After four months of real-world training at Xiaomi's auto factory, humanoid robots achieved a 98% success rate in bilateral self-tapping nut insertion, up from 90.2%. Two new tasks were added, marking a new stage in intelligent automotive manufacturing.
Technical Breakthrough: Four Months of Production Line Training Delivers Impressive Results
Xiaomi's robotics division recently released the latest progress of its humanoid robots after four months of training at Xiaomi's auto factory. Since entering the Xiaomi EV factory for production line operations early this year, these humanoid robots have expanded from single-station trials to multi-scenario industrial operations, with continuous improvement in precision, stability, and coordination.
After a quarter of real-world production line conditioning, the bilateral self-tapping nut insertion success rate climbed from 90.2% to 98% — just 1 percentage point shy of human qualification standards, with workstation stability effectively reaching mass production requirements.
Core Capabilities: From Single Tasks to Flexible Operations
While solidifying capabilities at established stations, Xiaomi's robots have successfully taken on two new roles in general assembly: sorting center console side covers and folding recycling totes, with both new tasks achieving a stable 90% success rate.
Sorting center console side covers represents the core technical breakthrough of this training phase. These components are large, irregularly shaped, and highly flexible, requiring a process involving multiple pick-and-place operations across various positions. The task demands extreme precision in whole-body control, dual-hand coordination, fine manipulation, and environmental adaptability. It marks the first time Xiaomi's robots have achieved stable, continuous operation over extended periods on flexible automotive parts.
Through whole-body degree-of-freedom control, active compliance strategies based on end-force perception, coordinated handovers, and in-hand pose adjustments, the robots can autonomously handle unexpected issues like jams or snags to execute precise placement.
Technical Architecture and System Integration
At the tote folding and recycling station, the robots overcame challenges such as fine fingertip control, dual-arm coordination, box stacking, and multi-robot pacing to achieve continuous, stable operation. Furthermore, the robots interface directly with the factory's digital system to read production tasks and material data, eliminating the need for paper documentation. Multiple units can synchronize their status in real time, while a remote human intervention mechanism remains in place to balance operational efficiency with safety.
| Training Metric | Initial Data | Latest Data | Improvement |
|---|---|---|---|
| Bilateral Nut Insertion Rate | 90.2% | 98% | +7.8pp |
| New Task Success Rate | — | 90% | New |
| Gap vs Human Standard | — | 1pp | Near达标 |
Mass Production Prospects and Industry Significance
Xiaomi's progression from single-station trials to multi-scenario coverage, and the leap from 90.2% to 98% success rate, demonstrates the rapid iteration capability of humanoid robots in automotive manufacturing scenarios. However, 98% success rate still has distance from true large-scale mass production — automotive production lines have extremely stringent requirements for consistency and zero defects.
For the global automotive manufacturing industry, humanoid robot production line applications remain in the early validation stage. The progress of Chinese automakers in this field is worth watching and may have profound implications for the global intelligent manufacturing landscape in the future.
