Although the market is sluggish, we must be active, on the one hand, actively grasp the opportunities of the medium- and long-term investment layout adjusted by the market, and on the other hand, bow our heads and rush to do a good job in investment research.
The analysis shares some of our research and views on intelligent driving.
First, understand intelligent driving
1. Classification of autonomous driving
There are multiple levels of autonomous driving technology, and different agencies have proposed different standards, but the most commonly used are those proposed by the U.S. Highway Safety Administration (NHTSA) and the Automotive Engineers Association (SAE).
Although the two standards differ in specific levels, they both believe that from level 3, the main responsibility for driving shifts from the person to the vehicle itself, which also makes level 3 a key node in the development of autonomous driving technology.
In the domestic "Classification of Automotive Driving Automation", L0 to L2 is regarded as an automatic driving assistance system (ADAS), while L3 and above is considered to be advanced automatic driving.
At present, the global autonomous driving is in the process of transformation from L2 to L3: At present, advanced autonomous driving is mainly used in limited scenarios, such as smart parks, port terminals and parking lots, including commercial vehicle logistics, automatic parking and other specific fields.
If Level 3 autonomous driving can be realized, it will be a big step towards full autonomous driving, or it will bring about the growth of demand for various high-end supporting equipment.
SAE grading standards for autonomous driving technology
2. The evolution direction of autonomous driving technology
Autonomous driving technology evolves along the two dimensions of intelligence and networking, just like people, one direction is to make the car itself smarter, the car can be smarter, can see the surrounding environment clearly through the camera, radar and other "eyes", and make decisions through the "brain" of the on-board chip;
The other direction is to enable cars to communicate with other people, cars, roads and even the cloud through "social networks" such as the Internet of Vehicles (V2X) to achieve better cooperative driving.
However, V2X relies on the construction of roadside cloud infrastructure, and it is expected that in the short to medium term, it will still be dominated by bicycle intelligent solutions.
Second, the intelligent driving industry chain
The intelligent driving industry chain involves multiple entities, including vehicle manufacturers, parts suppliers, in-vehicle platform developers and mobility service providers.
This huge industry chain brings together a variety of players, from new automakers to traditional automakers, to technology giants and startups.
Source: Intelligent Driving Industry Chain, Where is Autonomous Driving Going? 》Guojin Securities
1. Chip: the brain of intelligent driving
Autonomous driving chip is the most core hardware, the added value is also very high, the market competitors are mainly domestic and foreign data technology field of leading companies, which are either foreign companies, or not yet listed companies, only in the domain controller field has a small number of A-share companies involved.
2. Vehicle-mounted sensors: pure vision and lidar route dispute
The sensor is the core hardware of the autonomous driving perception layer, equivalent to the eyes of the smart car, mainly using cameras, lidar, ultrasonic radar and millimeter-wave radar and other equipment to perceive the environment around the vehicle in real time and obtain the distance and contour information of surrounding objects.
According to TF Securities' forecast in June 2023, by 2030, the domestic vehicle camera market will reach 38 billion yuan, and the global market size will reach 114.1 billion yuan, with an annual compound growth rate of 17.5%.
In terms of cost, CIS sensors accounted for 52%, and optical lenses and module packages accounted for 39%. (Source: TF Securities, 2023.06)
When it comes to automotive radar, lidar is a key technology.
Source: Comparison of the advantages and disadvantages of different types of sensors, "Triple Inflection Point Approaching L3 Landing Acceleration" West China Securities
From the perspective of technology development path, the autonomous driving perception scheme is mainly divided into two routes: pure vision and lidar. The pure vision solution mainly relies on cameras, supplemented by sensors such as millimeter wave radar and ultrasonic radar.
The overall cost of this solution is low, typical of a leading company in the United States. However, vision-only solutions highly rely on deep learning algorithms for the recognition of objects and their distances.
Therefore, the vision scheme has very high requirements for the algorithm and requires a large amount of training data to continuously improve the algorithm.
The lidar solution takes lidar as the core, and combines sensors, such as cameras, millimeter-wave radar, and ultrasonic radar.
This scheme is characterized by strong perception and low algorithm requirements.
Multi-sensor fusion solutions compensate for some of the shortcomings of pure vision solutions, but they are relatively expensive and often need to be combined with high-precision maps to achieve better results.
3. High-precision map: Reduction may become a trend
High-precision map is a map data paradigm for autonomous vehicles, with an absolute position accuracy of nearly 1 meter and a centimeter-level relative position accuracy, which can accurately and comprehensively express road characteristics.
In lidar perception scenarios, HD maps may be essential.
However, the current reduction of high-precision maps may become a trend.
