Interview with Zhaowei's Chen Yidong: For Dexterous Hands, the First Design Goal Is to "Look Good"

Gasgoo Munich- It took humanity millions of years to stand upright, but the trait that truly separates us from the rest of the animal kingdom isn't our legs — it's our hands.

Opposable thumbs, rich tactile nerve endings, and a feedback system capable of rapidly adjusting grip strength: this is the precision machinery that allows a human hand to pick up a needle or swing a hammer with equal ease.

If humanoid robots are to replace humans, they must solve the same riddle: how to build a hand that is dexterous enough?

To explore that question, we sat down with Chen Yidong — Deputy General Manager of Zhaowei Machinery & Electronics and General Manager of Zhaowei Dexterous Hand — at the recent Juhe Intelligent Industry Development Conference 2026, co-hosted by the Juhe Intelligent Industry Innovation Center and Hubei Sci-Tech Investment.

Zhaowei has spent 25 years refining micro-transmission systems, establishing itself as a domestic leader. Now, the company is transplanting its automotive-grade system capabilities into the embodied intelligence arena. It has already launched four generations of dexterous hand products, which are now operational on the production lines of top-tier international manufacturers.

Chen Yidong, Deputy General Manager of Zhaowei Machinery & Electronics and General Manager of Zhaowei Dexterous Hand

Over 30 Billion Yuan Floods In, Igniting the "Dexterous Hand" Boom

In the humanoid robot race, the dexterous hand is currently one of the hottest sub-sectors.

Just how hot?

Data from IT Juzi offers a stark illustration. By early August this year, the dexterous hand sector had seen 74 funding rounds involving 47 companies, with disclosed totals reaching roughly 28.51 billion yuan — already surpassing the full-year tally for 2025.

Add in the recently disclosed 4.5 billion yuan raised by Sharpa, and the sector's total funding for the year has breached the 30 billion yuan mark.

The money isn't pouring in without reason.

On a macro level, China's 15th Five-Year Plan has identified embodied intelligence as a strategic priority. Industrially, nearly 100 companies have clustered in the dexterous hand space. And capital continues to flow in real money. As Chen puts it, "The entire embodied robotics sector has achieved more in the past two years than in the previous twenty."

The data offers the most direct proof. A report from the China Humanoid Robot & Embodied Intelligence Council shows that in the first half of this year, Chinese humanoid robot shipments exceeded 40,000 units — accounting for 97% of the global total. For all of 2025, that figure was under 20,000. Gasgoo Automotive Research Institute projects that global humanoid robot shipments could approach 100,000 units this year.

By 2030, optimistic forecasts suggest global shipments will approach, or even exceed, 1 million units.

One humanoid robot requires two hands. If fully equipped with dual hands, that implies a theoretical market space of nearly 2 million sets. Of course, actual penetration rates will depend on the assembly ratio of five-finger solutions.

Why the hand?

The answer is simple: Humanoid robots rely on versatility to replace humans. Break it down: the legs transport the robot to the workstation, the "brain"""decides what needs to be done, and the hand ultimately executes the task.

Every action a robot takes to interact with the physical world must pass through these hands. One could even argue that the dexterous hand determines the functional ceiling and commercialization speed of the humanoid robot; the boundary of the hand's capability is the boundary of the robot's deployment.

For now, in highly structured industrial scenarios — like fixed loading and unloading or sorting standard materials — simple grippers suffice. But what comes next?

Inserting wire harnesses, tightening screws, polishing, and precision assembly — each demands millimeter-level accuracy and gram-level force control. Further down the line, as robots enter homes, they face objects of varying shapes, materials, and weights. With no standard workflows or fixed stations, generalization is impossible without a dexterous hand.

Only when the "hand" is capable enough can humanoid robots break free from the limitations of a single workstation, transitioning from specialized automation equipment to general-purpose intelligent labor — thereby unlocking vast commercial applications.

Capital is betting on the hand for exactly this calculation.

Data from Qianzhan Industry Research Institute indicates that China's dexterous hand market is expected to reach 960 million yuan in 2026, a 170% year-on-year increase. By 2031, it could surpass 20 billion yuan, with a compound annual growth rate exceeding 80%.

Breaking it down to a single robot, Chen offers a clearer quantitative judgment: The dexterous hand will eventually account for 10% to 15% of the total hardware cost.

