Xiaomi open-sourced the full post-training process for its visual-language-action model Xiaomi-Robotics-0 to advance embodied AI. The model can master high-precision tasks like inserting earphones into a case with only 20 hours of real-robot training, showcasing sub-millimeter alignment and real-time correction. By releasing model weights, technical reports, and source code, Xiaomi aims to lower development barriers and accelerate the integration of AI robots into real-world applications.