Xiaomi-Robotics-1 shows that more data beats bigger models when training robots to move
100,000 hours of motion data. Xiaomi just proved the scaling law that changes robot training forever: more data > bigger models.
Why it matters
Xiaomi's robotics foundation model challenges the prevailing assumption that model scale drives performance. By prioritizing data quantity over parameter count, they've demonstrated a different path to robot capability—with implications for how companies should allocate training resources.
The key facts
6 to knowXiaomi-Robotics-1 trained on 100,000+ hours of motion data
Data scaling outperformed model size scaling in performance gains
Training data collected via camera-equipped handheld grippers (human demonstrations), not deployed robots
Performance gains have not plateaued
Absolute success rates remain low (specific metrics not provided)
Implies different scaling law for robot foundation models vs. LLMs
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Xiaomi trained Xiaomi-Robotics-1 on more than 100,000 hours of motion data collected by people using camera-equipped handheld grippers rather than robots. Adding data improved performance far more than increasing model size. The gains haven't plateaued, though absolute success rates remain low.