When a robot hand gets too hot, the problem is not drama. It is downtime. Xiaomi’s latest redesign of the CyberOne humanoid hand turns that engineering constraint into the main story. The company shrank the hand by 60 percent to reach a human-scale form factor and then packed it with more dexterity, more sensing, and a cooling system modeled on sweating. That combination matters because robotic hands are among the most crowded pieces of hardware in humanoid machines: motors, transmissions, sensors, and structure all compete for the same few cubic centimeters, while heat quietly erodes precision and service life.
The new hand is built around a 1:1 human-scale geometry based on a 1.73-meter body model, a choice that helps robotic training transfer more cleanly from simulation into real-world tasks. Xiaomi also reported an 83 percent increase in active degrees of freedom, pushing the hand closer to the range associated with human manipulation. On the sensing side, the palm now carries about 8,200 square millimeters of tactile coverage, allowing the robot to register pressure and contact well beyond the fingertips. That matters in factory-style work, where cameras can be blocked by parts, fixtures, or the robot’s own motion, and where force control often matters more than a clean line of sight.
The unusual part sits inside the thermal design. Xiaomi says the hand uses 3D-printed metal liquid-cooling channels that behave like sweat glands, shedding heat generated by dense drive motors during sustained operation. In reported testing, the system could dissipate around 10 watts of heat, enough to address a common failure mode in compact robotic end effectors. This is not an isolated idea in robotics research. Cornell engineers previously demonstrated 3D-printed hydrogel actuators that sweat for thermal regulation, showing that biological cooling strategies can outperform bulkier internal airflow solutions in confined designs.
The deeper point is that heat management is becoming a defining systems problem for humanoids. As processors, actuators, and sensors get more capable, they also concentrate more thermal load into smaller spaces. Without careful control, components throttle, calibration drifts, and wear accelerates. A robotic hand may look like a manipulation problem, but in practice it is also a packaging problem, a materials problem, and a thermal problem at the same time.
Xiaomi’s durability figures suggest why this redesign is getting attention. The company said the hand survived more than 150,000 grasping cycles, well above the roughly 10,000-cycle threshold often associated with tendon-driven failures. In automotive nut-fastening work, it reportedly sustained a 90.2 percent success rate inside a 76-second factory cycle over three hours of operation. Those numbers do not make the hand human, but they do shift the conversation from impressive demos to repeatable duty.
There is also a research angle. Xiaomi said it gathered tactile data with gloves and released TacRefineNet along with 61 hours of raw tactile data. That kind of dataset matters because touch remains one of the least mature inputs in robot learning. Vision has dominated for years; touch is what keeps a machine working when vision becomes unreliable. A sweating robot hand sounds like a gimmick until the numbers point elsewhere. In humanoid robotics, staying cool is starting to look like a prerequisite for staying useful.
