(Report produced by Author: Zhongtai**, Feng Sheng, Cao Senyuan).
Sensors are the key to the perception layer of humanoid robots
The embodied intelligence of humanoid robots attaches more importance to the perception layer and the interaction layer. Humanoid robots emphasize embodied intelligence, that is, agents that can perceive and understand the surrounding environment and complete tasks through autonomous learning. Sensors and software are the key to embodied intelligence, and the perception and cognitive layers are the threshold for robots to move towards embodied intelligent robots, and the rapid iteration of machine vision and polymorphic language large models is expected to greatly improve the perception and cognitive ability of robots. The sensor is the core of "perception", and its quality and technical level will directly affect the performance and stability of the robot. Robot sensors can be divided into internal sensors and external sensors according to the different detection objects. The sensors are used to sense the state of the robot itself, such as speed, position, etc.;External sensors are used to perceive the surrounding environment of the robot, such as vision, hearing, touch, smell, pressure, temperature, humidity, distance, etc.
Sensors on Tesla's humanoid robots account for nearly 30% of the cost. In a robot, the amount of sensors is huge, in addition to cameras (image sensors), radar (laser, millimeter wave, ultrasonic), six-dimensional torque sensors, encoders, tactile sensors, there are position, speed, acceleration, balance, force, touch, vision, hearing, proximity, distance, smell and so on more than 10 kinds of sensors are used in humanoid robots. According to the data of the sensor expert network, the cost of various sensors on the Tesla Optimus robot accounts for nearly 30%.
Force, flexible haptics, inertial sensors have high barriers and large value, and domestic manufacturers are ready to go
The torque sensor is an important part of the robotic arm to perceive the force, and its application prospects are broad. According to the measurement dimension, the torque sensor can be divided into one to six dimensional torque sensors, in the humanoid robot, the dexterous hand or will use a miniature pressure sensor or a micro six-dimensional force sensor, the wrist and ankle with high requirements for flexible control may use a six-dimensional torque sensor, and other joints of the body will use a joint torque sensor. According to data from high-tech robots, force sensors that simulate the touch sensation of joints and limb ends, for example, account for about 15% of the cost of humanoid robots. The R&D and manufacturing process of six-dimensional force sensors is difficult, and the domestic production capacity has been obtained, and the space for cost reduction is expected to be opened.
Flexible tactile sensors are expected to further improve the perception capabilities of robots. The tactile sensor can provide pressure, temperature, and other information for the robot's perception, further improving the robot's perception abilityThe electron** is composed of a combination of point-like tactile sensors, which are mostly arranged in a matrix. Due to the characteristics of high flexibility, scalability and high elasticity, electronic ** is also known as flexible tactile sensor. At present, domestic companies have the production capacity of flexible sensors, but due to difficult processing technology, high cost and other problems, it is not yet possible to apply it in batches in robot solutions.
Inertial sensors are expected to become standard in robots. Inertial sensor, also known as inertial measurement unit (IMU: inertial measurement unit), can be fused with multi-sensor data such as cameras and force sensors on the robot to maintain body balance, speed and trajectory, and positioning and navigation, etc., which are expected to be standard on quadruped robots and humanoid robots. The technical barriers of inertial sensors are high, and the current price of a single high-precision IMU is about several thousand yuan to 10,000 yuan, and the value is high.
The MEMS market is in the process of rapid growth
The global MEMS market has an average CAGR of 8-9%. According to Yole Intelligence data, the global sales volume of MEMS increased from 201 in 20186.8 billion to 303 in 20215.9 billion and is expected to reach 487 in 20270.8 billion, with a CAGR of 8 from 2021 to 202720%。The global market size for MEMS has grown from 99 in 2018$9.4 billion increased to $13.5 billion in 2021$9.5 billion and is expected to reach $222 by 2027$5.3 billion, with a CAGR of 856%。
The growth rate of the domestic MEMS market is higher than the global average growth rate. Benefiting from the implementation of strategies such as industrial Internet of Things, intelligent manufacturing, and artificial intelligence, coupled with the acceleration of smart city construction, intelligent manufacturing, and smart medical development at all levels, the MEMS market has shown a rapid growth trend at home and abroad. According to the China Business Industry Research Institute;From 2018 to 2022, the CAGR of China's MEMS industry market size is 1482%, the growth rate in recent years is still improving, and the scale of China's MEMS market will reach 100.8 billion yuan in 2022, a year-on-year increase of 2029%, the growth rate of the domestic market is significantly higher than that of the overseas market.
