Humanoid robot|Depth 1: First principles, why is the Tesla chain most optimistic and what is optimistic

For data and models in this article, please contact: Sun Xiaoya/Zhang Tongtong

summary

Traditional robots are tools, and humanoid robots are the next generation of intelligent terminals

Traditional robots are a mature track with a scale of 100 billion, divided into industrial and service robots, and industrial/service robots are 100 billion level markets, with a global growth rate of about 10%/20% in recent years. Humanoid robots belong to the emerging track, the participants are mostly technology, Internet companies, due to the technical difficulty + high cost, delayed mass production, the industry is still in the state of exploration, recently we believe that the industry has undergone obvious changes, humanoid robots are expected to achieve a breakthrough of 0 to 1, that is, the supply side - obvious changes (TSL, X1 Technologies) + technology side - artificial intelligence technology has made significant progress.

Considering the current application of humanoid robots and economy, we believe that the first step of humanoid robots to penetrate may be the security, warehousing and logistics industries, the first to penetrate regions or countries with high labor costs such as Europe and the United States. From the perspective of long-term logic, we believe that traditional robots are only "tools", while humanoid robots are embodied intelligence, that is, intelligent systems that can understand, reason and interact with the physical world. We expect that the sales of humanoid robots are expected to reach 10,000 units in 24 years, 5.63 million units in 28 years, CAGR=387% in 24-28 years, and a market space of 1 trillion yuan in 28 years. Looking forward to the long-term space, according to the number of people Musk said and the number of robots, if it is made 2:1, the world's 8 billion people, 4 billion humanoid robots, and a unit price of 20,000 US dollars, there is 80 trillion US dollars of space.

Tesla is expected to replicate the achievements of electric vehicles and lead a new era of humanoid robots

Humanoid robot is the carrier of artificial intelligence, is a "input-computing platform-output" process, in the past robot only "output", new product positioning and business model, we are optimistic that Tesla to realize the dream of humanoid robots, the essential logic for its ability to complete the closed loop of business model, to core software and hardware self-research + low-cost, scale advantage of the supply chain as the moat:

1) FSD visual perception algorithms can be reused on robots and empowered by Know-How accumulated in the field of intelligent driving, and the implementation of FSD V12 and Dojo supercomputing will greatly improve the speed of AI training in the future;

2) Direct migration of tram supply chain (sensors, motors, reducers, etc.), strong large-scale cost reduction ability (with hardware manufacturing, cost reduction ability);

3) The self-owned factory creates application scenarios and continuously trains and optimizes robots in real scenarios in the form of machine learning + a large amount of data.

What do you think of the investment in the humanoid robot industry chain? The core looks at Tesla, and the core components are motors, reducers, lead screws, and sensors

The Tesla robot Optimus has 28 actuators in the torso joint (rotation 14 + linear 14) and 12 actuators on the hand. Rotary joint: the use of rotary drive, mainly composed of frameless torque motor + harmonic reducer (may be partially replaced by planetary) + torque sensor + encoder; Linear joint: linear drive, mainly composed of frameless torque motor + roller/planetary ball screw + torque sensor + encoder; Hand: Each hand has 6 drives, mainly composed of coreless motor + precision planetary gearbox + torque sensor + encoder.

Motor: We expect Tesla torso joints to choose frameless torque motors with a high probability and hollow cup motors for hands. At present, the frameless torque motor is doing well in the United States Kollmorgen, and there are domestic frameless torque motors that are doing well such as Buke shares. According to the caliber of Buke shares a single frameless torque motor price in a few hundred yuan -1,000 yuan, brushless hollow cup motor overseas to MAXON to do better, domestic Mingzhi electrical appliances, Dingzhi technology, Hengshuai shares can do,

Reducer: We expect that Tesla's rotating joint mainly uses harmonics, the hand adopts a precision reducer, and the planetary reducer has a certain probability of replacing harmonics in the lower body. Planetary reducer because of the impact resistance is stronger than harmonics + low cost, is expected to be used in the lower body joints of the robot (there are 4 rotating joints), the global leader of harmonic reducer is Hamonaco, domestic green harmonics, Laifu harmonics, Tongchuan precision (subsidiary of Hanyu Group) and so on, planetary reducer global leader is Saiwei transmission, etc., domestic enterprises have Zhongdaide, double ring transmission and so on. The current price of domestic harmonic reducer is 1200 yuan / piece, and the cost is expected to be reduced to 500-1000 yuan / unit in the future, and the cost of planetary reducer is expected to be reduced to 200 yuan / unit in the future.

Lead Screws: We expect Tesla to use planetary rollers in the lower body joints and ball/T-screws in the upper body. Planetary rollers have the advantages of strong bearing capacity, large axial stiffness, high transmission efficiency and long service life, but the price is higher & overseas monopoly (Schaeffler is currently priced at 20,000 yuan / piece), so we expect or use it in the lower body (8 pieces) with high requirements for carrying capacity. At present, Dingzhi Technology and Best are in research and development, and Hengli Hydraulics is in the process of expansion.

Sensors: Sensors for humanoid robots mainly include vision, force control, and position (encoder). Vision sensors are in the head, force control and encoder sensors are used in all joints of the body. Vision sensors are similar to the driverless supply chain, and the good ones in China are Hikvision (whether with an algorithm obviously affects the value amount, but T self-developed algorithm), force control sensors We expect Tesla robots or use single-axis torque sensors in body joints, and 6-axis force sensors in the feet, in addition, the IMU (inertial measurement unit) mentioned in the UCLA paper is worth noting, and the relevant domestic listing target has core kinetic linkage.

Investment Advice:

The investment of robots this year is still the 0-1 stage, the market is relatively extreme, and the confirmation of the prosperity of the 0-1 stage (establishing the direction) and the industrial space (establishing the height) are the most important. In terms of prosperity, overseas Tesla, X1 (behind OpenAI), domestic not only Xiaomi, many large manufacturers such as Huawei, Tencent, etc. are also joining, in addition, more domestic supply chains have emerged (join the T chain / want to join the T chain + really start to invest money into the production line). In terms of industrial space, Musk gave us an outlook, humanoid robots are a trillion-dollar market based on 8 billion people.

From an investment point of view, instead of spending energy to spread other downstreams, it is better to focus on Tesla's optimus changes and progress, which is the fundamental thing to promote the market. Tesla Optimus Joint Assembler Supply Value is as follows:

• Linear Joint Assembler A: We expect the value of 2-3W after mass production, mainly responsible for linear joint integration (rotation in research). We expect that the core components are motor (made by yourself, 500*14) + planetary roller screw / ball screw (do it yourself, it is expected that the planetary roller screw will have to reduce the cost to the level of 1,000 yuan, we use 1200*8+200*6).
• Rotation & Linear Joint Assembly Vendor B: We expect the value of 2-3W after mass production, responsible for rotation + linear joint assembly (rotation is more advantageous at present). We expect that the core parts of the rotary joint are motor (500*14) + harmonic reducer (joint venture with Green) + encoder (100*14), and the core parts of the linear joint are motor (500*14) + planetary screw/T-screw + encoder (to be made by yourself, 100*14).
• Hand Joint Assembler C: We expect a value of 1-1.1W after mass production, responsible for hand joint assembly. We expect that the core components include coreless motor (600*12) + precision gearbox (most likely to do 200*12) + encoder (100*12).

Certainty of comprehensive value quantity & barriers. We first recommend Tier1 suppliers [Tuopu Group], [Sanhua Intelligent Control], [Mingzhi Electric]. We believe that under the global sales of 1 million robots, assuming that the value of Tuopu & Sanhua & Mingzhi is 2.5/2.5/11,000 yuan / robot, the market share is 50%, the net interest rate is 15%, and the robot business gives 50X valuation, and the upward elasticity of the above three targets is 50%+.

Secondly, we recommend [Double Ring Drive] and Tier 2 supplier [Green Harmonic] that is expected to become a Tier 1 supplier. Shuanghuan has the ability to produce planetary/harmonic reducers, and we expect the mass production price of planetary/harmonic reducers to be 200/600 yuan per piece.

Finally, it is recommended that the elasticity is large+, but Tesla has not yet been clearly contacted, but there may be opportunities &is expected to enter other supply chains in the future. 1) Lead screw: [Dingzhi Technology], [Best]; 2) Reducer: [Zhongdaide]; 3) Motor: frameless torque-[Buke Co., Ltd.], Hollow Cup-[Dingzhi Technology] (Jiangsu Leili), [Hengshuai Co., Ltd.], [Weichuang Electric]; 4) Sensors [Keli Sensing], [Xindong Lianke], etc.

Risk warning: the development of humanoid robots is not as expected, the progress of Tesla humanoid robots is not as expected, domestic suppliers entering the supply chain of Tesla humanoid robots is not as expected, and the calculation is subjective

1. Traditional robots are tools, and humanoid robots are the next generation of intelligent terminals

1.1. The traditional robot industry is a mature track that has been developed for many years

The research of robotics began in the mid-20th century, mainly based on the rapid development of computer technology and automation technology.

According to the IFR classification, robots mainly include industrial robots and service robots.

• Industrial robot: is a kind of automatic operation through programming or teaching, with multi-joint or multi-degree of freedom, and has certain perception functions, such as vision, force perception, displacement detection, etc., so as to realize independent judgment and decision-making on the environment and work objects, and can replace manual completion of various heavy, tedious or harmful physical labor in the automatic machine.
• Service robots: Service robots are a collective term for various advanced robots other than industrial robots that are used in non-manufacturing industries and serve humans. According to different application scenarios, service robots can be divided into two fields: home service robots and professional service robots, among which, household service robots can be divided into cleaning robots, entertainment robots, educational robots, escort robots, home safety robots, etc.; Professional service robots can be divided into medical robots, security inspection robots, logistics robots, customer service robots and military robots.

