About this Resource
<p><i><span style="font-size: 16px;">Look at Wuji Hand 2, its 20-DOF robotic hand, human-motion data glove, and why better hands could unlock the next generation of embodied AI.</span></i></p><h2><br></h2><h2>The Most Important Robot at China | ICRA 2026</h2><p><span style="font-size: 16px;">Humanoid robots are becoming more advanced every year. They can walk, balance, run, dance and increasingly use AI to understand their surroundings. But there is still one major problem, <b>their hands</b>. A robot may have an extremely powerful AI model, but if its hands cannot safely and precisely interact with everyday objects, much of that intelligence remains trapped inside a computer. This is why the Wuji Hand 2, developed by Shenzhen-based Wuji Technology, attracted so much attention at ICRA 2026 in Vienna.</span></p><h2><br></h2><h2>The Problem With Robot Hands</h2><p><span style="font-size: 16px;">Human hands are incredibly difficult to reproduce. We use our fingers to control tiny amounts of force, manipulate flexible objects and constantly adjust our grip without consciously thinking about every movement. Robots have struggled to achieve the same level of dexterity. Some advanced robotic hands have many independently controlled joints, but they can cost well over $100,000. Others are cheaper and easier to manufacture but sacrifice the precise movement needed for delicate tasks. This creates a major problem for embodied AI, AI that operates in the physical world. A robot might be able to understand an instruction such as “take the money out of the wallet,” but actually doing it requires precise movements, force control and tactile feedback.</span></p><h2><br></h2><h2>Wuji Took a Different Approach</h2><p><span style="font-size: 16px;">Wuji Technology originally focused on building compact motors and actuators for robotics. Instead of using a traditional tendon-and-cable system, the company placed small motors and rigid gears directly inside the fingers. This is important because cables can stretch and create mechanical uncertainty. A direct-drive design can make the relationship between a computer's instructions and the finger's movement more predictable. The first Wuji Hand already offered 20 degrees of freedom while weighing less than 600 grams. It was also designed to be considerably cheaper than research-grade robotic hands that traditionally cost tens or hundreds of thousands of dollars. The second generation, <b>Wuji Hand 2</b>, takes this approach further.</span></p><h2><br></h2><h2>A Human-Sized, Highly Dexterous Hand</h2><p><span style="font-size: 16px;">Wuji Hand 2 has 20 actively controlled joints and weighs around 580 grams. Its dimensions are designed to closely match a human hand. That matters because robots with human-sized hands can interact with objects and environments designed for humans without requiring everything around them to be redesigned. The hand can also continuously monitor and adjust the force applied by its fingers at a rate of up to <b>1,000 times per second</b>. In simple terms, this allows the hand to make extremely rapid adjustments when holding an object. That type of control is important for tasks ranging from holding fragile objects to applying enough force to manipulate something resistant.</span></p><h2><br></h2><h2>The Bigger Breakthrough May Be the Glove</h2><p><span style="font-size: 16px;">The most important part of Wuji's system may not actually be the robotic hand. Wuji also demonstrated a <b>data-collection glove</b> equipped with sensors that track the movement of the human hand. The glove and Wuji Hand 2 use the same 20-degree-of-freedom structure. This creates a direct connection between human movement data and robotic movement. For AI researchers, this could be extremely valuable. Robotics has a major data problem. Large AI models can learn from enormous amounts of text, images and video available online. Robots do not have an equivalent internet-sized database showing exactly how humans physically manipulate objects. That data has to be collected.</span></p><p><span style="font-size: 16px;"><br></span></p><p><span style="font-size: 16px;">Traditionally, collecting robotic training data can involve expensive robots, specialized equipment and human operators controlling machines remotely. Wuji's approach suggests another possibility, <b>capture how humans naturally perform tasks and use that movement data to train robots with matching physical structures.</b></span></p><h2><br></h2><h2>Why This Matters for Embodied AI</h2><p><span style="font-size: 16px;">Imagine a factory worker wearing the data glove while performing normal tasks. Instead of simply recording a video, the system could capture the detailed movement of the worker's fingers and joints. That information could potentially become training data for an AI model. Because the robotic hand has matching joint movements, the gap between the human demonstration and the robot executing the task becomes much smaller. This is a major idea in embodied AI, teaching machines physical skills by learning from human behaviour. The long-term goal is to create a cycle where <b>h</b><b>umans perform tasks, sensors collect movement data, AI learns the skill, robots perform the task and more data is collected</b>. If this can scale, it could become an important part of building more capable general-purpose robots.</span></p><h2><br></h2><h2>The ICRA 2026 Demonstration</h2><p><span style="font-size: 16px;">At ICRA 2026 in Vienna, Wuji Hand 2 was demonstrated performing real-world manipulation rather than simply appearing in a promotional video. One of the notable demonstrations involved manipulating a <b>Rubik's Cube</b>, requiring multiple fingers to coordinate their movements, maintain contact and adjust force in real time. The demonstration does not prove that Wuji has solved every problem in robotics. But it highlights why dexterous hands are becoming one of the most important pieces of the humanoid robotics puzzle.</span></p><h2><br></h2><h2>Why the Hand Could Be More Important Than the Robot</h2><p><span style="font-size: 16px;">Today's AI systems can write code, generate images, reason through problems and control increasingly sophisticated machines. But physical interaction is much harder. Tasks such as pulling a banknote from a wallet, opening packaging, peeling a banana or handling delicate objects require a level of touch and precision that robots still struggle to reproduce. The problem isn't necessarily that the AI doesn't understand what it should do. The physical hardware may simply be incapable of doing it reliably. That makes the hand a crucial bridge between <b>AI intelligence and physical action</b>.</span></p><h2><br></h2><h2>What Wuji Could Change</h2><p><span style="font-size: 16px;">Wuji's most interesting contribution may therefore be bigger than a single robotic hand. By combining an affordable, highly dexterous hand with a matching human-motion data glove, the company is attempting to address two major robotics challenges at once:</span></p><p></p><ul><li><span style="font-size: 16px;"><b>Better hardware</b> for precise physical manipulation
</span></li><li><span style="font-size: 16px;"><b>More useful data</b> for training embodied AI</span></li></ul><p></p><p><span style="font-size: 16px;"><br></span></p><p><span style="font-size: 16px;">The approach could also make advanced robotic research more accessible. If high-dexterity hands become significantly cheaper and easier to deploy, more universities, startups and AI laboratories could experiment with physical AI. Wuji is not guaranteed to solve the robotics industry's biggest problems. Other companies are developing competing technologies, and humanoid robotics remains a rapidly changing field. But the Wuji Hand 2 demonstrates an important shift in thinking. The future of robotics may not depend only on building bigger AI models or more powerful humanoid bodies. <b>The ability to give those AI systems human-level physical control could be just as important. </b>And that is what made Wuji Hand 2 one of the most interesting pieces of robotics hardware showcased at ICRA 2026.</span></p>