<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Humanoid Training</title>
  <link href="https://humanoidtraining.ai/"/>
  <link rel="self" href="https://humanoidtraining.ai/feed.xml"/>
  <id>https://humanoidtraining.ai/</id>
  <updated>2026-10-04T00:00:00Z</updated>
  <author><name>Ram Agarwal</name></author>
  <entry><title>Humanoid Training — how humanoid robots learn</title><link href="https://humanoidtraining.ai/"/><id>https://humanoidtraining.ai/</id><updated>2026-10-04T00:00:00Z</updated><summary>How humanoid robots are trained in 2026: teleoperation, simulation, human video, vision-language-action models, and what it means on a factory floor. From the maker of FabSpeak.</summary></entry>
  <entry><title>How a humanoid is trained</title><link href="https://humanoidtraining.ai/how/"/><id>https://humanoidtraining.ai/how/</id><updated>2026-10-04T00:00:00Z</updated><summary>Why writing code for forty joints stopped working, and what replaced it: human video, simulation, teleoperation and continuous fleet learning, explained in order.</summary></entry>
  <entry><title>Where humanoid training data comes from</title><link href="https://humanoidtraining.ai/data/"/><id>https://humanoidtraining.ai/data/</id><updated>2026-10-04T00:00:00Z</updated><summary>Teleoperation, simulation and human video compared: what each is good for, what each gets wrong, and why 5 to 50 episodes per operator-hour is the number that governs the field.</summary></entry>
  <entry><title>Vision-language-action models</title><link href="https://humanoidtraining.ai/models/"/><id>https://humanoidtraining.ai/models/</id><updated>2026-10-04T00:00:00Z</updated><summary>Helix, GR00T N1, Gemini Robotics, pi-0.7, Large Behavior Models and Xiaomi-Robotics-1: who builds them, which are open, and how a VLA is put together.</summary></entry>
  <entry><title>Humanoids on the factory floor</title><link href="https://humanoidtraining.ai/floor/"/><id>https://humanoidtraining.ai/floor/</id><updated>2026-10-04T00:00:00Z</updated><summary>The verified deployments at BMW, Amazon and Hyundai, what Tesla actually said about Optimus, and why a semiconductor fab is a harder room than a car plant.</summary></entry>
  <entry><title>Where humanoid training goes next</title><link href="https://humanoidtraining.ai/future/"/><id>https://humanoidtraining.ai/future/</id><updated>2026-10-04T00:00:00Z</updated><summary>Robot-free demonstration, world models, fleet learning, skills as products, and the MES as the robot's supervisor. With an honest timeline and where I could be wrong.</summary></entry>
  <entry><title>Glossary</title><link href="https://humanoidtraining.ai/glossary/"/><id>https://humanoidtraining.ai/glossary/</id><updated>2026-10-04T00:00:00Z</updated><summary>Plain definitions of the robot-learning and semiconductor-fab terms used across this site: VLA, policy, teleoperation, sim-to-real, FOUP, AMHS, MES, Cpk and the rest.</summary></entry>
  <entry><title>References and sources</title><link href="https://humanoidtraining.ai/references/"/><id>https://humanoidtraining.ai/references/</id><updated>2026-10-04T00:00:00Z</updated><summary>The full source list behind every figure on Humanoid Training, with contested claims flagged and arXiv identifiers verified on 4 October 2026.</summary></entry>
  <entry><title>About and contact</title><link href="https://humanoidtraining.ai/about/"/><id>https://humanoidtraining.ai/about/</id><updated>2026-10-04T00:00:00Z</updated><summary>Why this site exists, who writes it, and how to send a correction. A FabSpeak project by Ram Agarwal, independent and unsponsored.</summary></entry>
</feed>
