A team of Chinese researchers has unveiled what they describe as the world’s first robotic hand capable of combining high-resolution touch sensing across its entire surface with full human-like range of motion, marking a significant step forward in embodied artificial intelligence and robotic manipulation.
A Hand Built to Feel, Not Just Grip
The system, called the F-TAC Hand — short for Full-hand TACtile-embedded Biomimetic Hand — was developed through a collaboration involving researchers from Peking University, the Beijing Institute for General Artificial Intelligence, and other partner institutions. Unlike most robotic hands, which rely heavily on external cameras or sensors limited to the fingertips, the F-TAC Hand is engineered to sense contact, pressure, and slippage across nearly its entire surface, much like human skin does.
The hand’s core innovation is its comprehensive tactile sensing capability, offering high-resolution coverage with roughly 0.1 millimeter spatial resolution across about 70 percent of the hand’s surface. This is achieved by embedding 17 vision-based tactile sensors across six different configurations on the hand, with the sensor housings doing double duty as both sensing components and structural parts of the hand itself — a design choice that preserves flexibility while maximizing coverage.
Mimicking the Human Sense of Touch
Explaining the motivation behind the design, lead researcher Zhao Zihang, a PhD candidate at Peking University, noted that human touch relies on a dense network of skin sensors paired with specialized neural processing that interprets that input in real time. The team aimed to replicate that same architecture mechanically, rather than treating touch as an afterthought layered onto a purely mechanical gripper.
Replicating this proved to be no small engineering task. The human hand contains 27 bones and 34 muscles, giving it roughly 24 degrees of freedom — a level of complexity that has long made dexterous robotic hands difficult to design and even harder to control.
Smarter Grasping Through AI
Beyond the physical sensor array, the research team also developed an algorithm capable of generating human-like grasping strategies. Co-lead author Li Yuyang, of the Beijing Institute for General Artificial Intelligence, explained that the probability-based model allows the hand to produce natural, human-resembling grasping motions covering 19 common grasp patterns used in everyday manipulation.
In testing, the hand demonstrated a high Kapandji score — a standard clinical measure of thumb mobility and opposition — and was able to perform all 33 recognized human grasp types. Across 600 real-world trials, researchers found that the tactile-equipped hand significantly outperformed non-tactile alternatives, particularly in noisy, unpredictable real-world conditions where objects shift or behave unexpectedly during handling.
Part of a Broader Race in Robotic Touch
The F-TAC Hand arrives amid intensifying global competition to solve the “sense of touch” problem in robotics, long considered one of the final major hurdles to truly dexterous humanoid robots. Chinese firms have been especially active in this space: brain-computer interface company Qiangnao recently unveiled its Revo 3 hand, featuring 21 degrees of freedom and full-palm tactile sensing with a grasping force of 70 newtons, while Xiaomi has integrated full-palm tactile sensing spanning roughly 8,200 square millimeters into its CyberOne humanoid platform.
Researchers say the broader significance of full-hand tactile sensing lies in giving robots the ability to manipulate objects reliably even when vision is obstructed or when fine force control is required — such as during delicate assembly work or handling fragile items — rather than depending solely on cameras to infer what their “hands” are doing.
What Comes Next
The development is expected to accelerate research into tactile-embodied AI systems, an area seen as critical for humanoid robots that must eventually operate in unstructured, real-world environments such as homes, warehouses, and hospitals. By solving the long-standing trade-off between sensing coverage and mechanical dexterity, the research team behind F-TAC Hand says its work opens the door to more sophisticated studies of how robots can learn to manipulate objects the way humans do — through touch as much as sight.