NVIDIA selected Unitree, but replaced Unitree's dexterous hands

Unitree has been selected by NVIDIA, but NVIDIA didn't choose Unitree's robotic hands.

While Unitree Technology passed the IPO review on the STAR Market, its H2 Plus robot was hand - picked by NVIDIA and became the body of the NVIDIA Isaac GR00T reference humanoid robot released by NVIDIA.

This powerful alliance has set the embodied intelligence circle and the AI circle ablaze. After all, NVIDIA's status in the industry needs no further elaboration, and Unitree has made it into the first - tier of global humanoid robots with its robots' excellent motion control capabilities.

What really comes as a surprise is that this reference robot doesn't use Unitree's own dexterous hands, but instead uses Sharpa Wave.

Sharpa is a dexterous hand company that has been established for only two years. Its headquarters is in Singapore, and its core team comes from the dexterous hand team formed by Hesai Technology in 2024. It doesn't have as high a profile as Unitree, nor does it frequently appear in financing news like Lingxinqiaoshou and Lingjiedian.

During this year's Spring Festival Gala, Sharpa's dexterous hands also appeared in the program "My Most Unforgettable Tonight" performed by Shen Teng and Ma Li. However, in the niche dexterous hand market, it remains quite low - key.

Now, it stands at NVIDIA's embodied intelligence table together with Unitree.

One is responsible for the body, and the other is responsible for the hands.

But the question is, what makes a two - year - old dexterous hand company catch NVIDIA's eye?

Unitree's hands lost this time

Actually, Unitree does have its own dexterous hands.

From its public product line, Unitree has launched the Dex series of dexterous hands, covering both three - finger and five - finger solutions, and their performance is not bad.

But what NVIDIA wants this time may not just be "a hand that can be installed on a humanoid robot".

From a technical perspective, Sharpa's dexterous hands do have their advantages.

In 2024, Hesai Technology formed a dexterous hand team, which is the predecessor of Sharpa. From Hesai Technology's lidar to dexterous hands, Sharpa seems to be "cross - border".

In many people's minds, autonomous driving and robotics are two completely different industries.

But for a company like Hesai, there is actually some common technical accumulation between the two.

Autonomous driving requires machines to perceive, understand, and make decisions in real - time in complex environments, which places high demands on sensor accuracy, system reliability, and engineering capabilities. When robots enter the real world to perform tasks, they also face similar problems.

The difference is that autonomous driving solves the problem of "how a car sees the world", while dexterous hands solve the problem of "how a robot interacts with the world".

That's why Sharpa transferred its past capabilities to another field. Soon after its establishment, Sharpa launched its flagship product, the Sharpa Wave dexterous hand.

In the field of dexterous hands, there has long been a contradiction: it's often difficult to achieve both high performance and mass production.

Some products use under - actuated solutions, controlling multiple joints with a small number of motors, which have relatively low costs but limited flexibility and control accuracy. Other products aim for capabilities close to those of human hands, with complex structures and high prices, making large - scale implementation difficult.

Sharpa has chosen a middle - ground approach.

The Sharpa Wave has 22 active degrees of freedom, and its overall size is close to a 1:1 ratio with a human palm. To achieve higher control accuracy, it uses a direct - drive transmission architecture to improve joint response speed and motion control capabilities.

But more importantly than the number of degrees of freedom is Sharpa's investment in the tactile system.

In the humanoid robot industry, an increasingly clear trend is that vision alone can no longer meet the needs of complex operations.

Sharpa has developed a tactile system called Dynamic Tactile Array (DTA). It integrates a miniature camera and more than 1000 tactile sensing units inside each fingertip, allowing the robot to sense pressure changes, recognize textures, sliding, and contact states, and obtain a "tactile feedback" similar to that of a human fingertip.

According to the data disclosed by Sharpa, its tactile sensing accuracy can reach the order of 0.005N, the refresh rate reaches 180Hz, the control frequency of the whole hand reaches 500Hz, and the output force of a single fingertip exceeds 20N.

All these parameters point to the same goal: to enable the robot to truly handle real - world objects.

This is not exactly the same direction as Unitree's own dexterous hands.

Unitree's Dex series is more in line with its own overall machine system. Whether it's the three - finger or five - finger solution, the focus is on enabling the robot to complete grasping and operations within its own body, motion control, and development ecosystem.

This means that it's not that Unitree's hands are not good; it's just that for NVIDIA's reference robot, Sharpa's hands are a better fit.

So, the value of Sharpa Wave lies in turning the act of "interacting with the world" into a data entry point that the robot can sense, feedback, and train with.

But parameters are just parameters. To prove that the dexterous hands really have the ability to "interact with the world", it still needs to be implemented in specific tasks.

Sharpa has attracted market attention precisely because it has turned these parameters into a series of operation demonstrations that have impressed its peers.

Sexy dexterous hands, dealing cards online

At the product exhibition of IROS 2025 (International Conference on Intelligent Robots and Systems), a demonstration by Sharpa impressed the industry: A Sharpa dexterous hand drew a card from a stack of playing cards held by another hand and placed it on the table.

