<?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>Kueue on Route179</title>
    <link>https://route179.dev/tags/kueue/</link>
    <description>Recent content in Kueue on Route179</description>
    <generator>Hugo</generator>
    <language>en-us</language>
    <copyright>2026 Sheng Chen</copyright>
    <lastBuildDate>Fri, 20 Nov 2026 09:00:00 +1100</lastBuildDate>
    <atom:link href="https://route179.dev/tags/kueue/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Physical AI on EKS: NVIDIA Isaac Lab training for the Unitree Go2</title>
      <link>https://route179.dev/2026/11/20/physical-ai-on-eks-isaac-lab-unitree-go2/</link>
      <pubDate>Fri, 20 Nov 2026 09:00:00 +1100</pubDate>
      <guid>https://route179.dev/2026/11/20/physical-ai-on-eks-isaac-lab-unitree-go2/</guid>
      <description>&lt;h2 id=&#34;introduction&#34;&gt;Introduction&lt;/h2&gt;
&lt;p&gt;Recently I&amp;rsquo;ve been researching Physical AI, and with a little (OK, a LOT of) help from Claude I
built an NVIDIA Isaac Lab setup that trains digital twins of the Unitree Go2 robot to walk, climb
and navigate different terrains. That&amp;rsquo;s roughly &lt;strong&gt;two months of practice, squeezed into under an
hour of training&lt;/strong&gt;, on Amazon EKS with 4× EC2 G6e (NVIDIA L40S) GPU nodes.&lt;/p&gt;
&lt;p&gt;The trick is simulation. Physical AI (robots that &lt;em&gt;learn&lt;/em&gt; a behaviour instead of having it
hand-coded) gets trained in a physics simulator first. A simulator runs thousands of robots in
parallel, nothing breaks when they fall over, and you can reset them a million times. Only once a
policy works in sim do you move it to real hardware.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
