Entrepreneurship by former core members of Tencent Robotics X from Tsinghua University: The industry's first dexterous hand capable of "using a mobile phone with one hand" has arrived | Emerging new projects

Text | Fu Chong

Editor | Su Jianxun

One-sentence Introduction

Yuansheng Intelligent Robotics (Shenzhen) Co., Ltd. recently released its first product, the Apex Hand. This product emphasizes a "hexagonal" balanced performance in multiple key indicators such as degrees of freedom and dynamic performance. It is the industry's first dexterous hand that can operate a mobile phone with one hand and can also complete hand tasks such as pulse-taking and mouse operation.

Yuansheng Intelligent, founded by Yang Sicheng, recently launched the Apex Hand, the industry's first dexterous hand that can operate a mobile phone with one hand. Photo: Provided by the interviewee

Team Introduction

Yang Sicheng, the company's founder and CEO, graduated from Beihang University with a bachelor's degree and from Tsinghua University with a master's degree. Yang Sicheng joined Tencent Robotics X in 2018 and was one of the earliest core members of the laboratory.

Li Wangwei, the co-founder and CTO of Yuansheng Intelligent, holds a doctoral degree from the National University of Singapore.

The founding team has published nearly 50 top-tier journal papers and holds over 100 patents in the fields of dexterous hands and tactile perception.

Financing Progress

In August 2025, Yuansheng Intelligent completed a multi-million-yuan "Angel +" round of financing, led by Qiancheng Capital, with Xunshang Venture Capital and existing shareholder Kunzhong Capital following.

Products and Business

The company recently released its first five-fingered dexterous hand, the Apex Hand. This product includes six core capabilities:

Degrees of Freedom: 21, covering the degrees of freedom required for human hand work;

Dynamic Performance: Response/acceleration close to that of a human hand;

Load Capacity: The fingertip force of a single finger is about 2.5KG, and the vertical lifting limit is about 30KG;

Robustness: It can withstand a certain degree of accidental impact and can be used stably even in unknown working conditions;

Precision: Precision ≤ 0.1mm, close to zero backlash transmission;

Tactile Perception: It realizes the perception of the physical world through self-developed electronic skin. The flexible surface is more compliant than a purely rigid hand.

The high degrees of freedom of the Apex Hand not only enable it to operate a mobile phone with one hand but also make it one of the few dexterous hands that have achieved a perfect score in the Kapandji test (opposition test) of hand flexibility.

In addition to pulse-taking and mouse operation, the Apex Hand can also perform difficult tasks such as stably grasping smooth objects in narrow spaces and using human tools such as scissors.

The fingertip force of a single finger of the Apex Hand is about 2.5KG, and it can pierce a 1cm-thick wooden board. Photo: Provided by the interviewee

Core Barriers

Yuansheng Intelligent's greatest technical advantage lies in its full-stack development experience from scratch in both dexterous hands and tactile sensors. The structure with a rigid interior and a flexible exterior ensures the product's load capacity, while also providing fault tolerance for grasping and safety when interacting with the environment.

In terms of tactile perception, Yuansheng Intelligent has pioneered a brain-inspired tactile processing technology with ultra-high spatio-temporal resolution. This technology has a sub-millisecond communication delay and supports the simultaneous transmission of tens of thousands of tactile points. At the same time, it has a high integration level and a refresh frequency > 1000Hz.

In Yuansheng Intelligent's solution, self-developed electronic skin is used. Currently, common electronic skins on the market sense touch through the deformation of rubber materials. However, it is difficult to achieve both the softness and durability of rubber. The self-developed material of Yuansheng Intelligent is flexible, avoiding the above problems.

Product photo of the Apex Hand. Photo: Provided by the interviewee

Founder's Thoughts

Since the 1960s and 1970s, top universities such as Stanford and MIT have been involved in the research of five-fingered dexterous hands. However, there has always been a lack of truly marketable products on the market.

The reason is that previous research on dexterous hands was mainly led by scientific research institutions. The academic community focused more on single-point breakthroughs in performance, lacked the motivation for productization, and had relatively limited engineering capabilities.

To ultimately become part of general-purpose robots and enter scenarios such as households, dexterous hands must ensure hardware stability, data collection and utilization, and model capabilities.

A product that can support model algorithm research and large-scale implementation needs to have comprehensive and powerful performance and perform well in terms of dexterity, practicality, and robustness (reliability).