We believe that traditional robots such as industrial robots are more similar to industrial equipment, and service robots are similar to consumer goods with home appliances, so the development of traditional robot (industry + service) industry is more related to the growth rate of industrial manufacturing and service industries.

Chart: Global robot sales and forecast 2017-2024 (billion yuan)

Source: Green Harmonic Prospectus, IFR, Chinese Institute of Electronics, Tianfeng Securities Research Institute

1.2. Industrial/service robots are both hundreds of billions of markets, and the global growth rate in recent years is about 10%/20%

Industrial robot market space: In 2021, the global industrial robot market rebounded strongly, with a market size of 119 billion yuan, a year-on-year increase of 26%; In 2021, the installed volume of industrial robots worldwide reached 520,000 units, a year-on-year increase of 35%. With the continuous release of market demand and the further popularization of industrial robots, industrial robots are the terminal equipment for automated production, and the downstream is automobile manufacturing, 3C industry, chemical industry, metal processing industry, food manufacturing, etc.

Service robot market space: In 2021, the global service robot market reached 117 billion yuan, a year-on-year increase of 25%. Service robot application scenarios and service models are constantly expanding, and it is estimated that the global service robot market will reach 197.2 billion yuan in 2024, with a five-year CAGR of 20%.

Chart: Global Industrial Robot Market Size and Forecast 2017-2024 (100 million yuan)

Chart: Global Service Robot Market Size and Forecast 2017-2024 (100 million yuan)

Chart: Global Industrial Robot Installation and Forecast 2015-2025 (10,000 units)

Chart: 2018-2021 Global Industrial Robot Installation by Industry and Forecast (10,000 units)

Source: Chinese Institute of Electronics, Wind, IFR, Tianfeng Securities Research Institute

1.3. The four major families of industrial robots occupy half of the country, and the service robot pattern is scattered

Industrial robot competition pattern: From the perspective of the global competition pattern, the "four major families" Japan Fanuc, Switzerland ABB, Japan Yaskawa, Germany KUKA occupy half of the country, the total global market share in 2020 reached 58%, of which Fanuc's sales accounted for the highest 17%, and Chinese brand industrial robots accounted for only 11.9%.

Service robot competition pattern: There are many downstream subdivisions of service robots, divided into household robots, logistics robots, leisure and entertainment robots, medical robots, etc., and the application scenarios are more diversified and complex, and the overall pattern is relatively scattered.

Figure: Global industrial robot competition landscape in 2020

Figure: Application structure proportion of China's service robot market in 2018

Source: Huajing Industry Research Institute, Zhiyan Industry Information, Tianfeng Securities Research Institute

1.4. Humanoid robots are still in the concept stage of 0-1

There is currently no standard definition of humanoid robots, which generally refer to intelligent robots designed and manufactured that imitate human form and behavior. It has a more complex structure, sensing, drive and control system than ordinary robots, and is designed and manufactured to interact with artificial tools and the environment, thereby assisting or even replacing human production and life.

The humanoid robot has lasted more than 50 years since it was proposed, and has entered the initial stage of industrialization:

• 1973-2000: Technological start-up. Features: Can initially imitate human upright walking, but slow mobility.
• 2001-2015: Technological breakthrough phase. Features: technological development, functional performance breakthrough, has the initial ability to act.
• 2016-present: Initial test of industrialization. Features: The technology has reached a certain foundation, and the landing scene is initially explored, and the main problem is practicality + cost.

Source: Guanyan Tianxia, the official websites of various companies, Tianfeng Securities Research Institute

1.5. Different from industrial robots, the four major families occupy a core position, and humanoid robots are mostly technology and Internet companies

The four leading families of traditional industrial robots are not very competitive in the humanoid robot track, and there is no obvious progress in the research and development of humanoid robots, only some humanized two-arm robots, such as ABB's two-arm collaborative robot Yumi.

Humanoid robots on the track of science and technology, Internet leaders bet, of which the Internet explores application scenarios. In recent years, humanoid robots have ushered in high attention, in addition to traditional robot manufacturers, car companies, Internet companies have settled in, the exchange of multiple forces has enriched the availability of robots, through technology migration, continuous penetration into many fields. 1) The attention of leading technology companies to the humanoid robot track can empower their main businesses such as consumption and entertainment, AI can empower the exploration of downstream application scenarios, and promote the application of technology to industry and other fields. 2) Relying on the advantages of business understanding, Tesla integrates multiple resources to accelerate cost reduction and large-scale mass production.

Table: Commercialization of mainstream products of humanoid robots

Source: Gaogong Robot, Guanyan Tianxia, official websites of various companies, Zhidong, 21st Century Economy, Tianfeng Securities Research Institute

1.6. Boston Dynamics Atlas, hydraulic drive pursues extreme athletic ability, high cost is difficult to commercialize

Boston Dynamics History: Boston Dynamics was founded in 1992 as a laboratory for the development of military robots for the U.S. military. It has been acquired by Google, SoftBank and Hyundai Motor Group, and currently holds 80% of the shares.

Atlas hydraulic structures and perception for superior athletic performance: Atlas can run, flip, jump and soar like an athlete and quickly jump over obstructions with a certain inclination, known as a "parkour" robot. Behind Atlas' extreme athletic capabilities are powerful hardware performance and perception and control technologies:

• Hydraulic drive: Atlas robot limbs are driven by hydraulics, the whole body has a total of 28 degrees of freedom, the hydraulic system is composed of servo valves, actuators, hydraulic pipelines, etc., embedded in the limb structural parts through 3D printing. Compared with the motor drive: 1) The hydraulic drive has the characteristics of high-load driving, and the core principle is to generate high-pressure liquid through the liquid compression pump, and the high pressure acts on the cylinder to generate huge thrust, driving the joint movement of the robot, so as to make difficult actions such as backflips; 2) The hydraulic structure also makes the robot more elastic, in addition to the compressibility of the oil itself, if the accumulator is used, it can also play a shock absorption role, so that the limb is elastic.
• Motion control: The actions performed by Atlas are derived from the action templates in its behavior library, which can autonomously select the corresponding actions from the library according to the target situation. During the parkour process, Atlas' Model Predictive Controller (MPC) adjusts details such as its power, posture, and timing of movements to respond to various factors that may occur in real time, such as the environment and foot slippage.
• Environmental awareness: Atlas uses the RGB camera and TOF depth sensor in the head to obtain environmental information, and uses the multi-plane segmentation algorithm to extract planes from the point cloud to build 3D models of different objects, plan the path based on the model, and calculate the foothold of each step.

Disadvantages: the hydraulic system is noisy, easy to leak, sensitive to pollution, high requirements for the precision and quality of hydraulic components, and high requirements for the maintenance team, resulting in high manufacturing costs, a single unit worth about 2 million US dollars, it is difficult to go out of the laboratory and towards commercialization.

Source: Zhidong, Technology Neighbor, China Design Intelligent Manufacturing Award Official Account, Tianfeng Securities Research Institute

1.7. Ameca robot: access to GPT, realistic facial expression + strong interaction ability + weak motor ability

Manufactured by robotics company Engineered Arts, Ameca has facial expressions and movements that are extremely close to real people, and is mainly designed as a platform for the further development of human-computer interaction robot technology, which is expected to be used in entertainment and communication scenarios in the future. Ameca uses embedded microphones, binocular-mounted cameras, chest cameras, and facial recognition software to interact with the public, with the underlying system using Tritium+Mesmer. Interactions can be managed by GPT-3 or human telepresence.

• Tritium: connects software, equipment and the cloud, mainly responsible for intelligence and the linkage between various mechanical structures;
• Mesmer: Provides rich data on facial expressions and body movements. After using 36 cameras to 36 cameras to 360 degrees of the human body, Mesmer reconstructed the 3D model by comparing the color of pixels and anchor point positioning, so as to provide the robot with the ability to smile, giggle, open mouth, raise eyebrows, surprise and other expressions, and reproduce changes in human faces.

Cons: Not yet walking, no clear commercialization plan, according to Reuters, can be sold for $133,000.

Pictured: Ameca has realistic facial expressions

Pictured: Ameca robot head assembly

Source: Play Machinery, VRAR5, Guanyan World, Tianfeng Securities Research Institute

1.8. Tesla Optimus: Breaking the game with cost-reducing mass production, synergy with the car end, and fast research and development

In September 2022, Tesla's humanoid robot Optimus prototype debuted, Optimus is 1.72 meters tall, weighs 56.6 kilograms, and has a fast research and development process, which can already complete walking, climbing stairs, squatting, holding objects and other actions, and the overall exercise ability needs to be further developed.

• Joint actuators: Optimus has a total of 28 structural actuators in the body, 57 degrees of freedom throughout the body, and the joints are designed with bionic joints, which are divided into rotary joints and linear joints. Unlike the Atlas, Optimus opts for a more stable, cost-effective motor drive solution, which focuses more on practicality and travels at a speed of 5 miles per hour (about 8 kilometers per hour).
• Leading hand freedom: Optimus draws on biology and has 11 degrees of freedom in one hand to handle objects weighing about 20 kilograms.
• Electrical system: The robot is equipped with a 2.3kWh, 52V battery pack at the torso, and highly integrated charging management, sensors, and cooling systems, which can support the robot to work all day.
• Brain: Optimus' "brain" uses the same chip as Tesla vehicles, supports Wi-Fi, LTE links and audio communication, allowing it to process visual data, make action decisions based on multiple sensor inputs, support communication, etc., and the security of system hardware and software has been improved again. The Dojo supercomputing platform based on the FSD system gives Optimus powerful computer vision capabilities.

So far, Optimus' research and development is not very mature, and Tesla's vision is to make robots serve factories and homes. In the principle of robot design, Musk emphasizes high reliability and low cost, believing that the final price of the robot will be less than 20,000 US dollars, and the output may reach millions of units.

