On March 29, 2024, the “China Automotive New Supply Chain Configuration Data Product Launch Conference” organized by the World Automotive Research Institute was successfully held in Shanghai. The World Automotive Research Institute aims to be a leading think tank in the automotive industry, focusing on intelligent automotive insights throughout the industry chain, providing comprehensive research results and industry data resources to support commercial decision-making. At the event, we launched and released three sets of original database products – Passenger Vehicle Intelligent Driving Configuration Database, Passenger Vehicle Intelligent Cockpit Configuration Database, and Passenger Vehicle Electrification Configuration Database – along with market analysis monthly reports. Currently, autonomous driving technology is rapidly advancing, from basic vehicle control assistance to advanced autonomous driving assistance, with continuous upgrades and improvements. The penetration rate of Level 2 driving assistance is rapidly increasing, driving the rapid development of local smart driving solutions and suppliers in China. The penetration rate of Level 2 driving assistance is rapidly increasing, with urban NOA accelerating implementation. ADAS plays a crucial role in vehicle safety, with the 2024 version of C-NCAP optimizing existing scenarios and adding new active safety testing items. As the performance of functions like lane-keeping assistance and adaptive cruise control improves, and the cost decreases due to mass production and technological advancements, more vehicles can offer driving assistance features at lower prices, especially in the 100,000-200,000 yuan (27640$) price range, becoming the main market for Level 2 driving assistance. According to the World Automotive Research Institute’s Intelligent Driving Configuration Database, the market penetration rate of L2-level standard equipped new energy passenger vehicles reached over 51%, with 3.7 million units in 2023. In the 200,000-500,000 yuan (69110$) price range, the penetration rate exceeds 85%.
As the penetration rate of low-level driving assistance functions increases, high-level intelligent driving systems are reaching a turning point in development. In November last year, four ministries jointly issued a notice on carrying out pilot work on the admission and road passage of intelligent connected vehicles, specifying the requirements for the admission norms of L3/L4 autonomous driving and improving related rules. High-level autonomous driving in China has clear policy support and responsibility delineation. BYD, Changan, Zhiqi, BMW, Mercedes-Benz and many other domestic and foreign OEMs with L3 technical strength actively applied, and currently, many OEMs have successively announced the acquisition of conditional automatic driving test licenses on highways or expressways. It is not difficult to foresee that with more car companies joining in the development of L3 autonomous driving, high-level intelligent driving is expected to truly undergo a “qualitative change.” Of course, currently, car companies mainly focus on NOA as the core driving force. At present, high-speed NOA has entered mass production, and urban NOA is becoming the focus of competition among car companies, aiming to seize the initiative. Data shows that by 2023, there will be 940,000 vehicles equipped with navigation assistance driving. From 2023 to 2024, high-level intelligent driving functions represented by urban NOA will accelerate landing, with brands such as Huawei and Xiaopeng successively releasing high-level intelligent driving models, and many OEMs launching landing plans.
Huawei is one of the most aggressive forces in smart driving. Its urban NCA is now available nationwide. Xiaopeng’s XPILOT system has entered the 4.0 phase, with its city NGP covering 243 cities. By 2024, Xiaopeng’s XNGP will cover the entire national network, including internal roads, unnamed roads, and parking lots. Even Xiaomi, which has just released its first car model, is accelerating the expansion of urban NOA. Xiaomi Group Chairman and CEO Lei Jun introduced that the Xiaomi SU7 fully supports valet parking, narrow parking spaces, and high-speed navigation. The city NOA will start user testing in April, open in 10 cities in May, and be fully available nationwide in August. It is worth noting that the NOA function is mainly supported by autonomous mid-to-high-end models, with new energy brands mainly offering standard NOA function models. The penetration rate of Ideal and Avita has reached 100%. Among them, the highest proportion of models is in the 350,000-400,000 price range, with a market share of 33.1%.
Currently, over 70% of vehicles equipped with NOA navigation function are priced above 300,000 yuan (41470$). However, we can already see the emergence of mid-to-low-end models like Xiaopeng P5 and Baojun Cloud. The Global Automotive Research Institute predicts that by 2024, the number of domestic vehicles equipped with NOA function is expected to exceed 1.8 million, with a projected volume of over 8.5 million by 2030.
