Xiaomi-Robotics-U0-4B packs unified embodied world generation into a compact 4B model. 📜 Apache 2.0. 🤖 📄 🏆 The full U0 system ranks #1# on World Arena and lifts π0.5’s real-world OOD success from 36.9% to 63.2% with generated data. 🌍 Generate or edit consistent multi-view scenes across robot embodiments. It also supports text-to-image and instruction-guided generation. 🛠 Change objects, layouts, robot bodies, and camera views while preserving task structure and cross-view consistency. 🧠 Text, images, and embodied observations share one token space and next-token objective. Inference code and Gradio are included.