Pictured: Optimus robot delivery delivery

Source: Tesla official website, Play Machinery, VRAR5, Guanyan World, Heart of the Machine, Tianfeng Securities Research Institute

1.9. The pain point of the development of humanoid robots lies in the contradiction between immature products and high-demand application scenarios

For a long time, the pain points of the development of the humanoid robot industry are embodied in 1) the exercise ability is not up to standard: 2) the cost remains high; 3) Lack of application scenarios that can be implemented.

With the development of recent decades, humanoid robots have declined from millions of dollars in the early days (such as ASIMO fraud of $3-4 million) to the current tens of thousands of dollars (such as EVE priced at 320,000 yuan), and there are scenarios that can be commercialized (EVE can be applied to the security industry), we believe that the core contradiction of the current product is that the motion control ability is not up to standard, which is embodied in

• Motion control technology is not mature: lower limb motion control is the biggest difficulty, in the past industrial robotic arms, collaborative robots only replace upper limb movement, and lower limbs have higher requirements for strength, stability, and action consistency, and the current research and development of lower limb movement still needs key breakthroughs.
• Poor force perception and easy falling: Humanoid robots have insufficient force perception, and if they step on small pebbles, they will fall due to uncontrollable strength, and falling will damage joints and have extremely high maintenance costs. Let's talk about how to integrate into the unstructured world to achieve diversified applications.
• Joint torque technology needs to be developed: at present, the joint drive can not do the human level, the mainstream scheme has its own advantages and disadvantages, the harmonic reducer is light in weight but limited in torque, and the RV reducer has large torque but large volume.

In general, humanoid robots have passed the stage of "show-based + gimmick", the current technology has a preliminary foundation, the current stage focuses on solving technical and cost pain points, once the technology is close to the critical point, the general performance of humanoid robots in various unstructured scenarios will create a lot of demand.

Table: Humanoid robot motion control and human analogy

Source: OFweek, Gaogong Robot, Tianfeng Securities Research Institute

1.10.1. Humanoid robots are expected to achieve a breakthrough of 0 for reason 1: Significant changes on the supply side (TSL, X1)

Tracking technology and cost problems mainly look at supply-side changes, at present, there are some heavy players in the humanoid robot track, such as Tesla and X1 Technologies have shown/will show a large marginal change in technology/cost side, and the humanoid robot industry is growing rapidly driven by excellent enterprises:

• Player 1: Tesla. 1) The speed of iteration and evolution is amazing: The humanoid robot that appeared on Tesla AI Day in September 22 is the result of less than a year after the official establishment of Tesla's humanoid robot project, and Boston Dynamics has exchanged for more than a decade for today's top performance. This is because it can reuse the same FSD and Dojo supercomputers of Tesla Motors, train a neural network model every 8 minutes, and achieve a 30% improvement in training speed in code and design. 2) Pry open the market at low prices: ASIMO and Atlas cost $250 and $2 million each, while Tesla's humanoid robot directly controls the target price at $20,000 (RMB 140,000), which is expected to become a phenomenon and product. Tesla can reduce the cost of humanoid robots, mainly thanks to the innate conditions accumulated by its automatic driving: Optimus' visual recognition, scene construction, etc. all use the same neural network learning technology as FSD (full self-driving capability), and the chip implanted in the "brain" of Optimus is also the same as Tesla vehicles. 3) Ambitious long-term planning. Tesla's 2023 shareholders' meeting Musk spent a lot of space on robots, on the one hand, the imagination of the vast market space (if compared with the number of people and the number of robots, if it is made 2:1, the demand for humanoid robots may far exceed the demand for cars), on the other hand, it is an expression of Tesla's confidence in humanoid robots (robot motors, controllers and electronic equipment, all independently designed, Tesla has a world-class motor and electronics team, It is possible to design a robot that does not exist in this world), and even think that Tesla's long-term value is given by humanoid robots.
• Player 2: X1 Technologies. 1) Humanoid robots achieve mass production and application in the field of security: Norwegian humanoid robot company X1 Technologies, formerly known as Halodi Robotics, was established in 2014, mainly producing two humanoid robots: EVE (two-wheeled) and NEO (bipedal), of which EVE humanoid robots have achieved mass production and successfully found application scenarios, and delivered 140 robots to ADT Commercial in 2022. Applied to night patrol work, January '23 said that the goal for this year is to deploy 150 to 300 robots (that is, 23 years of sales), on May 23, the CEO of OpenAI-powered robotics company 1X revealed that its humanoid robot has been put into use in the United States and Europe, this robot works as a security guard at two industrial sites: unlike other security robots, it has a head, a face, two arms, and can navigate autonomously. 2) OpenAI invests in R&D and expansion: In 2023, OpenAI led the $23.5 million investment in 1X Technologies to increase NEO's R&D efforts and expand EVE production.

1.10.2. Humanoid robots are expected to achieve a breakthrough of 0, reason 2: Significant progress in artificial intelligence technology - AI assistance

Humanoid robot industrialization or riding the east wind of AI. Humanoid robots contain three core capabilities of human-computer interaction, environmental perception and motion control, and many venture capital institutions say that the progress of humanoid robots depends on the overall development of artificial intelligence. Therefore, the multi-modal building model and natural language capability behind ChatGPT are important factors to promote the recent industrialization trend of humanoid robots, specifically, it can help humanoid robots break through some pain points:

• Solve force perception problems. According to Yijiahe's answer to investors' questions, CV and other technologies represented by deep learning can allow robots to better understand the environment, identify objects and defects, and solve general intelligence problems to meet the adaptability and task completion ability of semi-structured environments.
• Accelerate technology iteration and solve motion control problems faster: The key to motion control lies in the algorithm iteration speed. 1) Boston Dynamics, which has the best motion control technology, uses model predictive control (MPC), which is also a common algorithm in the industry, that is, to build a robot dynamics model to predict the future actions of the robot. Previously, Tesla expected to solve the motion control problem in 18-36 months, but at this stage, large models can better train learning algorithms and large data sets and accelerate iteration; 2) Once there is a breakthrough in software and hardware technology (such as joint modules, intelligent bionic materials), large models can also accelerate technology integration.
• Bring high-value downstream scenarios. 1) The multimodal large model helps humanoid robots understand human emotions more naturally and accurately, better adapt to human needs, and have scarce value in the context of accelerating population aging and increasing demand for mental and physical companionship. 2) AI intelligent models can mine datasets from other walks of life to accelerate the application of more valuable commercial scenarios.

Figure: How Model Predictive Control (MPC) works

Source: CSDN, CloudMinds, KLCII Community, TF Securities Research Institute

1.11. What is the demand scenario for humanoid robots?

Considering the current application of humanoid robots and economy, we believe that the first step of humanoid robots to penetrate may be the security, warehousing and logistics industries, the first to penetrate regions or countries with high labor costs such as Europe and the United States.

• Operational: On May 23, 2023, the CEO of OpenAI-powered robotics company 1X revealed that the humanoid robot it has developed has been put into use in the United States and Europe, and this robot works as a security guard at two industrial sites: unlike other security robots, it has a head, a face, two arms, and can navigate autonomously. It is understood that
• Economy: Previously, Musk said that the future goal is to reduce the cost of humanoid robots to 20,000 US dollars (about 140,000 yuan), and the current price of humanoid robots on the market is often millions of yuan, and it is difficult to reduce from millions to hundreds of thousands of yuan. But our recent calculations found that the starting wage of Tesla factory workers in the United States is 19 US dollars / h (McDonald workers are 14-19 US dollars / h), and European and German laws require a minimum hourly wage of 13 euros (14 US dollars).

From the perspective of long-term logic, we believe that traditional robots are just "tools", and humanoid robots are embodied intelligence, that is, intelligent systems that can understand, reason and interact with the physical world, and the application field of humanoid robots starts from the to B scene, and eventually enters the to C field of home and other scenarios through continuous iterative optimization.

1.12. Humanoid robot market space: The space is expected to reach trillions in 2028, which can match the electric vehicle in 2021

It is expected to overcome the cost problem and achieve mass production, and eventually form a closed loop of the entire business model.

• Historical experience shows that the initial cost of humanoid robots is high, but the cost is reduced rapidly: taking the humanoid robot EVE of 1X Technologies as an example, the price of EVE was as high as 1.4 million yuan when it was first sold to Toyota Research Institute, and the price dropped to 1.05 million yuan when the third version of EVE was pre-sold in 2019, and the price of 140 EVEs in small-scale mass production has dropped to 320,000 yuan by 2022.
• The non-structural environment adaptability of humanoid robots can reduce the cost for B-end and C-end users. 1) Humanoid robots have high flexibility and strong environmental adaptability, and can move freely in complex human life scenes and human living environments, and B-end and C-end users do not need to specially modify the environment for humanoid robots, reducing application costs. 2) The humanoid robot has strong migration, higher service life and residual value than other robots, and reduces the annual use cost.
• The versatility of humanoid robots brings large-scale cost reduction effect: humanoid robots are a multi-task, general-purpose robot platform, in the future it will have a lot of application scenarios in industry, services, special and other fields, is expected to break the barriers between various use scenarios, the ceiling is extremely high, once the industrial trend is formed, the lower penetration rate can also bring huge market space, and large-scale mass production can also quickly reduce costs.

We expect sales of humanoid robots to be 10,000 units in 24 years, 5.63 million units in 28 years, CAGR=387% in 24-28 years, and a market space of 1 trillion in 28 years.

Long-term space: Musk said that the number of people and the number of robots are compared, if it is made 2:1, the world's 8 billion people, humanoid robots 4 billion, a unit price of $20,000, there is 80 trillion US dollars of space.

