Gasgoo Munich- In July 2026, temperatures in Shanghai briefly hit 40 degrees Celsius. But the heat inside the World Expo Exhibition Hall was even more intense. Last year, just over 80 companies focused on embodied intelligence attended WAIC; this year, that number surged past 200 — more than doubling.
Even more striking is the fundamental shift in the narrative. In this year's official WAIC preview, "AI dextrous hands" were listed alongside humanoid robots as a major new product category for the first time. A consensus is rapidly forming: what determines whether a robot can actually work is the point where it touches the world — the hand.
From "able to walk" to "able to work," from "able to chat" to "able to act," China's AI supply chain is executing a collective pivot. And dextrous hands — the critical "fingertip power" in this transition — are standing in the spotlight, facing scrutiny from industry, capital, and technology alike.
Yet a fundamental question remains unresolved: how much longer until dextrous hands are truly "good enough"?
The "Hand" Landscape at WAIC 2026
Walk into this year's embodied intelligence pavilion, and you see robots screwing bolts, assembling parts, sorting materials, and folding clothes everywhere. At the booths, audiences ask the same three questions: "How many degrees of freedom?" "Does it have tactile sensing?" "How much does it cost?"
Behind these questions lies an ever-growing list of contenders.
The direct-drive camp is making a massive splash.
Lingxinqiaoshou (Linker Robotics) unveiled its flagship Linker Hand O30, a 20-degree-of-freedom (DoF) fully direct-drive dextrous hand. Each joint features an independent drive unit, achieving full-link direct-drive control from the finger base to the tip. Weighing just 740 grams, it delivers a rated load of 30 kg in industrial settings.
Linker Hand O30; Image Source: Linker Robotics
Linjiedian (Critical Point) showcased the domestic debut of its direct-drive flagship, the OmniHand 3 Ultra-M. It boasts 20 active degrees of freedom, five fingertip visuo-tactile sensors, and over 300 3D tactile perception points on the palm.
Tashizhihang's DexHand adopts a 21-DoF quasi-direct-drive solution, structurally replicating the human skeletal anatomy at a 1:1 ratio.
Stardust Epoch (Xingdong Jiduan) presented the XHAND 1 PRO, featuring 21 fully active degrees of freedom, 18 full-hand tactile sensors, and a tactile sensitivity of 0.01 N.
Cable-driven solutions are equally impressive.
Pantheon Hand 20 Cable-driven Dextrous Hand; Image Source: Interstellar Light Year
Interstellar Light Year's Pantheon Hand 20 employs a world-first "drive-hand separation" architecture. Weighing a mere 290 grams, the entire hand can output a gripping force of 500 N.
Lingqiaozhineng's DexHand021 Pro, a wrist-hand integrated unit, features 22 degrees of freedom, bidirectional cable drive, and full-palm perception capabilities.
Unitree Robotics' Dex5-1 is equipped with 94 tactile sensors and boasts a minimum gripping diameter of just 10 mm.
Hybrid drive and biomimetic routes are rising rapidly.
Joyson Electronics' debut "Lingxi" TeleHand Professional Edition features an industry-first "in-palm integration + hybrid drive" solution, incorporating direct drive, tendon cables, and linkages within the palm simultaneously.
"Lingxi" Dextrous Hand; Image Source: Joyson Electronics
Xinuoweilai's Xynova Flex 2 places high-power drive units at the rear (biomimetic tendons) while retaining micro direct-drive motors at the palm's end, creating a hybrid drive configuration.
MoonSpring Biomimetics' Yingshou Y-Hand M2 holds the top spot globally among similar products, boasting 38 ultra-high degrees of freedom.
Tactile perception is becoming standard.
Qianjue Robot released the X-TouchMind V1, a VTLA (Vision-Touch-Language-Action) embodied tactile model that integrates vision, language, touch, movement, and robot body state into a single system.
Image Source: Tashan Technology
Tashan Technology is attacking the problem at the chip level. Its dynamic tactile sensing chip, the "Emerald E10A," draws a standby current of just 0.9 μA and achieves a time resolution of 2.9 μs.
Making its debut, Huaweike showcased two series of flexible electronic skin, "Dragon Scale" and "Lingxi," capturing a 70% market share in China's tactile sensing niche.
Incomplete statistics show over 20 companies exhibited dextrous hand products at this WAIC. From robot integrators and specialized hand manufacturers to upstream module suppliers, the entire supply chain made a collective appearance. As Su Yang, co-founder of Linker Robotics, put it: "You see that almost every robot in the hall has hands now. That’s a very clear market signal."
