Intelligent driving multi-sensor fusion has become the mainstream technology solution, introducing laser radar and high computing power chips. In 2021-2022, new forces will significantly increase the number of perception hardware such as cameras and introduce LiDAR as a security redundancy; The increase in the number of sensors has raised the demand for computing power in intelligent driving chips, and the Nvidia Drive Orin has become the mainstream choice instead of the early Nvidia Xavier and Mobileye Q4. (1) Xiaopeng: In 2020, Xiaopeng P7 was launched, with the intelligent driving chip using NVIDIA Xavier and a computing power of 30 TOPS. In 2021, Xiaopeng P5 was launched and began to introduce laser radar configuration. The Xiaopeng G9 will be launched in 2022, with the high-end model equipped with two Nvidia Drive Orins, a computing power of 508 TOPS, and two laser radars. (2) NIO: Taking the NIO ES8 as an example, the 2020 Mobileye EyeQ4 intelligent driving chip will be used in 2022, but the number of cameras will increase from 7 to 9. In 2023, the intelligent driving chip will be upgraded to four Nvidia Drive Orins with a computing power of 1016 TOPS, and one new LiDAR will be added, increasing the number of cameras to 11. (3) Ideal: In 2021, the Ideal ONE will upgrade its intelligent driving chip from Mobileye EyeQ4 to two Horizon Journey 3 chips, with a computing power of 10 TOPS and an increase in the number of millimeter wave radars. The L series will be launched in 2022, and the L7/L8 low-end models will upgrade their intelligent driving chips to Horizon Journey 5, with a computing power of 128 TOPS; The L9 and L7/L8 high-end models have been upgraded to two NVIDIA Drive Orins, with a computing power of 508 TOPS. In terms of perception hardware, a new LiDAR has been added, and the number of cameras has been increased to 11.

　　At the algorithmic level, BEV+Transformer and Occupancy Network are widely used. In 2020 and 2022, Tesla respectively introduced BEV+Transformer and Occupancy Network into its intelligent driving system. BEV achieved 3D reconstruction of sensor information based on the original 2D perception algorithm, generating panoramic perception scenes from a top-down angle, improving computational efficiency and detection accuracy, and enhancing information collaboration in the intelligent driving system; Occupancy Network divides the spatial environment into countless 3D grids, and identifies obstacles in the space by determining whether each grid is occupied, providing more accurate obstacle location and environmental information. Last year, domestic car companies began to upgrade their intelligent driving systems to related algorithms. For example, Huawei's ADS 2.0, which is equipped on models such as the Wenjie M7, uses BEV and GOD networks to achieve high-speed NCA and urban NCA functions that do not rely on high-precision maps, and can recognize irregular obstacles outside the whitelist. At the functional level, city navigation and advanced parking functions have been successively implemented.

　　(1) Driving scenarios: Since the second half of last year, Huawei and new forces have accelerated the implementation of urban navigation and assisted driving functions, further expanding navigation and assisted driving from closed scenarios such as highways and expressways to more complex urban scenarios. Some car companies have achieved map free city navigation through real-time perception and reconstruction of road conditions using BEV models, freeing themselves from dependence on high-precision maps.

　　(2) Parking scenario: Upgrade to advanced parking functions such as RPA and AVP. Tesla FSD V12 introduces end-to-end neural network, which has been launched in North America. In November 2023, Tesla will push the FSD V12 to employees. Unlike the previous version of the modular architecture that uses programming to set a variety of manual rules to control vehicle driving, the new version will introduce end-to-end neural networks, and a single model will realize the output from sensor input to driving trajectory. The FSD V12 is similar to the process of human learning to drive, using Tesla's self-developed supercomputer Dojo and real driving data from global car owners for algorithm training. The number of lines of code has decreased from over 300000 to around 2000, and the large number of real scenarios has also enabled the model to better monitor and respond to corner cases (extreme scenarios) that were not adequately addressed by previous modular architectures. Since the beginning of this year, domestic new forces have gradually implemented the application of end-to-end large-scale models. In May, Xiaopeng's end-to-end large model was mass-produced and put into operation; In July, NIO upgraded its intelligent driving architecture to an end-to-end architecture that incorporates world models; In October, Wanjie M5 and M7 will upgrade to HUAW EI ADS 3.0, adopting an end-to-end network architecture and ideally adopting a new autonomous driving technology architecture with an end-to-end large model to enable full push; At the end of the year, Xiaomi will launch a pioneer version of end-to-end intelligent driving. The penetration rate of L2 has increased, and pure visual solutions such as Xiaopeng and Huawei ADS SE may accelerate the sinking of advanced intelligent driving. According to Gaogong Intelligent Automobile, from 2020 to 2023, 753000, 2.292, 5.86, and 7.94 million new L2 vehicles will be delivered in the national passenger car market, with a total of 3.798 million vehicles delivered in the first half of 2024 and a loading rate of nearly 40%. According to the China Association of Automobile Manufacturers, in the various segmented markets of new energy passenger vehicles in the first half of 2024, the total penetration rates of L2 and L2+for models with 80000 to 160000 yuan, 160000 to 240000 yuan, 240000 to 320000 yuan, 320000 to 400000 yuan, and over 400000 yuan are 31.8%, 74.1%, 85.5%, 88.4%, and 81.5%, respectively.

