← Retour au blog
tech 20 July 2026

Xiaomi-Robotics-1: A Revolution in Predictive Robotics Modeling

Xiaomi-Robotics-1 marks a shift in robotic policy modeling, integrating massive embodiment-free pre-training and real-world data to enhance robot capabilities.

Article inspired by the original source
Xiaomi-Robotics-1 ↗ robotics.xiaomi.com

Introduction

In the ever-evolving world of robotics, Xiaomi-Robotics-1 stands out as a pioneer. This ready-to-use robotic foundation model is trained on over 100,000 hours of real-world manipulation trajectories. In an era where language and vision models follow empirical scaling laws, robotics has lagged behind. But why? The answer lies mainly in the scarcity of large-scale, high-quality data, a barrier that Xiaomi-Robotics-1 seeks to overcome.

The Xiaomi-Robotics-1 Model

Xiaomi-Robotics-1 integrates large-scale embodiment-free (UMI) pre-training with a modest amount of real robot data in a post-training stage. This model studies how it behaves as it scales.

Large-Scale Pre-training

Pre-training focuses on exposing the model to as much of the real world as possible. Using 100,000 hours of UMI data, covering over 1,700 scenarios ranging from household tasks to industrial environments, Xiaomi-Robotics-1 builds a general representation for action generation.

Post-training with Real Data

After pre-training, the model is refined with real robot data, including over 7,200 hours of data collected in real homes. This data is crucial for aligning the model with real embodiment capabilities and instruction-following.

The Method

Following the training paradigm of LLMs (large language models), Xiaomi-Robotics-1 uses a two-step method. The first step learns general representations for action generation from large-scale UMI data, while the post-training stage aligns the model with real robot embodiments and instruction-following capabilities.

Practical Applications

The applications of Xiaomi-Robotics-1 are vast. From household tasks like tidying a sofa, sorting a shoe cabinet, to complex industrial operations, this model paves the way for more sophisticated and versatile automation.

Concrete Examples

Consider an example: in a domestic environment, a robot trained with Xiaomi-Robotics-1 can navigate and perform complex tasks like preparing a dinner or cleaning up after a meal, while adapting its behavior to user preferences.

Conclusion

Xiaomi-Robotics-1 represents a significant advancement in predictive robotics. By combining large-scale pre-training and real data, this model offers tremendous potential to transform how we design and use robots.

Let's discuss your project in 15 minutes.

robotics automation Xiaomi policy models machine learning
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call