Heavy perception, light map is still mainstream, rapid expansion period for lidar The competition for urban NOA continues, focusing on achieving larger scale data collection at lower costs for algorithm optimization and model iteration. The impact of price wars cannot be ignored, requiring higher cost control from car companies. “Heavy perception, light map” has become the prevailing solution, largely based on cost considerations. Car companies aim to accelerate the expansion speed of urban NOA, with high costs of high-precision map collection and freshness posing obstacles. The application of algorithms like BEV+Transformer enhances real-time perception on vehicles, reducing dependence on maps, making light maps mainstream. In early February, Wenjie launched the first batch of “NCA without map” push, and in the OTA upgrade in March, all models of Wenjie added city intelligent driving assistance that does not rely on high-precision maps, achieving “can drive everywhere, can drive on any road” City NCA. Recently, Zhiji Automobile also announced that it will start the public testing of “NOA without map” in Shenzhen, Guangzhou, and Suzhou in April, and has started recruiting public testing users. Zhiji Automobile is scheduled to officially launch the IM AD mapless urban NOA in mid-2022. The Gaisei Automotive Research Institute believes that with the rapid development of autonomous driving algorithm technology, the future will evolve into a technology solution mainly based on real-time perception capability + SD Pro map/navigation map. As one of the key sensors to enhance vehicle intelligent driving perception, lidar is gaining more favor from car companies. Data shows that by 2023, among vehicles with standard L2 level functions, the 5V1R/5V3R sensor solutions are mainly used, with most new vehicles with NOA function adding lidar sensors to enhance vehicle perception capabilities.
Brands like Xiaopeng and Weilai have confirmed that LiDAR is a key sensor to ensure the comfort and safety of intelligent driving products. Since 2024, with the launch of models like Xike 001 and Zhiji L7, the trend of standardization of LiDAR in the field of smart cars has become more apparent.
Competition in city NOA and the development of L3 and above high-level autonomous driving models are driving rapid growth in the adoption of LiDAR on vehicles. Companies like Surestar and Hesai Technology may have a broader stage for development. The smart driving supply chain is developing rapidly, with local players becoming important drivers. Among the many industry players, local manufacturers are gradually becoming a strong force. Currently, the advanced route of ADAS technology can be divided into two categories: driving and parking. Driving includes some active safety functions, high-speed NOA, and urban NOA; while parking functions are used in parking scenarios, such as memory parking and valet parking. In the past, foreign Tier 1 companies dominated the market share in the driving ADAS and parking fields. However, in the past two years, with local suppliers entering the market from independent brands, their market share has been continuously increasing and is expected to further rise in the future. According to Gaishi Automotive Research Institute, by 2023, the market share of local suppliers in the driving ADAS field will increase to 14.4%, while the market share of local suppliers in the parking field has already exceeded 30%.
Integrated parking and driving is one of the important solutions in the current field of intelligent driving. Today, the market share of integrated parking and driving developed by OEMs and outsourced is close to 50%. Local suppliers are leading foreign Tier with advantages such as full-stack development of software and hardware, rapid product iteration, and multi-chip platform adaptation. Looking at it from the perspective of components, besides LiDAR, some domestic suppliers have stood out in areas such as air suspension and high-precision maps. In the field of intelligent driving domain control chips and wire-controlled actuators, some domestic companies have begun to achieve scale development.
Domestically produced cameras and millimeter wave radars have great growth potential. In the advanced intelligent driving field, the domestic supply chain is developing rapidly, with many outstanding domestic manufacturers in software algorithms, maps, chips, and overall solutions.
