On October 27, Xiaomi disclosed through its official technical account a bionic dexterous hand aimed at industrial use. The spec sheet arrived in one go: 21 degrees of freedom, hybrid drive from 26 motors, and 1,500-point high-density tactile sensing across the whole palm, with hybrid drive plus tendon tension control managing grip force. Xiaomi did not say which robot the hand will be mounted on, and gave no word on mass production timing or price.
A hand reverse-engineered from a car factory
Where the requirements came from is more concrete than the specs themselves. The pilot-production center of Xiaomi Auto's intelligent manufacturing department compiled 2,579 SOPs (standard operating procedures) across 16 typical workstations in an automotive plant. The team picked 1,827 of them for data collection and validation, and by Xiaomi's account the motion-capture data covers more than 70.8% of the manual operations at those typical workstations.
The captured motion data was classified, sliced, cleaned and retargeted, then compared against human-hand data and used in simulation. Tools, parts and materials actually used on the factory floor were also turned into simulation assets for multi-angle grasping tests.
On the tactile side, resolution is 1 mm at the fingertips, finger pads and finger bases, and 2 mm in the palm. The tactile skin is replaceable; Xiaomi stresses this is for easier maintenance and also makes it simpler to collect tactile data that is isomorphic with the human hand. The design follows on from the tactile glove Xiaomi built earlier, and heat dissipation and full-palm sensing were also optimized at the system level. Xiaomi calls this route “ultimate biomimicry.”
Doing the math on that 70.8%
Divide the two published figures and 1,827 ÷ 2,579 comes to roughly 70.84%, almost exactly the “more than 70.8% of procedures” Xiaomi cites. The back-of-the-envelope conclusion is that the coverage figure is most likely counted by number of SOPs, that is, the share of procedures chosen out of all procedures.
That way of counting has limits. One SOP might be a ten-second screw fastening; another might be a multi-minute wiring-harness assembly. Coverage counted by SOP number can differ widely from coverage counted by working hours or by motion difficulty. What the remaining nearly 30% of unselected SOPs are, whether too hard, too rare, or simply not work for a hand, Xiaomi has not said.
Another figure that can be worked out is the gap between motors and degrees of freedom: 26 motors for 21 degrees of freedom leaves 5 extra. Combined with the “hybrid drive plus tendon tension control” description, one possibility is that some joints are driven directly by motors while others run through tendons, with additional motors adjusting tension. This is an inference from the public description only; Xiaomi has not published the actual drive allocation.
Built for its own factory first
Plenty of teams in China build dexterous hands, and most start from general grasping demos, showing moves like picking up an egg, spinning a pen or twisting a bottle cap. Xiaomi has picked a different starting point this time: it first defined the workstations in its own car plant, then worked backward from the motions at those stations to decide how many degrees of freedom the hand needs and how fine the touch sensing should be. The upside is clear requirements, a controllable data source, and a test site right inside its own factory.
The cost is that generality is not yet visible. Grasp targets in automotive assembly are relatively fixed, with part sizes, weights and materials all documented in drawings. Move to logistics sorting or the home, where objects vary endlessly, and how much of this SOP-derived design carries over will need to be settled by future data.
The replaceable tactile skin is a distinctly engineering-minded choice. On a factory floor a single hand grasps thousands of times a day, so wear on the tactile layer is routine, and whether the skin can be swapped quickly on the line directly affects downtime. By writing “easy to maintain” into the announcement, Xiaomi signals that it is thinking about long-duration deployment.
What Xiaomi has published so far is described as “exploration results.” It gives no production metrics such as grasp success rate, cycle time or continuous running time, and public information offers no answer on how far this hand is from actually entering the workshop.
Sources: Xiaomi official technical account, ITHome, Sina Tech, CocoLoop; figures for degrees of freedom, motor count, tactile points and SOP count follow Xiaomi's disclosed numbers.