Table: Humanoid Robot Market Space Estimation

Source: NewAtlas, newsbeezer, Tesla.com, Wind, Tianfeng Securities Research Institute

2. Tesla is expected to replicate the achievements of electric vehicles and lead a new era of humanoid robots

2.1. The core technical requirements of humanoid robots are significantly higher than those of traditional robots

Traditional robots such as industrial robots emphasize the ability to execute actions, the degree of freedom is generally 2-6, service robots have more downstream application scenarios, generally have higher requirements for environmental perception, and humanoid robots are carriers of artificial intelligence, with a high degree of generalization + integration + intelligence, motion control and perception computing have higher requirements, from the degree of freedom is significantly higher than industry, such as Tesla humanoid robot in 57 degrees of freedom.

According to UBTECH's prospectus, the core technologies of humanoid robots include:

• Servo control: high-performance servo drive control, which can be used as a finger and foot drive mechanism, providing precise and repetitive movements to achieve a wide range of flexible movements, including walking, dancing and push-ups;
• Computer vision: A technology that enables robots to recognize and recognize human vision-like faces, objects, and environments, helping them pick details of objects for the next action, action, or human-computer interaction decision.
• Voice interaction: is a comprehensive technology based on voice-based information carriers, enabling robots to interact with humans in a human-like way. Integrate technologies such as Automatic Speech Recognition (ASR), Natural Language Processing (NLP), and Text-to-Speech (TTS).
• Motion planning and control: Motion planning refers to the movement task method, and control is the execution process. Motion planning and control ensures that the robot can accurately perform the motion instructions provided and realize the functions of operation, movement and movement.

2.2. Relying on the synergy advantages of electric vehicles, TSL humanoid robots are expected to achieve mass production in 24 years and commercialization in 26 years

The first step in the industrialization of Tesla's electric vehicles took 5 years, and it took 15 years from establishment to mass delivery. Tesla was founded in 2003, with a history of 5 years in the launch of the Roadster in 2008, followed by another 4 years of launching the Model S in 2012, and then the Model X in 2015, truly achieving high-volume delivery in 2018.

Tesla Robot At Tesla's AI Day in August 2021, Musk announced the creation of the humanoid robot "Tesla Bot", and released a prototype of the humanoid robot "Optimus" in September 22, which weighs 73 kilograms and is 1.72 meters tall, capable of carrying goods, watering plants and moving metal rods.

From the current progress, the progress of robots is obvious or faster than electric vehicles, we think the reason may be that Tesla was just established at that time, funds, management are not in place, etc., and humanoid robots can be based on the existing R & D team, management team of electric vehicles, and even the industrial chain of trams, algorithms can be replicated to robots, so the progress is faster.

Musk believes that the price of Tesla robots will be cheaper than Tesla's cars in the future, the price is aimed at 20,000 US dollars, and the planned sales volume exceeds one million units, and it is expected to achieve mass production in 3 years, commercial use in 5 years, and 21 years as the forecast period, that is, 24 years to achieve mass production, 26 years to achieve commercial use.

Table: Tesla Electric Vehicle Development Timeline

Table: Progress of Tesla humanoid robots

Source: Tesla official website, Tianfeng Securities Research Institute

2.3.1. Why are you optimistic about Tesla realizing the humanoid robot dream? Reason 1: Direct migration of vehicle-side algorithms

The evolution speed of AI models is a key part of the iteration of humanoid robots, and algorithms are an important tool for training models. The humanoid robot captures the scene through the camera, and then the algorithm calculates the size of the item, plans the road, etc., and guides the robot's robotic arm to make a holding action, or drives the robot to make a road planning forward. The key is to allow the robot to accurately identify a large number of items and classify the data, which is undoubtedly inseparable from the training of high-quality data. How to use AI algorithms to shorten training time, effectively identify key data, and complete simulation is the key to accelerating the iterative progress of robots.

Tesla built FSD and Dojo to accelerate data training and model iteration. 1) FSD Beta has been tested on more than 160,000 vehicles, training 75,000 neural network models in a year, equivalent to one training every 8 minutes, and the V12 version to be launched in the future is expected to achieve a complete end-to-end large model, greatly optimizing the speed of iteration. 2) The supercomputer Dojo is a training accelerator, which is expected to land in July 23, which is expected to increase Tesla's training speed in code and design by 30% (if manual labeling is used, it takes a week or two or even months to train a neural network model), and has cost and energy advantages.

Tesla humanoid robot migrates vehicle-side algorithms. Tesla's humanoid robot's visual recognition, scene construction, etc. all use the same neural network learning technology as FSD, using deep learning, big data analysis, Dojo training, automatic labeling and other algorithms, Dojo's huge computing power can provide cloud training support. In the future, with the further expansion of users, the amount of data and model volume will be larger, and Tesla's advantages in visual AI algorithms, artificial intelligence chips, automatic driving application practices, and AI training platforms will be further expanded, which is expected to break the shackles of the development of humanoid robots.

The world as Optimus sees it, using the occupancy network also used in FSD

Source: Tesla, CSDN, Cloud-piercing Star, Hard Tech Planet, Tianfeng Securities Research Institute

2.3.2. Why are you optimistic about Tesla realizing the dream of humanoid robots? Reason 2: Direct migration of supply chain, strong ability to reduce costs on a large scale

The tram supply chain can be migrated to the robot supply chain:

• The value of humanoid robot parts accounts for a high proportion: 1) The number of joints of humanoid robots is much higher than that of industrial and collaborative robots, which greatly increases the cost of humanoid robot parts, and important components such as servo motors, reducers, control systems, drives and machine vision account for more than 70% of the cost of humanoid robots. 2) The emergence of humanoid robot technology is not revolutionary, but the continuous evolution of existing robot/automation, automatic driving, artificial intelligence and other technologies, which is still strong in common with the upstream of the original industry chain.
• Tesla robots can share its mature automotive supply chain, which is conducive to greatly reducing manufacturing costs: humanoid robots are similar to the vehicle-end supply chain, some are the same as the vehicle-end, such as the supply chain of batteries, cameras, and chips can be shared, and several large parts in the rest of the new energy vehicles, motors, speeds, and battery thermal management, can be migrated to humanoid robots after simple transformation.

Tesla's large-scale cost reduction ability has been verified on the vehicle side, and we have reason to believe that it will be replicated on the robot side. Through calculations, we find that Tesla began to make a profit when its sales reached 370,000 units, and from 2017 to 2021, the fixed cost and variable cost of bicycles continued to decline, and the scale effect was significant. Because we predict that the cost composition of humanoid robots is similar to that of automobiles, if humanoid robots become terminal products with sales scale comparable to automobiles in the future, the cost reduction effect can be expected.

Table: Tesla Bicycle Profit Measurement Table

Source: Tesla earnings report, robot lecture hall, geek park, tianfeng securities research institute

2.3.3 Why are you optimistic that Tesla will realize the dream of a humanoid robot? Reason 3: Have your own application scenarios

Tesla mentioned the first application of its humanoid robots in its own factories, replacing repetitive labor. The first version of Tesla's robot will focus on simple, repetitive tasks, with the first use being to move parts around its own factories. According to new job postings from Tesla's robotics program, Tesla plans to deploy "thousands of humanoid robots" within its factories. Factory production is one of the few clear use cases for humanoid robots today, and Optimus has already put simple work on Tesla's Fremont factory.

Similar to autonomous driving, Tesla's approach to the humanoid robot track also adopts a "progressive" route, starting with simple actions and iterating in the application. By synergizing with the original business and creating its own practical application scenarios, Tesla is likely to replicate the subversive impact in the automotive field in the field of humanoid robots, changing its original, slow linear growth trajectory.

The speed of progress in training optimization can be referred to Tencent Robotics X robot. By applying cutting-edge pre-trained AI models and reinforcement learning technology to the field of robot control, Tencent has greatly improved the flexibility and autonomous decision-making ability of Tencent Robot Dog Max. The first stage is to collect data on the exercise posture of real dogs, including walking, running, jumping, standing and other movements, Tesla has its own factory as an application scenario to better complete training.

Table: Sales of humanoid robots at Tesla's Gigafactory

Source: Tesla, Veike Robot, Chezhizhi, Tesla, Robot Lecture Hall, Tianfeng Securities Research Institute

3. What do you think of the investment in the humanoid robot industry chain? The core looks at Tesla, and the core hardware components are motors, reducers, lead screws, and sensors

From the perspective of investment, it is more optimistic about the motion control module of Tesla's industrial chain, that is, the hardware part

The humanoid robot industry chain mainly includes the software side (environment perception module, intelligent AI chip module, operating system module) and the hardware side (motion control module). We believe that the software side, especially the algorithm-related high probability Tesla does it itself, so from the perspective of A-share investment opportunities, domestic hardware suppliers have greater opportunities.

Since there is currently no mass-produced humanoid robot, there is no detailed cost splitting in the industrial chain, we look at the cost composition of industrial robots and service robots + combined with the current information available to Tesla Robot, and roughly calculate the value distribution of the humanoid robot industry chain.

From the perspective of the large industrial chain, robots can be divided into upstream core components (accounting for the majority), midstream robot ontology, downstream system integrators, considering that the humanoid robot industry is still in the early stage + high barriers of ontology manufacturing, we believe that the opportunities of domestic enterprises are mostly in the core components, so the following article focuses on the analysis of core components.

Figure: Industrial robot industry chain

Figure: Cost Structure of Industrial Robots

Source: Green Harmonic Prospectus, Huajing Industry Research Institute, Tianfeng Securities Research Institute

The core components of industrial robots are reducers, servos, controllers, and service robots, and the core components are sensors

The three core components of industrial robots are reducer (cost accounting for 25-30%), servo system (cost accounting for about 25%), controller (cost accounting for about 20%), of which the servo system is mainly composed of drives, motors and encoders.

Compared with the characteristics of industrial robots such as specific environment, pursuit of efficiency, and high requirements for accuracy and torque, the working environment of service robots is in the field of greater uncertainty, the application scenarios are more complex, and the requirements for environmental recognition, control algorithms, artificial intelligence, etc. are higher.