The robot acquires its "sense of touch" through a force control system
The force control system gives the robot a "sense of touch". Corresponding to the human sense of touch, the tactile sensing system of a robot is a device or system that can measure a given property of an object through contact. In general, the tactile perception of a robot is related to the measurement of forces in a predetermined area. In order to improve the application effect of robots, robots should also be equipped with advanced tactile perception systems to enable them to perceive their surroundings, stay away from potentially damaging effects, and provide effective information for subsequent tasks such as hand operations.
Force position control with force sensor as the core is one of the main control methods of the robot. Force control generally refers to the force hybrid control technology that uses the force sensor as a feedback device in the field of robot application, combines the force feedback signal with the position control (or speed control) input signal, and realizes the force position mixing control technology through the relevant force position mixing algorithm. It is also known as force position hybrid control technology, referred to as force control.
Force transducers are the best solution for precise and direct force control
Direct force control is achieved by means of different force feedback methods, of which the use of multi-dimensional torque transducers is the best solution. a.Current loop: The external force is estimated by the dynamic model of current feedback and identification, which is suitable for direct drive motors or application scenarios with small reduction ratios, such as small impedance control of human-computer interaction robotic arms and small quadrupeds. For robots with large reduction ratios, the friction of the joint back-drive is huge, and the motor current is not enough to accurately estimate the external force received on the end effector. The advantage is that there is no need for additional sensors, and the disadvantage is that the force control accuracy is poor due to friction, and the application scenarios are limited. b.Elastomers - strong resistance to external impacts: SEA (series elastic drive) uses elastic deformation and high-precision position sensors to indirectly measure the size of the force distance, which is often suitable for applications with high integration of humanoid robots and high requirements for drive output torque. c.Force sensor: This is achieved by attaching a force sensor to the end of the robotic arm. Compared with the one-dimensional force sensor, the six-dimensional force sensor has more force control dimensions, can measure all the forces of the object in three-dimensional space, can measure the application load in any direction and any axis, and can withstand the overload of 5 to 20 times the rated measurement range, and the accuracy can even reach 001n, correspondingly** is also more expensive. Typical application examples include robotic arms such as the KUKA LBR IIWA manipulator and Tesla's Optimus humanoid robot.
Classification of six-dimensional force transducer structure - Strain gauges are the mainstay, and MEMS technology is expected to become the mainstream
Strain gauges are the heart of resistive strain gauge multidimensional force transducers. Strain gauges are composed of high-resistance metal wires, high-resistance metal foils, or semiconductors arranged in a grid shape and glued to an insulating substrate. Strain gauges include metal resistance strain gauges and semiconductor strain gauges (i.e., silicon strain gauges).
Metal foil resistance strain gauges are currently more widely used. The resistance change is caused by the change of size (geometry), and the types include wire, foil and metal thin film strain gauges, among which the most widely used at present is the foil strain gauge, which has the characteristics of small transverse effect, large allowable current and high production efficiency. However, such strain gauges still need to be polished manually, which takes a long time, has low production efficiency and high cost.
The comprehensive performance of silicon strain gauges is better. In semiconductor materials, the piezoresistive effect is about 50 times more significant than that of metal strain gauges, and the silicon strain coefficient is as high as tens to 100 (2 3 for metal strain gauges). After comparing the five dimensions of stability, stiffness, dynamic characteristics, cost and signal-to-noise ratio, the stability, signal-to-noise ratio and dynamic characteristics of silicon strain gauges are better than those of metal strain gauges, and the difference between the two is not big in stiffness, and the metal is slightly better in cost, but in recent years, the process of silicon strain gauges has been improved and improved, and the comprehensive cost is also greatly reduced.
