Shulou(Shulou.com)11/24 Report--

Original title: "Why do humanoid robots still want to see China?"

Which country will have more development opportunities for humanoid robots in the future? The answer is China.

A few days ago, the Tesla Bot humanoid robot "Optimus" prototype made its world debut in Shanghai.

It is 1.72m tall and weighs 57kg. It can carry bricks, deliver goods and water flowers. Musk said it could also play the role of Catwoman in the future and would bring about "a fundamental transformation of civilization."

▲ network map | Musk and Tesla humanoid robot TeslaBot caused widespread repercussions in the science and technology and investment circles when it was unveiled on Tesla 2022 AI Day in early October. Although there are different views, many people who are looking for future growth choose to believe.

Our country regards robot as the next focus of national economic development. In April this year, the 14th five-year Plan for the Development of Robot Industry was issued, pointing out that China will become the global source of robot technology innovation, high-end manufacturing agglomeration and new highland for integrated applications in 2025.

In China, foot robots (including biped and quadruped), including humanoid robots, have gained a certain degree of public awareness through the performance and presentation of Spring Festival Gala, Beijing Winter Olympic Games and other occasions. The entry of big companies such as Tencent and Xiaomi also gives quadruped robots such as robot dogs a chance to enter the public.

At what stage has China's foot robot industry developed, what is its position in the world, and what are the development opportunities for young people in this industry?

We talked to Zhang Wei, professor, doctoral supervisor and professor of Pengcheng Scholars, Department of Mechanical and Energy Engineering, Southern University of Science and Technology.

Professor Zhang Wei's core ideas are as follows: 1. Compared with the communication and electronics industry, the robot industry is still in the stage of BB machines, Business Communications and other special machines; the humanoid robot with strong versatility will be the iPhone of the robot world. Just sending a message, the iPhone is far less cost-effective than the BB, but when it can support multiple functions, it will quickly reshape the entire industry.

2. General-purpose robots do not necessarily need to be universal to have commercial value. It may take a decade or so for humanoid robots to enter thousands of households. But industrialization and commercialization do not require foot robots to achieve universal capabilities before they can begin, and this node may be achieved in the next two or three years.

3. It is difficult and great to study the foot robot. Great because it has great subversion, mobile phone is the general terminal of personal communication and entertainment, intelligent car is the general terminal of people on the structured road in the future, humanoid robot will be the general terminal to connect the digital world and the physical world when facing all kinds of unstructured scenes.

4. In order to learn the foot robot well, we must lay a good mathematical foundation and give consideration to hands-on practice, both of which are indispensable. There is an acute shortage of talents in this emerging industry, which is a good time for young people who dare to face the challenges.

The following is a transcript of Professor Zhang Wei's conversation: Q1: what is your opportunity to research and enter the field of foot robot?

A1: it's hard to say that I've been fond of robots since I was a child, which is unrealistic. It is only in recent years that I have a great interest in humanoid robots.

The first is the in-depth study of control. Since 2005, I have been working on the control and trajectory optimization of hybrid dynamical systems (systems with the coupling of discrete events and continuous dynamics). Later, with the maturity of deep learning, it is also related to the reinforcement learning algorithm of this kind of systems. The legged robot system will switch between different dynamic models with discrete touchdown events, which is essentially a kind of hybrid system. My previous theoretical research in this area provides a very important basis for subsequent application research.

Secondly, the hardware of the foot robot has gradually matured in the past two years, and the algorithm can be verified and iterated quickly on the hardware. the maturity of the hardware is very important to the development of the industry.

The research on robots began about five or six years ago. Before that, the experimental platform was very scarce, and we were all laying the foundation, because the industry had high requirements for the foundation, and the past research was not wasted. Instead, it can be directly introduced into the following research.

After long-term research and matching of learning and technical background, there is an opportunity to go deep into the research of foot robot.

Q2: what are the needs of human beings for foot robots?

A2: on the whole, robots mainly address two types of needs:

The first part is to solve the mobile demand, so that the robot can do one thing from A to B, including the unmanned car, the food delivery robot and so on.

The second part is to solve the operational requirements, let the robot operate on each scene from A to B, such as adding a robotic arm on the basis of movement to complete more tasks.