The core components and technologies of service robots mainly include lidar, visual recognition, SLAM navigation, servo servos, voice technology, etc., according to the internal data of a service robot we know, the majority of the cost components are sensors (22%), machine parts (20%).

Chart: Cost Composition of Industrial Robots (%)

Figure: Cost Composition of Service Robots (%)

Source: iResearch, TF Securities Research Institute

Tesla humanoid robot torso design & configuration

The humanoid robot unveiled by Tesla on AI Day 22 weighs 73 kilograms and is 172cm tall. Among them, the color is the mechanical part, containing 28 joint actuators and hands, and 57 degrees of freedom in the whole body; The gray contains the brain and torso, which uses the same chip at the end of the car, and the torso integrates thermal management, battery pack, external sensors, etc.

The torso of the Tesla robot Optimus has 28 actuators (divided into 14 actuators for rotary joints and 14 actuators for linear joints) + hands (a total of 12 actuators for both hands).

Photo: Tesla humanoid robot configuration

Figure: Tesla robot drive design (28 joints designed with 6 types of drives are identified in different colors)

Figure: Distribution of Tesla robot drives

Source: Tesla official website, Tianfeng Securities Research Institute

Tesla humanoid robot dexterous hand design & configuration

The reason Tesla robots use humanoid hands is to ensure that objects in our factories can be grasped, and this logic is consistent with why robots have to make human shapes. The world around us is designed to be ergonomic, and scenes and objects are designed according to human characteristics, making them look like humans, which can make robots better suitable for application scenarios.

The Tesla Optimus dexterous hand has 6 drives and 11 degrees of freedom. It has an in-hand controller that drives fingers and receives sensor feedback, which is important for understanding the object we are grasping and proprioception, which is the ability to recognize the spatial location of our hand.

Pictured: Tesla humanoid robot dexterous hand

Source: Tesla official website, Tianfeng Securities Research Institute

Tesla's Optimus components: The core is the motor (hollow cup, frameless torque), reducer/lead screw, and sensor (position & torque)

• Rotary joint: Rotary drive, consisting of frameless torque motor + harmonic reducer (possibly partially replaced by planetary) + torque sensor + position sensor (encoder).
• Linear joint: linear drive, composed of frameless torque motor + roller/planetary ball screw + torque sensor + position sensor (encoder).
• Hand joints: Each hand has 6 drives, consisting of a coreless motor + precision planetary gearbox + encoder + torque sensor.

Figure: Tesla actuator configuration

Source: Tesla official website, Tianfeng Securities Research Institute

The value of the three core parts of TSL robot is split

From the perspective of joints, we expect the value distribution of a single Tesla robot after mass production to be:

• Rotating joints: The total value is 22,000 yuan. Among them, motor 0.7 + harmonic reducer 0.8 + encoder 0.1 + torque sensor 06,000 yuan.
• Rotating joints: The total value is 27,000 yuan. Among them, motor 0.7 + planetary lead screw 1 + ball/T-type lead screw 0.1 + encoder 0.1 + torque sensor 08,000 yuan.
• Hand: The total value is 27,000 yuan. Among them, motor 0.7 + planetary reducer 0.2 + encoder 0.1 + torque sensor 05,000 yuan.

From the perspective of parts, we expect that the value distribution of a single Tesla robot after mass production is: frameless torque motor 14,000 yuan, hollow cup motor 07,000 yuan, reducer (harmonic, planet) 10,000 yuan, lead screw (planetary, ball/T type) 11,000 yuan, encoder 04,000 yuan, torque sensor 18,000 yuan (ordinary, six-dimensional).

Source: TF Securities Research Institute

3.1. Motors (frameless torque, coreless cup)

Frameless torque motor: compact structure, suitable for high-power scenarios, humanoid robot body joints are likely to be adopted

Frameless motors consist of separate rotor and stator components that are individually mounted into the mechanical equipment to transfer torque to the load. They are usually integrated with other components, including housings, shaft trains, bearings, feedback devices and other elements to form the entire drive system. Since the frameless motor eliminates the housing and bearing, the stator and rotor are directly fixed to the mechanical equipment through careful structural design, thereby simplifying the structure of the equipment and reducing the overall space. Ordinary rotary direct drive motors have many points such as saving volume, reducing weight, energy saving and environmental protection, high precision, and maintenance-free. In addition to these advantages, frameless motors further optimize the overall size by reducing volume and weight, and improve system performance by reducing system inertia and increasing system bandwidth.

At present, frameless torque motors are mainly used in collaborative robots. At present, the unit price of frameless torque motor products ranges from several hundred yuan to one thousand yuan, and the price will be different for different power segments and sizes, and the single amount of the product is reflected in the collaborative robot, generally a collaborative robot will use 6 to 7 frameless torque motors.

We believe that the motor used at the joint of the humanoid robot's body is most likely a frameless torque motor. Frameless torque motor is designed for applications with small demand, light weight, low inertia, compact structure and high power, and its adaptability is strong in robot joints, medical robots, etc. The public research minutes of Buke Co., Ltd. said that based on the existing information, frameless torque motor products can meet its functional needs in the use of large joints of humanoid robots.

Figure: Frameless torque motor for collaborative robots

Source: Buke stock announcement, Kollmorgen official website, Buke stock public number, Tianfeng Securities Research Institute

Frameless torque motor: The overseas leader is Kollmorgen in the United States, and domestic suppliers include Buke Co., Ltd. and Haozhi Electromechanical

At present, the frameless torque motor is doing well in the United States Kollmorgen, and its representative product is TBM2G frameless torque motor. TBM2G has the characteristics of high performance, small size, stable performance, and customization. The TBM2G has a maximum torque of 15.1N/m, and the torque density is about 40% higher than that of competitors, and the power density is about 150% higher than that of competitors.

Domestic frameless torque motors do well such as Buke Co., Ltd. and Haozhi Electromechanical.

• Buke shares: In 2016, the first generation of Buke frameless motor was listed with performance comparable to a brand outside the United States, in 2022, Buke launched the third generation of frameless torque motor FMC series based on more user pain points, compared with the same industry products, Buke frameless torque motor currently has 52mm-132mm 10 frame sizes, covering 3-25kg load demand, under the same torque, the size is smaller, and the temperature rise is lower.
• Haozhi Electromechanical: According to its 2022 annual report, the frameless Liju motor developed by the company has the following characteristics: 1) torque fluctuation ≤ 1%, which is more conducive to robot torque control 2) 3.5 times overload capacity, which greatly strengthens the load capacity of the robot.

Photo: Kollmorgen TBM2G frameless torque motor

Photo: Buke frameless torque motor

Source: Buke stock announcement, Kollmorgen official website, Buke stock public number, Haozhi Electromechanical Annual Report, Tianfeng Securities Research Institute

Coreless motor: divided into brushed and brushless, brushless because of higher speed and longer life suitable for application in humanoid robot hands

Brushed coreless rotor motor: The rotor structure breaks through the rotor structure of the traditional motor, and the ironless rotor is adopted, also called the coreless rotor. This novel rotor structure completely eliminates electrical energy losses due to eddy currents formed by the core.

Brushless coreless motor: also known as brushless cogging motor, its stator part uses hollow cup winding, and adopts a coggingless core design.

The difference between brushless and brushed coreless motors: 1) Life: The life of brushed coreless motors lies in the commutator (once damaged, the motor should be replaced), and the life is generally 1000-3000h; The brushless coreless motor adopts bearings, which has a longer life and can reach 10000h+; 2) Speed: brushless coreless motor speed is higher; 3) Control: brushless coreless motors need to be matched with motor drivers to achieve drive control; There is brush core, the motor current control can be.

Brushless coreless motors are mainly used in biomedical, optical fields, robotics and automation (including electric grippers, robot joints, robot hands, semiconductor ZR modules, logistics shuttles, deep well oilfield valves, rail hanging robots), aerospace fields.

Pictured: Brushless coreless motor

Table: Brushless and brushed coreless motors

Source: Electrical Design Process Engineering, Mingzhi Electric Official Website, Tianfeng Securities Research Institute

Brushless coreless motor: high-end product barriers are high, overseas leaders lead the domestic in stability and mass production

Brushless coreless cup motor has compact structure and small volume, wound into a cup, with a size of 6, 8, 10mm; High power density, speed up to 5, 10, 150,000 rpm; The micro motor has the highest efficiency, fast dynamic response, low vibration and low noise, and can realize precise control form.

Brushless coreless cup motor overseas to MAXON to do better, domestic Mingzhi electrical appliances, Dingzhi technology, Hengshuai shares can do. 1) FOREIGN SUPPLIERS: SWISS MAXON, GERMANY FAULHABER, SWISS PORTESCAP, AMERICAN AMP; 2) Domestic suppliers: Mingzhi Electric, Dingzhi Technology (subsidiary of Jiangsu Reilly), Weichuang Electric, Hengshuai Co., Ltd., etc.

The localization rate of high-end coreless motors is low, and the pain point is

• Production technology: the design performance cannot reach the international advanced level, and the manufacturing capacity is not consistent and long-life. 1) Design, Switzerland MAXON has five winding technologies for precision DC motors, and German FAULHABER adopts the diagonal winding design designed by Dr. Fritz Faulhaber; 2) Manufacturing, the main equipment requirements are high, MAXON, FAULHABER production equipment self-developed and self-made, external supply, domestic Mingzhi electrical winding machine purchased from Japan Odawara & Tanaka Seiki, Dingzhi Technology self-developed equipment;
• Mass production capacity: domestic supply can not be supplied in large quantities. MAXON IN SWITZERLAND AND FAULHABER IN GERMANY HAVE ACHIEVED AUTOMATED MASS PRODUCTION, WITH A PRODUCTION CAPACITY OF 20 MILLION UNITS; Domestic manufacturers mainly use semi-automatic / manual production, high rework rate (30 per 10,000 in China, 5 per 10,000 abroad), performance deviation, Dingzhi Technology said that it can achieve automated batch production, but has not yet entered the mass production stage.
• The drive system (control chip, software algorithm) failed to follow up in time.