What Does "Good Enough" Mean?
Coinciding with WAIC, another bombshell dropped in the dextrous hand sector. On July 18, Future Human Intelligence released the world's first dextrous hand with 47 degrees of mobility. It uses a hybrid drive combining cables, linkages, and gears, equipped with 21 self-developed joint actuators. The hand achieves fully driven, independent control with no under-actuated blind spots; each finger can rotate a full 360 degrees, fingertips can bend backward by 10 degrees, and it possesses 42 independent degrees of freedom.
Image Source: Future Human Intelligence
47 degrees of mobility — what does that number signify? For context, the highest DoF hand at this WAIC was MoonSpring's Y-Hand M2 with 38 degrees, while a normal adult human hand typically has around 20 to 22 functional degrees of freedom. With a single product launch, Future Human Intelligence raised the industry's parameter ceiling by nearly a quarter.
But a deeper question surfaced: Does higher freedom of movement equate to being "good enough"?
For years, degrees of freedom were the most intuitive metric for progress in dextrous hands. From 10 to 16, 20 to 38, and now 47, the numbers kept climbing. Yet this year's WAIC sent a clear signal: the pure race for degrees of freedom is fading, and the definition of "good enough" is returning to systemic rationality.
This isn't to dismiss the value of degrees of freedom. MoonSpring's Y-Hand M2, with its 38 degrees, can perform 33 types of human-like dexterous manipulations with a 96% success rate. Future Human Intelligence's 47-degree product further expands the boundaries of independent finger movement — the ability to rotate each finger a full circle means it can theoretically achieve poses physically impossible for traditional hands.
But as industry analysts point out, "higher degrees of freedom usually mean more actuators, sensors, and control variables, which in turn brings increased weight, power consumption, heat, cost, and potential points of failure". The "human-like hand" on a spec sheet is still a full set of engineering capabilities away from being a "human replacement" on the production line.
So, what is the standard for "good enough"? Judging by the exhibits at WAIC, industry discussions, and the new variable introduced by Future Human Intelligence, it involves at least four dimensions:
First, tactile perception capability. In his keynote at WAIC, 2024 Turing Award winner Richard Sutton declared that AI is moving from a "human data era" relying on static text to an "experience era" of autonomous interaction. Dextrous hands with delicate, full-field tactile sensing are the indispensable hardware base for AI to step out of pure text training and achieve autonomous physical experience learning.
In Qianjue Robot's dual-arm collaborative box-folding demo, a staff member disturbed the folding cardboard. The robot sensed the deformation and force changes through tactile and visual feedback, yet still completed the process. This closed-loop capability of "perceive-adjust-execute" is far closer to the definition of "good enough" than a simple degree-of-freedom number.
Second, reliability and lifespan. A "good enough" hand must withstand the scrutiny of real-world scenarios. The industry generally requires a lifespan of 1,000 to 2,000 hours, or about 200,000 operations; in industrial settings, the requirement is often over 300,000 operations. Ruiyan Zhikong's RY-H2 has undergone multiple generations of iteration and ultra-long lifespan testing exceeding one million cycles; MoonSpring's Xinshou X-Hand M1 boasts a lifespan in the millions. These data points are drawing the bottom line for "good enough."
Third, acceptable cost. At this year's WAIC, many companies have pushed prices down to the 10,000 yuan level, or even the 1,000 yuan level. Ruiyan Zhikong's RY-H2 brought the price of a high-quality hand below 10,000 yuan. But "good enough" isn't just about unit price; it's about ROI. Linker Robotics' Linker Hand O30 costs 59,000 yuan without tactile modules and 69,000 yuan with them — for scenarios requiring mass deployment, there is still room for prices to fall.
Fourth, scenario adaptability. Definitions of "good enough" vary wildly by scenario. Precision assembly values force control accuracy and repeatability; logistics handling prioritizes load, durability, and cost; research platforms might chase degrees of freedom and data interfaces. Zhenghe Robot simultaneously launched the TriFlex three-finger hand for general industry and the DuoPick two-finger hand for standardized scenarios; Wanna Robot built a full matrix from the Pro20A flagship to the Eco12 lightweight anthropomorphic model. This "scenario-defines-hardware" mindset is providing a more precise annotation for "good enough."
Even Future Human Intelligence, which just shattered the DoF record, isn't simply stacking parameters. Its choice of a hybrid "cable-linkage-gear" drive over a single technological route shows that even in the pursuit of upper limits, the industry realizes no one solution fits all.