On the one hand, the coverage, cost, and frequency of updates of high-precision maps limit their application in the mass production of autonomous driving at the city level.
On the other hand, high-precision map surveying and mapping belongs to the field of strong supervision, and the promulgation and implementation of relevant policies and regulations have also increased the cost of enterprises using high-precision maps.
As a result, several companies have announced in recent years that they are looking for alternatives to high-precision maps.
In April, some companies announced an advanced autonomous driving system that does not rely on high-precision maps and installed it in certain models, and several other companies have proposed plans to completely abandon high-precision maps by 2023.
4. Domain controller: from distributed to centralized
An automotive domain controller is an electronic control unit that integrates multiple functions to manage and control various systems of a vehicle.
With the development of automotive electronics technology and the increase of functions, the traditional single control unit can no longer meet the demand.
As a result, automakers began to use multiple control units to manage the system, but the communication and coordination between them required complex software algorithms and hardware design, which increased development costs and complexity.
Domain controllers, on the other hand, integrate the functions of multiple control units into a single control unit, simplifying system architecture and reducing costs.
From the perspective of market share in 2022, the proportion of domestic enterprises is relatively small, but the layout is very positive, and it is expected to achieve a leading position in the domain controller market.
5. Drive-by-wire chassis: a key technology towards L3 and above autonomous driving level
The traditional automobile chassis is mainly composed of four parts: transmission train, driving system, steering system and brake system, which are interconnected and complement each other.
The by-wire chassis is a by-wire modification of the chassis signaling mechanism, replacing mechanical signaling with electrical signaling, making it more suitable for autonomous vehicles.
Specifically, the driver's operating commands are transmitted to the electronic controller through the by-wire chassis, and then transmitted by the electronic controller to the actuator, and finally complete the functions of steering, braking, and driving.
The by-wire structure replaces the mechanical connection between the steering wheel, brake pedal and chassis, and realizes multi-source electrical signal operation, so that the by-wire chassis has the advantages of high precision, high safety and high response speed.
It can be said that the by-wire chassis is a necessary condition for achieving advanced autonomous driving at L3 and above.
The by-wire chassis mainly includes brake-by-wire, steer-by-wire, suspension, throttle-by-wire and shift-by-wire. According to the forecast of Toubao Research Institute in April 2023, by 2026, the scale of China's by-wire chassis industry is expected to grow from 18.06 billion yuan in 2021 to 115.95 billion yuan in 2026, with a compound annual growth rate of 45.0%.
Scale of China's by-wire chassis industry (Source: IFIND, Toubao Research Institute, April 2023)
1) Brake-by-wire: 30% penetration rate of new energy vehicles expected in 2025
The park-by-wire brake (EPB) has undergone a development process from cable type to integration.
It is expected that by 2025, the domestic EPB market will reach 22 billion yuan, and there is a large space for domestic alternatives.
Drive-by-wire service brakes include electro-hydraulic braking systems (EHB) and electro-mechanical braking systems (EMB).
The market prospect is broad, and it is expected that by 2025, the penetration rate of drive-by-wire driving brakes in new energy and fuel passenger vehicles will reach 30% and 16%, respectively.
Based on the value of a bicycle of 2,300 yuan, the corresponding market size will reach 12.9 billion yuan, and the compound annual growth rate (CAGR) from 2021 to 2025 will reach 87%. ("Automotive Industry In-depth Report on By-wire Chassis: 100 Billion Blue Ocean under the Wave of Intelligent Electric", Guohai Securities, 2023.5.31)
2) Steer-by-wire: It is expected that the penetration rate of SBW may reach 15% in 2025 ("Automotive Industry In-depth Report By-wire Chassis: 100 Billion Blue Ocean under the Intelligent Electric Wave", Guohai Securities, 2023.5.31)
Automotive steering systems have progressed from mechanical steering (MS) to mesohydraulic power steering (HPS), to electro-hydraulic power steering (EHPS) and electronic power steering (EPS), and finally to by-wire steering (SBW).
With the increasing degree of electrification, by-wire steering systems are gradually attracting attention.
The biggest difference between SBW and EPS is that there is no mechanical connection between the steering wheel and the actuator. SBW has obvious advantages in cost control, chassis design flexibility, driving experience, driving safety and space utilization.
However, due to high technical barriers and limitations by old national standards, the current penetration rate of SBW is still low.
With the lifting of restrictions on steer-by-wire by national standards in January 2022 and the gradual mass production of related products from domestic and foreign suppliers, the penetration rate of SBW will accelerate.