These two parts summarize 90% of the robot's capabilities. It can be said that these two requirements are also the needs that we hope the foot robot can achieve.

Q3: the general public will have a question: why do we have to build robots with feet (whether biped or quadruped)?

A3: most of the previous robots were wheeled robots. In solving the mobile demand, there is a big difference between the foot robot and the traditional wheeled robot, and there are more advantages.

First of all, the wheeled robot moves by turning the wheel, and the efficiency of moving on the flat ground is relatively high; the legged robot walks according to the way people move, has a stronger ability to adapt to the terrain, and can reach where anyone can reach in principle.

Second, wheeled robots can only be used in scenes built for it, such as building approach bridges, paving roads or building tracks. It is difficult for wheeled robots to go to uneven places such as outdoors, stairs and steps, but they are encountered everywhere in people's lives.

The legged robot can make up for the terrain adaptability that the existing wheeled robot does not have, which is very important to solve the mobile demand.

In terms of operational ability, the leg-foot mechanism can provide a huge advantage for the operation, which is often overlooked.

▲ diagram | 3D point-and-foot biped robot developed by Southern University of Science and Technology now generally solves the mobile operation by installing a robotic arm on a wheeled site. The wheeled chassis used is generally very heavy, and it is difficult to adjust the center of gravity by adjusting the attitude of the chassis. Once the manipulator exceeds the center of gravity, it is difficult to maintain balance. At the same time, because the chassis is large and the attitude is fixed, the mobile operating space of the wheeled robot is also relatively limited.

The foot robot (especially the humanoid biped robot) is relatively consistent with the human degree of freedom, and can complete the operation task that the wheeled chassis + manipulator can not do in the narrow space.

For example, a humanoid foot robot can bend over to pick up something like a human, or raise its foot to keep its balance when picking up something on the opposite table. It can be said that when the technology is fully mature, this form of robot can solve any mobile operation needs that human beings can solve.

Q4: will foot robots, like humans, be able to flexibly face all kinds of situations and solve all kinds of needs?

A4: if you only focus on a certain type of scene, then usually the foot robot is not the best choice. Then why do you need a foot robot? In essence, it is the difference between special plane and general purpose.

A special plane refers to a machine that takes the place of a person to complete the corresponding work in a fixed scene. But the problem is that many applications can not design a specific special plane corresponding to a scene.

The development trend of the whole robot is essentially a process from special plane to gradually universal. For example, most automated production lines can be thought of as robots, but people are more likely to call them automated devices than robots because of their strong oneness.

Later, with the increase of demand, the cost of changing a set of equipment for each production is too high, so it is inevitable to introduce industrial robots and cooperative robots, and the manipulator can meet the operational needs of more flexible manufacturing.

However, the mechanical arm can only be deployed in a fixed position, which still can not meet some needs, so people study to add a mechanical arm to the mobile chassis, that is, the current composite robot, which not only has the ability to operate, but also has the ability to move at the same time, to be able to complete a task from A to B. This is the process of gradually improving versatility.

To put it simply, if you want to solve a single task, the answer is not a foot robot in most cases; if you want to solve multiple tasks and the tasks are rich enough, the answer is a foot robot.

Versatility is the greatest advantage of the foot robot, you do not need to design a special plane for any staircase or any scene to complete the task, the foot robot itself can complete multiple tasks. When there are enough such tasks in the future and the cost of the robot is low enough, the legged robot will be of great value.

Q5: does this mean that foot robots don't need to create an environment where they can walk?

A5: the previous design concept is focused on special aircraft, and this logic is appropriate under a large number of, single and repetitive requirements. A train is essentially a special plane, as long as people need to lay a track before they can run.

It is not these requirements that legged robots need to replace. In a variety of situations where there is not so much demand, the versatility of robots is needed to accomplish things that originally need people to do. If you want to replace a person to complete the task, you must be able to go where you can go. You don't need to look like a person, but you must move in the same way as a person.

Our world is built according to the characteristics of human activities, including how wide the steps are, how high the threshold is, how narrow the door frame is, which are all customized according to the shape of the human body. We have built the world, and if we have to transform the whole world in order to introduce mechanization, it is not designed for people, but for robots.