The standard for "good enough" is shifting from a single-dimensional parameter race to a multi-dimensional assessment of system capability.
Who Is Betting on What?
The fiercest debate in the dextrous hand sector right now concerns the choice of technology roadmap.
Tendon (cable) solutions mimic human tendons, using flexible cables to pull joints. Their advantage lies in the ease of placing motors in the palm or forearm, allowing for lighter fingers suitable for high degrees of freedom and compliant grasping.
Zhou Chen, co-founder and CEO of Lingqiaozhineng, starts from the logic of "software-defined hardware" — "Cable drives are naturally closer to the human hand's driving method and respond better to the uncertainties of the physical world."
Interstellar Light Year is also bullish on the "performance ceiling" of cable drives. Its co-founder, Zhou Jiaqi, stated bluntly: "Cable drives are technically orders of magnitude harder than direct drive or linkage solutions, but their performance ceiling is also much higher."
As far back as the middle of last year, Tesla announced a full pivot to the cable-driven route, adding weight to this approach.
But the shortcomings of cable drives are equally obvious: cable friction, slack, and wear introduce hysteresis, and tension consistency and maintenance difficulty increase after long-term use.
Direct drive or quasi-direct drive solutions place power units as close to the joint as possible, reducing friction, backlash, and elastic deformation caused by long-distance transmission. The concentrated appearance of fully direct-drive products at this WAIC made them a market focus.
The direct-drive structure of Linker Robotics' Linker Hand O30 fundamentally eliminates transmission backlash, delivering sub-millimeter repeatability and millisecond-level dynamic response speeds.
More importantly, the motion model of direct drive is more deterministic and predictable, showing a natural advantage in Sim2Real transfer training — strategies trained in simulation deploy more smoothly onto real robots.
But the cost is equally direct: micro-actuators must simultaneously meet requirements for power density, volume, heat dissipation, and cost. Once multiple power units are packed into a palm, weight and heat quickly become constraints.
Linkage solutions rely on rigid mechanisms to transmit motion, with relatively deterministic structural relationships, making repeatability, load capacity, and maintenance easier to engineer.
Lin Tong, Director of the Perception Acquisition Center at Parsini, believes that "tendon solutions have strong explosive power but limited lifespan; direct drive offers excellent control but insufficient output limits. By comparison, linkage solutions are the optimal choice for balancing high performance, long life, and high stability".
However, linkages occupy internal finger space, increase weight, and their mechanical coupling may sacrifice some independent movement capability.
Hybrid drives attempt to combine the best of all worlds.
Joyson Electronics' "Lingxi" TeleHand integrates direct drive, tendons, and linkages within the palm. The integrated micro-actuators are half the size and 30% lighter, with torque density 2 to 3 times higher than industry competitors.
Xinuoweilai's Xynova Flex 2 places high-power drive units at the rear (biomimetic tendons) while retaining micro direct-drive motors at the palm end. As Xia Yuxuan, CEO of Xinuoweilai, put it: "Pure direct drive and pure cable drive are like two sides of a coin — they shine in the vacuum dynamics of a lab, but each faces a high wall for industrial application. True breakthroughs are often born from fusion."
Biomimetic tension-compression bodies represent a more fundamental underlying innovation.
MoonSpring Biomimetics relies on the "Biomimetic Tension-Compression Body Robot Theory and Technology" proposed by Professor Ren Lei, completely mimicking the logic of human bone, muscle, and ligament coordination. Its Yingshou Y-Hand M1 palm weighs just 299.7 grams yet outputs a gripping force of 200 N. This route, starting from underlying mechanical logic, attempts to fundamentally solve the contradiction of traditional rigid structures: "stiff when powerful, weak when lightweight."
Image Source: MoonSpring Biomimetics
Technology roadmaps are far from converging. As industry analysis points out, "tendons, linkages, direct drive, and hybrid drives will coexist in different scenarios for a long time, and even the same hand may combine multiple solutions". What is truly becoming confirmed is an industrialization path: scenario defines hardware architecture, tactile sensing enters the control loop, and lifespan, cost, and batch consistency determine whether a product can be delivered. Precision assembly values force control, logistics values load and durability, research platforms chase degrees of freedom — different scenarios demand different optimal configurations.
How Far Is "Good Enough"?
Answering this requires examining three sets of data.
