---
id: 20260609-T0-09
title: "OpenSkill：让AI代理在真实世界自主学习进化"
title_en: "OpenSkill Enables Self-Evolving LLM Agents"
url: https://ai.daily.yangsir.net/daily/20260609-T0-09
issue_date: 2026-06-09
publish_date: 2026-06-08T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2606.06741
---

# OpenSkill：让AI代理在真实世界自主学习进化

arXiv论文OpenSkill提出开放世界自进化框架，解决AI代理部署后适应难题。现有方法依赖预设学习条件，但真实环境往往缺乏这些要素。新框架通过无监督学习让代理在开放环境中持续进化，无需人工标注数据或成功轨迹验证。研究指出，这使代理能应对未知场景，将推动自主AI系统在复杂现实世界中的应用。

## English Version

**OpenSkill Enables Self-Evolving LLM Agents**

OpenSkill framework on arXiv solves post-deployment adaptation challenges for AI agents. Existing approaches rely on curated learning conditions, but real-world environments often lack these. The new framework enables unsupervised evolution in open worlds without labeled data or success trajectory verification. This allows agents to handle unknown scenarios, advancing autonomous AI systems in complex real-world applications.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2606.06741)

**详情页**：https://ai.daily.yangsir.net/daily/20260609-T0-09

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*