More than a year ago, we released DYNA1, “the first robot foundation model built for round-the-clock, high-throughput dexterous autonomy.” Looking back at the DYNA1 videos now, it’s funny how we thought of it as “high-throughput dexterous” back then. With DYNA1, we showed for the first time that a robot foundation model could be robust enough for true 24/7 automation. But it still wasn’t enough to fully meet customer needs — customers care about throughput, and they care about quality of outcome too (they even want us to remove napkins with stains on them...) It’s hard, and some people may see napkin folding as a “small” use case. But we thought it was important to go through the full deployment cycle to figure out what the real research problems in robotics are. Through the DTF deployment, we explored large-scale pretraining, RL, data-collection hardware, and a lot more. So here it is: a real “round-the-clock, high-throughput dexterous autonomy” system that creates value for customers. If you care about solving the real problems in robotics, join us!