<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>🤖 编程Agent on Elon&#39;s AD Insight</title>
    <link>https://auto-driving-blog.pages.dev/tags/-%E7%BC%96%E7%A8%8Bagent/</link>
    <description>Recent content in 🤖 编程Agent on Elon&#39;s AD Insight</description>
    <image>
      <title>Elon&#39;s AD Insight</title>
      <url>https://auto-driving-blog.pages.dev/images/share.png</url>
      <link>https://auto-driving-blog.pages.dev/images/share.png</link>
    </image>
    <generator>Hugo</generator>
    <language>zh-cn</language>
    <lastBuildDate>Thu, 22 Oct 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://auto-driving-blog.pages.dev/tags/-%E7%BC%96%E7%A8%8Bagent/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>精读 EmbodiedEvo：让编程 Agent 在一个 Playground 场景里「长经验」，冻住交给新场景——RoboDojo-Sim 37.82% 超全部官方已验证结果</title>
      <link>https://auto-driving-blog.pages.dev/posts/paper-reading/embodiedevo%E7%B2%BE%E8%AF%BB/</link>
      <pubDate>Thu, 22 Oct 2026 00:00:00 +0000</pubDate>
      <guid>https://auto-driving-blog.pages.dev/posts/paper-reading/embodiedevo%E7%B2%BE%E8%AF%BB/</guid>
      <description>EmbodiedEvo（Navers Lab / Einsia.AI / 清华，arXiv 2610.18867）精读：不是训练 VLA，而是让 coding agent（Astra/Sol/Luna）在单个 Playground 场景里边跑代码边沉淀可复用的 Skills（控制过程）与 Lessons（失败模式/恢复经验），把 harness 冻结后交给全新 Test agent，在未见过的布局上直接用。RoboDojo-Sim：平均 44.48 分 / 37.82% 成功率，超越当时全部官方已验证结果；Memory 类任务 71.90/69.67%（下一名 VPP2 仅 55.56% SR）。Astra 的经验还能迁给更弱的 Sol/Luna（+8.86 / +5.50 pp）。Skills/Lessons 相比原始交互历史，同等性能下 Test 阶段执行成本低 53%。逐张讲解全部 15 张插图，覆盖总览、方法框架、失败轨迹、能力演化、经验迁移与代码示例。</description>
    </item>
  </channel>
</rss>