　　(1) Xiaopeng MONA: The Xiaopeng MONA series, formerly known as Didi's intelligent car development business, focuses on the price range of around 150000 yuan and launches multiple models based on the Fuyao architecture that support intelligent driving for young customers. The first model MONA M03 is positioned as an A-class pure electric sedan, equipped with over 20 perception hardware including 7 cameras, 2 millimeter wave radars, 12 ultrasonic radars, etc. It comes standard with LCC, ACC, 360 imaging system, and full scene intelligent parking. The M03 Max is equipped with XNGP, which enables high-speed NGP, urban NGP, AI driving and other functions.

　　(2) Huawei ADS SE: Huawei has established deep cooperation with companies such as SAIC, Chery, Jianghuai, and BAIC under the HarmonyOS model. Other cooperative brands include Avita and Shenlan under Changan, Lantu and Mengshi under Dongfeng, and GAC Trumpchi. At present, in addition to the low-end versions of Hongmeng Zhixing models such as Wenjie M7 and Zhijie R7, models such as Shenlan S07/L07 have also been equipped with Huawei Qiankun Intelligent Driving ADS SE on the market, using a sensing hardware combination of 10 cameras, 3 millimeter wave radars, and 12 ultrasonic radars, which can achieve functions such as ACC, LCC, and urban NCA.

　　L3: TJP traffic congestion navigation function pilot for boarding, which can achieve eye and hand detachment in specific situations. Starting from Level L3, the dominance of intelligent driving is transferred from the driver to the auto drive system, allowing the driver to get out of his/her way under certain circumstances. For example, TJP traffic congestion navigation can automatically center the lane, maintain the speed, and maintain the distance under certain congestion scenarios. After the Ministry of Industry and Information Technology issued the "Notice on Carrying out Pilot Work for Intelligent Connected Vehicle Access and Road Traffic" in November last year, the first batch of pilot lists for intelligent connected vehicle access and road traffic were announced in June this year. Nine car companies, including Changan, BYD, GAC, SAIC, BAIC Blue Valley, FAW, SAIC Hongyan, Yutong, and NIO, were included in the pilot list. Among them, pilot models such as the Deep Blue SL03i and Haobo GT have been equipped with TJP function.

　　L4: Apollo Go is about to make a profit. Tesla released Cybercab to verify the robotaxi business model. As the representative of L4 level intelligent driving, Robotaxi has no steering wheel or brake pedal in the car. Many brands have been tested and operated worldwide, including Waymo abroad, Apollo Go, Wenyuan Zhixing, Xiaoma Zhixing, etc., of which Apollo Go has been operating in 11 cities in China. It is planned to achieve balance of income and expenditure in Wuhan by the end of the year, enter the profit period in 2025, and begin to layout in overseas markets; Tesla recently released Cybercab, which will continue to adopt a pure visual solution and be equipped with the latest in car computing platform AI5. It is planned to be put into production in 2026, and the vehicle cost will be reduced to less than $30000. The operating cost will be reduced from the current $1/mile to $0.2/mile.

