Discover how Xiaomi’s humanoid robot achieved a 90.2% success rate assembling electric‑car components, marking a historic step in AI manufacturing. Learn more now!
Xiaomi has taken a decisive leap in the race for intelligent manufacturing. For the first time ever, a fully autonomous humanoid robot worked inside an electric‑vehicle (EV) factory, assembling complex components with a 90.2% success rate.
Record‑breaking three‑hour test
In a live demonstration shared on Weibo, the robot operated continuously for three hours at the bolt‑installation station of Xiaomi’s EV casting plant. During that time it completed the fastening of self‑tapping bolts on both sides of the work cell, meeting the line’s fastest cycle time requirement of 76 seconds per unit.
What the robot actually does
The task required the robot to:
- Accurately pick self‑tapping bolt belts from an automated feeder.
- Place each belt onto a positioning jig.
- Synchronise with a sliding conveyor to tighten the bolts automatically after casting integration.
Achieving precise alignment and reliable fastening proved to be the biggest challenge because the internal spline geometry, non‑fixed grasp posture, and magnetic interference all increased assembly complexity.
AI‑driven solution: Vision‑Language‑Action (VLA) model
Xiaomi tackled these hurdles with a data‑centric control approach powered by a 4.7‑billion‑parameter VLA model called Xiaomi‑Robotics‑0. The model combines visual perception, natural‑language reasoning, and action planning, reinforced by continuous learning through simulation.
This hybrid training pipeline reduces reliance on real‑time remote supervision. The robot quickly adapts to varied operating conditions and constantly refines its behavior from physical interactions.
Multimodal perception for robust performance
Beyond vision, the system fuses tactile feedback and joint proprioception, dramatically lowering the chance of mis‑evaluating complex states. The result is a more stable and resilient operation even when faced with unpredictable disturbances.
Ultra‑fast whole‑body movement control
At the core is a hybrid controller that blends optimisation‑based planning with reinforcement‑learning policies. Each optimisation loop finishes in under 1 ms, guaranteeing real‑time responsiveness. The RL component was trained on hundreds of millions of simulated noise patterns, enabling the robot to maintain balance under extreme disturbances and transfer zero‑shot to the physical hardware.
Scaling up the deployment
The successful bolt‑assembly station is just the first step. Xiaomi is already piloting the humanoid on additional workcells—ranging from parts picking in crates to installing front‑badge components—while focusing on two critical metrics: cycle‑time reduction and yield improvement.
Industry outlook
Xiaomi’s founder, Lei Jun, forecasted that dozens of humanoid robots will be operational across its factories within the next five years. The move mirrors a broader industry push, with rivals like Tesla planning mass‑production versions of their Optimus robots by early 2026.
As AI‑driven robotics accelerate, Xiaomi’s breakthrough signals a new era where flexible, human‑like machines can reliably perform high‑precision tasks on automotive assembly lines.
What’s next?
Stay tuned for updates as Xiaomi expands its humanoid fleet, refines the VLA technology, and aims to set new standards for speed and accuracy in EV manufacturing.