There is no denying that both ideas are beneficial. However, in the case of small scale of demand, it is impossible for people to transform the whole environment for robots, and the value of foot robots is reflected here.

Q6: what is the development stage of foot-standing robot in our country at present? Compared with the world, are we synchronized or lagging behind?

A6: as far as academic circles are concerned, there are not many researches on foot robots, and not every school and every major can have dozens of teachers to do this.

In terms of the overall level, it has improved a lot in recent years, but there is still a certain gap with foreign countries (especially the United States).

In 2019, MIT in the United States opened up a set of quadruped robots, including hardware and software. Everyone's technical lower limit has been raised to the same starting line, and everyone can start to make a robot dog in a short time, but this open source solution is essentially open source at the level of academic papers, not mature product solutions. So most robot dogs can only take two steps or turn a somersault, which is still a long way from solving the real landing needs.

Now this gap is mainly reflected in the software algorithm ability, the domestic hardware is relatively mature, but the functionality is very little. So as you can see, most applications are still performance-based, with fixed choreography of the action scene in advance and then playback, including robots performing in some grand occasions.

How to define whether the level is high? It depends on whether the robot makes decisions and controls based on real-time information, or plays back the actions that are scheduled in advance.

At present, our domestic research on action based on real-time perceptual information is still in its infancy, and even in the world, few companies have a fully mature plan. Companies like Boston Dynamics didn't let robot dogs into real situations until 2018, demonstrating scenarios such as going up and down stairs and patrolling factories, but so far they haven't seen a steady operation in factories around the clock, so the technology should not be fully mature.

Just like people can walk with their eyes closed, but this is called blind walking. Under the premise of open source, the technology of "blind walking" is basically at the same level. Although there are still some gaps among the research groups, they are not much different. When you really open your eyes for a walk, you have to feel it in various scenes such as stairs, and then plan where to step on and how much force to apply according to the perceived situation. at present, many robot companies in China do not have a mature plan. the research in colleges and universities is also relatively few.

▲ diagram | LimX is going down the stairs to show real-time visual feedback Q7: what are the industrial and technological advantages and shortcomings of the development of foot robots in China?

A7: at present, our biggest challenge is the technical deficiency, but compared with foreign countries, we may still be a winner in the end. The advantages are as follows:

The first is the cost advantage. Due to the mature industrial chain and relatively low labor cost, the hardware cost of foot robot has a huge advantage in China.

The second is the advantage of the scene, the whole industrial chain of robot development and landing scene can be found in China. Whether it is the scene of ToC or ToB, the richness in China is hundreds of times that of foreign countries. Some industry scenarios do not even exist abroad, from this scale, there are great advantages at home.

The technical deficiency is mainly due to the lack of original innovation ability of robot algorithm. The algorithm ability of robot software can be divided into three layers.

The first is the top-level AI layer, which takes the place of people to make decisions. For example, what the robot should do now, what objects there are in the environment, where to go, what objects to grab, and so on. Fortunately, after the recent popularity of autopilot, this kind of AI capability has an advantage in China. Although the technology is not so mature now, at least there are no obvious disadvantages both domestically and internationally. This field is not exactly the field of robotics, what we need to solve is to provide a general mobile and operating platform.

The second is the ability of positioning and navigation. AI (or human) decides where the robot should go and what to do, and then the legged robot needs to know which point An is and which point B is in the process of moving, and walk from A to B without deviation, which belongs to the ability of positioning and navigation. This is similar to unmanned cars and drones. At present, this is not very difficult, and there are many domestic teams that have done excellent work in this area.

Finally, there is the ability of motion control algorithm, that is, how to complete the movement from A to B, from this step to that step. How to do this has nothing to do with the upper AI, because AI has already told you to go from this point to that point. In this process, you need the ability to combine perception and control, perceive the surrounding environment in real time, and control 20 or 30 motors in real time according to the perceptual information to complete the required mobile operation tasks.

Among these capabilities, the control ability of multi-joints based on real-time perception is relatively scarce in China, which is the reason why our team focuses on this core technical point.