"While the overall cost of hardware is dropping very fast, and the price of the hand will fall accordingly — causing this ratio to adjust dynamically — the general direction won't change. It will hold steady at 10% to 15%."

Notably, foreign firms that have analyzed the bill of materials (BOM) for Tesla's Optimus estimate the hand accounts for about 17% of the value — broadly aligning with Chen's assessment.

By this calculation, the dexterous hand is set to become one of the independent modules with the highest value share on a humanoid robot.

Image source: Zhaowei Machinery & Electronics

How Do You Design a Good Dexterous Hand?

"My understanding might differ from others, but I believe the first goal is that it has to look good."

Chen's logic is straightforward. The hand is part of the human body, and people are naturally drawn to beautiful things. Guided by this philosophy, Zhaowei has consistently pursued a five-finger design. In March, their dexterous hand won a Red Dot Design Award for its anthropomorphic design.

But good looks only solve the problem of acceptance.

Digging deeper, the dexterous hand must ultimately boost productivity. The second hurdle, therefore, is reliability.

"The essence of reliability lies in the components — from the silicone and plastics matched with the hand to the bearings and even the assembly process. Everything must be reliable. That's why we define reliability as a critical aspect of the dexterous hand," Chen explains.

Looking further ahead, once the robot's "brain" becomes generalized, a single hand must adapt to various tasks — making high degrees of freedom (DOF) crucial.

So, the third hurdle is multi-DOF.

Anthropomorphic design, high reliability, and multi-DOF: these are the three goals Chen has set for a "good hand."

But building such a hand is no easy feat.

"Last year, my understanding was that dexterous hands were defined by ‘three smalls' and ‘three bigs.' The smalls were small size, light weight, and low price; the bigs were big data, high DOF, and intelligence. Now, my understanding is that a dexterous hand needs to be a ‘decathlon' athlete," Chen notes.

He also points out an easily overlooked detail: The development of the dexterous hand is tied to the investment in the entire embodied hardware system. Often, when a dexterous hand malfunctions, the root cause isn't the hand itself — it's the "brain" or the body.

A common example: A robot falls because its "brain" isn't capable enough to keep it stable. If the dexterous hand doesn't adjust during the fall and hits the ground directly, it gets damaged.

"So how does the hand close quickly during a fall? That involves a technical problem: high-speed motion under weak magnetic control. Electric vehicles rely on this control technology to reach speeds of 150 or 200 km/h. Zhaowei has transplanted it from the automotive sector."

Materials present another hurdle. If a robot malfunctions and flails about, the machine shouldn't break. This requires the dexterous hand to possess anti-collision capabilities.

Beyond materials, there are heat, vibration, noise, and jitter — each requiring a dedicated solution.

Heat is a "hard bone" for the entire industry. A high-DOF dexterous hand needs to cram over twenty motors into a palm-sized space. With heat sources stacked together, dissipation is incredibly difficult. Moreover, high temperatures cause tactile signals to drift.

Behind this, Chen attributes the bottleneck to two factors: chip performance and the "cerebellum" control strategy. Power consumption simply won't come down in high-temperature environments.

Vibration is often tied to motor selection. Stepper motors, for instance, jitter at low speeds and lack power at high speeds, easily losing steps during operation. Zhaowei replaced all finger joints with iron-core motors, smoothing out residual jitter through angle error compensation performed once per revolution.

Noise usually hides in gears and electromagnetics. Gear friction creates a dull thud, while electromagnetic harmonics produce a sharp whine. Robots are entering factories, shops, and eventually homes — the barrier of noise cannot be bypassed.

A major source of sway is backlash, or "lost motion," in the industry. There are always tiny gaps between gears and bearings. When a finger reverses direction, it "air walks" for a segment first, causing positioning to float. Over time, wear increases, making the sway more pronounced. This requires high-precision gear grinding and anti-backlash structures to suppress — which happens to be Zhaowei's core business from 25 years of making micro gearboxes.

Furthermore, dexterous hands need enhanced perception capabilities — specifically, skin sensing to achieve precise tactile identification of objects.

"Therefore, making a good dexterous hand requires a group of craftsmen, working solidly, day and night. It is extremely challenging," Chen says. Yet, in his view, if Zhaowei can provide a reliable hand to help the industry navigate the bubble cycle, that is a meaningful endeavor in this era.

