Humanoid robots can walk, gesture, and look astonishingly capable in a polished demo. Then Jan Liphardt bought one for his house, along with three quadruped robots, and discovered how quickly the illusion breaks. After charging the batteries and getting the humanoid to stand, it simply stood there. It could not become the R2-D2-like helper, companion, teacher, or coworker he had imagined. In this episode of Automated, Brian Heater speaks with Jan Liphardt, founder and CEO of OpenMind and associate professor of bioengineering at Stanford University, about the software, intelligence, and trust systems that robots still need before they can become useful parts of everyday life. Jan explains why the final 3% of humanoid deployment can consume nearly all the work. A robot that performs well in a lab still has to handle traffic, animals, wet pavement, moving leaves, different languages, and countless other edge cases before it can operate safely in an unstructured environment. The conversation explores why OpenMind is deliberately not trying to solve every robotics problem. Rather than building hands, arms, or high-speed control systems, the company combines its own models with technology from other teams. Its OM1 runtime and FABRIC coordination layer are designed to help different robots become more capable, understandable, and useful around people. Brian and Jan also discuss why social intelligence requires much more than speech. A compelling humanoid has to track attention, move its head and shoulders, use its hands, understand body language, remember preferences, and respond in ways that feel natural. Jan even argues that a robot’s imperfections can strengthen the bond between the machine and the person helping it. They also dig into one of OpenMind’s most distinctive ideas: connecting sensors and models through natural language so people can inspect how a robot reaches a decision. That approach trades some speed for intelligibility, giving developers a clearer way to debug behavior, add guardrails, and improve governance. Jan also challenges the idea that robotics has one universal data problem. Teaching math, folding a T-shirt, and completing an assembly task require very different data and system designs. In some cases, he argues, a robot may learn more effectively from physical constraints than from watching millions of videos of humans. The conversation also covers Jan’s path from physics and bioengineering into robotics, why elite degrees do not always predict great engineers, what a 17-year-old calling from a robot field trial taught him about hiring, and why combining academia with the real world can produce better research questions. Connect with Jan Liphardt https://www.linkedin.com/in/jan-liphardt Learn more about OpenMind https://openmind.com/ Learn more about Jan’s work at Stanford https://profiles.stanford.edu/jan-liphardt Register for the Automated Happy Hour in Mountain View on September 22 https://luma.com/n688dcl5 Learn more about the Advanced Vision and AI Conference on September 23 and 24 https://www.automate.org/events/advanced-vision-and-ai-conference/register We’d love to hear from you. Have thoughts or guest suggestions? Reach us at podcast@automate.org You can find the transcript and more episodes of Automated at automated.fm Unlock full access to Automated and explore everything automation. Subscribe today and leave a review on YouTube, Apple Podcasts, Spotify, and the Automated Newsletter. https://www.youtube.com/@automatedpodcast https://podcasts.apple.com/us/podcast/automated-with-brian-heater/id1837762221 https://open.spotify.com/show/60olq6brlBEIJWggx2fMR6 https://www.automate.org/automation/automated-newsletter LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3/ Instagram https://www.instagram.com/automatedpod/
