Xiaomi Open-Sources Robotics-U0 Embodied World Model and Training Stack

Xiaomi has announced the release of the open-source model Xiaomi-Robotics-U0, focusing on embodied intelligence. This model features approximately 38 billion parameters and includes additional versions like Xiaomi-Robotics-U0-4B and FlashAR. It is designed to unify text-to-image generation, enhancing robotic capabilities in multi-view scene processing. Xiaomi improves inference efficiency through technologies like FlashAR+, achieving up to 83 times faster generation rates. The model also boasts higher transfer win rates compared to other models like GPT-Image-2, reflecting its superiority in handling multi-image processing.