<?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>🔄 测试时训练 on Elon&#39;s AD Insight</title>
    <link>https://auto-driving-blog.pages.dev/tags/-%E6%B5%8B%E8%AF%95%E6%97%B6%E8%AE%AD%E7%BB%83/</link>
    <description>Recent content in 🔄 测试时训练 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>Wed, 19 Aug 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://auto-driving-blog.pages.dev/tags/-%E6%B5%8B%E8%AF%95%E6%97%B6%E8%AE%AD%E7%BB%83/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>论文精读｜DriveVLA-M0：失败感知的记忆增强，让 VLA 学会吃一堑长一智</title>
      <link>https://auto-driving-blog.pages.dev/posts/paper-reading/drivevla-m0%E7%B2%BE%E8%AF%BB/</link>
      <pubDate>Wed, 19 Aug 2026 00:00:00 +0000</pubDate>
      <guid>https://auto-driving-blog.pages.dev/posts/paper-reading/drivevla-m0%E7%B2%BE%E8%AF%BB/</guid>
      <description>DriveVLA-M0 给 VLA 自动驾驶加上了&amp;rsquo;失败记忆&amp;rsquo;：离线把模型表现差的场景连同路网/交互结构特征写进 latent memory，在线用专用的 Retrieve Model 检索结构相似的失败案例，再用解耦 LoRA 做测试时训练（TTT）逐场景纠正，让模型&amp;rsquo;吃一堑长一智&amp;rsquo;。NAVSIMv1 上 94.1 PDMS、NAVSIMv2 上 47.0 EPDMS，TTT 后向开销仅 26.44ms。它与同系 DriveVLA-W0 走了完全相反的路：W0 用世界模型换&amp;rsquo;稠密监督&amp;rsquo;，M0 用失败记忆换&amp;rsquo;场景自适应&amp;rsquo;。</description>
    </item>
  </channel>
</rss>
