Home
Xiaomi Open Sources Full Post-Training of VLA Large Model for Sub-Millimeter Robotic Operations
Xiaomi continues to make significant strides in robotics. After releasing and open-sourcing its VLA large model, Xiaomi-Robotics-0, in February, the company has now disclosed the complete post-training process for the model. This move is intended to bridge the "last mile" gap between laboratory robots and practical production, transforming AI robots into truly productive tools that are ready to deploy out of the box.

In a recent capability demonstration, the robot equipped with this model exhibited impressive precision manipulation skills. With only 20 hours of reinforcement training on task data, Xiaomi-Robotics-0 mastered the highly challenging action of sequentially organizing earphones. The video shows the robot smoothly and accurately placing small earphones one by one into compact charging slots.
This seemingly straightforward action involves significant technical complexity. According to official technical details, the task presents two core challenges: precision and stability. The tolerance between the earphone and the charging case slot is extremely tight, requiring sub-millimeter-level spatial perception for accurate alignment. Additionally, the extremely smooth surfaces of the earphone and case (surface roughness as low as Ra 0.03μm) make slipping likely during contact. The model must have high-speed real-time feedback and action correction capabilities to prevent assembly failure.

Notably, within its first month of release, Xiaomi-Robotics-0 ranked among the top six on the global VLA model download list on HuggingFace, reflecting strong industry interest. To further support the developer ecosystem, Xiaomi has fully open-sourced the project's technical report, model weights, and source code.
Developers can now access these resources through the Xiaomi Robotics website and open-source platforms like GitHub. By open-sourcing the complete post-training process, the industry expects a substantial reduction in the development barrier for high-performance robotic tasks, accelerating the deployment of intelligent robots in complex, precision-demanding scenarios.
Technical reference: https://robotics.xiaomi.com/xiaomi-robotics-0.html#pack-earbuds
Related article
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
Related Special Topic Recommendations
Comments (0)
0/500
Xiaomi continues to make significant strides in robotics. After releasing and open-sourcing its VLA large model, Xiaomi-Robotics-0, in February, the company has now disclosed the complete post-training process for the model. This move is intended to bridge the "last mile" gap between laboratory robots and practical production, transforming AI robots into truly productive tools that are ready to deploy out of the box.

In a recent capability demonstration, the robot equipped with this model exhibited impressive precision manipulation skills. With only 20 hours of reinforcement training on task data, Xiaomi-Robotics-0 mastered the highly challenging action of sequentially organizing earphones. The video shows the robot smoothly and accurately placing small earphones one by one into compact charging slots.
This seemingly straightforward action involves significant technical complexity. According to official technical details, the task presents two core challenges: precision and stability. The tolerance between the earphone and the charging case slot is extremely tight, requiring sub-millimeter-level spatial perception for accurate alignment. Additionally, the extremely smooth surfaces of the earphone and case (surface roughness as low as Ra 0.03μm) make slipping likely during contact. The model must have high-speed real-time feedback and action correction capabilities to prevent assembly failure.

Notably, within its first month of release, Xiaomi-Robotics-0 ranked among the top six on the global VLA model download list on HuggingFace, reflecting strong industry interest. To further support the developer ecosystem, Xiaomi has fully open-sourced the project's technical report, model weights, and source code.
Developers can now access these resources through the Xiaomi Robotics website and open-source platforms like GitHub. By open-sourcing the complete post-training process, the industry expects a substantial reduction in the development barrier for high-performance robotic tasks, accelerating the deployment of intelligent robots in complex, precision-demanding scenarios.
Technical reference: https://robotics.xiaomi.com/xiaomi-robotics-0.html#pack-earbuds
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber











