"Ruiman" Zheng Suibing: The rush of robots into factories to work is led astray by Musk | Emergence 36 People

Welcome to follow "Intelligent Emergence"

Written by | Wang Fangyu

Edited by | Su Jianxun

In 2024, the popularity of humanoid robots has soared, with capital and talents flocking in, but so far, no whole machine manufacturer has announced mass production, and the commercialization progress is slow.

Some industry insiders commented that "The commercialization progress of more than 80 domestic humanoid robot startups is almost zero."

In contrast, the "shovel sellers" of humanoid robot startups are ushering in a dividend period of development - key component manufacturers such as motors, mechanical arms, and dexterous hands are becoming the first beneficiaries of the embodied intelligence wave.

In the mechanical arm and joint sector, there is a manufacturer - "Realmotion" whose products are quite well-known in the industry. Among the more than 100 embodied robot companies exhibited at the World Robot Conference in 2024, 26 are using Realmotion's mechanical arm products.

According to Realmotion, it has a significant market share advantage in the domestic embodied robot field, The amount of orders on hand this year has exceeded hundreds of millions of yuan.

Being able to obtain a high market share is closely related to the technical route that Realmotion has adhered to for a long time - Zheng Suibing, the founder of Realmotion, called the title of the project plan when he started his business in 2018 "Let Robots Enter Life". Different from the large and heavy industrial mechanical arms, Realmotion makes lighter and smaller general-purpose products.

"This was a risky choice at that time." Zheng Suibing recalled to 36Kr. At that time, specialized robots were popular in the industry, and the mainstream view was that "general-purpose is useless", and Realmotion's technical route was questioned.

Zheng Suibing endured the pressure and concentrated on R & D. It was not until three years later that he presented the first lightweight mechanical arm product, and it made its first appearance at the World Robot Conference in September 2021. This product has the advantages of high power density, high precision and high reliability, and quickly opened the market.

Realmotion's mechanical arm products         Picture source: Authorized by the enterprise

The more important turning point occurred in 2023.

The AI large model has set off an upsurge of embodied intelligence, and both new and old robot companies have turned to general-purpose intelligent robots. This is highly consistent with the route that Realmotion has adhered to since its establishment, "Let Robots Enter Life".

With the long-term accumulation in the operational ability and generalization ability of the mechanical arm in the past few years, Realmotion's mechanical arm products have met the needs of embodied intelligence enterprises and have been widely welcomed by the market. Many humanoid robot enterprises use its products for large-scale training.

The reason why Realmotion can catch the trend lies in the logical judgment that Zheng Suibing insisted on at the beginning of his entrepreneurship. He firmly believes that in the context of the aging era, the demand for robots to enter life is certain.

In the current embodied intelligence wave, Zheng Suibing has his own judgments on the trends of many scenarios and directions.

The mainstream view believes that it will take five to ten years for humanoid robots to enter the C-end family scene, but he believes that within the next 2 - 3 years, embodied robots can enter the family like floor-sweeping robots.

Regarding the large-scale commercialization bottleneck of embodied robots, he believes that it is not in software and hardware, but that enterprises dare not apply them. It should accumulate data and optimize iterations in applications like autonomous driving.

Regarding the enthusiasm of humanoid robots entering the factory, he believes that Chinese robot enterprises blindly following Tesla are led astray. The first landing scene of humanoid robots is probably not automobile manufacturing, and bipedal is not the best form.

However, no matter whether embodied robots enter life quickly or slowly, in the process of truly moving towards general-purpose intelligent robots, "shovel seller" Realmotion will be a difficult role to bypass in the industrial chain.

The following is a conversation between "Intelligent Emergence" and Zheng Suibing, the founder of Realmotion. The content is slightly edited:

Part 1: It took 3 years and 9 months to develop the first product

Intelligent Emergence: When Realmotion started in 2018, why did it choose the direction of ultra-lightweight humanoid mechanical arms?

Zheng Suibing: Our vision is to make robots with operational capabilities truly integrate into daily life. Therefore, we have anchored this goal and established the company in January 2018, focusing on solving the generalization ability, operational ability and cost problems of robots. In the initial stage of entrepreneurship from 2018 to 2019, we confirmed the direction of ultra-lightweight humanoid mechanical arms.

The reason for choosing the "humanoid" design is that we hope that robots can adapt to the living and working environment of humans. The consideration of "ultra-lightweight" comes from the fact that humans can easily carry their own arms to move and work, which means that if the mechanical arm is light enough, it can be placed wherever it is needed.

