Xiaomi showcases its humanoid robots working autonomously in factory settings with 90.2% success rate using a VLA + model that fuses vision with fingertip sensor data, approaching human-level performance on the production line.

Xiaomi has released a significant reality check for the robotics industry, showcasing three hours of autonomous operation data from its humanoid robots working in its Beijing electric vehicle factory. The specific task involved the bilateral installation of self-tapping nuts on large integrated die-cast parts—a precise, dexterous operation critical to automotive assembly. The robots achieved a 90.2% success rate with a consistent 76-second cycle time, a metric that directly meets the factory's required 'production beat.' This demonstration moves beyond controlled lab environments and validates Xiaomi's 'factory-first' strategy, proving that humanoid robots can perform economically viable work on real production lines. The benchmark for commercial viability is now clearly set for 2026.

The technical core enabling this performance is a sophisticated Vision-Language-Action (VLA) model that uniquely fuses visual data from onboard cameras with high-fidelity tactile feedback from fingertip sensors. This multi-modal approach allows the robot to understand its environment and manipulate objects with near-human precision and adaptability. The 90.2% success rate on a complex, multi-step mechanical task signals a major leap from research prototypes to practical industrial tools. For manufacturers, this translates to a potential solution for labor-intensive, repetitive assembly jobs, with the promise of 24/7 operation. The next phase will involve scaling this technology across more varied and challenging tasks to drive down the failure rate and further close the gap with human workers.