<?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>表示学习 on Elon&#39;s AD Insight</title>
    <link>https://auto-driving-blog.vercel.app/tags/%E8%A1%A8%E7%A4%BA%E5%AD%A6%E4%B9%A0/</link>
    <description>Recent content in 表示学习 on Elon&#39;s AD Insight</description>
    <image>
      <title>Elon&#39;s AD Insight</title>
      <url>https://auto-driving-blog.vercel.app/images/share.png</url>
      <link>https://auto-driving-blog.vercel.app/images/share.png</link>
    </image>
    <generator>Hugo</generator>
    <language>zh-cn</language>
    <lastBuildDate>Wed, 22 Jul 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://auto-driving-blog.vercel.app/tags/%E8%A1%A8%E7%A4%BA%E5%AD%A6%E4%B9%A0/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>论文精读｜TransDiffuser：去相关多模态表示驱动的扩散式轨迹生成</title>
      <link>https://auto-driving-blog.vercel.app/posts/paper-reading/%E8%AE%BA%E6%96%87%E7%B2%BE%E8%AF%BB-2505-09315-transdiffuser/</link>
      <pubDate>Wed, 22 Jul 2026 00:00:00 +0000</pubDate>
      <guid>https://auto-driving-blog.vercel.app/posts/paper-reading/%E8%AE%BA%E6%96%87%E7%B2%BE%E8%AF%BB-2505-09315-transdiffuser/</guid>
      <description>LiAuto 提出 TransDiffuser，用多模态表示去相关（Decorrelation）解决扩散轨迹生成中的模式坍缩问题。核心思路是强制多模态条件特征的维度去相关，让条件信号更丰富地覆盖潜空间，从而无需任何锚点轨迹或场景先验即可生成多样化轨迹。ResNet-34 + LiDAR 配置在 NAVSIM 上达到 PDMS 94.85。本文将用一张图一张图对应讲解的方式，把这篇论文的动机、架构、核心创新全部讲清楚。</description>
    </item>
  </channel>
</rss>