Our ultra-lightweight humanoid mechanical arms have subsequently been widely used in the market, covering various fields such as new catering, new retail, big health, home services, intelligent inspection, scientific research and education, which also validates that our initial direction is correct.

We also emphasized "four" times "lighter and smaller", which means that compared with the traditional collaborative robot, under the same output and load capacity, the weight and volume of our mechanical arm is only one-fourth of that of the collaborative mechanical arm. Such a design not only improves the portability of the mechanical arm, but also expands its application range.

Finally, we put forward the concept of "ten-thousand-yuan-level price". In 2018, we envisioned that if the price of the mechanical arm could be controlled within 10,000 yuan, then it might enter daily life on a large scale and be widely adopted. This is our initial judgment and the goal we are committed to achieving.

Through these strategies, Realmotion is committed to promoting the popularization of robot technology, allowing robots to enter thousands of industries and thousands of households.

Intelligent Emergence: It was too advanced to let "robots enter life" at that time, wasn't it?

Zheng Suibing: Yes, but I think everything should return to the underlying logic, that is, to see whether there is a demand for this in the market and whether this demand can be realized. During the period from 2017 to 2018, many peers have begun to try to apply industrial robots and collaborative robots to different industries, such as unmanned pharmacies, unmanned supermarkets, and coffee making, which makes us judge that the market demand is mature.

Since 2010, we have focused on the research and development of mechanical arms, with the goal of making them lighter and smaller. From a technical perspective, we believe that there is a feasible implementation path. Because there were no successful cases at home and abroad at that time, our idea was questioned by many.

In 2017, we consulted many experts and proposed a plan to develop ultra-lightweight humanoid mechanical arms, hoping to develop mechanical arms with high power density. The question of one expert impressed me deeply. He asked: For so many years, industrial robots and collaborative robots have been trying to become lighter and smaller, but failed. Why can we achieve this breakthrough?

In the face of these questions, we gave a response, but we also admitted that these logics have not been verified in practice and are still in the theoretical stage. We further proposed the concept of a general-purpose mechanical arm that integrates with humans and the environment. This was a considerable controversy in the industry in 2018, because the prevailing view at that time was that general-purpose robots were not practical, and the market preferred specialized equipment.

But we still believe that the direction of the general-purpose mechanical arm is correct. If the industrial robot is a big truck, what we want to create is a car, but the challenge is that the core components of the car - the engine, gearbox, etc. - are not yet mature. Our four core components - reducer, motor, driver and controller - these core components have to be made by ourselves, but there were only five people in the company at that time.

In retrospect, it was our youth and arrogance at that time, as well as our firm belief in technology, that made us persist. Our goal is to make robots not just tools for industrial production, but an indispensable part of daily life. Although this goal may seem somewhat advanced at that time.

Intelligent Emergence: Finally, you still made it.

Zheng Suibing: Yes, but it took us nearly 4 years to come up with the first product. Compared with the collaborative mechanical arms on the market, the volume of the ultra-lightweight humanoid mechanical arm is halved, but the power density output is twice.

When it was first displayed at the World Robot Conference in September 2021, many peers found it incredible. We demonstrated the load capacity, speed and other performance of the product on the spot.

At the same time, we participated in the 30,000-hour average trouble-free operation test of the China Robot Testing Center and passed the verification. Currently, we are sprinting for the 50,000-hour average trouble-free operation test, and it is expected that the verification report can be obtained in February next year, which fully proves that the Realmotion mechanical arm has the characteristics of a long service life.

However, it takes nearly 4 years to develop a product, and this business progress speed is generally unacceptable to investment institutions.

Part 2: "Embodied robots are expected to enter the household within 2 - 3 years"

Intelligent Emergence: Have you solved the three bottlenecks you judged?

Zheng Suibing: At first, we thought they were all solved, but now we think we have solved 2.5 of the three bottlenecks.

At the beginning, we identified three major bottlenecks that robots need to overcome to enter life: generalization ability, operational ability and cost. We have successfully solved the operational ability and cost problems. In 2022, we established AI_Lab, aiming to solve one of the key problems of integrating robots into daily life, the generalization ability.