Flexible tactile sensors: Helping robots take their perception to the next level
The application of "electronics**" is expected to give robots a richer sense of touch and further upgrade their perception capabilities. There are many receptors composed of different types of neurons in the human body**, among which Merkel cells and Ruffini bodies mainly sense continuous low-frequency pressure, and the two-touch sensor is an imitation of these two tactile receptors in the human body. The principle of the tactile sensor is to convert tactile stimuli into electrical signals to provide pressure, temperature, and other information for robot perceptionThe electron** is composed of a combination of point-like tactile sensors, which are mostly arranged in a matrix. At the same time, because the electronic ** is made of flexible materials, it has the characteristics of high flexibility, expandability, and high elasticity, so it is also called a flexible tactile sensor.
Demand side: The demand in the consumer electronics and medical fields drives the rapid growth of the flexible sensor market
The global flexible sensor market space is expected to exceed 50 billion yuan. According to the data of Hanwei Technology's WeChat***, according to the statistics of relevant institutions, the global flexible sensor market CAGR from 2021 to 2028 will reach 68%, and the market space is expected to reach 84 in 2028$700 million. Among them, the field of flexible electronics is expected to become the largest demand increment, according to Frost & Sullivan**, the global flexible electronics market CAGR is expected to reach 144 from 2019 to 202571%, and the market size will reach 3049 in 2025$400 million.
The scale of China's flexible sensor market is showing a rapid growth trend. In 2017, the market size of China's flexible sensor industry was about 700 million yuan, and the market size of China's flexible sensor industry in 2022 will be 211.2 billion yuan, with a CAGR of more than 23% from 2017 to 2022. Among them, the market is mainly distributed in North China, Central South China, and East China, accounting for respectively. 79%。In 2022, the domestic demand for flexible sensors will increase by 16% year-on-year2%, production increased by 26% year-on-year9%，Supply side: The flexible sensor industry is accelerating, and domestic companies are gradually following up
The processing technology has been upgraded, and the development of flexible sensors has become more and more mature. Silicon piezoresistive sensors based on MEMS have the characteristics of high precision and small size, and in recent years, MEMS sensors with flexible substrates have made great breakthroughs, and flexible substrates have good flexibility, impact resistance and mechanical properties. At the same time, the development of polymer microfabrication technology, electronic printing technology, and 3D printing technology has subverted the traditional design and manufacturing methods, and specific conductive materials can be injected into polymer materials, which proposes a new solution for flexible sensor manufacturing.
The domestic electronics industry has gradually followed up. The development of electronics in the world is still in the early stage, but on the whole, overseas companies in the sensor industry are leading. The market share of overseas companies such as Novasentis, Tekscan, and JDI is still in the top position, and according to QY Research data, the world's top eight flexible tactile sensor companies will account for 57 in 20221% market share, high market concentration;Among the domestic enterprises, Hanwei Technology's subsidiary, Nengstar, has independent intellectual property rights in the field of flexible piezoelectric sensors, and flexible micro-nano sensors have been clearly applied in the field of intelligent robots. In addition, Pacini Sensing has been deeply involved in the field of tactile sensors, and the flexible sensing array of the multi-dimensional tactile sensor PX-6AX developed by Pacini Sensing can provide additional information such as sliding, friction, texture, and temperature for robots.
What are inertial sensors and IMUs?
MEMS accelerometers typically consist of a mass, a damper, an elastic element, a capacitive plate (for capacitive accelerometers), and an ASIC chip (application-specific integrated circuit chip). Depending on the measurement dimension, accelerometers are divided into three types: uniaxial, biaxial, and triaxial. Compared with single-axis and two-axis accelerometers, which can only detect changes in the motion state of the plane, three-axis accelerometers can realize the measurement of acceleration in three-dimensional space with a single product, so as to meet the application needs of miniaturization and more fields.
Classification: According to the sensing principle, MEMS accelerometers can be divided into piezoresistive, capacitive, and thermal types. Capacitive MEMS accelerometers dominate the market due to their high sensitivity, high accuracy, and low temperature sensitivity.
Working principle: Capacitive MEMS accelerometer is a comb-like structure that combines a mass and a movable plate into a comb-like structure, and forms a parallel capacitive plate structure with a variable distance from the fixed plate. When the change of the motion state of the measured object drives the mass in the chip to move, it will drive the movable plate to move, so that the distance between the movable plate and the fixed plate changes, resulting in the change of the capacitance value between the two plates. At this time, the ASIC chip converts the change in capacitance value into an electrical signal and calculates the acceleration physical quantity.
What role does IMU play in humanoid robots?