▲ diagram | LimX based on visual stability continuous steel hollowed-out staircase Q8: what is the level of talent supply in the field of foot robot in China? Which domestic colleges and universities have strong strength in the development of foot robot?

A8: there are a lot of teams doing it, and recently it has attracted a lot of new research teams to enter this field. In short, everyone's progress is still obvious, but everyone's energy still needs to be focused.

At present, in addition to our Southern University of Science and Technology, there are Zhejiang University, Beijing Institute of Technology, Harbin University of Technology and other representative institutions.

Q9: how do you evaluate the products and the technical strength behind some foot robot companies in our country?

A9: personally, I don't think there is a very good landing at present. Many robots that use robot dogs as toys are bought for a walk or used for performance, which are essentially out of curiosity or entertainment. It is difficult to classify robot landing.

Internationally, Boston Dynamics and Anybotics are promoting the landing of industrial inspection, and there are also examples of factory inspection in China, but at present, most of them can not solve the problems of getting up and down stairs and complex terrain, so they do not have many advantages compared with wheeled robots, mostly out of curiosity and the need of publicity and display.

People's interest in robots comes from the expectation that robots will replace people to complete tasks in the future, but this expectation has not been realized at present.

Q10: will the commercialization of a legged robot not be realized until it becomes omnipotent?

A10: we hope to push the legged robot to the ground, and not only the quadruped, we also attach great importance to the humanoid biped robot. On this point, we share the same idea with Tesla.

When it is mature and cheap enough, there will be humanoid robots in every family in the future. Humanoid robots are the mobile phones of the future, equivalent to the iPhone of the robot world, but we are still in the stage of mobile phones and BB machines. If you just send information, the BB will be enough, and you don't need iPhone, just like the difference between a special plane and a general-purpose robot.

Some companies will directly layout to the form of iPhone, such as Agility Robotics, Tesla and our team will directly develop humanoid robots. It is not as cost-effective as the BB machine in a short time, because it can only accomplish simple tasks, such as sending a message.

However, the commercialization process of humanoid will be gradual, and all we need to do is to fix the shape first, and then add functions similar to App.

With the emergence of the first App, we can replace the BB machine, and then there are two or three App, which can gradually replace the functions of making phone calls, sending Email, taking notes, playing games, and so on. When there are enough of these functions, iPhone can replace other specialized electronic devices.

Of course, this is a simple analogy, in short, when a humanoid robot can complete multiple tasks, its business inflection point will appear, and it will usher in a disruptive development. The most important thing is that no matter how much App is added in the future, its basic form will not change, and there may be an Internet-like development of the robot world that was previously unimaginable.

So my view is that general-purpose robots don't need universal to have commercial value. This is just the ultimate product form, we just have to keep adding functions to it. The first function is not enough, its performance-to-price ratio is not comparable to the special plane, so add the second function, which is gradually more cost-effective than two special planes. Tesla is also similar to this way of thinking. Instead of a BB machine on the left and a mobile phone on the right, each special plane field has its own work, it is better to invent a final form of iPhone.

Of course, it may be a bit exaggerated to make a mobile phone in the BB era at first, it seems unnecessary to do such an expensive thing, but we are the first people to iterate over this product. When we have the functions of making a phone call, sending Email, and watching video, the performance-to-price ratio exceeds the special plane on the market, and our industry can begin to develop rapidly. After gradually forming the market scale, the market no longer needs special planes such as BB machines.

There are different dimensions for dividing robots. In a certain dimension, I tend to divide robots into tool robots, such as cars and airplanes, which belong to human vehicles. Robots can replace people to drive cars, but they will not replace cars. Because this is a human tool; the other category is human robots, which are most likely to be replaced by humanoid robots or foot robots in the future.

For example, Musk said that humanoid robots in the future will be larger than cars, and we think so, too.

In the past, there were concerns in the market, first, because the technology was not yet mature. Another is that the cost is too high, but Tesla said at the press conference that it could be less than 20,000 US dollars. Although the price is still expensive at first glance, it is enough in some commercial and industrial scenarios. In the future, there is no need for a sweeping robot on the left and a stir-frying robot on the right, and one form can complete the task of babysitting.

