Company News

Rotary joints versus linear joints: who will ultimately win? This question cannot be simply answered by declaring one will eliminate the other. The final scenario will likely be one of mutual prosperity, differing only in their respective proportions and flourishing in different domains. Today, I will provide a detailed explanation of the future development trends of rotary and linear joints and the balanced state that will emerge after their competition.

As you can see, companies like Unitree, DeepRobotics, or Zhongqing all use rotary joints (they are all very, very excellent leading players). Few manufacturers are using Tesla's linear actuators. The only domestic company using a linear solution that can actually walk properly is XPeng (note the condition: properly walkable ). This overwhelming preference actually stems from the maturity advantage of rotary motors. Rotary motors were first used in industrial robots over 20 years ago. Therefore, RV reducers, harmonic reducers, planetary gear reducers, etc., are relatively mature technologies—simply put, they're ready to use out of the box. If a robot's entire body joints use planetary reducers, it can start turning with a simple power connection. As long as it's built in a human-like form, can it be called a humanoid robot? Any humanoid robot starts with a valuation of 1 billion yuan.

This is understandable because companies need to develop, secure financing, tell stories, and show people things that move and look human-like. Therefore, any company that can make rotary motors, one by one, claims to be part of the humanoid robot industry chain. This is why our embodied robot industry has developed particularly rapidly. Of course, when many companies enter the field, a selection process occurs, and only those that do well and innovate will survive. Looking back now, relatively early humanoid robot companies have already entered the first wave of bankruptcies. Does this remind you of the similar historical trajectory in the new energy vehicle sector, where some players were eliminated after fierce competition? I believe more cases will emerge this year and next. Because everyone's solutions are too similar, too close, too homogeneous. Many embodied robot本体 companies still outsource components because the investment is quite large—it is, after all, a capital-intensive, heavy-industry manufacturing sector requiring significant equipment.

You buy, he buys, I also buy. In the end, the assembled humanoid isn't made by themselves, so what are they competing on? That's right—the brain. What we call algorithms or, more fashionably, AI. Only with powerful computing algorithms, massive databases, models, and proprietary training know-how can we train good robots. Its hands won't shake, its legs won't keep stepping to find balance, and it won't fall over when walking. OK, at this point, everyone is actually competing on software. After this round, some embodied companies should be eliminated. Because if their algorithms aren't powerful enough, they can't control the robot's limb movements. They can't even walk upright normally or can only walk for a short time before failing. Naturally, such robots won't sell. If there's no breakthrough, the company is quickly heading towards collapse.
Next, after overcoming numerous challenges and passing the algorithmic hurdle, we arrive at the hardware component. Good hardware is actually extremely important. Imagine having Lin Dan's brain but not his hands. Does that make it easier to understand? Your algorithm might be incredibly powerful, capable of directing with pinpoint accuracy. But if your hands don't obey your commands—if your hardware fails to execute your brain's instructions precisely—then the ball will never land exactly where your brain intended it to. This is why powerful hardware is absolutely necessary.

So what constitutes good hardware?

Good hardware must have extremely high mechanical processing precision, excellent durability, wear resistance, and product lifespan guarantees. The motor part must have powerful and stable performance, capable of stable output under various load conditions. It requires reliable drive control that can perform functions such as recognition, compensation, and judgment during operation. Only such an all-around performer can match a good algorithm and train a good robot. Otherwise, no matter how good your AI is, training a pile of garbage hardware is no different from training a speaker.

Alright, so the question is gradually becoming clear: between rotary joints and linear joints, who is the king? This questionmust be examined from its essence. As mentioned earlier, most rotary joints are gear structures, and the biggest problem with gears is backlash. This is determined by their mechanical structure—zero backlash equals seizure (we must be scientific, don't rush), and it's very difficult to eliminate. Of course, you might say we have motors, drive controls, sensors, etc., to solve this. OK, indeed, we can solve part of the problem. But the biggest issue lies in the fundamental structure itself. Compensating for this is never the optimal solution. Why compensate when we could avoid the problem from the beginning?Looking at linear actuators, lead screws themselves are high-precision products that can achieve high-precision grinding. Therefore, they have a natural advantage like a straight-A student. Combined with motors and other peripheral components (which are quite mature), their chances of success are actually greater. It could even be said that rotary joints are the "catalyst" for linear joints.
OK, the above analysis is from the perspectives of hardware precision and lifespan. From the perspective of functionality, neither can do without the other. Humans have muscles and bones. When you do pull-ups or deadlifts, you don't rely on your elbows and knees to generate force, right? Gym-goers know that trainers often tell you to consciously engage your thigh muscles during exercises, not your knees, to avoid knee injuries. Similarly, you shouldn't do parkour (this should make sense). In other words, rotary joints shouldn't bear heavy loads—load-bearing should be handled by muscles—linear joints. What happens if you frequently subject rotary joints to heavy loads? Knee effusion . And what about rotary motors? You can imagine. The correct answer is: rotary joints change the direction of movement, while linear joints apply tremendous force in that direction with precision control, guided by AI commands from the brain to achieve pinpoint accuracy. Doesn't this sound like a living person standing in front of you? What can't such a robot achieve?