What makes it impressive is that the action of dealing playing cards places extremely high demands on the force - control accuracy of the dexterous hand and the ability to predict the sliding of the playing cards.

Moreover, Sharpa has also released a series of demonstration videos: autonomously peeling eggshells, peeling apples, dealing playing cards, folding paper windmills, and even assembling a computer host, including precisely inserting a graphics card and tightening the fixing screws.

These tasks may seem like just interesting demos, but for the robot industry, they represent completely different levels of technical difficulty. Because it's not difficult to grasp an object; the real challenge is to control the contact process.

A robot can easily pick up an egg, but it may not know when to increase or decrease the force. It can recognize a playing card, but it's difficult to ensure that the paper won't slip or deform.

Many of the capabilities demonstrated by Sharpa essentially point to the same question: Can a robot adjust its actions through tactile feedback like a human?

In this process, the DTA tactile system starts to play its role.

When the robot touches an object, the fingertips can sense pressure changes, friction states, and the sliding trend of the object in real - time and feed this information back to the control system for dynamic adjustment. Soft or fragile objects such as eggs, paper, and fruits best demonstrate the value of this system.

Meanwhile, Sharpa is not satisfied with just being a dexterous hand supplier. In 2026, the company officially launched its first full - body humanoid robot, Sharpa North.

At CES 2026, North completed demonstrations such as playing table tennis, taking photos with a selfie stick, and dealing playing cards. The most representative one was an autonomous paper windmill assembly task with more than 30 steps.

From identifying parts, grasping materials, to folding, splicing, and finally completing the assembly, the whole process lasted several minutes and involved a large number of two - hand collaborative operations and continuous motion planning. This means that Sharpa's robot has the possibility of completing long - sequence, multi - step tasks.

From dexterous hands to humanoid robots, from a hardware supplier to a full - stack system developer, Sharpa's path is becoming increasingly clear:

It is not satisfied with being just a component of a robot. What it really wants to do is to become part of the next - generation embodied intelligence platform.

And the platform is precisely the keyword for NVIDIA's reference robot this time.

If the previous demonstrations proved that Sharpa's hands can perform complex operations, then the more crucial question is: What can such hands bring to NVIDIA?

Peers are chasing and causing a stir

For Sharpa, being selected by NVIDIA is undoubtedly a landmark moment.

But the more crucial point is that in this reference robot, Sharpa occupies a rather critical position.

Because when NVIDIA creates the "reference robot", it wants to build a reusable development base for the embodied intelligence industry, allowing developers, research institutions, and robot companies to conduct training, verification, and development around this solution.

In this solution, Unitree provides the body. With its motion control capabilities, Unitree solves the problems of how the robot stands up, walks, and moves.

What Sharpa supplements is how the robot can really reach out and do work after approaching an object.

But for NVIDIA, this is not all.

More importantly, Sharpa's products have been incorporated into Isaac Lab, which is the most core open - source simulation training framework in NVIDIA's robot system.

In the tele - operation process, a human operator can control the 22 - degree - of - freedom dexterous hand through a data glove, mapping hand movements to robot joint movements in real - time. After these movements are recorded, they become data for imitation learning and strategy training, and can be used as samples for subsequent training, reuse, and expansion.

That's why Sharpa has obtained more than just an ordinary hardware position. It has integrated into NVIDIA's entire process from tele - operation data collection, simulation training, strategy evaluation to real - world deployment, which is the core value of its cooperation with NVIDIA.

Of course, being selected by NVIDIA doesn't mean that Sharpa has already secured victory, because the dexterous hand market is changing too fast.

In the past year or so, capital has been chasing this "hand" forward: Companies such as Lingxinqiaoshou, Lingjiedian, Yingshi Robot, Aoyi Technology, and Pacini Sensing are all accelerating their iterations around high degrees of freedom, tactile feedback, force - control accuracy, and mass - production capabilities.

Some are competing in terms of financing speed and product implementation.

For example, Lingxinqiaoshou has completed multiple rounds of financing since 2025. After the Series B+ round in 2026, its valuation was reported to have reached $3 billion, and the target valuation for the next round of financing is expected to reach $6 billion. Yingshi Robot also completed hundreds of millions of yuan in Series C1 and C2 rounds of financing in 2026, continuing to bet on the research and development of dexterous operation technology, innovation of core components, and product delivery capabilities.

Some are competing in terms of tactile technology.

Pacini Sensing's product line covers multi - dimensional tactile sensors, the DexH series of tactile dexterous hands, and the humanoid robot TORA. It completed a Series B financing of over 1 billion yuan in March 2026.

Aoyi Technology's new - generation dexterous hand has highlighted selling points such as high - density dot - matrix tactile sensors and pressure - sensing capabilities from 0.1N to 25N.

This means that today it's 22 degrees of freedom, the DTA tactile system, and NVIDIA's reference design, but tomorrow there may be new hands catching up with lower costs, higher stability, or stronger data closed - loops.

But at least for now, NVIDIA has made its judgment in its own way:

As the industry shifts from "who can make the robot walk" to "who can make the robot do work", a smart enough pair of hands has become as important as the legs.

This article is from the WeChat official account "Blue Word Plan", author: Chester, published by 36Kr with authorization.