　　The whole vehicle EE A is evolving from distributed to centralized. Under the trend of intelligence, the number of ECUs inside cars continues to increase, and the extensive installation of perception hardware related to assisted driving also leads to a geometric growth in the amount of data generated. Traditional distributed EEAs are no longer sufficient to meet the requirements. In order to improve data processing capabilities, reduce computational redundancy, and simplify the use of ECUs and wiring harnesses, EEAs are developing towards higher integration. In 2016, Bosch proposed the development trend of automotive EEA, which was divided into three stages: distributed, domain centralized, and centralized.

　　The functional domains of body, power, chassis, etc. have achieved cross domain integration first. The domain centralized EE architecture is usually divided into five domains: body domain, power domain, chassis domain, intelligent driving domain, cabin domain, etc. With the enhancement of the performance of the main control processor and further concentration of functional domains, functional domains with lower computing power requirements such as the vehicle body domain, power domain, and chassis domain in the domain centralized EEA are first integrated to form a cross domain centralized EEA; The intelligent driving domain and cockpit domain, which require high computing power, still use independent domain controllers.

　　The integration of cabin and driving is imminent, and domestic car companies are gradually transitioning to regional architectures. Tesla has gradually transitioned from the distributed EEA of the 2012 Model S to the multi domain fusion of the 2015 Model X and the quasi regional architecture of the 2018 Model 3, achieving central integration of the EEA. In 2020, new forces began to focus on domain integration architecture and gradually develop towards cross domain fusion. With the upcoming application of cabin driving fusion chips such as NVIDIA Thor and Qualcomm 8775, intelligent driving and cabin domain functions can be achieved with a single SoC. Mainstream new energy vehicles will further transition from cross domain fusion to central computing.

　　The sinking of intelligent driving functions is expected to enhance the application of lightweight computing and integrated parking solutions. The integrated driving and parking domain controller enhances the intelligent driving experience and reduces the cost of the host factory by reusing sensors, sharing computing resources, simplifying interfaces, reducing wiring harnesses, and improving development efficiency. The integrated solution of light computing power and navigation is mostly based on Texas Instruments TDA4 and Horizon Journey ? Third class low to medium computing chips mainly use 5V5R or 11V5R sensor configurations, which can achieve high-speed navigation function in driving scenarios and APA automatic parking, HPA memory parking and other functions in parking scenarios. According to GAC Motor, the delivery volume of pre installed vehicle parking integrated domain control in the domestic passenger car market in 2021 and 2022 was 460000 and 780000 respectively, and it is expected to reach 2 million vehicles in 2023, corresponding to penetration rates of 2%, 4%, and 9%.

　　Wire controlled braking: EHB penetration rate continues to increase, and EMB is expected to achieve mass production by 2025. The braking system has evolved from mechanical braking, hydraulic braking, ABS/ESC to line controlled braking, which can be divided into hydraulic line controlled braking (EHB) and mechanical line controlled braking (EMB). EHB can be further subdivided into Two Box and One Box schemes according to the degree of integration. (1) EHB: According to Gaogong Intelligent Automotive, 3.07 million, 4.97 million, and 7.96 million passenger cars equipped with wire controlled braking systems were delivered in the domestic market in 2021, 2022, and 2023, with front loading rates of 15.0%, 25.0%, and 37.7%, respectively. Overseas suppliers such as Bosch, Continental Group, Wandu, ZF, and Adidas have certain first mover advantages, while domestic suppliers such as Freddie Power, Bethel, Lick, and Feige continue to benefit from the increasing intelligence of domestic brand models and the trend of domestic substitution. (2) EMB: Cancel the hydraulic device, and the braking torque is generated by a brushless motor installed on the caliper, which can reduce weight, improve energy recovery efficiency, reduce braking reaction time and braking distance, and meet the braking system requirements of L3 and above autonomous vehicles. Brembo, Bosch, and other companies plan to mass produce by 2025. Continental Group, ZF, and domestic suppliers such as Bethel, Great Wall Precision, Wanan Technology, Bobost, Grubo, Nathan, and Lik also have layouts for EMB. Both domestically and in Europe are promoting the revision of relevant regulations, and EMB is gradually being applied.

