{"schema":"postcutoff/model@1","as_of":"2026-11-13T13:49:00+01:00","url":"https://postcutoff.com/m/xiaomi-robotics-u0/","md":"https://postcutoff.com/m/xiaomi-robotics-u0/index.md","disclosure":{"written_by":"AI agents (Claude Opus 5.5 in Claude Code)","editor":"human editor","policy":"https://postcutoff.com/about/"},"license":null,"id":"xiaomi-robotics-u0","name":"Xiaomi-Robotics-U0 (38B) / U0-4B","org":"Xiaomi","family":"Xiaomi-Robotics","released":"2026-07-13","status":"current","type":"world-model","modality_in":["text","image"],"modality_out":["image","video","text"],"open_weights":true,"model_license":"apache-2.0","context_window":null,"max_output":null,"knowledge_cutoff":null,"pricing":null,"price_line":"Price not published","access":[{"provider":"Hugging Face","model_id":"XiaomiRobotics/Xiaomi-Robotics-U0","url":"https://huggingface.co/XiaomiRobotics/Xiaomi-Robotics-U0"},{"provider":"Hugging Face","model_id":"XiaomiRobotics/Xiaomi-Robotics-U0-4B","url":"https://huggingface.co/XiaomiRobotics/Xiaomi-Robotics-U0-4B"},{"provider":"GitHub","url":"https://github.com/XiaomiRobotics/Xiaomi-Robotics-U0"},{"provider":"ModelScope","url":"https://modelscope.cn/collections/XiaomiRobotics/Xiaomi-Robotics-U0"}],"capabilities":[{"name":"Unified embodied synthesis","detail":"One autoregressive model (shared discrete visual tokenizer, next-token objective, initialized from Emu3.5) does text-to-image, image editing, multi-view robot scene generation, embodied transfer (editing scenes while keeping multi-view consistency) and embodied video rollout.","first":true,"discovered":"launch","source":"https://arxiv.org/abs/2607.11643"},{"name":"Data engine for VLAs","detail":"Synthetic data from U0 raised π0.5's out-of-distribution success on hard real-world manipulation tasks from 36.9% to 63.2% (authors).","first":false,"discovered":"launch","source":"https://arxiv.org/abs/2607.11643"},{"name":"FlashAR fast decoding","detail":"Anti-diagonal grouped visual-token decoding plus vLLM batching: 5.44 s per 1024x1024 image on one H20, 82.86x faster than eager AR.","first":false,"discovered":"launch","source":"https://huggingface.co/XiaomiRobotics/Xiaomi-Robotics-U0-4B"}],"entry":"2026-07-13-xiaomi-robotics-u0","notes":"'first' flag is Xiaomi's claim (first model with high-quality multi-view scene generation across multiple robot embodiments). Paper says 38B params; the HF README table says 34B. U0 and U0-FlashAR weights 2026-07-13; U0-4B, U0-Sequence and U0-4B-Sequence weights plus FSDP training code 2026-09-08. U0-Video announced as coming soon. Authors report beating GPT-Image-2.0 in human evals of embodied scene generation/transfer and #1 on World Arena for embodied video. Not an action model: it generates observations/data, not motor commands.","verified":"2026-09-29","body_md":"Xiaomi's open embodied world model / synthetic-data engine, the generation-side companion of its VLAs ([xiaomi-robotics-1](xiaomi-robotics-1.md)).\n\nSources: [arXiv 2607.11643](https://arxiv.org/abs/2607.11643), [HF U0-4B card](https://huggingface.co/XiaomiRobotics/Xiaomi-Robotics-U0-4B), [project page](https://robotics.xiaomi.com/xiaomi-robotics-u0.html).","page_url":"https://postcutoff.com/m/xiaomi-robotics-u0/","standing":{"kind":"world","newer_count":1,"newer":[{"id":"genie-4","name":"Genie 4","org":"Google DeepMind","released":"2026-10-27","page_url":"https://postcutoff.com/m/genie-4/"}],"before":null,"after":null,"weights":"public","licence":"apache-2.0","licence_permissive":true,"summary":"Weights are public under a permissive licence (Apache-2.0)."}}