Table: Comparison of production technologies of coreless motors by major manufacturers

SOURCES: MINGZHI ELECTRIC, MAXON, FAULHABER OFFICIAL WEBSITE, TIANFENG SECURITIES RESEARCH INSTITUTE

Overseas suppliers: the industry leader is Switzerland MAXON, Germany FAULHABER, established early, deep technology

MAXON IN SWITZERLAND AND FAULHABER IN GERMANY ARE INDUSTRY LEADERS, WITH ANNUAL SALES (CORELESS MOTORS + GEARBOXES + ENCODERS) OF ABOUT 3.2 BILLION EUROS

• MAXON, Switzerland: Founded in 1961, it employs around 3,200 people worldwide and is headquartered in Sachseln, Switzerland. 1) Products, 1.2W varieties, including DC brushed and brushless motors, gearboxes, sensors, encoders, servo amplifiers, position controllers; 2) Output, about 5 million pieces; 3) revenue, 626 million Swiss francs (equivalent to 5 billion yuan) in 2021; 4) used in high-demand fields, such as NASA's Mars rover, the driving force source is maxon motor.
• FAULHABER, Germany: Dr. 1947 Fritz Faulhaber GmbH is founded near Stuttgart, Germany, with around 2,300 employees worldwide and is headquartered in Germany. 1) Products, including high-performance DC motors, brushless DC motors, linear motors, stepper motors; 2) R&D and production bases, including Germany, Switzerland, the United States, Romania, Hungary.

Figure: Introduction to Maxon

PICTURED: FAULHABER BRUSHLESS DC MOTOR

SOURCE: MAXON, FAULHABER OFFICIAL WEBSITE, JIANGSU REILLY ANNOUNCEMENT, TIANFENG SECURITIES RESEARCH INSTITUTE

Domestic suppliers: Mingzhi and Dingzhi have cost-effective advantages, and the product performance is close to the overseas level

Product performance: Through the comparison of product performance parameters, it can be seen that the domestic coreless motor is basically benchmarked with the Swiss MAXON and the German FAULHABER. 1) SPEED: MAXON > DINGZHI> FAULHABER> MINGZHI; 2) TORQUE: MAXON IS BETTER THAN FAULHABER, MINGZHI, DINGZHI; 3) Response speed (mechanical time): Mingzhi, MAXON, Dingzhi are better than FAULHABER; 4) WEIGHT: MINGZHI, DINGZHI, FAULHABER ARE BETTER THAN MAXON; 5) Motor efficiency: MAXON, Mingzhi, Dingzhi are above 80%, FAULHABER is slightly lower.

Price: the average overseas price is 3000-5000 yuan / piece, the average domestic price is 1500 yuan / piece, Mingzhi said that the price can be reduced to 600+ yuan / piece during mass production, which has a cost-effective advantage.

Price reduction space: 1) production line investment, according to Dingzhi, the investment in automated production line is about 500-600w/piece; Mingzhi said that the investment amount of a single semi-automatic production line equipment is 280W (winding machine 40W/set *3 + others); 2) Production line output, Mingzhi said that the average annual output of a single device is nearly 3w (one off the line in 5 minutes), and the annual output is 4500w yuan. According to the input-output ratio, domestic manufacturers also have room for further price reduction.

Table: Performance comparison of the same size coreless cup motor from major manufacturers

DATA SOURCE: MAXON, FAULHABER, MINGZHI ELECTRIC, DINGZHI TECHNOLOGY OFFICIAL WEBSITE, DINGZHI TECHNOLOGY PROSPECTUS, TIANFENG SECURITIES RESEARCH INSTITUTE

Mingzhi Electric: The advantage lies in technology accumulation + drive supporting + localized services, which is expected to become the main supplier

The company's main advantages are: 1) Technology accumulation: Mingzhi began to layout hollow cup motor since 2018, up to now the company has 30 hollow cup motor invention patents, product lines include coreless cup motor (body + drive), reducer, encoder; 2) Driver: The driver drives the motor by controlling voltage and current at high frequency, and the technical and algorithm threshold is high. , the technology is benchmarked against Israeli ELMO, which can be combined with the coreless cup motor to form a microdrive system; 3) Localized service: domestic production of coreless motors, can directly connect with North American customers to provide services.

Figure: T motion's motor driver can be combined with a coreless motor

Figure: Company acquisition history

Source: Mingzhi 22nd Annual Report, Mingzhi official website Tianfeng Securities Research Institute

Potential suppliers: Dingzhi Technology/Jiangsu Rayleigh, Weichuang Electric said that there are relevant technologies

Dingzhi Technology/Jiangsu Rayleigh: Micro motor leader, the main advantage is:

• R&D strength: 1) self-made core components and accessories, the company self-made screws, nuts, to achieve independent and controllable core components, R & D and production of high-speed precision planetary gears, break foreign monopoly; 2) Continuous new product development, H1 launched voice coil motors in 20 years, breaking foreign monopolies, entering the supply chain of Mindray invasive ventilators, and launching coreless motors in 22 years, which can realize automated mass production.
• Production capacity: 1) Automated production: the first domestic manufacturer that can realize automated batch production, the company's self-developed high-speed slotless brushless motor manufacturing integration technology, can realize automatic batch production of hollow cup motor, reduce the rework rate, manufacturing capacity has consistency, long life; 2) Production capacity: The production equipment is self-developed and self-made, which can avoid the outbreak of demand, and the enterprise urgently needs capacity expansion but is limited by equipment supply.

Weichuang Electric: The company established a robot industry department in 22Q2, with coreless motor products. At present, it is mainly based on the robot supporting as the starting point, and the execution end is mainly based on low-voltage servo, coreless motor, special frameless torque motor, harmonic reducer and other such products.

Pictured: Dingzhi Technology coreless cup motor products

Source: Dingzhi Technology official website, Dingzhi Technology prospectus, Weichuang Electric announcement, Tianfeng Securities Research Institute

3.2. Reducers (harmonics, planets)

The reducer includes RV, harmonics, planets, and it is expected that Tesla will use harmonics with a high probability, and the lower body may use planets

The reducer connects the power source and the intermediate mechanism of the actuator, which has the function of matching the speed and transmitting torque. According to the control accuracy, the reducer can be divided into general transmission reducer and precision reducer. Generally, the control accuracy of the transmission reducer is low, the precision reducer has a small return gap, high precision, long service life, and is more reliable and stable, which is used in high-end fields such as robots and CNC machine tools.

Due to its small size + high transmission efficiency, the harmonic reducer is suitable for the joints of humanoid robots. The harmonic reducer has the advantages of large single-stage transmission ratio, small size, small mass, high motion accuracy and can work normally in confined space and medium radiation. Compared with the general reducer, when the output torque is the same, the volume of the harmonic reducer can be reduced by 2/3 and the weight can be reduced by 1/2, which makes it have a strong advantage in the robot forearm, wrist, hand and other components.

Planetary reducers are expected to be used in the lower body joints of robots because of their strong impact resistance than harmonics + low cost. According to the announcement of Zhongdaide: the harmonic reducer has the advantages of small size and large transmission ratio, but the impact resistance is slightly worse; The planetary reducer has the advantages of large rigidity, strong impact resistance and low price, but the speed ratio directly affects the size, so the space is limited. From a technical point of view, we believe that both can be applied to humanoid robots, but both will make special design adjustments, not individual parts, but integrated in a modular form, with higher precision and smaller size.

Table: Harmonic and planetary gearboxes comparison

Source: Green Harmonic Prospectus, Zhongdaide Announcement, Tianfeng Securities Research Institute

Harmonic reducer: the global leader is Hamonaco, and domestic green harmonics, Laifu harmonics, etc. can be done

Green harmonics took the lead in breaking the monopoly of overseas harmonic reducers, and now, domestic enterprises have realized the mass production of harmonic reducers. 1) From a global perspective: the leader in 21 years is Hamonaco accounting for 82%, and the green harmonic is 7%; 2) From the domestic point of view: Hamonako accounted for 38% in 21 years. In addition, Zhongda De, Fengli Intelligence, and Haozhi Electromechanical also said that they can do it.

The price of domestic harmonic reducer is declining year by year, currently at 1000-1500 yuan / piece, and the price is expected to be reduced to 500-1000 yuan / unit in the future under the large-scale mass production of humanoid robots. According to the green harmonic prospectus, in 2018 and 2019, the unit price of the company's harmonic reducer was 1885 yuan / unit and 1632 yuan / piece, respectively, and the subsequent company did not separately batch the revenue of harmonic reducer, we expect that the current unit price of domestic harmonic reducer is 1500 yuan / piece, and if the humanoid robot is mass-produced, the price is expected to drop to 500-1000 yuan / piece.

Green Harmonics and Sanhua Intelligent Control announced joint venture expansion in Mexico. The two parties agreed to jointly invest in the establishment of a joint venture in Sanhua's industrial park in Mexico, the main business of the joint venture is the research and development, manufacturing and sales of harmonic reducer related products, and the joint venture is occupied by Green Group.

Figure: China's harmonic reducer market pattern in 2021

Figure: Green harmonic harmonic reducer gross profit margin (data at the beginning of 20 years includes other metal parts, but the proportion is not large)

Source: Huajing Industry Research Institute, Green Harmonic Annual Report, Tianfeng Securities Research Institute

Planetary reducer: the pattern is relatively scattered, and domestic vigorous virtue can be done

According to QYResearch survey, the global planetary gearbox market size was $960 million in 2022 and is expected to reach $1.3 billion in 2029, with a compound growth rate of 5.02%. Planetary gearboxes are mainly concentrated in Europe, Asia and North America, and China will account for 36.68% of the global market share in 2022.

