Xiaomi open-sources its first VLA model for robots

Xiaomi open-sources its first VLA model and shakes up robotics

Xiaomi has made a decisive move in autonomous systems by open-sourcing Xiaomi-Robotics-0, its first Vision-Language-Action model for robots. The announcement, made on February 12, means developers worldwide will get access to a model with 4.7 billion parameters that combines visual perception, natural language understanding and real-time robot control. This is not just another open project: it signals that the Chinese tech giant is betting seriously on multimodal AI for physical automation and is ready to share its work with the global developer community.

Until now, VLA models have been the preserve of a select few. The most powerful systems in the field came from companies such as Tesla, with its object-manipulation hardware, and Google DeepMind, with its robotics experiments. These models learn to link what robots see with natural-language commands and concrete real-time actions. The problem is that training such systems takes huge volumes of video data, substantial computing resources and a deep understanding of robot mechanics. Xiaomi's release changes that dynamic. By publishing the source code and weights of its model, the company democratizes access to a technology that used to be the domain of large corporations with enormous budgets.

Xiaomi-Robotics-0 is designed to be efficient and practical. A model with 4.7 billion parameters is a middle ground between speed and performance. It is compact enough to run on robot platforms with limited computing power, yet capable enough to handle complex manipulation tasks. The architecture combines three key components: a visual encoder that analyzes images from the robot's cameras; a language component that processes the user's natural-language instructions; and an action module that generates control commands in real time. This integration is critical because a robot must do more than understand a task (for example, "pick up the red cube"); it must immediately translate it into movement coordinates and grip force.

Xiaomi's open-source strategy has a clear logic. In the robotics ecosystem, competition plays out less at the model level than in hardware, software and the application ecosystem. By opening its VLA model, the company lets thousands of developers experiment with new use cases and adapt the technology to their own robots. This creates a network-value effect: the more people improve the model and find new uses for it, the more likely it is that solutions emerging from that community will work with Xiaomi's in-house products.

For the industry, this means faster development. Startups, universities and corporations will be able to experiment with multimodal robot control without starting from scratch. The model is already trained on real-world data and has a sensible architecture that can be adapted to different platforms. That matters especially now, as robotics stands on the threshold of moving from labs into production and everyday life.

Xiaomi's move reflects a broader trend: AI leaders increasingly understand that opening part of their work builds a stronger ecosystem than tight control. This does not mean the company is abandoning its own robots; on the contrary, they will remain closed and proprietary. But the VLA model becomes a foundation on which an entire industry can grow, and Xiaomi is already laying bricks at its base.