Hangjiashuo Robotics
During the 2026 World Artificial Intelligence Conference, ZTE unveiled a lightweight, full-size humanoid robot named "ZTE Xingzai." Standing 1.6 meters tall and weighing a mere 30 kilograms, Xingzai is currently the lightest humanoid product of its size specification in the industry.
The key to its lightweight design lies in the whole-machine tendon-driven lightweight architecture. By centralizing drive components such as motors and reducers in the torso and transmitting power to the joints of the limbs via slender tendons, this architecture significantly reduces the mass and moment of inertia of the arms and legs.
The emergence of Xingzai has once again put the debate over humanoid robot drive technologies in the spotlight. Tendon-driven actuation—a solution that mimics the human tendon traction mechanism—enables robots to "slim down," but what exactly makes it work, and what challenges does it face?
What Exactly Is Tendon-Driven Actuation?
To understand tendon-driven actuation, it helps to look at the human movement mechanism. Our muscles are not located on the fingers or toes; instead, they are concentrated in the torso and the proximal parts of the limbs, transmitting force to the distal joints via long tendons.
The logic of tendon-driven actuation is identical: relocating bulky and heavy actuators, such as motors, to the main body or the proximal ends of the joints, and transmitting power to the end effectors or the distal ends of the limbs via flexible cables (tendons).
Taking ZTE Xingzai as an example, it is precisely this whole-body lightweight architecture—which centralizes drive units in the torso and uses tendon transmission—that allows it to maintain a weight of just over 30 kilograms at a height of 1.6 meters. This is not just a numerical advantage; a lighter body means lower energy consumption, faster response, and higher safety, making it particularly suitable for open public spaces with dense crowds.
Advantages: Lightweight, Flexible, and Agile
The first advantage of tendon-driven actuation is lightweighting. Traditional solutions place motors directly at joints such as fingers, wrists, elbows, and knees, resulting in bulky limbs and high inertia. In contrast, by centralizing the actuators, the tendon-driven approach significantly simplifies the limb structure, reduces mass and moment of inertia, and enables more agile motion responses with lower energy consumption.
The second advantage is high biomimicry. Tendon transmission mimics the distribution of human tendons, enabling more dexterous manipulation closer to that of the human hand, and allowing for more natural gaits such as walking and running. The next-generation dexterous hand of Tesla's Optimus has shifted from relying on torsion springs to tendon-driven transmission. Its structure is closer to human tendons, significantly enhancing finger dexterity and movement naturalness. Currently, the Optimus dexterous hand has achieved 22 degrees of freedom (DOF), which is on par with the human hand.
The third advantage is flexible spatial layout. Tendons themselves possess flexible characteristics, allowing the transmission layout to be adjusted flexibly according to spatial requirements. This is particularly suitable for complex transmission scenarios like dexterous hands, which have limited internal space and require the actuation of multiple DOFs. It also facilitates the centralized arrangement of multiple actuators within the torso, achieving a compact design.
Industry insiders note that tendon transmission has the "highest ceiling" in terms of human-like manipulation and lightweighting. The cable-driven solution can structurally resolve the "impossible tetrahedron" of size, DOF, payload, and speed for dexterous hands. For this very reason, tendon-driven actuation has become the mainstream technical direction for humanoid robot dexterous hands and shows immense potential in whole-body joint actuation.
Shortcomings: Lifespan, Payload, and Assembly
However, tendon-driven actuation is not without its flaws.
Lifespan is its biggest weakness. After prolonged use, tendons are prone to irreversible tensile deformation (known in the industry as "creep") and wear, which directly leads to the dexterous hand "failing to grasp accurately or exert sufficient force," or a decrease in the motion accuracy of whole-body joints. Analyses point out that the tendon-driven solution has a limited lifespan, with high costs and insufficient durability for a single dexterous hand. Some solutions even require tendon replacement every 2,000 hours.
The payload capacity is relatively weak. Although tendon transmission is flexible, its load-bearing capacity is inferior to linkage and gear solutions. In industrial scenarios requiring high gripping force or heavy payloads, the advantages of tendons are not obvious. Consequently, current industry trends are diverging: consumer-grade scenarios adopt the tendon route, high-performance scenarios use gears, and high-payload scenarios rely on linkages.
Assembly and control complexity is high. The routing, tensioning, and calibration of tendons require precise design and algorithmic support. Especially when multiple joints across the body use tendon-driven actuation, the coupling, friction, and nonlinear characteristics among the cables pose extremely high demands on control algorithms. One company pointed out directly that cable-driven technology is "several orders of magnitude more difficult" than direct drive and linkage solutions, and the requirements for control algorithms increase exponentially.
Divergent Paths: Each Showing Its Strengths
The technological paths for the dexterous hands and even the whole-body actuation of current humanoid robots are far from converging, with the three major solutions—tendon (cable-driven), linkage, and direct drive—each having their own supporters.
Pacini Sensing has chosen the linkage solution. Lin Tong, Director of its Data Collection Center, believes that while the tendon solution offers strong explosive power, its lifespan is limited; the direct drive solution provides excellent control but lacks output ceiling; the linkage solution is the optimal choice that balances high performance, long lifespan, and high stability. BrainCo is betting on the "fully direct drive + back-drivable" route, believing it better suits the training needs of imitation learning and reinforcement learning. Meanwhile, companies like Xingji Guangnian and Tesla remain firmly optimistic about the performance ceiling of cable-driven systems.
Cable-driven systems pursue the ultimate upper limit of physical performance, linkages bet on current engineering reliability, and direct drives sacrifice some performance space for the trainability of algorithms.
However, the shortcomings of tendons are not unsolvable. Given the application of new materials such as Ultra-High-Molecular-Weight Polyethylene (UHMWPE) fibers, the strength, wear resistance, and fatigue resistance of tendons are steadily improving. The strength of UHMWPE fibers is 15 times that of high-quality steel, yet their density is only 1/7 that of steel. Benchmark companies like Tesla are also mitigating tendon wear through automatic tensioners and floating designs. The industry generally believes that the disadvantages of tendons in terms of lifespan and payload are expected to be alleviated as materials continue to be optimized.
Returning to ZTE Xingzai—it proves with its 30-kilogram weight that tendon-driven actuation can make humanoid robots lighter, more dexterous, and safer. However, it also reminds us that challenges such as lifespan, payload, and assembly remain to be solved on this path.
The humanoid robot industry is advancing at an unprecedented pace. Technological solutions across various directions have not yet converged, and no single route can dominate the market. Just as the human hand has tendons, robots can also have "tendons." This evolution from "being able to walk" to "knowing how to use hands" may well begin with a slender tendon.
The data is sourced from publicly available information on the internet. Corrections and additions are welcome. This article represents only the author's views and does not constitute any investment advice.
END
To join the [Hangjiashuo Robotics Exchange Group] / for business cooperation, please contact: Rachel Ye Zhiqiu, WeChat ID: hangjiashuo888