The first set is market expectations. Fang Hainan, CMO of Yinshi Robot, revealed that the company delivered over 10,000 dextrous hands in 2025, with a target of 30,000 to 50,000 for 2026. Linjiedian achieved operating net profitability in Q1 2026, with cumulative deliveries of the OmniHand series exceeding 8,000 units. In terms of shipment volume, dextrous hands are moving from "lab toys" to "industrial tools."
The second set is capital heat. Total financing in the domestic dextrous hand sector was 16.877 billion yuan in 2025; in just the first half of 2026, financing exceeded 25 billion yuan. In Q1 2026 alone, sector financing approached 5 billion yuan, a 70% surge over the full year of 2025. Two phenomenal star companies emerged: Linjiedian completed four rounds of intensive financing in five months, with the latest round near 1 billion yuan and a post-money valuation exceeding 1 billion USD; Linker Robotics raised seven consecutive rounds, reaching a valuation of 3 billion USD, with its new round targeting a valuation of 6 billion USD (about 410 billion yuan). According to public data, total financing in the first half of 2026 has already surpassed 25 billion yuan, far exceeding the full-year scale of 2025.
The third set is reality bottlenecks. Beneath the surface of prosperity lie fatal hidden dangers: most companies have not yet figured out a profitable model, product prices are caught in a "race to the bottom" compressing margins, and four core bottlenecks — hardware, data, algorithms, and scenarios — remain unbroken. Tesla's Optimus has "slipped" multiple times due to hand bottlenecks, making the battle for mass production of dextrous hands far harder than imagined.
The temperature gap between these three data sets reveals the true answer to "how far is 'good enough'?"
On the hardware level, the gap is rapidly narrowing. From degrees of freedom to tactile perception, from drive schemes to material processes, domestic dextrous hands have approached or reached international advanced levels in multiple dimensions. The China Industrial Internet Institute has released standards such as the "General Technical Requirements for Humanoid Robot Dextrous Hands"; Wuxi Kaiqi Robot, jointly with an academician team, developed China's first comprehensive dextrous hand testing equipment. The establishment of standards and testing systems is a key marker of industrial maturity.
On the data and algorithm level, challenges remain arduous. Dexterous manipulation is viewed as the "last centimeter" of embodied intelligence landing. In this track, hardware reliability, multimodal perception, data quality, algorithm generalization, and scenario diversity are highly coupled; a weakness in any link could become a bottleneck for robots entering the real world.
The TacVerse 1k dataset released by Qianjue Robot contains 1,000 hours of real physical interaction records; Linjiedian's DUET dual-layer embodied contact intelligence architecture stratifies long-sequence task planning and fine contact control — these explorations are attempting to fill the data and algorithm void.
On the cost level, the downward trend is clear but has not yet reached the threshold for mass deployment. The cost of a dextrous hand accounts for 18% to 25% of a humanoid robot's total cost, directly dictating the unit price. From the 10,000 yuan level to the 1,000 yuan level, prices are falling fast, but there is still distance to the "affordable" mass-market line.
On the scenario level, the leap from demo to delivery is happening. This year's competitive focus has expanded from single-step operations to long-horizon tasks. The simulated real-world operational environments seen everywhere at the booths reflect the anxious battle for commercial landing. From twisting balloon dogs to folding boxes, from swapping graphics cards to dual-arm collaboration, these displays are no longer showing off tech, but publicly responding to industry-wide problems.
Conclusion
Returning to the opening question: how much longer until dextrous hands are truly "good enough"?
Judging by this WAIC, the answer is neither far nor near.
Not far, because breakthroughs in hardware capability are happening at a visible speed. From the DoF race to tactile sensing, from single drive to hybrid solutions, from lab prototypes to 10,000-unit deliveries, the dextrous hand industry has traveled a road in less than three years that might take other hardware sectors a decade. Sales in China's dextrous hand market are expected to reach 70,200 units in 2026, a number that is accelerating upward.
Not near, because every step from "can demo" to "can deliver," from "can grasp" to "can perceive," from "lab-ready" to "factory-ready" to "home-ready" implies a fresh set of engineering challenges. Tactile perception precision, long-horizon task stability, mass production consistency, continuous operation reliability — each of these baseline metrics for "good enough" takes time to polish.
The dextrous hand industry stands at a critical crossroads. On one side are the heat of capital and market expectations; on the other, the real barriers of engineering landing. As one practitioner said at the booth: "The core change in the dextrous hand industry this year is from 'having one' to 'being able to work.'"
And the answer to "can it work" will ultimately be written by real scenarios — not on spec sheets or in financing news, but in every tightened screw, every folded box, and every stable grasp.
