Xiaomi Unveils Second-Generation Humanoid Robot, Degrees of Freedom Increased from 21 to 66
Nashnova编辑部
Xiaomi debuted its second-generation humanoid robot at the 2026 World Robot Conference, with degrees of freedom leaping from 21 to 66 and a factory-tested bolt-tightening success rate of 98% — the industry's first published performance metric from a real production line, though no mass-production timeline has been announced.
What exactly got upgraded?
The robot stands 1.7 m tall and weighs 66 kg. Degrees of freedom jumped from 21 to 66, with 33 concentrated in the hands for finer grasping and manipulation.
This means → the threefold increase in freedom was bet almost entirely on dexterity. Xiaomi sees hand precision as the first gate to factory deployment.
The control system uses a "brain + cerebellum" layered architecture: the brain — a large language model — handles task comprehension, planning, and voice interaction; the cerebellum handles motion control and dynamic balance. In plain terms = the brain decides *what to do*; the cerebellum decides *how to move*.
Four months on a factory floor — how did it perform?
Xiaomi says the robot completed four months of live testing on its car-assembly line, performing bolt-tightening and bin-carrying tasks with a bolt-tightening success rate of 98%.
This means → it is the first time any maker has published quantified humanoid-robot performance from a real industrial setting. The signal matters more than the number itself.
Yet hand movements still looked stiff during the live demo, and Xiaomi has not announced a mass-production plan. The gap between "high single-task success rate" and "deployable at scale" remains wide.
Who dominates the global market?
DIGITIMES analyst Shen Zouhao disclosed that 2026 global humanoid-robot shipments are projected at 38,000 units, 93% from Chinese makers; 2027 is forecast at 45,000 units.
By 2030, shipments could reach 183,000 units, with Chinese makers still holding roughly 71% share. This reflects an early-mover advantage and a more complete domestic supply chain.
Shen noted that Taiwanese suppliers looking to enter the market would need to leverage non-China supply-chain entry points for differentiation.
What does the evolving hardware stack mean for chips?
Humanoid-robot hardware is converging on a "brain, cerebellum, peripheral nerves" three-layer architecture, each layer demanding a different class of chip.
The brain layer requires compute approaching top-tier on-device AI servers — Tesla's Optimus Gen 3 already hits 2,000 TOPS. The cerebellum layer shares computing characteristics with automotive systems, giving chipmakers with automotive backgrounds an edge.
The peripheral-nerve layer is where rising degrees of freedom drive surging demand for joint micro-controllers (MCUs — tiny chips that control each joint's motion) and sensors. Nvidia has released its Holoscan sensor-bridging protocol; STMicroelectronics and Microchip Technology have already adopted it. In plain terms = more joints mean more "nerve-ending" chips — a market that scales linearly with joint count.
How far away is real commercialisation?
Xiaomi's factory test data is a milestone signal that humanoid robots are moving from showcase to industrial deployment — but it does not equal commercial readiness.
Two nodes will define the timeline: when a mass-production schedule is announced, and whether hand precision can break past its current ceiling — the stiffness visible in the live demo shows that gate has not yet been cleared.
This reflects the broader industry reality: single-task success rates can be impressive, but the distance from lab results to volume production remains the core hurdle.
Content is for reference only, not financial advice.