Physical AI Isn't About Bigger Models [Part 1] | Ted Lutkus Onyx Robotics

Robotics and physical AI get challenging in the real world because we start dealing with unmodelled forces. This is a non-trivial problem across all dynamic systems: cars, drones, collaborative arms, humanoids and much more. In this conversation with Ted Lutkus, the Founder and CEO of Onyx Robotics, we talk about the problems in controls and why it’s so hard to get your modeling right. Onyx is developing the controls and simulation layer and the tooling to empower controls engineers. We talk about what companies struggle with when building their own controls stack and system identification to understand where you need AI and learning.Whatever your background, I’m sure you’ll have a lot to take away and if you’re a developer in this space, you’ll probably nod along. Timestamps:00:00 - Intro01:50 - PhD to Building A Product06:10 - Challenges of Hardware8:15 - The Pivot To Controls, Simulation & Tooling12:04 - Onyx Robotics Now15:30 - Where does AI come in?Check out Onyx Robotics: https://www.onyx-robotics.com/Follow Ted Lutkus:   / ted-lutkus-787904137  ______ References:Marco Hutter & ANYMal, ETH Zurich: https://arxiv.org/abs/2505.16084 https://raisim.com/sections/Articulat...   • Planning and Control of Skating Motions fo...  System Identification: https://www.mathworks.com/help/ident/...