The role of IMU in humanoid robots is mainly reflected in three aspects: precise attitude control and balance maintenance, navigation and positioning, action execution and path planning, and its application direction lies in inertial navigation and measurement and stable control. Attitude control and balance: Inertial stability control is to continuously monitor the attitude and position changes of the system, and use the servo mechanism to dynamically adjust the attitude of the system to help the robot maintain a stable attitude and balance. Humanoid robots need to precisely control their posture and balance in performing various actions. Navigation and positioning: Through the time integration and superposition of diagonal rate and linear acceleration, it can dynamically determine its own position change, which has the advantage that it can be used independently without the help of external information. At the same time, IMU can be combined with satellite positioning to provide accurate indoor and outdoor positioning information, and the ability of humanoid robots to navigate and locate in complex environments is greatly improved. Action execution and path planning: Based on IMU's measurement data, humanoid robots can perform various actions more precisely, such as walking, turning, jumping, etc. In addition, IMU data can also be used for path planning to help robots avoid obstacles and plan more reasonable motion trajectories.
Keli Sensing: a leading enterprise of strain gauge sensors, with a market share of the first place in China
Founded in 1995, Keli Sensing is a leading enterprise in the intelligent sensor industry, mainly developing and producing various types of physical quantity sensors, as well as different industrial Internet of Things systems and multi-scenario application solutions. Starting from mechanics, the company integrates multi-physics technology, develops multi-variety sensing, and carries out multi-dimensional industrial layout and operation. At present, the company has become one of the world's largest steel sensor manufacturers and one of the leaders in the expansion of industrial Internet of Things applications.
The main products include strain gauge sensors, miniature, torque, multi-dimensional force and other high-end force sensors, and the market share of domestic force sensors is the first. Focus on following up and reserving a number of photoelectric sensors, temperature and humidity sensors, and high-end mechanical sensors, and accelerate the transformation and development from a single physical sensor enterprise to a platform enterprise with multi-physical sensor fusion. Samples and trial production of a variety of torque sensors, multi-dimensional force sensors and other products have been carried out (in the field of humanoid robots). The product range is complete, with high precision, good stability and other characteristics.
AVIC: The first defense enterprise to establish a strain gauge production line
Founded in 2002, AVIC is an enterprise mainly engaged in providing military and civilian intelligent measurement and control products and system solutions. Relying on the self-developed strain gauge technology, through the network testing and management of the production line, the world's leading sensitivity high and low temperature detection and control methods have realized the compensation and adjustment of the sensors one by one, and occupy a dominant position in the domestic high-end strain gauge and sensor market, and the products are sold to North America, Europe and other parts of the world. It focuses on serving the defense market such as aviation, aerospace, weapons, shipbuilding, and nuclear industry, as well as civilian industries such as universities, scientific research institutes, electric power, and high-speed rail.
The main products are resistance strain gauges, strain gauge sensors, and automotive comprehensive performance testing equipment. It has independently developed more than a dozen kinds of high-precision weighing and load sensors with independent intellectual property rights, and has established a leading position in the field of domestic weighing and load cell manufacturing and research and development.
Innosilicon: a domestic manufacturer of high-precision MEMS sensors
The company developed the first generation of MEMS gyroscopes in 2015 and the third generation of mass production in 2019, and expanded the field of MEMS sensors on this basis, mainly downstream in the fields of high-end industry, unmanned systems and high reliability. At present, the core parameters of the inertial sensors produced by the company have reached the international advanced level, which can compete with overseas brand products such as Honeywell and ADI, and some of the company's products have the best advantages. From 2020 to 2022, the company's comprehensive gross profit margin will maintain more than 85%, and the net profit margin will maintain 45%-55%, which also proves that the company's products have strong competitiveness in the market.
The company continues to invest heavily in R&D, and the structure of core technical personnel is stable. The company's R&D expenditure continues to account for more than 20% of the overall revenue, and the number of R&D employees will account for more than 50% of the company's total number in 2022, which has the characteristics of a R&D company. The company's core personnel have rich work experience in the field of semiconductor and chip design, and hold the company's equity, and the personnel structure is relatively stable.
This article is for informational purposes only and does not represent any investment advice from us. To use the information, please refer to the original report. ）
Featured Report**: Think Tank for the Future].