Jonathan Tompson

Super excited to finally share our work on “Self-Improving Embodied Foundation Models”!! (Also accepted at NeurIPS 2025) • Online on-robot Self-Improvement • Self-predicted rewards and success detection • Orders of magnitude sample-efficiency gains compared to SFT alone • Generalization enables novel skill acquisition 🧵👇[1/11]

@coolboi95 @generalistai Amazing work guys! Buttery smooth policies :-)

I’m really excited to announce @generalistai ! Our mission is to make general-purpose robots a reality. Getting to this “ChatGPT moment” of robotics requires thinking deeply about the entire system from the ground-up. We’ve spent the past 1+ years focusing on dexterity, pushing the boundaries of data, models, and hardware. Our end-to-end models (pixels in, actions out) exhibit a step-function change in robot dexterity. They are fast, reactive, graceful, and precise. We’re now pumped to share a preview of our robots doing some very difficult tasks! The same training recipe for all tasks. So many little moments of magic in the videos we've shared ✨ From micro-adjustments and untangling behaviors, to inter-hand coordination and high-velocity maneuvers. I encourage you to watch them all on our website! There is still so much more to build, and I could not have asked to be part of a better team towards creating the future of Generalist robots! Reach out if you’re interested in working together!