Dexcel Robotics, a Shenzhen-based startup, has introduced its first product, the Apex Hand. The company describes it as versatile in freedom of movement, dynamic performance, and other benchmarks. It is also presented as the first dexterous hand in the industry capable of operating a smartphone with one hand, along with performing tasks such as pulse diagnosis and mouse control.
The company was founded by CEO Yang Sicheng, who earned his undergraduate degree from Beihang University and a master’s degree from Tsinghua University. Yang joined Tencent Robotics X in 2018 as one of its earliest core members. Co-founder and CTO Li Wangwei holds a doctorate from the National University of Singapore. Together, the founding team has published nearly 50 papers in leading journals and holds more than 100 patents in dexterous hands and tactile sensing.
In August, Dexcel closed an angel plus round worth an eight-figure RMB sum. The funding round was led by Fibonacci VC, with participation from Xunshang Venture Capital and existing investor Kinzon Capital.
The Apex Hand is a five-finger robotic system designed to replicate the dexterity and functionality of the human hand. It offers 21 degrees of freedom (DoF), enabling movements that cover most everyday tasks. Its dynamic performance, including response speed and acceleration, approaches that of a human hand. Each finger can apply about 2.5 kilograms of force, while the hand as a whole can lift objects weighing up to 30 kilograms.
Built for durability, the Apex Hand can withstand impacts and operate in unfamiliar environments. Its precision is measured at a deviation of 0.1 millimeters, with near-zero backlash in transmission. The hand is also equipped with a proprietary electronic skin that provides tactile sensing through a flexible surface, an alternative to conventional rigid materials.
Its high DoF not only enables single-handed smartphone use but also makes it among the few dexterous hands reported to achieve a perfect score on the Kapandji test, which measures finger-to-finger mobility. Beyond pulse diagnosis and computer use, the Apex Hand can reportedly grip smooth objects in tight spaces, cut with scissors, and manipulate tools. Its design combines rigidity inside with flexibility outside, balancing load-bearing strength with tolerance for error and safe interaction with the environment.
On the tactile side, Dexcel has developed brain-inspired, ultra-high spatiotemporal resolution tactile processing. This enables sub-millisecond communication latency and simultaneous transmission from tens of thousands of tactile points, with refresh rates above 1,000 hertz. Unlike conventional electronic skins that rely on rubber deformation and often compromise durability, Dexcel’s proprietary material is described as both flexible and resilient.
Research into dexterous hands dates back to the 1960s at Stanford University, the Massachusetts Institute of Technology, and other institutions. However, no product has reached widespread commercial adoption. Many academic projects focused on breakthroughs in specific performance areas while leaving gaps in product engineering.
For dexterous hands to become part of general-purpose robots and household applications, they must combine hardware reliability, data collection and modeling, and scalable deployment. That requires comprehensive performance across dexterity, practicality, and robustness. Dexterity depends on DoF, proportionate sizing, and tactile perception. Practicality involves load capacity, responsiveness, and precision. Robustness requires adaptability in unstructured environments, where actions cannot always be preplanned.
According to Yang, 21 DoF is sufficient to replicate nearly all human hand functions. He added that exceeding this number increases costs and complicates control systems without delivering meaningful gains. He also emphasized modular design as key for mass production, ensuring reliability and consistency while avoiding unnecessary complexity in assembly and maintenance.
Even so, Yang noted that deploying dexterous hands in complex environments remains challenging, as hardware stability requires further refinement. He predicted that within two to three years, applications would first emerge in semi-structured settings, such as factory tasks too complex for traditional automation but less variable than household environments.
Dexcel has resisted narrowing its focus to a single use case too early, warning that this could reduce the product to another piece of automation equipment rather than a true dexterous hand.
Dexterous hands have long been a bottleneck for embodied intelligence. Drive mechanisms, whether cable, direct-drive, or linkage, remain unsettled. Elon Musk has said that hand development accounts for nearly half of the engineering effort behind Tesla’s humanoid robot, Optimus. The market still lacks a stable, high-performance dexterous hand, leaving space for startups. Since 2024, breakthroughs in artificial intelligence, particularly reinforcement learning, have shown potential in controlling high-DoF hands. Reinforcement learning allows trial-and-error learning of complex control strategies, addressing the challenge of programming such systems manually. This convergence of technology and market opportunity, Yang said, motivated the founding of Dexcel.
He also noted that the hand is the main interface for robots to interact with the environment and collect data, which can help overcome the problem of data scarcity in embodied intelligence.
Although Dexcel entered the field later than some peers, Yang believes the industry remains in its early stages, with significant room for growth. According to QYResearch, the global market for dexterous hands could exceed USD 5 billion by 2030. The competition, Yang said, will hinge on who can achieve productization first.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Fu Chong for 36Kr.