Image source: Zhaowei Machinery & Electronics

Cable, Linkage, or Direct Drive: Which Path Is the Answer?

Technologically, there are currently three main paths for dexterous hands: cable-driven, linkage, and direct drive. Direct drive is further divided into linear and rotary motion.

Each path has its pros and cons.

Cable drive mimics the tendons of a human hand. Motors aren't placed on the fingers but moved to the forearm or palm, using steel wires or fiber ropes to pass through the wrist and pull the joints — somewhat like a "marionette".

The benefit is that the fingers themselves are light with low inertia, allowing for fine and flexible movements. DOF can be stacked high, and the motion is closest to a human hand. Tesla's Optimus follows this route.

But the weakness also lies in that cable. Under tension, the rope slowly elongates — known in the industry as "creep." Sliding back and forth inside a sheath creates friction, so when the motor sends a command, the finger lags behind. Achieving accuracy and stability requires massive algorithmic compensation. Over time, the rope wears and loosens, accuracy drops, and troubleshooting or replacing it is a hassle.

Linkage takes the rugged route. Motors are mounted in the palm, using rigid links to push the fingers to bend. It offers good rigidity, strong gripping force, simple structure, and low cost. Precision remains stable over long periods of repetitive work, and maintenance is easy.

The cost, however, is obvious: DOF is hard to increase, fingers tend to be clumsy and heavy, movements are stiffer, and anthropomorphic feel and fine operation take a hit.

Direct drive involves stuffing the motor directly into the finger joint. The motor shaft is rigidly connected to the joint, with no cables or long links. With a short transmission chain, response is fastest, control precision and force feedback are best, wear is low, and maintenance is easy — giving it the edge for fine work.

The price is that the finger is only so big. Motors, reducers, and heat dissipation are all crammed together. Design difficulty is high, and both heat and weight become headaches. Costs are hard to bring down.

Because each route has trade-offs, some companies are pursuing a "hybrid" approach.

One type is product-level "hybrid drive." For example, the Xynova Flex 2 released by Xinuo Future in May claims to be the world's first mass-produced "tendon + direct drive" hybrid dexterous hand, boasting 23 DOF. The Flex 2's logic: high-power-density micro electric cylinders handle rapid, precise responses, while tendons manage smooth movements, dividing labor based on joint needs to capture both the compliance of cable drive and the precision of direct drive.

Joyson Electronics showcased its "Lingxi" dexterous hand at WAIC 2026. The professional version features 20 DOF and even integrates direct drive, tendon, and linkage drive methods within the palm simultaneously.

Another type is company-level technological "fusion." Lingxin Qiaoshou is a typical representative; its product matrix covers the full spectrum of tendon, direct drive, and linkage. Models like the L6, O6, and L20 target different scenarios — a corporate-level hybrid approach that doesn't bet on a single route.

Zhiyuan Robotics' Critical Point, established as a spinoff in January 2026, is also laying out multiple paths. Its OmniHand series has delivered over 8,000 units cumulatively, with its latest valuation reaching $1 billion — a representative case of an OEM entering the fray to build hands in-house.

Zhaowei Machinery & Electronics, for its part, has laid out both direct drive and linkage technological routes and has launched different products for each.

Notably, hybrid drive is the direction heating up the fastest this year. The coexistence of these multiple technological paths also suggests the sector is still in its infancy.

That said, Chen believes the current mainstream route is still linkage. The reason sounds counter-intuitive: the "brain" isn't smart enough yet, and we still need to balance lightweighting with longevity.

But looking long-term, Chen argues that while cable drive is currently limited to specific scenarios due to creep, reliability, lifespan, and maintenance issues, if these problems are solved in the future, cable drive will certainly be a strong contender. The future of the dexterous hand ultimately points toward high DOF and high reliability.

As the market battles over routes, standards are simultaneously taking shape.

According to Chen, the national standard for dexterous hands is set to take shape by the end of this year. This means the industry will transition from "wild growth" to a new stage of regulated competition.

Image source: Zhaowei Machinery & Electronics Prospectus

From Auto to Dexterous Hand: What Is Zhaowei's Edge?

The first source of confidence is technological homology.