Third, the intelligent driving industry is about to usher in a triple inflection point of technology, policy and products
1. Large models accelerate the landing of L3 intelligent driving
The generation capacity of large models continues to improve, combined with multi-modal input and output capabilities, not only in the field of NLP, but also the application of large models in image processing, language processing, video processing and other dimensions is constantly strengthened, and it is also expected to further empower the entire automotive intelligence trend.
1) Modular & end-to-end, intelligent driving two algorithm frameworks
There are two main categories of intelligent driving algorithm frameworks: 1) modular intelligent driving algorithm scheme; 2) End-to-end solution.
As of 2022, the most adopted autonomous driving architecture in the industry is still modular autonomous driving system, which has the advantage that the entire architecture is very simple, using a three-layer architecture of perception layer, decision planning layer, and execution control layer to solve the problems of perception, decision-making and execution of tasks.
Each layer can be further subdivided into various modules and submodules, so that the development team can develop different submodules in parallel and improve development efficiency.
However, this modular system also faces complex and huge challenges, requiring manual design of all modules, and limited computing power, which makes it impossible to upgrade all submodules to deep learning networks.
Therefore, the modular autonomous driving system can handle most of the autonomous driving scenarios, but it may require a lot of effort to standardize when dealing with a few special scenarios.
Another system solution for autonomous driving is end-to-end (End to End), the core idea of which is to directly input the collected information (such as image data, point cloud data, etc.) into a unified deep learning network, and output the driving commands of autonomous vehicles (such as steering wheel angle, steering wheel speed, vehicle speed, etc.).
Compared with the modular system, the end-to-end architecture design is simple, the hardware cost is low, and it is easier to achieve generalization in different scenarios, which is suitable for the landing of autonomous driving scenarios on urban roads, thereby upgrading to the L3 level.
2) With the TransFormer model, the perception system is fully upgraded
As a representative enterprise of pure visual solutions, a global leading electric vehicle company uses real-time mapping to complete the construction of the environment in order to better understand the surrounding environment through cameras and get rid of the maintenance costs caused by frequent updates of high-precision maps, so as to assist vehicles in decision-making.
Its BEV perception algorithm uniformly converts the image data collected by cameras with different viewing angles to God's perspective to realize real-time generation of live maps of vehicles.
This method complements the road topology information required for subsequent decision-making in autonomous driving, making it possible to get rid of the dependence on high-precision maps.
BEV(2D)+Occupancy Network (3D)+TransFormer algorithm is used to encode the relevant information of the lane, including the location of lane nodes, attributes, bifurcation points, convergence points and geometric parameters of the lane spline.
This information is similar to the word encoding in the language model, and the topological connection between lane lines is obtained through time series processing, so as to realize driving path planning.
The entire vector map generation process involves 75 million parameters.
3) Leading the technology direction, high-level autonomous driving solutions are expected to achieve cost reduction
Once the new autonomous driving architecture of leading enterprises was launched, it was quickly recognized by various enterprises in the industry, and domestic mainstream car companies have adopted BEV+TransFormer to model a new generation of autonomous driving perception solutions.
Although based on safety considerations, the current urban navigation assisted driving functions are equipped with lidar as a redundant perception configuration, but with the maturity of the algorithm, it is expected to reduce the dependence on lidar, greatly reduce the cost of the entire system, and further promote the penetration rate of high-level autonomous driving.
2. Domestic policies are gradually improving and L3 is accelerating
On November 2, 2022, the Ministry of Industry and Information Technology (MIIT) and the Ministry of Public Security issued the Notice on Carrying out the Pilot Work of Intelligent Connected Vehicle Access and Road Traffic (Draft for Comments), which will select intelligent networked vehicle products equipped with autonomous driving functions (L3-L4 levels) that meet the requirements for mass production, carry out access pilots, and carry out road traffic pilots in the restricted public road areas of pilot cities.
On June 21, a meeting mentioned that it will launch pilot access and road access of intelligent networked vehicles to support the commercial application of L3 and higher level autonomous driving functions. "I believe that in the near future, more products with intelligent level will run on the road."
On June 30, the Ministry of Industry and Information Technology and other five departments issued the "Implementation Opinions on Manufacturing Reliability Improvement", focusing on three industries: machinery, electronics and automobiles.
Again mentioning the autopilot system of the car. It is pointed out that the automotive industry focuses on by-wire chassis systems such as steer-by-wire, brake-by-wire, automatic shifting, electronic throttle, suspension system, and autonomous driving systems such as high-precision cameras, lidar, basic computing platforms, and operating systems.
On July 7, the Beijing High-level Autonomous Driving Demonstration Zone Work Office announced the official opening of the "unmanned in the car" commercialization pilot of intelligent networked passenger vehicles.
