arXiv cs.AI· Wei Xie, Kaituo Feng, Shiming Fu, Yong Liu, Xiangyu Yue, Shuangjia Zheng·· 3 天前AI 评分36
ReTerm：面向自我改进终端智能体的恢复训练
ReTerm: Recovery Training for Self-Improving Terminal Agents
AI 导读
基于 Qwen3.5-9B 训练的终端智能体 ReTerm-9B 在 Terminal-Bench 2.0 上达到 34.8% Pass@1，较基座模型提升 14.6 分。研究团队构建 ReTerm-10K 数据集用于 SFT 冷启动，并提出在线 RL 算法 ReTerm-GRPO，从智能体自身失败状态分叉生成多次独立自我恢复尝试，依据任务是否完成进行学习。代码、数据与模型均已开源。
正文 · 原文
Abstract:Terminal agents have shown strong abilities in complex coding tasks. However, they often struggle to recover from errors during long-running execution (e.g., command errors, faulty implementations), eventually leading to task failures. To this end, we present ReTerm, a self-improving terminal agent trained by learning how to recover from its own failed terminal tasks. To support its training, we construct a customized data pipeline and produce ReTerm-10K, a high-quality dataset that provides effective recovery demonstrations for SFT cold start. Furthermore, we propose ReTerm-GRPO, an online RL algorithm that forks from the agent's own failure states to generate multiple independent self-recovery attempts. By learning from whether these attempts successfully complete the task, the agent self-improves through its own recovery experience. Built on Qwen3.5-9B, our ReTerm-9B achieves 34.8% Pass@1 on Terminal-Bench 2.0, improving over the base model by 14.6 points. Further analysis shows that our method achieves consistently higher recovery rates across diverse error categories compared with standard GRPO training. All code, data and model are released at this https URL.
| Comments: | Project page: this https URL |
| Subjects: | Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Software Engineering (cs.SE) |
| Cite as: | arXiv:2610.18484 [cs.AI] |
| (or arXiv:2610.18484v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.18484 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Kaituo Feng [view email]
[v1]
Tue, 13 Oct 2026 14:44:24 UTC (2,091 KB)
来源：arXiv cs.AI · arxiv.org