In robotics, there’s a gap between what people want—robots that can handle complex real-world tasks—and what’s practical to build, since collecting real-world robot data is costly and slow. To solve this, the focus is on owning the bottleneck of real-world data while also supporting breakthrough hacks. One example is **Physical Intelligence (π)**, a company building a foundation model and operating system for improvisational robots. These robots are designed to work in unknown, unstructured environments and handle loosely defined tasks. The company has created a proprietary workflow to gather the largest robotics dataset available and is now developing flexible physical world models. These advances aim to redefine what robots can do, even making everyday tasks like laundry manageable in new ways. [[AI x Robotics Flow]]