Xiaomi has open-sourced Xiaomi-Robotics-U0, a 38-billion-parameter multimodal autoregressive foundation model that unifies four distinct robot task categories within a single framework1,2. Code and weights are available on GitHub, HuggingFace, and ModelScope.
The release, described as the first unified generative model in embodied AI capable of handling four task types simultaneously, establishes a complete pipeline for generating and editing robot image and video training data.
Four capabilities in one framework
Xiaomi-Robotics-U0 covers embodied scene generation, embodied transfer, robot interaction video generation, and general-purpose image generation and editing. Embodied scene generation creates multi-view initial scenes for specified robot hardware from text descriptions, spanning environments from tabletops and kitchens to warehouses and open worlds. Embodied transfer migrates existing robot trajectories to new environments, changing lighting, background, surface materials, target objects, or workspace style while preserving original arm poses and scene layout. Robot interaction video generation produces subsequent video frames based on initial observations and operation instructions, maintaining motion coherence and physical consistency with zero-shot generalization to unseen scenarios. General text-to-image and image editing capabilities remain intact, allowing internet-scale visual knowledge to transfer into embodied AI tasks.
The model can also generate entirely new scenes covering hazardous, extreme, or long-tail environments inaccessible to physical robots.
Benchmark and real-robot results
Xiaomi-Robotics-U0 achieved top scores on the WorldArena benchmark across 126 participating global models. In real robot evaluations under out-of-distribution conditions, strategy task completion rates improved an average of 26% when trained on data augmented by the model.
A UNIS inference acceleration architecture improves generation efficiency by approximately 83 times compared to the raw autoregressive paradigm.
ANALYSIS The 26% completion-rate gain on out-of-distribution tasks points to a practical use case: synthetic data augmentation that reduces the cost of physical data collection for robotics labs. By open-sourcing weights and code, Xiaomi positions the model as shared infrastructure rather than a proprietary advantage, a move that could accelerate external adoption and community iteration on embodied-AI training pipelines.
The release follows Xiaomi's robot factory deployment. The open-source package includes the full project page, GitHub code repository, HuggingFace model weights, and ModelScope collection.