Autonomous Driving System Architecture and Technology Evolution
The global autonomous driving industry entered a phase of rapid commercialization in 2023, with the global shipment of passenger vehicles equipped with L2 and above autonomous driving systems reaching 18 million units, and the penetration rate exceeding 20% according to the statistics of SAE International. As the opening chapter of this book, this chapter systematically deconstructs the core technical architecture and evolution context of autonomous driving systems, clarifies the technical boundaries and landing requirements of different autonomous driving levels based on the SAE J3016 classification standard, analyzes the functional logic and interaction mechanism of each layer of the autonomous driving system, sorts out the evolution path from rule-driven to data-driven technology, and discusses the core pain points faced by the current industry, laying a complete theoretical and practical framework for the in-depth interpretation of subsequent chapters.
SAE J3016 Autonomous Driving Classification System: Technical Boundaries and Landing Requirements
The SAE J3016 standard released by SAE International in 2014 and updated in 2021 is the most widely recognized classification system in the global autonomous driving industry, which defines six levels of autonomous driving from L0 to L5 based on the degree of takeover of the Dynamic Driving Task (DDT) by the system. DDT refers to all operational and tactical operations required for vehicle driving, including lateral and longitudinal control of the vehicle, environment monitoring, and response to unexpected events. The core classification logic of SAE J3016 focuses on two dimensions: whether the system performs the DDT, and whether the human driver is required to take over the DDT when the system fails.
The 2021 updated version of SAE J3016 further clarifies the boundary between "driving assistance" and "automated driving": L0-L2 are classified as driving assistance systems, where the driver is always responsible for the DDT, even if the system provides auxiliary control; L3 and above are classified as automated driving systems, where the system is responsible for the DDT within the defined Operational Design Domain (ODD), and the human driver is only required to take over when the system requests. ODD is a core concept for L3 and above systems, which defines the applicable scope of the system, including geographic range (such as a specific city or highway section), road type (such as urban roads, highways, parking lots), speed range (such as 0-120km/h), environmental conditions (such as sunny days, light rain, no heavy snow), and lighting conditions (such as daytime, nighttime with street lights). The system can only be activated within the ODD, and once the vehicle leaves the ODD or the system encounters a scenario it cannot handle, it will send a takeover request to the human driver.
Definition and Landing Requirements of Each Level
- **L0 (No Driving Assistance):** The system does not provide any control of the vehicle, and the human driver is fully responsible for all DDTs. Typical scenarios include traditional fuel vehicles without any auxiliary functions.
- **L1 (Single Driving Assistance):** The system provides single-function auxiliary control for either lateral or longitudinal DDT, such as Adaptive Cruise Control (ACC) for longitudinal control or Lane Keeping Assist (LKA) for lateral control. The human driver is still fully responsible for the DDT. L1 systems have been widely popularized in mid-to-low end passenger vehicles, with a penetration rate of over 90% globally in 2023.
- **L2 (Combined Driving Assistance):** The system provides combined auxiliary control for both lateral and longitudinal DDTs at the same time, such as ACC + LKA, or automatic lane changing, automatic parking and other functions. However, the human driver must always monitor the driving environment and be ready to take over at any time. L2 systems are the current mainstream mass-produced products, with global penetration rate exceeding 22% in 2023, and domestic penetration rate in China exceeding 35%. It should be noted that many car companies' marketing terms such as "L2.5", "L2.9" are not recognized by the SAE standard, and their essence is still L2, with no transfer of driving responsibility.
- **L3 (Conditional Automated Driving):** Within the defined ODD, the system completes all DDTs, and the human driver does not need to monitor the driving environment, but must be able to take over the DDT immediately when the system sends a takeover request. The responsibility for accidents within the ODD belongs to the system provider (car company or algorithm supplier). The landing of L3 systems requires strict ODD definition, responsibility division mechanism, and safety certification. In 2023, Mercedes-Benz's L3 system DRIVE PILOT became the first L3 system approved for mass production in Germany, which can be activated on highways with speed limit of 60km/h or below, and the driver can engage in non-driving activities such as watching mobile phones. However, as of 2024, no L3 system has been mass-produced and listed in China, mainly due to the imperfect laws and regulations on responsibility division and the complex traffic environment.
- **
- **L4 (High Automated Driving):** Within the defined ODD, the system completes all DDTs, and the human driver does not need to monitor the driving environment or be ready to take over at any time. When the system encounters a scenario outside the ODD or cannot handle it, it will automatically execute a Minimum Risk Maneuver (MRM) such as pulling over to the side of the road and parking, without requiring human takeover. The responsibility for all accidents within the ODD belongs to the system provider. The core technical requirement of L4 systems is full coverage of all scenarios within the ODD, with zero human intervention required under normal operating conditions. Typical landing scenarios of L4 systems include Robotaxi operating in specific urban administrative areas, autonomous driving of heavy trucks on designated highways, autonomous operation of port container trucks, and autonomous delivery vehicles in closed campuses or industrial parks. As of 2024, Waymo's L4 Robotaxi service has covered more than 10 cities in the United States, with a cumulative safe driving mileage of more than 32 million kilometers, and the accident rate is 85% lower than that of human drivers. In China, Baidu Apollo's L4 Robotaxi service "Luobokuaiping" has covered 10 cities including Wuhan, Beijing, and Chongqing, with a cumulative order volume exceeding 100 million as of 2024, and fully driverless service has been opened to the public in Wuhan and other cities. It should be noted that the ODD of current L4 systems is still strictly limited: for example, Waymo's L4 system currently does not support operation in heavy snow, heavy fog, or unstructured roads without clear lane markings, and the operating speed is limited to 65km/h or below in most urban areas.
- **L5 (Full Automated Driving):** The system can complete all DDTs under all road and environmental conditions that human drivers are capable of handling, without any ODD restrictions, and does not require human takeover under any circumstances. The responsibility for all accidents belongs to the system provider. As of 2024, no L5 system has been practically realized, and it remains a long-term strategic goal for the global autonomous driving industry. The core challenge of L5 systems is the coverage of unlimited long-tail scenarios, including extreme weather (blizzards, sandstorms, heavy fog with visibility below 50 meters), unstructured roads (country roads without lane lines, temporary construction sites, rural areas with mixed human and animal traffic), and extreme traffic scenarios (road rage of other drivers, pedestrians running red lights, sudden falling objects from trucks).