Paper 2607.15330
Xiaomi-Robotics-1: Scaling Vision-Language-Action Models with over 100K Hours of Real-World Trajectories
- Published
- Jul 2026
- Research lab
- Independent
- Citations
- 5
- GitHub
- 454 stars
01 In brief
Summary
Xiaomi-Robotics-1 is a foundational vision-language-action (VLA) model trained on over 100,000 hours of real-world manipulation trajectories collected via UMI devices.
It uses a two-stage training recipe: pre-training on UMI data with auto-labeled state-transition language prompts (generated by a VLM) to learn generalizable action generation, and post-training on about 10,000 hours of cross-embodiment data (including 7,200+ hours of in-house robot data) to align with robot embodiments and imperative instructions.
The model uses a Mixture-of-Transformers architecture combining a VLM (Qwen3-VL) and a diffusion transformer with flow matching, plus an auxiliary action-generation supervision via Choice Policies.
Experiments show scaling behavior: validation action error decreases with more data and larger model sizes (2B, 5B, 10B variants).
Post-training success rates in unseen environments increase monotonically with pre-training data scale (26% without pre-training to 75% with 100% of 20k hours) and model size (61% for 2B to 79% for 10B).
Fine-tuning on novel tasks with less than 10 hours per task achieves 75% average success, outperforming π0.5 (40%).
On simulation benchmarks, it achieves state-of-the-art results: 74.5% on RoboCasa, 57.4% on RoboCasa365 (previous best 46.6%), and 20.07 average score on RoboDojo (previous best 13.07).
02 From the paper
Abstract
We present Xiaomi-Robotics-1, a foundational vision-language-action (VLA) model capable of (1) following diverse language instructions to perform a wide range of mobile manipulation tasks in unseen environments out-of-the-box, and (2) efficiently adapting to novel downstream tasks with minimal fine-tuning data. We propose a two-stage training recipe consisting of pre-training and post-training. During pre-training, we imbue the model with broad and generalizable action-generation capabilities by training on over 100k hours of real-world manipulation trajectories collected via UMI devices. Crucially, we develop a scalable auto-labeling pipeline that annotates trajectory clips with natural languages describing scene state transitions, providing rich and precise conditioning for action learning. During post-training, we aim to align these capabilities with robot embodiments and imperative instructions that humans naturally use to prompt robots. Extensive experiments demonstrate strong scaling behavior. Xiaomi-Robotics-1 consistently improves with increased data scales and model sizes during pre-training. This scaling behavior directly transfers to post-training, where a stronger pre-training model yields better out-of-the-box real-robot performance in unseen environments. Furthermore, Xiaomi-Robotics-1 serves as a strong robot foundation policy that can be efficiently fine-tuned on complex, dexterous tasks with high data efficiency. Across multiple simulation benchmarks, Xiaomi-Robotics-1 outperforms state-of-the-art methods. Notably, it establishes a new state-of-the-art with a 57.4% success rate on RoboCasa365, surpassing the previous best of 46.6%. Furthermore, it achieves an average score of 20.07 on RoboDojo, significantly outperforming the prior state-of-the-art (13.07). Code and model checkpoints will be released. Project page: https://robotics.xiaomi.com/xiaomi-robotics-1.html