Xiaomi Open Sources Full Post-Training Process for High-Precision VLA Model
Xiaomi has unveiled the complete post-training process for its visual-language-action (VLA) large model, Xiaomi-Robotics-0, making it open-source for real-world applications. This initiative represents a significant advancement in Xiaomi's embodied intelligence efforts, designed to help robots learn complex operational skills efficiently with minimal data.
Mastering "Needle-in-a-Straw" in 20 Hours
Building on a pre-trained foundation, the research team utilized just 20 hours of task-specific data for real-robot post-training. This enabled the robot to execute the challenging maneuver of precisely inserting earphones into a case. The task demands exceptional spatial perception accuracy and must compensate for displacement caused by surfaces with extremely low friction.
The model achieves alignment within sub-millimeter tolerances and can correct action deviations in real time. This smooth, continuous execution highlights the strong potential of Xiaomi-Robotics-0 for high-precision assembly operations.

Open-Source Ecosystem Accelerates Productivity
To transform this model into a practical, ready-to-use tool, Xiaomi has open-sourced not only the model weights but also the technical report and full source code. This comprehensive, end-to-end release significantly lowers the entry barrier for developers exploring embodied intelligence.
The model has already demonstrated strong performance on leading international benchmarks, ranking among the most downloaded globally. By opening the post-training process, Xiaomi enables developers worldwide to collaboratively refine robotic perception and execution logic, speeding up the integration of AI robots into real-world production and daily life.
Project Website: https://robotics.xiaomi.com/xiaomi-robotics-0.html
Open Source Code: https://github.com/XiaomiRobotics/Xiaomi-Robotics-0
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 has unveiled the complete post-training process for its visual-language-action (VLA) large model, Xiaomi-Robotics-0, making it open-source for real-world applications. This initiative represents a significant advancement in Xiaomi's embodied intelligence efforts, designed to help robots learn complex operational skills efficiently with minimal data.
Mastering "Needle-in-a-Straw" in 20 Hours
Building on a pre-trained foundation, the research team utilized just 20 hours of task-specific data for real-robot post-training. This enabled the robot to execute the challenging maneuver of precisely inserting earphones into a case. The task demands exceptional spatial perception accuracy and must compensate for displacement caused by surfaces with extremely low friction.
The model achieves alignment within sub-millimeter tolerances and can correct action deviations in real time. This smooth, continuous execution highlights the strong potential of Xiaomi-Robotics-0 for high-precision assembly operations.

Open-Source Ecosystem Accelerates Productivity
To transform this model into a practical, ready-to-use tool, Xiaomi has open-sourced not only the model weights but also the technical report and full source code. This comprehensive, end-to-end release significantly lowers the entry barrier for developers exploring embodied intelligence.
The model has already demonstrated strong performance on leading international benchmarks, ranking among the most downloaded globally. By opening the post-training process, Xiaomi enables developers worldwide to collaboratively refine robotic perception and execution logic, speeding up the integration of AI robots into real-world production and daily life.
Project Website: https://robotics.xiaomi.com/xiaomi-robotics-0.html
Open Source Code: https://github.com/XiaomiRobotics/Xiaomi-Robotics-0
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





Home






