Xiaomi-Robotics-1: More Data Matters More Than Model Size in Robot Training

Xiaomi trained its Xiaomi-Robotics-1 model on more than 100,000 hours of motion data collected by people using handheld grippers with cameras, not by the robots themselves. In July 2026, the company showed that quality gains come primarily from the amount of data, not from model size.

How the data was collected

The data for Xiaomi-Robotics-1 was recorded by people, not robots. They held grippers with built-in cameras and performed ordinary household manipulations. This approach is cheaper and faster than teleoperating physical robots, and it made it possible to collect more than 100,000 hours of motion recordings.

• Model: Xiaomi-Robotics-1 from Xiaomi
• Training data volume: over 100,000 hours of motion
• Data was collected by people using handheld grippers with cameras, not by robots
• Key finding: growth in data matters more than growth in model size
• The absolute share of successfully completed tasks remains low

Why data matters more than model size

Adding training data improved Xiaomi-Robotics-1's results far more than increasing the model's parameter count. This directly contradicts the logic familiar from large language models, where the bet is often on the scale of the network itself. In robotics, where what is scarce is recordings of real physical actions, the bottleneck is the volume and diversity of data, not the model's computational capacity.

More data delivers a bigger gain than increasing model size — the key finding of

Xiaomi's research on Xiaomi-Robotics-1.

Where the approach hits its limits

Xiaomi-Robotics-1's quality gains have not yet plateaued: the curve keeps rising as more data is added, which means the potential has not been exhausted. But the absolute numbers remain low — the model successfully completes only a fraction of tasks. In other words, scaling with data works, but a reliable household robot that handles manipulation tasks with confidence is still a long way off.

What this means

Collecting data by hand, with people using cheap gripper-cameras, could become a way around robotics' main shortage: a lack of recordings of real actions. Absolute quality is still low, but since the gains are not hitting a ceiling, the bet on data rather than model size sets the direction for the next generation of robotic manipulators.

FAQ

How did Xiaomi collect data for Xiaomi-Robotics-1?

People recorded the data: they performed household actions with handheld grippers with built-in cameras instead of operating robots. This made it possible to collect more than 100,000 hours of motion recordings faster and more cheaply than with physical robots.

What matters more for training a robot — data or model size?

According to Xiaomi's results, data. Adding training recordings improved Xiaomi-Robotics-1 significantly more than increasing the model's size, and this gain has not yet plateaued.