The global leader of planetary reducer is Saiwei Transmission, etc., and domestic enterprises include New's Dart, Hubei Kefeng, Zhongdade and so on.

Chart: Global planetary gearbox market landscape in 2022

Photo: Zhongdaide planetary reducer products

Source: QYResearch, Zhongdaide official website, Tianfeng Securities Research Institute

3.3. Lead screws (T-shapes, balls, rollers)

Lead screws are divided into slide/rolling lead screws, and rolling lead screws are divided into ball/roller screws

Lead screw is a kind of high precision parts, it can accurately determine the coordinate position of the table, convert the rotary motion into linear motion, and transmit a certain amount of power, so there are high requirements in terms of precision, strength and wear resistance.

Lead screws can be divided into three categories according to their friction characteristics: sliding lead screws and rolling lead screws. The tooth shape of the sliding lead screw is mostly trapezoidal. This tooth type has the advantages of higher effect, good transmission performance, high precision and convenient processing than triangular tooth payoff. Rolling screws are divided into two categories: ball screws and roller screws. Compared with roller screws, ball screws have small friction, high transmission efficiency and high precision, so they are more commonly used, but their manufacturing process is more complicated.

There are many manufacturers of rolling functional parts in the mainland, but most enterprises can only produce medium and low-end products with a single variety and low technical content, and there is a certain gap between the manufacturing process level and foreign countries, and the brand with international influence has not yet been formed, and the independent brand needs to be further broken. The world's major rolling screw manufacturers include Bosch Rexroth Hiwin Technologies in Taiwan, Kuroda Precision Industry in Japan, NSK in Japan, Schaeffler in Germany and other companies.

Figure: Lead screw classification

Source: Hengli Hydraulics Announcement, Tianfeng Securities Research Institute

Planetary roller screw: high technical barrier screw, Tesla adopts reverse type

A planetary roller screw is a precision mechanical transmission mechanism that converts the rotational motion of a servo motor into linear motion. It mainly relies on a plurality of rollers and screws, nuts between the thread meshing transmission to transmit power, compared with traditional screws, roller screws and other similar mechanisms, it has strong bearing capacity, large axial rigidity, high transmission efficiency, long service life, good dynamic performance, strong applicability and convenient installation and maintenance and many other advantages, is an ideal functional part of intelligent high-end equipment, in precision CNC machine tools, heavy machinery, medical equipment, weapons and equipment, aerospace and other fields have broad application prospects.

Standard planetary roller screw: the lead screw and nut are triangular multi-head threads, the rollers are spherical single-ended threads with a certain spiral rising angle, and there are straight teeth machined at both ends, and the inner ring gear is fixed at both ends of the nut and meshed with the spur gears at both ends of the rollers.

Reverse planetary roller screw (Tesla scheme): The structural form is similar to the standard type, except that it has no inner ring gear, and the two ends of the lead screw are machined with straight teeth meshing with the gears at both ends of the roller, and the nut as an active part, its length is much larger than the standard type. In general, the nut of the reverse planetary roller screw is the active part, the lead screw is the output component, there is no relative axial displacement between the roller and the screw, which is mainly used for small and medium loads, small stroke and high-speed application scenarios, its biggest advantage is that the nut can be used as a motor rotor to achieve the integrated design of motor and lead screw, forming a compact integrated electromechanical actuator.

Figure: Composition of standard planetary roller screw structure

Figure: Composition of reverse planetary roller screw structure

Source: CNKI, Tianfeng Securities Research Institute

Planetary roller screws: high barriers, stable suppliers are mostly overseas enterprises

The structure of planetary roller screw is similar to that of ball screw, the difference is that the load transfer element of planetary roller screw is threaded roller, which is a typical line contact; The ball screw load transfer element is a ball, which is a point contact. The main advantage is that there are numerous contact points to support the load. Threaded rollers instead of balls allow the load to be quickly released through numerous contact points, resulting in higher impact resistance. The advantage of planetary roller screws and ball screws in terms of load carrying capacity and life is that they can provide higher dynamic and static loads than ball screws, and the replacement of balls by threaded rollers will allow the load to be quickly released through numerous contact lines, resulting in higher impact resistance.

Barriers: 1) Design: Planetary ball screws have certain design barriers (overseas patent restrictions): 2) Raw materials: mainly steel, different suppliers of materials vary greatly, there are differences in formula (trace element addition), there is a large price difference (imported materials are 2-3 times the domestic price); 3) Process know-how: such as the heat treatment stage after material procurement, there are differences in temperature, process and so on.

Suppliers: Domestic enterprises only have a single production capacity, and there is still a big gap with overseas large-scale mass production enterprises, and the gap between domestic and imported products is reflected in: transmission efficiency, transmission accuracy, transmission stability and service life.

• Overseas: Schaeffler in Germany, Exlar in the United States, Rollvis in Switzerland, SKF in Sweden, NTN in Japan.
• Domestic: Dingzhi Technology and Best are in research and development, and Hengli Hydraulics is in the process of expansion. Non-listed companies include Shandong Bote Precision, Changzhou Cisco Drive Technology, and Nanjing Technology.

Table: Planetary roller screw production process

Source: CNKI, Tianfeng Securities Research Institute

Dingzhi Technology, Best and Hengli Hydraulics are all laying out rolling lead screws

Dingzhi Technology can currently make T-type lead screws, and planetary roller screws are under development.

• T-shaped lead screw: independent design of T-shaped screw tooth shape and rolling wheel, using international advanced rolling equipment and production technology, product surface finish Ra<0.2, lead accuracy ± 0.002mm, stroke error <0.1mm@300mm, with good screw accuracy, finish and transmission efficiency, according to the parent company Jiangsu Reilly public platform, the product has been recognized by customers in the sample testing stage, has completed development and achieved a breakthrough in cost, mass production time according to customer progress determination.
• Planetary roller screw: The company's technical department is divided into four platforms according to the research and development field, of which the stepper motor platform is equipped with an independent screw nut transmission group, which is specially dedicated to the research and development of screw and nut. T-screw and planetary roller screw have a certain degree of synergy in technology and process.

Best plans to enter the ball screw track, which is currently in the process of sending samples, with an existing production capacity of 500 million (output value, including guide rails), and next year is considering investing another 1 billion. In January 2022, the company established a wholly-owned subsidiary, Yuhua Seiki to comprehensively layout linear rolling functional parts. The global ball screw (including guide rail) market space is 15-20 billion yuan, the current localization rate is about 10%, the company intends to achieve domestic substitution.

Hengli Hydraulics plans to expand the production of ball screws, planetary roller screws, etc. Hengli Hydraulics announced in 2021 the non-public discovery of share plans, one of the fundraising projects is a linear drive project, with a total investment of 1.5 billion yuan, after the project reaches production, it will form an annual output of 104,000 standard ball screw electric cylinders, 4,500 heavy-duty ball screw electric cylinders,

3.4. Sensors (vision, force control)

Sensors for humanoid robots mainly include vision, force control, and position (encoder)

The sensor is a kind of detection device, which can feel the measured information, and can transform the felt information into an electrical signal or other required form of information output according to certain laws to meet the requirements of information transmission, processing, storage, display, recording and control.

Generally speaking, the joint module is the highest proportion of humanoid robot value, accounting for more than 40%, by combing the possible application location and number of humanoid robot sensors, humanoid robot sensor value proportion or second only to the joint module, of which vision, force control, hearing belongs to the general robot necessary sensors, smell, taste, etc. are non-essential sensors, only specific work scenarios need to be equipped.

• Vision: The point of robot vision is moving the scene, detailed depiction, and high integration.
• Force control: Multi-dimensional torque/force sensing is the optimal solution at present, and electronic skin may be the ultimate solution for touch. The basic feature of electronic skin is that various electronic components are integrated on a flexible substrate to form a skin-like circuit board with high flexibility and elasticity like skin. Electronic skin is a kind of flexible sensor, and domestic Hanwei Technology has been laid out since 2013.

In addition, according to the UCLA paper, in addition to joint current sensors and joint angle encoders, the main sensors used are stereo vision cameras, inertial measurement units (IMUs), and foot contact force sensors.

Figure: Classification of robot sensors

Source: "Design of a Highly Dynamic Humanoid Robot", Robot Lecture Hall, Hanwei Technology Public Account, Tianfeng Securities Research Institute

Vision: Similar to automotive vision, the leading domestic company is Hikvision

Similar to automotive vision, robot vision needs to be depicted in detail in moving scenes, which puts forward high requirements for the perception unit. At the same time, robots, especially humanoid robots, are limited by internal space, and their perception units need to have a higher degree of integration.

The three key technologies common to robot autonomous navigation and automotive autonomous driving are environmental perception, planning decision-making and execution control, autonomous driving and robots mainly obtain surrounding environment information through sensors, and will also expand environmental awareness capabilities through high-precision maps and IoT technology. Commonly used visual and environmental perception sensors are cameras, radar (millimeter wave, laser, ultrasonic, etc.), infrared sensors, GPS, etc.

• Camera: The robot's eyes, including normal monocular cameras and depth cameras.
• Radar: Lidar, millimeter wave radar, ultrasonic radar.
• Infrared sensor: The basic principle of ranging of infrared sensors is that the light-emitting tube emits infrared light, and the light-sensitive receiving tube receives the reflected light from the object in front of it, according to which it is judged whether there is an obstacle ahead. According to the intensity of the emitted light can judge the distance of the object, its principle is that the light intensity received by the receiving tube changes with the distance of the reflected object, the distance is close to the reflected light intensity, the distance is far away is the reflected light is weak. P Overseas companies are mostly comprehensive suppliers with a comprehensive layout, while mainland sensors started relatively late, and the development of high-end sensors lagged behind developed countries such as Europe, the United States, Japan and South Korea, but leaders such as Hikvision have appeared. From the business point of view, excellent enterprises in the field of domestic visual sensing can provide a complete set of visual sensing solutions from hardware to algorithms.

