Xynova Flex 2 puts Xiaomi’s humanoid bet on the robot hand layer

Xynova has become one of the robotics component companies to watch because it is not only selling a dexterous hand. It is building the stack around the hand: high-degree-of-freedom fingers, compact actuators, joint modules, sensing, control, and developer-facing integration for humanoid robots.

XRoboHub reported that Xynova has launched Flex 2, a second-generation hybrid dexterous hand with 23 degrees of freedom, a 400 gram palm, plus or minus 0.1 mm repeatability, 0.05 N force control, a 12 kg single-hand grasp load, and multimodal sensing for adaptive grasping and slip detection. RoboHub has not yet found an official public Flex 2 product page carrying every one of those specifications, so those launch specs should be treated as reported specs until Xynova publishes a full data sheet.

What is publicly supported is the broader company direction. Xynova describes itself as a humanoid robotics component supplier focused on high-DoF dexterous hands, high-torque-density rotary actuators, and linear actuators. Its own company page says it operates an integrated production line spanning machining, motor winding, module assembly, and dexterous hand assembly.

The Xiaomi connection is also important. Public financing reports from late 2025 and early 2026 described Xiaomi Strategic Investment as an investor in Xynova, first around the angel-stage financing and then as an existing shareholder participating again around the Pre-A round. That does not prove Xiaomi’s own humanoid hand uses Xynova technology. It does show Xiaomi is putting capital into the manipulation component layer instead of only chasing full robot demos.

That distinction matters for humanoid robotics. Walking demos and whole-body motion get the attention, but useful work usually fails at the hand. A general-purpose robot has to hold soft objects, tools, handles, cables, packages, and irregular parts without crushing them or dropping them. That requires more than finger count. It needs force control, tactile feedback, repeatable actuation, low mass, software integration, and enough durability to survive repetitive work.

Flex 2, if the reported numbers hold up, sits directly in that bottleneck. A 23-DoF hand with fine force control and multimodal sensing is not just an end effector. It is a platform decision for humanoid builders who do not want to design every actuator, tendon path, sensor interface, and control loop from scratch.

The strategic read is simple: the humanoid race is splitting into layers. Some companies will sell complete robots. Others will own the components that every serious robot needs. Xynova is trying to be in the second group, where the winning product may not be the flashiest demo but the hand that reliably makes robots useful.

For buyers and developers tracking humanoid robots, Xynova Flex 2 belongs on the same watchlist as Tesla Optimus hand updates, Figure’s manipulation stack, Unitree’s developer humanoids, and China’s growing ecosystem of dexterous-hand suppliers. The key question is no longer whether a robot can wave. It is whether the hand can do boring physical work, thousands of times, without supervision.

Sources checked by RoboHub include Xynova’s official company page, Gasgoo coverage of Xynova’s Pre-A financing and Flex 2 roadmap, and earlier financing coverage naming Xiaomi Strategic Investment among Xynova’s backers.