I think it may take more than a decade for humanoid robots to enter thousands of households. However, industrialization and commercialization do not require foot robots to start after realizing superpowers. this node may be realized in the next two or three years. Tesla's prediction may be skeptical. But we are very confident about this direction and its commercialization process according to our own understanding.

What state are our legged robots in now? Compared with another area, it is probably the autopilot phase around 2010. Is there a real autopilot now? It hasn't really landed yet, but the research and commercial exploration of self-driving has been extended so much that it has even begun to upend the entire auto industry.

Therefore, the commercial node is also coming step by step, the function is also gradually increasing, and the market scale is also gradually opening.

Q11: what is the development space of foot robot in the future?

A11: there are generally two ideas for the development of the entire robot industry. The first development idea is to focus on a vertical field, all kinds of special planes must exist, as long as the scale of the field is large enough and the scene is specific enough. However, in some fields that are small in scale, special in demand and not worth designing special planes, what they have in common is that they will become application scenarios for legged (mobile operation) robots.

Q12: is the legged robot expected to become a family companion entertainment role in the future?

A12: now foot robots can meet some of the companionship needs, including anthropomorphic or cute-looking robots, but now they are not particularly successful, not particularly large-scale into every family, because the cost is still high.

What is the value of companionship? How do you define this market? It's not very clear yet. I think there will be a branch in the future, but now the market and products are not very mature. I think it depends more on the development of general AI technology than on the development of general robots.

Q13: what is the reason why the legged robot has attracted capital attention? Do you think capital can be patient enough to wait for the development of foot robots?

A13: we define this as a difficult and great thing. The difficulty is that compared with the existing mature robots, the software and hardware of foot robots not only need great innovation to achieve function, but also need to keep the cost as low as possible. great because of its breadth of application and impact on human society has a great subversion. That's what the capital interested in this track should be interested in.

This type of capital is generally an investment institution with a long-term vision and long-term attention to the development of cutting-edge science and technology, and the legged robot is a field with enough imagination and absolute subversion.

Especially this year is the first year of the real industrialization of humanoid robot. Amazon invested 150 million US dollars to support the landing of Agility Robotics in the logistics scene. Tesla's huge investment in humanoid and the release of AI Day also let domestic investors at least see the feasibility of humanoid robot industry. At the same time, Musk's expectation and ability to reduce costs also let us see the potential economic benefits. So the whole capital is now very concerned about this track.

At present, domestic investment institutions such as Mingshi Capital, Fengrui Capital, Kunzhong Capital and other investment institutions are very concerned about foot robots. They are willing to accompany scientists in the next decade, they do not expect to produce value immediately, and we all agree on this point. Because this thing is difficult but great, if it is not difficult, it cannot be great, anyone can do it.

In addition, it should also be noted that most investors in the market do not understand this emerging track. Due to the emergence of Tesla, they pay more attention to the two feet than the four feet. In fact, the core technologies of these two types are the same for us, and we define them as general-purpose mobile robots.

Q14: do you have any suggestions for young people who are interested in the foot robot industry and may be interested in joining the industry in the future?

A14: lay a good mathematical foundation and hands-on practice, both are indispensable.

There are also many people in China who are very enthusiastic about doing all kinds of robots, but in essence they only do all kinds of repetitive development and integration at a technical level, and most students'in-depth mastery is not enough.

Many robot industries mainly rely on engineering experience and technology integration ability of practitioners, but foot robot is a difficult and great thing, which requires not only engineering experience, but also in-depth understanding of basic theories, including robotics, nonlinear control, optimal control, optimization theory and algorithms, multi-rigid-body dynamics, machine learning, computer vision and so on. It takes a long time to accumulate, but there is no hurry.

The development and iteration in this field is also very fast, and the technology three years ago is completely different from what it is today, but as long as you understand it deeply, you will find that the underlying principles and logic are the same. Only with the ability of in-depth understanding of basic principles can we respond to changes without change, and a superficial understanding of technology is far from enough. At present, there is a relative shortage of talents for foot robots in the market, which is a good time for young people who dare to face the challenges.

This article comes from the official account of Wechat: true stories Research Room (ID:zhengulab), by Gong Zheng and Zheng Sifang, and edited by Uncle Mian

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.