Xiaomi Open-Sources VLA Large Model Xiaomi-Robotics-0

2026-04-27

On April 27, Xiaomi unveiled a significant upgrade to its open-source Vision-Language-Action (VLA) architecture, Xiaomi-Robotics-0, which initially debuted in February. Within its first month of release, the framework rapidly secured the sixth position on HuggingFace’s worldwide VLA model download charts.

Aiming to transition the technology into a plug-and-play industrial asset, the company has officially launched a comprehensive real-robot post-training pipeline, accompanied by fresh capability demonstrations.

Building upon its pre-trained foundation, Xiaomi demonstrated that merely 20 hours of physical interaction data enabled Xiaomi-Robotics-0 to master the intricate process of placing wireless earbuds into their charging case. The system now executes this multi-step maneuver with seamless, continuous precision.

Engineers highlighted that this particular task introduces two critical engineering obstacles:

  • The tolerance gap between the earbuds and the docking slots is exceptionally tight, demanding sub-millimeter spatial perception from the AI to guarantee exact alignment.
  • Both the earbud and casing surfaces exhibit a roughness level as low as Ra0.03μm, creating a high risk of slippage during contact. To counter this, the model must instantly recalibrate its motion trajectories to prevent assembly errors.

RECENT AI NEWS

RECENT AI TOOLS