Founded in 2001, Zhaowei specializes in micro-transmission and drive systems. Its clients span smart vehicles, consumer electronics, healthcare, industrial manufacturing, and embodied robotics, with customized solutions for different scenarios.

The automotive business is Zhaowei's foundation, currently contributing about 60% of the group's output value. This revenue provides the financial backing for Zhaowei's expansion into embodied intelligence.

In the automotive industry, using legacy businesses to "feed" new ones is standard practice.

More importantly, the technological underpinning of Zhaowei's dexterous hand layout is homologous to its auto business: micro motors, micro transmissions, and micro controls. These are Zhaowei's "root technologies."

"We've been doing this for 25 years. It is also the only difference between us and the rest of the industry: we are the only company capable of full-stack self-development of core modules," Chen asserts.

Zhaowei is currently transferring its existing equipment, personnel, facilities, processes, and workflows from the automotive sector to embodied intelligence, building a full-process guarantee capability from requirement to delivery.

In tactile sensing, for instance, Zhaowei is attempting to integrate perception technologies from the automotive sector into the tactile system. "Vehicles already possess mature sensing systems for temperature and pressure, with natural advantages in reliability and anti-decay. We are extending this advantage into the field of tactile perception to achieve more stable, long-lasting sensing capabilities," Chen says.

The second source of confidence comes from understanding demand.

Chen distills this understanding into three sentences: Real scenarios, real demand, real products. Only when all three "reals" are validated does the business hold water.

The approach is to stand in the OEM's shoes: What scenarios will the robot enter? What kind of hand does that scene require? Then, reverse-engineer the product definition from that demand.

"We've watched this wave of embodied robotics development in China from the beginning — from a few people and a valuation of a few hundred million, to now tens of billions and teams of hundreds. We've seen it all." In his view, this shared journey grants them the authority to define products.

The third source of confidence is letting the product speak.

After all, to survive the bubble cycle and the industrialization cycle, you ultimately have to compete with hard products.

In the dexterous hand field, Zhaowei has built a multi-layered product matrix including the A17, B21, B06, and C06, covering both full-drive and under-actuated technological routes.

Its commercialized high-DOF dexterous hands employ a full direct-drive solution, with finger joints equipped with self-integrated micro-drive modules. This supports independent motion control for each joint, achieving both dexterous operation and stable, reliable performance.

The low-DOF series adopts a linkage under-actuated scheme. Through an ingenious linkage transmission structure, it achieves stable gripping with fewer drive units, effectively reducing system complexity and hardware costs while adapting to high-intensity operational scenarios.

At the core drive module level, a highly integrated mechatronic design is used to achieve high transmission ratio output within a compact space, providing a solid guarantee for the fine control and stable operation of the dexterous hand.

Through continuous iteration, Zhaowei's dexterous hand performance has evolved from the first generation's 17 DOF and 2 kg weight to a current weight of under 400 grams, with a static gripping force of 20 kg and an opening/closing lifespan exceeding 1 million cycles without failure. It can differentially adapt to humanoid robots of varying heights, such as 1.4 meters, 1.7 meters, and 1.8 meters.

Zhaowei's dexterous hands have already entered the production lines of top-tier international manufacturers for collaborative use, though the company has not disclosed client names or order volumes.

To better embrace the wave of large-scale dexterous hand deployment, Zhaowei announced this year that it plans to invest 800 million yuan to build a dexterous hand industrial park, preparing capacity in advance for the ecosystem's growth.

However, beyond the excitement, the financial statements tell a different story.

In the first half of this year, Zhaowei Machinery & Electronics achieved revenue of 816 million yuan, a slight increase of 3.8% year-on-year. Net profit attributable to shareholders was 74.6457 million yuan, a decline of 34.1%. Revenue from dexterous hands has not yet been listed separately in financial reports.

In other words, for Zhaowei, the dexterous hand is currently more like an "option" on a second growth curve, while the automotive foundation continues to provide the "blood transfusion."

Ultimately, Chen has a very specific answer for what this "option" will grow into — using his own dexterous hands to assemble his own dexterous hands on the production line.

That sounds like a tongue twister, but it is the most honest acceptance criterion.

Whether a pair of hands is dexterous and reliable enough? Let it assemble itself once, and you'll know.