Xiaomi-Robotics-1-5B
This repository contains the post-trained Xiaomi-Robotics-1 (XR-1) checkpoint for further post-training with the Xiaomi-Robotics-1 codebase. XR-1 is a Vision-Language-Action (VLA) foundation model for mobile manipulation, initialized from a model pre-trained on over 100K hours of embodiment-free UMI trajectories and subsequently post-trained on over 10K hours of cross-embodiment data.
Requirements
The reference environment requires Python >= 3.9, a CUDA GPU, transformers==4.57.1, and FlashAttention 2.
The Qwen3-VL backbone is loaded from Qwen/Qwen3-VL-4B-Instruct.
Post-training
Clone the source repository and install the dependencies:
git clone https://github.com/XiaomiRobotics/Xiaomi-Robotics-1.git
cd Xiaomi-Robotics-1/xr1
pip install -e .
pip install flash-attn --no-build-isolation
Download model_states.pt from the Files and versions tab and place it at:
pretrained_ckpt/model_states.pt
Download the XR-1 post-training demo dataset, then launch post-training:
wandb login
RESOURCE_GPU=1 bash scripts/train.sh \
trainer.project="xiaomi-robotics-1" \
trainer.exp_name="posttrain" \
trainer.default_root_dir="outputs" \
data=load_washer \
model=posttrain \
model.params.pretrained="pretrained_ckpt/model_states.pt"
The training configuration expects the demo episodes under data/. See xr1/README.md for the data format, normalization statistics, and multi-GPU training options.
Deployment
After post-training, serve the resulting checkpoint with:
bash scripts/deploy.sh outputs/project_xiaomi-robotics-1/posttrain 1 1
The runtime client accepts three camera images, a natural-language instruction, and the current robot state, and returns the predicted action sequence. See the deployment documentation before connecting a post-trained checkpoint to a robot.
Citation
@article{team2026xiaomi,
title={Xiaomi-Robotics-1: Scaling Vision-Language-Action Models with over 100K Hours of Real-World Trajectories},
author={Team, Xiaomi Robotics and Guo, Jun and Jin, Piaopiao and Li, Jason and Li, Peiyan and Li, Yingyan and Liu, Futeng and Peng, Wanli and Qin, Optimus and Su, Yifei and others},
journal={arXiv preprint arXiv:2607.15330},
year={2026}
}