In order to enable robots to play a role in a wider range of environments, the generalization ability needs to be urgently improved. In many industry applications, such as power inspection and factory operations, we found that the traditional procedural programming methods are often not efficient enough. Therefore, we are thinking whether we can endow the mechanical arm with higher intelligence so that it can remember and reproduce the actions of the buttons, switches and valves that have been operated.

Our AI_Lab team focuses on improving the ease of use of the mechanical arm. Our research returns to the traditional methods of deep learning and reinforcement learning, that is, the so-called embodied intelligence. By the second half of 2022, we often mentioned in public speeches that Realmotion has solved the three bottlenecks for robots to enter life.

However, in 2023, with the rise of large models, our views have changed. The emergence of large models has shown us new possibilities, and we have begun to conduct in-depth tests on them.

Finally, we divided the generalization ability into two parts: the cerebellum and the cerebrum. In terms of recognition, positioning and operation, our technology is very mature, and it can operate fine items such as transparent glass fragments without additional training, which solves the operational ability problem of the cerebellum. Since then, we no longer say that we have solved the three bottlenecks, but that we have solved 2.5 of the bottleneck problems.

Intelligent Emergence: So before the large model came out, the generalization ability we imagined was actually the ability of the cerebellum.

Zheng Suibing: Indeed, before the emergence of large models, our understanding of generalization ability was mainly focused on the function of the cerebellum. Our positioning at that time was "to recognize, position and operate everything", which emphasized that our technology can accurately identify objects, precisely position and perform operations in various situations.

In addition, we also position ourselves as "the entrance of the interaction between the real world and the virtual world", because we realize that bringing the complexity of the real world into the virtual world is not simply achieved by collecting a large amount of data.

We believe that the key to collecting data is not to buy a large number of data sets, but to the large-scale application of the product. Through the wide deployment of the product, we can collect the data in the actual operation, and these data in turn can promote the progress of our technology. This method ensures that we can learn from the diversity of the real world and continuously improve its generalization ability.

Intelligent Emergence: Many industry insiders judge that embodied robots will first land in the B-end and then the C-end. Because C-end users have high expectations, and the current general ability and generalization ability of robots cannot meet them.

Zheng Suibing: The market potential is huge. We predict that in the next two to three years, robots will gradually become a household appliance product that is well-known to every household, as popular as today's floor-sweeping robots, televisions and air conditioners.

The vision of robots truly integrating into daily life and providing various services for families, such as taking care of pets, looking after the elderly, remote operation, and performing tasks such as serving tea, heating food and feeding medicine, is no longer far away. Of course, the service robots I mentioned do not refer to bipedal ones, and the popularity of bipedal ones may take longer.

Intelligent Emergence: Is the bipedal more difficult?

Zheng Suibing: Bipedal robots face challenges in energy supply. Musk's choice of the bipedal robot route, although technically forward-looking, still has bottlenecks in battery life.

For example, there was a news about a humanoid robot participating in a marathon some time ago. We learned that the continuous walking time of a general humanoid robot does not exceed 15 minutes. Even if we double or triple this time, it can only maintain activities for up to two hours, which puts extremely high requirements on the battery capacity and the carrying capacity of the robot.

Intelligent Emergence: Now we see that bipedal robots have entered the factory. How do you view this phenomenon?

Zheng Suibing: Musk made bipedal humanoid robots enter the factory, and everyone followed suit. In my opinion, this is completely led astray. Everyone should not blindly imitate Musk's practice in this regard.

In actual situations, bipedal robots have a slow moving speed, and the power supply problem also needs to be overcome. The factory environment is not the most suitable for humanoid robots.

In addition, we cannot ignore the connection between Musk's space exploration project and his development of bipedal humanoid robots. At the technical level, the engineering and technical experience accumulated in the space exploration project, such as rocket manufacturing, the application of artificial intelligence and automation technology in space missions, and electric vehicle manufacturing, have provided support and reference for the research and development of bipedal humanoid robots.

From a business perspective, this is one of Musk's strategies to expand his business territory. Both space exploration and the research and development of bipedal humanoid robots can enhance his brand image and build a larger business ecosystem. More importantly, from the vision level, by using bipedal to find a new space for human life, humans have the ability to achieve multi-planetary habitation.

I personally believe that humanoid robots will be implemented and applied in a chassis form in the near future, but this scene is probably not likely to appear in automobile factories, and its application scope is likely to expand to all walks of life. We must not be misled by the current trend of following the trend, but should rationally explore the truly suitable development direction and application scenarios for