Is this 10,000 Hour World Model the Next AI Breakthrough? | MIRA by General Intuition
The Robotics StackShow Notes
The Robotics Stack Podcast Episode 8 starts with Flexion achieving remarkable results for long range horizon tasks with their new VLM (Vision Language Model) Architecture. Then we follow up with General Intuition and their new AI World Model called MIRA. It was trained on gaming data, specifically Rocket League. We also cover the "Github for Robotics" Tinkr AI & the recent news about Boston Dynamics an also their presence at the FIFA World Cup as the official robotics partner.
#ai #model #gaming #podcast
0:00 Podcast Overview 0:32 Flexion 15:21 MIRA World Model 26:06 Tnkr - "Github for Robots" 35:49 Boston Dynamics 48:05 Outro
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!
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.
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!