WRC 2026: Dexterous Hand Tech Routes Clear Up, Touch Becomes Standard, Industry Focus Shifts to Deployment

According to a report from Leiphone.com, more than a dozen dexterous-hand manufacturers attended the event, including Lingxin Qiaoshou, Xinuo Future, BrainCo, InTime Robot, Lingqiao Intelligent, Aoyi Tech, ZWHAND from Zhaowei Electric, Dahuan Robot, Paxi, Zhongke Xiji, and Lingzhang Robot. Robot-body makers such as Octopus Power, Chaowei Power and Xingdong Era also brought their own hands. Three changes stood out: roughly 20-degree-of-freedom (DOF) hands have become common; tactile sensing has moved from optional to standard; and the products actually shipping in volume are not the high-DOF flagships but six-DOF five-finger hands, followed by three-finger models.

On high-DOF hands, three drive routes are taking shape: direct drive, tendon-driven, and hybrid drive. Direct-drive hands, such as Lingxin Qiaoshou's Linker Hand O30 (20 DOF) and Xinuo Future's Prima 1 (22 DOF), transmit force transparently and are mainly sold to universities and research institutes. Tendon-driven designs place motors in the forearm or wrist, allowing thinner fingers; examples include Lingqiao Intelligent's DexHand021 Pro (22 DOF) and Octopus Power's OctoH-Hand (20 DOF). Hybrid drive, which combines direct drive with tendons, is visibly gaining traction. InTime Robot's RH524J1 has 24 active DOF, and Chaowei Power's KAIHand has 37 DOF (20 active). According to Chaowei's dexterous-hand lead Gao Song, the hybrid approach balances tendon compliance with direct-drive force control, but wiring within a tiny space is one of the hardest engineering challenges.

Tactile sensing was nearly ubiquitous at the booths. Four sensor routes coexist: piezoresistive, capacitive, visuotactile, and Hall-magnetic. Piezoresistive sensors are the most common due to their thinness and low cost; Octopus Power covers the whole palm with 1,900 touch points. Aoyi Tech uses capacitive sensors, which offer high sensitivity and design flexibility. Visuotactile sensors, which use a miniature camera to capture elastic-skin deformation, provide very high information density and are preferred by high-end hands; both Xinuo Future and BrainCo offer this option on their flagship products. Paxi follows the Hall-magnetic route, claiming 0.01N micro-force resolution. Manufacturers have not converged on a single solution; many offer several options depending on cost, size, accuracy and use case.

Weitai Robotics CEO Li Rui, a longtime researcher in visuotactile sensing, said the industry's question is no longer 'yes or no' but 'when' for touch. He likened current embodied AI to a person whose fingertips are anesthetized: they can see and act but fail at delicate tasks. Li said visuotactile sensors can reach tens of thousands to hundreds of thousands of touch points per square centimeter, far above conventional array sensors, and can measure tangential force, which is essential for grasping and insertion. He said the field is converging on visuotactile technology, and his company claims to be the only one currently mass-shipping visuotactile sensors for five-finger hands.

Despite the exhibition's high-DOF enthusiasm, actual market demand tells a more conservative story. Zhongke Xiji CEO Zhang Tianyi said the company targets annual capacity of 30,000 units, with six-DOF products accounting for 70 to 80 percent and the rest mostly three-finger models. Dahuan Robot's three-finger hand carries a 15kg load and a 50N gripping force; Zhongke Xiji's three-finger G series is rated at 30kg in normal use and up to 50kg in the lab. For many industrial tasks, a 20-DOF human-like hand is unnecessary; what matters is stable grasping and handling. High-DOF hands still face reliability and engineering barriers that keep them from large-scale production.

On the broader question of embodied AI landing in industry, Luming Robotics unveiled NexCore, a platform that connects data collection, data governance, model training, skill generation, evaluation, deployment and operational feedback. Founder and CEO Yu Chao said the industry should not define progress through robot performances but through real scene requirements. He argues that the true scaling law for embodied AI lies in accumulated task experience—real factories running thousands of tasks produce unique data, skill models and failure records. NexCore supports local deployment and works with non-Luming robots, aiming to give factories a tool to create their own robot skills. The company's cooperation with Mitsubishi Electric, which started with PLC quality inspection and expanded to screw driving and material handling, deepened to the point that Mitsubishi became an investor.

The report also highlighted a gap between demos and deployment. An industry insider said Figure AI's grab-and-place success rate on a BMW line was below 50 percent, and the displayed work scene was customized for the robot. A system integrator, Chen Hong, noted that many robot makers refuse to lend prototypes, and that a humanoid can cost 800,000 to 900,000 yuan after deployment, while factories are only willing to pay 300,000 to 400,000 yuan. These examples illustrate why the industry is moving from universal humanoids toward task-specific, data-driven solutions.