Specifically, in terms of dexterity, it needs to have both degrees of freedom, an appropriate overall size, and tactile perception ability. Practicality includes load capacity, response speed, and precision.

These two aspects refer to the performance in a structured environment, that is, how stable the performance of the dexterous hand is under fixed conditions or within a fixed time. All automated robots need to consider these characteristics.

Robustness refers to the adaptability in an unstructured environment (such as impacts and disturbances). After all, not all actions are pre-planned. As an embodied intelligent system, it must have the ability to handle unstructured problems.

Although there is still a debate about the exact number of degrees of freedom of a dexterous hand, I believe that if 21 degrees of freedom are reasonably allocated, they are sufficient to reproduce most of the capabilities of a human hand.

I previously studied many anatomical monographs and found that the reason for the debate about the number of degrees of freedom is that there are several areas on the hand where the degrees of freedom are not clear.

However, the number of most visible effective degrees of freedom is clear.

The obvious degrees of freedom of the active space of a human hand are about 22 - 5 degrees of freedom for the thumb, 4 degrees of freedom for each of the four fingers, and there is also 1 degree of freedom between the metacarpal bone of the little finger and the carpal bone, but the active space is already very small.

A hand with 21 degrees of freedom can basically reproduce all the operations of a human hand, with a maximum of 22.

Therefore, considering the cost and benefit, having more than 21 degrees of freedom doesn't make much sense as it will increase the number of motors, and an increase in motors will also bring control difficulties.

Only by adhering to modular design can a complex mechatronic product like a dexterous hand move towards mass production more quickly.

For a complex system like a dexterous hand, it must be divided into relatively independent modules to ensure the consistency and reliability of mass production. If dozens of degrees of freedom are mechanically "bundled together," problems will arise in mass production and maintenance.

This year, it is still very difficult to directly introduce dexterous hands into complex scenarios, and it will take time to improve hardware stability. The objective law of the development of dexterous hands is to solve the problems of hardware performance, data bottlenecks, and model capabilities. Large-scale implementation is only possible when these three aspects develop in synergy to a certain extent.

In the short to medium term of 2 to 3 years, it will first be applied in semi-structured scenarios. For example, tasks in factories that are difficult to complete with traditional automation but are not as unstructured as those in households.

Yuansheng Intelligent does not want to be prematurely confined to a single scenario, otherwise, it is easy to become an automated device, which deviates from the original intention of developing dexterous hands.

The Apex Hand catches a falling pen. Photo: Provided by the interviewee

Notes from the "Intelligent Emergence" Interview

Dexterous hands have long been the main bottleneck for the implementation of embodied intelligence. Multiple drive schemes such as cable-driven, direct-driven, and linkage, as well as the design of degrees of freedom, are still in the stage of technological divergence. Elon Musk also pointed out that the development of the hand accounts for nearly half of the overall engineering volume of Tesla's Optimus humanoid robot.

There is still a lack of high-performance products that can be stably delivered in the market, which brings opportunities for dexterous hand startups.

Since 2024, the capabilities of AI models have continuously improved, and the potential of reinforcement learning to control high-degree-of-freedom dexterous hands has emerged. This is the reason why Yang Sicheng believes that the timing for entrepreneurship is ripe: the technology and business of dexterous hands have reached a matching point.

Reinforcement learning (RL) provides a paradigm for automatically learning complex control strategies through "trial and error," thus solving the core pain point of high-degree-of-freedom dexterous hands, which are "difficult to control through manual programming." The industry has begun to regard dexterous hands not only as research projects in the laboratory but also hopes to productize them so that a large number of robots can have dexterous hand operation capabilities.

At the same time, data is generally considered to be the bottleneck in the research and development of embodied intelligence, and the hand is the core data entry point for interacting with the environment. Before "doing real work," robots can also collect data through tactile hands and learn about the physical world.

Yuansheng Intelligent was founded at the end of 2024. Compared with some earlier-established dexterous hand companies that have already received orders, this timing seems a bit "late."

However, Yang Sicheng believes that there is still a lot of room for improvement in the technical level of the entire industry, and it is still a good time for dexterous hand startups.

QYResearch predicts that the market size of dexterous hands will exceed $5 billion in 2030. Facing the huge market prospects, dexterous hand companies are competing to achieve productization breakthroughs first.