In March this year, Beijing's unmanned vehicles entered the stage of "unmanned in the car" manned demonstration application. Up to now, there are a total of 116 unmanned test vehicles, with a total test mileage of nearly 2 million kilometers and a cumulative order of more than 1.5 million person-times. (Source: Global Finance Network, 2023.7.8)
3. Overseas: Accelerate the construction of a policy and regulatory framework to regulate the development of autonomous driving
Japan and South Korea are accelerating the commercialization of autonomous driving.
In May 2019, Japan passed a new Road Traffic Law that allows Level 3 autonomous vehicles on the road.
In October of the following year, the National Police Agency announced amendments to the Road Traffic Law to allow Level 4 autonomous vehicles on the road, which came into effect on April 1, 2023.
In September 2022, the South Korean government released the Mobile Innovation Roadmap, which lays out a "three-step" plan for the popularization of autonomous driving.
South Korea became the third country after Japan and Germany to allow Level 3 (conditional automation) autonomous vehicles on the road.
Germany is speeding up the improvement of top-level laws on autonomous driving. In May 2021, Germany officially passed the world's first Autonomous Driving Act, which provides compliance guarantees for autonomous vehicles on the road.
Starting in 2022, Germany will allow vehicles with a high level of autonomous driving to drive on public roads and designated areas.
Under the framework of the Autonomous Driving Act, a German company obtained permission from the German Federal Motor Transport Authority to sell Level 3 autonomous vehicles.
The United States emphasizes motor vehicle safety standards and accelerates the adjustment of regulatory policies. In March 2022, the National Highway Traffic Safety Administration issued passenger safety technical regulations for driverless vehicles, emphasizing that autonomous vehicles must provide the same level of occupant protection as human-driven conventional vehicles.
As of May 2023, 40 states in the United States have established regulations and policies related to autonomous driving. In May 2023, California approved a German company's L3 autonomous driving system application on the road.
H2 2023 and 2024 is the window period for the acceleration of intelligent new cars of car companies, urban NOA function has become the key ability to distinguish car companies from intelligence, at present, major electric vehicle companies at home and abroad are testing and accumulating, and strive to promote NOA high-end assisted driving capacity building, industrialization progress will be expected to continue to exceed expectations, the singularity of unmanned driving is slowly coming!
I am silent, insisting on researching and analyzing funds every day,
One investment observation and reflection per trading day,
One "Fund Review" video per trading day,
Weekly strategy report published every Sunday,
Maintain efficient communication and exchange at all times, a fund investor who insists on striving to improve research capabilities.
If you think the content is valuable and feel that the communication of consultants is very important, just like it, leave a message, and forward it. Thank you. #头条文章养成计划#
Featured Articles:
1. Finally you can follow the vote
2. About the upgrade instructions of Yinhua Dimensity --- quietly
3. Explanation for the upgrade of Yinhua Dimensity --- Year Red
4. [Important] Combination subscription/upgrade tutorial
5. Back to fundamentals: 2023 strategy report
6. Upward trend and main line, gradually surfacing : strategy report for the second quarter of 2023
7. The dawn of the dawn (strategy report for the second half of 2023)
Tips: This article does not constitute investment advice, the market is risky, investment needs to be cautious.
Managed fund investment advisory services are provided by Yinhua Fund Management Co., Ltd. Regular fixed investment is a simple and easy way to guide investors to make long-term investments and average investment costs. However, regular fixed investment does not avoid the inherent risks of fund investment, does not guarantee investors to obtain returns, and is not an equivalent financial management method to replace savings. The Investment Advisory Portfolio Proposal may include fund products managed by Yinhua Fund and other fund products managed by fund managers. Before using the fund portfolio service, investors are requested to carefully read the relevant agreements, business rules and strategy statements, fully understand the details of the portfolio and the fund allocation of the portfolio, and confirm that the portfolio meets its own risk tolerance, investment period and investment objectives. Investors investing in fund investment portfolio strategies should follow the principle of "buyer's own responsibility", and on the basis of a comprehensive understanding of the risk-return characteristics, operational characteristics and suitability matching opinions of the fund investment advisory portfolio strategy, select appropriate fund investment portfolio strategies based on their own circumstances, make prudent investment decisions, and independently bear investment risks. Yinhua Fund does not guarantee that the fund's portfolio strategy will be profitable and minimum return, nor does it make a commitment to capital protection. The risk profile of a fund's portfolio strategy differs from that of a single fund product. The past performance of the investment advisory service is not indicative of its future performance, and the income generated for other investors does not constitute a guarantee of performance. The fund investment advisory business is still in the pilot stage, and there is a risk that the fund investment advisory institution will not be able to continue to provide services due to the cancellation of the pilot qualification. Funds are risky and investment needs to be cautious.