Xiaomi-Robotics-1 is a robot foundation model trained on over 100,000 hours of real-world manipulation trajectories, allowing it to perform various tasks such as tidying, sorting, and packing. The model uses a two-stage training process, with pre-training on large-scale embodiment-free data and post-training on real-robot data. This approach enables the model to scale neatly with data volume and model size, resulting in improved performance on real-robot tasks. The model can be used for downstream applications such as efficient adaptation to new tasks and simulation benchmarks.