain gaps and limited fidelity, leading to poor generalization. Existing work mainly scales the quantity and diversity of synthetic environments, while reward-signal quality and the mechanisms governing generalization remain under-explored. We study how RL improves terminal agents and propose the Agentic Compositional Generalization hypothesis: rather than teaching new domain-specific skills from scratch, RL primarily shapes high-level decision-making behaviors that compose and route low-level skills acquired during pre-training and supervised fine-tuning (SFT). This account is consistent with our empirical results and suggests that verifier quality, which determines which behaviors are reinforced, is more important than simply increasing environment quantity or diversity. Motivated by this insight, we propose River, a simple training recipe that improves reward quality by filtering low-quality environments and augmenting outcome rewards with process-level behavior regularization. Using this recipe, our RL-trained agent achieves the best performance among evaluated open-source RL-trained 8B models across four terminal-agent benchmarks. River also generalizes across model families, scales, agent harnesses, and RL objectives. Using fewer than 30% of the TMax training environments, River improves RL gains by 106% and 30% on average for models ranging from 2B to 27B on Terminal-Bench-Lite and Terminal-Bench-v2.1, respectively.
Subjects:
Machine Learning (cs.LG)
Cite as:
arXiv:2608.22631 [cs.LG]
(or
arXiv:2608.22631v3 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.22631
Focus to learn more
arXiv-issued DOI via DataCite
Submission history
From: Yihang Yao [view email]
[v1]
Sun, 23 Aug 2026 22:34:40 UTC (8,737 KB)
[v2]
Wed, 26 Aug 2026 21:18:13 UTC (8,736 KB)
[v3]
Thu, 24 Sep 2026 17:50:36 UTC (8,736 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled Learning Generalizable Behaviors for Terminal Agents, by Yihang Yao and 5 other authors
View PDF
HTML (experimental)
TeX Source
view license
Current browse context:
cs.LG
new
|
recent
| 2026-08
Change to browse by:
cs
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
export BibTeX citation
Loading...
BibTeX formatted citation
×
loading...
Data provided by:
Bookmark
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Code, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos
Demos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers
Recommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
IArxiv recommender toggle
IArxiv Recommender
(What is IArxiv?)
Author
Venue
Institution
Topic
About arXivLabs
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Which authors of this paper are endorsers? |
Disable MathJax (What is MathJax?)