The amount of vision sensors for a single humanoid robot is generally 2-4, and whether it has an algorithm has a greater impact, such as Hikvision's sensor with a vision algorithm The price is about 10,000 yuan / piece. But Tesla has opened up the underlying modules of FSD and robots to achieve a certain degree of algorithm multiplexing. FSD algorithms use sensor data for environmental perception, and these sensors can also help robots perceive their surroundings, identify objects, people, and obstacles.

Force control: Multi-dimensional force sensors have a large gap at home and abroad, and there are no leading enterprises

At present, the robot force control scheme is roughly divided into three categories, of which multi-dimensional torque/force sensor force control is the current optimal solution, and we expect Tesla robots to use single-axis torque sensors in body joints and 6-axis force sensors on feet.

• Current loop force control: force closed-loop feedback control through the current closed loop of the motor, which is suitable for direct drive motors or application scenarios with small reduction ratios. Its advantages are low cost and do not require the growth of additional sensing equipment, while the disadvantages are low accuracy, slower response, and limited application scenarios.
• Torque/force sensor force control: The force sensor in the robot is mainly related to the [single-axis torque sensor] of the joint part and the [6-axis force sensor] at the end of the robot actuator, which measures the internal force of the robot and the force of the interaction between the end effector and the external environment. Tesla Optimus equipped a torque sensor at the rotating joint actuator, and at the shareholder day, it showed a video of the manipulator controlling the force without breaking the egg, showing excellent motor torque control ability.
• Powered (elastomer): An elastomer is added to the end effector to measure the moment by the deformation of the elastomer, or to apply a constant force to the terminal. This solution is commonly used in low-cost and high-integration scenarios.

In the field of force control sensors, there is a large gap between domestic enterprises and overseas leaders, and at present, domestic Yuli is doing the best, but there is still a certain gap with overseas leaders. 1) Overseas: TDK, ST, ADI; 2) Domestic: Unicom Instruments (unlisted), Kunwei Technology (unlisted), Hyperson (unlisted),

Figure: Force transducer

Source: Robot Lecture Hall, Keli Sensing Announcement, Tianfeng Securities Research Institute

Force control: IMUs for balanced and stable walking deserve attention

The UCLA paper mentions the IMU as a key sensor for balanced and stable walking. Typical IMU (Inertial Measurement Unit) sensors include a three-axis accelerometer and a three-axis gyroscope. Optionally, a three-axis magnetometer for true north reference, commonly known as the attitude and heading reference system (AHRS), is also included.

IMU is an inertial sensor that has emerged with the continuous development of sensor technology and microelectronics technology, which belongs to a kind of MEMS and is a technology with a high degree of integration.

According to Yole statistics, in 2021, the global high-performance MEMS inertial sensor market size is about 4.5 billion yuan (including MEMS inertial sensor systems), and the share is concentrated in the hands of industry giants such as Honeywell, ADI, Northrop Grumman/Litef, and the top three companies together account for more than 50% of the share.

Table: MEMS Product Types and Market Structure

Chart: Global MEMS Inertial Sensor Market Pattern (100 million yuan, %)

Source: Design of a Highly Dynamic Humanoid Robot, Silicon Link's Prospectus, Tianfeng Securities Research Institute

4. Investment advice

Investment advice: The humanoid robot track has long slopes and thick snow, catalyzing succession, and Tesla is expected to mass produce next year

The humanoid robot industry has high barriers (high requirements for enterprise capital, technology and resource integration) + high ceiling (based on the global population of 8 billion) + long industrial chain, which is a rare track in the future that is expected to surpass the previous consumer electronics-Apple industry chain and Tesla-electric vehicle industry chain.

Standing at the current point in time, we are optimistic about this industry for reasons:

• Software level: The next wave of artificial intelligence proposed by NVIDIA founder Jensen Huang is "embodied intelligence", that is, intelligent systems that can understand, reason, and interact with the physical world; Different from traditional robots that only have "physical" execution, large models can introduce multimodal input by jointly training images, text, and embodied data, thereby improving the interaction and learning ability of the large model and environment, and injecting "soul" into the robot.
• Excellent supply level: Tesla's robot Optimus motion control ability evolution is obvious + Musk is optimistic about the long-term vision of humanoid robots. Optimus can currently pick up objects, discover and remember the environment, and imitate human actions based on AI, and can complete complex tasks of sorting items.

Tesla progress: According to its Tier 1 supplier feedback, Tesla robots are now in the B sample test, it is expected to end the B sample in May next year, set the C sample & price in July, pull the production line in September, and if all links go well, 10,000 units are expected to be shipped next year.

Investment advice: Tesla's Optimus is the main line of this round of the market

Different from last year's investment in household savings (ultra-long honeymoon period with good money-making effect), the prosperity is upward - the performance is revised up - the stock price is rising, the robot's investment this year is still the 0-1 stage, the market is more extreme, our investment ideas and views are as follows.

What is most important in the 0-1 phase? The confirmation of prosperity (establishing direction) and industrial space (establishing height), landing in the robot industry chain, many investors interpret that the fundamentals have not changed compared with last year, just the spillover of AI investment. You may wish to listen to the voice downstream of the industry chain:

• Domestic not only Xiaomi, many large manufacturers such as Huawei, Tencent, etc. are also joining, but as an important encryption industry, there is less communication with the secondary market. The benefits brought by external AI large models are to lower the threshold of simulation and simulation, on the one hand, from rule-based programming to image, video-based training, that is, training to improve cognition, and internal data iteration will be discussed later. On the other hand, AI lowers the threshold for robot research and development (for example, a small team, 1 software programmer replaces 3 electrical engineering doctors, and solves the problem that required several years of research and development in 2-3 months), which is conducive to accelerating evolution.
• Another big change this year, more domestic supply chains have emerged (joining the T chain / want to join the T chain + really start investing in the production line), and the academic cost reduction plan to feed back the industry has begun to appear.

From an investment point of view, instead of spending energy to spread other downstreams, it is better to focus on Tesla's optimus changes and progress, which is the fundamental thing to promote the market. It has also been repeatedly said before that Tesla has data migration (algorithm), car supply chain migration (cost reduction), and the closed loop of the scene is the most likely to run a large item. The recent addition of attention is the release of FSD V12 (the demo version has begun to be trialed internally V11.4), because Optimus will use the same neural network "occupation network" as FSD to identify actionable areas, analyze and plan according to the external environment perception information input collected by the sensor, find out a reasonable trajectory of action and avoid obstacles, and finally complete the action through the actuator.

Investment advice: How to choose Tesla's Optimus supply chain?

The specific supply value of Tesla Optimus joint assembly is as follows:

• Linear Joint Assembler A: We expect the value of 2-3W after mass production, mainly responsible for linear joint integration (rotation in research). We expect that the core components are motor (made by yourself, 500*14) + planetary roller screw / ball screw (do it yourself, it is expected that the planetary roller screw will have to reduce the cost to the level of 1,000 yuan, we use 1200*8+200*6).
• Rotation & Linear Joint Assembly Vendor B: We expect the value of 2-3W after mass production, responsible for rotation + linear joint assembly (rotation is more advantageous at present). We expect that the core parts of the rotary joint are motor (500*14) + harmonic reducer (joint venture with Green) + encoder (100*14), and the core parts of the linear joint are motor (500*14) + planetary screw/T-screw + encoder (to be made by yourself, 100*14).
• Hand Joint Assembler C: We expect a value of 1-1.1W after mass production, responsible for hand joint assembly. We expect that the core components include coreless motor (600*12) + precision gearbox (most likely to do 200*12) + encoder (100*12).

Certainty of comprehensive value quantity & barriers. We first recommend Tier1 suppliers [Tuopu Group], [Sanhua Intelligent Control], [Mingzhi Electric]. We believe that under the global sales of 1 million robots, assuming that the value of Tuopu & Sanhua & Mingzhi is 2.5/2.5/11,000 yuan / robot, the market share is 50%, the net interest rate is 15%, and the robot business gives 50X valuation, and the upward elasticity of the above three targets is 50%+.

Secondly, we recommend [Double Ring Drive] and Tier 2 supplier [Green Harmonic] that is expected to become a Tier 1 supplier. Shuanghuan has the ability to produce planetary/harmonic reducers, and we expect the mass production price of planetary/harmonic reducers to be 200/600 yuan per piece.

Finally, it is recommended that the elasticity is large+, but Tesla has not yet been clearly contacted, but there may be opportunities &is expected to enter other supply chains in the future. 1) Lead screw: [Dingzhi Technology], [Best]; 2) Reducer: [Zhongdaide]; 3) Motor: frameless torque-[Buke Co., Ltd.], Hollow Cup-[Dingzhi Technology] (Jiangsu Leili), [Hengshuai Co., Ltd.], [Weichuang Electric]; 4) Sensors [Keli Sensing], [Xindong Lianke], etc.

Source: TF Securities Research Institute

5. Risk warning

The development of humanoid robots is not as expected: This article discusses the development and volume of humanoid robots, if the industry progress is not as expected, it will affect the logical deduction of the whole text.

Tesla humanoid robot progress is less than expected: The optimism of this article is that Tesla leads the development of humanoid robots, if Tesla's progress is not as expected, it will affect the subsequent recommendation logic.

Domestic suppliers entering Tesla's humanoid robot supply chain is not as expected: Tesla's current humanoid robot has not yet been mass-produced, suppliers are also to be determined, this article discusses suppliers that may enter, if these suppliers enter Tesla's supply chain is not as expected, it will affect our recommendation.

Estimation has a certain subjectivity: this article involves a large number of space, value amount, profit margin measurement, the core assumption has a certain subjective component, if not as expected, will affect the logic of the whole text.