Show Notes
In this conversation, we sit down with Varun Nair to dig into the high-stakes world of data for robotics and physical AI. We talk about what actually matters for training robot intelligence. From egocentric video to UMI over teleop, simulation and the gaps between them. You will also learn why the field is still figuring out which data scales and the limits of human-operated data, evaluation challenges, and what it will take to move from demos to real deployment. If you’re building, investing in, or just obsessed with robotics and embodied AI, this one’s for you.
#robot #data #ai
0:00 Podcast Overview 0:30 Data Market 1:30 Varun's Background 3:01 Robotics & Crypto 6:27 Physical AI Definition 8:32 Simulation for Deployments 10:45 World Models 11:45 What is good data ? 12:42 GPT 6 Astra Evaluation 13:50 Data Life Cycle 17:01 Modalities & Pseudo Labeling 19:40 Depth for LLMs 20:19 How labs use data 21:38 Deployment Data = Post Training Data 22:17 Teleoperation 24:00 Patient Capital vs Impatient Capital 24:47 Benchmarks 26:35 UMI Data for Mid Training 27:35 Has Openai been training on robot data? 30:59 Humanoid vs other Embodiments 31:36 Hard vs Easy Tasks for Robots 32:51 Simulation 33:59 Niantic Spatial 34:42 Is robotics data broken? 36:17 Model Companies 37:38 Robot Everest 38:11 Avea 38:47 What Varun looks for 40:17 Outro
Guest https://x.com/_varunnair Léo https://x.com/LeoKharon Oskar https://x.com/hundreddaysx Podcast https://x.com/roboticsstack
More from The Robotics Stack
In this conversation, we sit down with Matt Trossen, founder of Trossen Robotics. We speak about the history of his company and what it takes to scale Physical AI with Teleoperation and AI Models. He shares his unique perspective on the robot market and presents the new product lineup which includes reworked Robotic Arms but also UMI Grippers and software like Trossen Cloud. His journey is one of a kind: Coming from USB controlled hardware for sensing and automation to working close with researchers on the ALOHA Project. The recent advancements in Artificial Intelligence opened a completely new world to robots for him and his company. Enjoy!
Asimov is an Open Source Humanoid Robot and today we have the CPO of the company behind it on the podcast. His name is Matthew Moore and he shares unique insights from Menlo Research. You learn what it really takes to build a robot from scratch, what you can do with Asimov and what the future holds for open robotics. #humanoid #robot #asimov
GPT 6 Astra is OpenAI's most powerful model. So can it be used for robotics? In this episode we look at various demos that show people implementing GPT 6 Astra to control robotic arms and more. Some work in simulation, some work in real environments. And the results are actually surprising. The LLM even starts showing up on various benchmarks and sometimes outperforms SOTA Physical AI models. Of course, there are caveats to this but the performance is seriously impressive. #chatgpt #astra #gpt #podcast
In this episode of the robotics stack podcast, we explore tactile data capture hardware with Matt Wulff from 6thSense. He is one of the co-founders and joins us to discuss the specialized robotics hardware required to give robots a realistic sense of touch. We examine how current systems struggle with physical dexterity and why new hardware approaches are essential for advancement. We break down the entire data capture pipeline from deployments to engineering the gloves & camera involved and how to turn this into useful robotics training data. We also speak about what the future holds for hands, humanoids & more!
In this episode of the robotics stack podcast we have our second interview. The guest is Randall Briggs who is working on Steel Bot, an open-source robot platform that is made in America. We are discussing the product, the history and alternatives like the Unitree G1 or Asimov. Randall also shares some interesting historical insights and explains why he has such strong conviction for the project.