The mass production landing of City NOA has greatly driven the demand explosion of upstream industries such as high computing chips, lidar, and cloud services. Many suppliers have taken steps towards listing. On December 20th last year, Zhixing Technology, a major representative of domestic autonomous driving domain controllers, successfully listed on the main board of the Hong Kong Stock Exchange. Earlier this year, TuSimple rang the bell for its Hong Kong Stock Exchange debut, becoming the world’s largest lidar enterprise by market value. Many other smart driving companies are planning IPOs. On March 28th, Horizon Robotics submitted its prospectus to the Hong Kong Stock Exchange, intending to list on the main board. As for the automotive industry, with the current trend of intensified competition in smart driving, independent car companies are accelerating the research and application of intelligent driving technology. Some powerful OEMs choose to independently develop to maintain technological leadership and independence. Most car companies choose to cooperate with suppliers to directly purchase mature solutions to reduce costs and achieve rapid implementation. Currently, suppliers providing smart driving solutions in the market include automotive component giants, tech giants, and professional startups, forming a diversified force for the development of smart driving. With continuous investment in technology research and development by independent car companies and the expansion of market share by domestic suppliers, the two sides mutually promote each other, providing strong support for the rapid development of smart driving technology. In the future, this trend is expected to continue to strengthen, further driving the popularization and application of intelligent driving technology in the Chinese and global markets. The integration of cockpits is gradually being mass-produced, and autonomous driving is moving towards end-to-end. When it comes to the future of intelligent driving, one cannot avoid the “keywords” such as cockpit integration, large computing chips, large models, and end-to-end. The concepts of integrated driving and parking are gradually maturing, ultimately leading to cockpit integration or cockpit integration in vehicle manufacturers’ electronic/electrical architectures upgrading to domain-centralized architectures, with driving and parking sensors and domain controllers sharing technology becoming increasingly mature. The development of cockpit integration technology is currently in a positive transformation period, with car manufacturers and suppliers laying out cockpit integration domain control. Some companies have introduced related products, such as Bosch’s central computing platform and Changxing Intelligent Driving’s cockpit integration platform. It is expected that by 2024, cockpit integration domain control will achieve small-scale production, and with the maturity of technology and market demand, it may be widely used in the near future. High computing chips are crucial for achieving cockpit integration, supporting increasingly complex algorithms and processing large amounts of data, as well as ensuring the security and reliability of the system. However, the focus of intelligent driving is currently on the integrated technology of parking and cruising. Most of the current low to medium computing power integrated parking and cruising solutions use multiple SOC solutions, while the new generation platforms all use single SOC solutions. Due to the need to support higher-level intelligent driving functions, the high computing power integrated parking and cruising domain controller mainly adopts multiple SOC solutions due to considerations of immature computing power embedding and system redundancy.
With the improvement of AI algorithms and chip design capabilities, even low-power SOC platforms can provide enough performance to support basic functions of autonomous driving. High-performance, cost-effective single SOC low-power platforms are expected to lead the way. Another crucial factor is the iterative innovation of autonomous driving algorithm technology. Many leading car companies choose to follow Tesla’s algorithm iteration method, mainly in three stages: First stage: Tesla introduced BEV and Transformer technology in its autonomous driving system, achieving “map-free” and marking the application of large models in onboard systems; Second stage: Upgrade the network to achieve “lidar-free”; The third stage is end-to-end autonomous driving.
On March 18th of this year, Tesla began rolling out the FSDV12.3 version in North America, introducing the “end-to-end neural network” technology, sparking industry discussions once again. The current intelligent driving system uses a modular model, dividing perception, prediction, and planning into three separate modules with significant technical differences. This modular architecture may result in information loss during information transmission, making the system complex, difficult to maintain, and unable to effectively handle complex road environments. In contrast, the end-to-end model has clear advantages in intelligent driving solutions, integrating perception, prediction, and planning into a single model, simplifying the structure of the solution and improving computational efficiency. Compared to traditional rule-based models, end-to-end models are easier to scale and achieve performance breakthroughs. However, the drawbacks of end-to-end models are also evident. End-to-end models are often seen as “black boxes,” merging multiple steps into one model, making it more difficult to understand how the model works internally. This opacity may lead to issues with interpretability and explainability. Additionally, end-to-end models require more computational resources, significantly increasing the cost and time of training and deploying models. Furthermore, end-to-end models may face challenges in debugging, optimization, and overfitting. Nevertheless, many companies believe that end-to-end autonomous driving is one of the most promising ways to achieve driverless driving in the future. Companies like WeRide, DeepRoute, and others have chosen to follow Tesla’s technological path and deepen their exploration on this road. Currently, the autonomous driving industry has reached a turning point, and we may see more exciting innovations and progress in the coming years.