　　The retail structure of passenger cars has been upgraded, and the mid to high end market continues to expand. With the increase of trade in purchases surpassing first-time purchases as the main force of automobile consumption, the trend of consumer upgrading in the passenger car market continues. According to the China Association of Automobile Manufacturers, the retail sales of passenger cars in China increased from 20.686 million units to 21.702 million units from 2019 to 2023, corresponding to a CAGR of 1%, and remained stable overall; The retail sales of passenger cars with a value of over 200000 yuan increased from 4.116 million units to 7.024 million units, corresponding to a CAGR of 14%, far exceeding the overall growth rate, with the proportion rising from 20% to 32%.

　　Tesla, new forces, and others are gradually replacing BBA. By brand, German luxury brands represented by BBA have a leading market share in the domestic mid to high end market. According to data from various car companies, Mercedes Benz sold 765000 vehicles in China in 2023, BMW sold 825000 vehicles in China, Audi sold 729000 vehicles in China, and according to the China Association of Automobile Manufacturers, BBA sold a total of 7.024 million vehicles at a price of 200000 yuan in 2023, with a total market share of 33%. In recent years, Tesla, along with emerging brands such as NIO, Huawei, and Xiaomi, have made breakthroughs in the mid to high end market. In 2023, Tesla sold 527000 vehicles in China, with a market share of 8%. Tesla's sales are concentrated in the B-class car market, while new forces have a richer layout of models in various segmented markets, which is expected to continue to increase their market share in the mid to high end market.

　　SUVs priced between 200000 and 400000 yuan: The sales of brands such as Wanjie, Ideal, and NIO are comparable to those of BBA. BBA models include Mercedes Benz GLC, BMW X3, Audi Q5L, etc. Currently, the terminal selling price of Mercedes Benz GLC and BMW X3 is 300000 to 400000 yuan, and the terminal selling price of Audi Q5L is 250000 to 300000 yuan. In August, the sales of the three models were 14000, 5000, and 13000 respectively. The new energy vehicle models mainly include Tesla Model Y and new energy brands such as the WENJIE M7, IDEAL L7/L8, NIO ES6/EC6, Xiaopeng G9, etc. In August, the WENJIE M7, IDEAL L7+IDEAL L8, NIO ES6+NIO EC6 all reached over 10000 units.

　　SUVs priced over 400000 yuan: The sales of the Wanjie M9 have surpassed those of BBA models. BBA models include Mercedes Benz GLE, BMW X5, Audi Q6/Q7, etc. Currently, the terminal selling price of Audi Q6 has dropped to less than 400000 yuan, while the terminal selling price of other models is around 500000 yuan. In August, BMW X5 sold 4000 units. New energy vehicle models include Tesla Model X, as well as new energy brands such as Wanjie M9, Ideal L9, NIO ES8, etc. The delivery in August was 17000 units, 7000 units, and 1000 units respectively. Wanjie M9 sales have surpassed BBA benchmark models, while Ideal L9 sales are comparable to BMW X5.

　　Sedan priced between 200000 and 400000 yuan: Xiaomi SU7 is expected to surpass BBA models in terms of monthly delivery after its launch. BBA models include Mercedes Benz C-Class, BMW 3 Series, Audi A4L, etc. Currently, the starting price at the terminal is around 200000 yuan, and the sales in August were 16000, 9000, and 9000 respectively. New energy vehicle models include Tesla Model 3 and new forces such as NIO ET5/ET5T, Xiaopeng P7, Xiaomi SU7, and Zhijie S7. Among them, the monthly delivery of Xiaomi SU7 remains above 10000 units, and with the increase in production capacity, the delivery target for October has been raised to 20000 units; NIO delivered a total of 7000 units of ET5 and ET5T.

　　Sedans priced over 400000 yuan: Currently dominated by BBA, domestic brands need to break through. BBA models include Mercedes Benz E-Class, BMW 5 Series, Audi A6L/A7L, etc., with a starting price of around 300000 yuan. The sales in August were 9000, 4000, and 13000 respectively; Mercedes Benz S-Class, BMW 7 Series/8 Series, Audi A8L and other D-class sedans are priced at millions of yuan. In addition to the Tesla Model S, high-end sedan models from domestic brands include the NIO ET7 and Xiangjie S9, which cost 400000 to 500000 yuan. However, there is still a certain gap in sales compared to BBA models, and higher priced models such as the NIO ET9 and Wangshang U7 have not yet been launched.