Chapter 1: Evolution and Core Value of Autonomous Driving Technology

For the first 90 years of the automobile’s existence, driving was a purely mechanical task: the driver controlled the engine, steering, and brakes with no digital assistance, relying entirely on their own perception, judgment, and physical reaction. The first electronic driver aids appeared in the 1950s, with anti-lock braking systems (ABS) introduced by Chrysler in 1971, followed by electronic stability control (ESC) in the 1990s, which could automatically adjust brake force and engine power to prevent skidding. These early systems were the embryonic form of autonomous driving, as they took over single, specific driving tasks that were previously fully controlled by humans. The true leap from assisted driving to autonomous driving came with the integration of multiple sensors, high-performance computing, and advanced algorithms, enabling vehicles to perceive their environment, make decisions, and control motion without continuous human input. This evolution did not happen overnight; it is the result of 40 years of incremental research, countless failed prototypes, and breakthroughs in artificial intelligence and hardware. To understand how far we have come, it is first necessary to clarify the standardized classification system for autonomous driving, established by the Society of Automotive Engineers (SAE) in 2014 and now adopted globally: the six-level L0 to L5 framework. Each level represents a step change in the degree of human involvement and system autonomy, as shown in the comparison table below.

| SAE Level | Official Name | Core Functionality | Human Responsibility | Real-World Application Examples | |-----------|---------------|--------------------|---------------------|---------------------------------| | L0 | No Driving Automation | Only provides warnings (e.g., lane departure alerts, forward collision warnings) | Full control of all driving tasks | Basic warning systems in most entry-level vehicles | | L1 | Driver Assistance | Controls a single driving dimension (speed or steering) | Must monitor the driving environment at all times and take over immediately when the system fails | Adaptive Cruise Control (ACC), Lane-Keeping Assist (LKA) | | L2 | Partial Driving Automation | Controls both speed and steering simultaneously in specific scenarios | Must remain fully attentive and ready to take over at any time | Tesla Autopilot, Xpeng NGP, Li Auto AD Max, the most common advanced driver assistance system (ADAS) in new vehicles sold today | | L3 | Conditional Driving Automation | Controls all driving tasks in predefined scenarios (e.g., highways, congested traffic) | Not required to monitor the environment continuously, but must take over immediately when the system issues a request to intervene | Mercedes-Benz S-Class with DRIVE PILOT (approved for use on German highways at speeds up to 60 km/h), Honda Legend with Traffic Jam Pilot (approved for use in Japan) | | L4 | High Driving Automation | Controls all driving tasks in predefined operational design domains (ODDs) without human intervention | No driving responsibility required within the ODD; human intervention is only needed if the vehicle exits the ODD | Waymo Robotaxi (operating in Phoenix, San Francisco, and Los Angeles), Baidu Apollo Go (operating in Beijing, Wuhan, Shanghai, and other Chinese cities), campus and industrial park delivery robots | | L5 | Full Driving Automation | Controls all driving tasks under all road, weather, and traffic conditions, with no need for a human driver | No human driving responsibility required at any time | Not yet commercially available; remains a long-term research goal for the industry |

The most critical distinction between adjacent levels lies in the boundary of responsibility. For L2 systems, the driver is always legally responsible for the vehicle’s actions, as the system is defined as an "assistant" rather than the "controller"; even if the system makes a mistake, the driver is expected to intervene immediately. For L3 and higher levels, the system becomes the primary controller within its designated operating scope, and the manufacturer bears legal responsibility for accidents caused by system failure within that scope. This distinction is why L3 commercialization has lagged far behind L2: it requires not only technical maturity but also complete legal and regulatory frameworks to define responsibility boundaries.

The evolution path of autonomous driving follows a clear, incremental logic: starting with single-function control of specific driving dimensions, expanding to multi-scenario combined control, and ultimately achieving full-scenario no-handover operation. The earliest driver aid, cruise control, was born in 1945 and only controlled vehicle speed; it could not adjust speed based on surrounding traffic, so it was classified as L0. The next step was adaptive cruise control (ACC), introduced in the 1990s, which used radar to detect the distance to the vehicle ahead and automatically adjust speed to maintain a safe following distance, achieving single-dimension speed control and entering the L1 stage. The combination of ACC and lane-keeping assist (LKA), which uses cameras to detect lane lines and automatically adjust steering to keep the vehicle centered in the lane, achieved simultaneous control of speed and steering, marking the arrival of the L2 stage. Today, L2 ADAS has become standard on most new vehicles priced above 150,000 RMB, with a penetration rate of over 40% in China’s new vehicle market as of 2023. The next step is L3: the system can handle all driving tasks in predefined scenarios (such as highways or congested traffic at low speeds), and the driver does not need to monitor the road continuously, but must take over immediately when the system requests it. The final step is L4 and L5: the vehicle can operate completely without human intervention, either in specific areas (L4) or in all scenarios (L5).

Core Value of Autonomous Driving: Beyond Convenience to Social Impact

The global investment in autonomous driving exceeds 100 billion USD annually, not just because it is a "cool technology", but because it brings transformative value to three core areas: road safety, transportation efficiency, and special scenario applications.

1. Revolutionizing Road Safety

According to the World Health Organization’s 2023 Global Road Safety Report, approximately 1.35 million people die in road traffic accidents every year, with over 90% of accidents caused by human error. The main human factors include fatigue driving (accounting for 22% of accidents), distracted driving (e.g., using a phone while driving, 18%), drunk driving (8%), and speeding (10%). Autonomous driving systems have no fatigue, distraction, or emotional interference, and their reaction speed is dozens of times faster than that of humans, which can fundamentally reduce the accident rate.

We can quantify this advantage with a simple physical calculation. The total braking distance of a vehicle is determined by two parts: the reaction distance (the

distance (the distance the vehicle travels before the driver or system begins to brake) and the physical braking distance (the distance the vehicle travels after the brakes are applied until it comes to a complete stop). For a human driver with an average reaction time of 1.5 seconds – a figure verified by multiple traffic safety studies across different age groups and driving experience levels – driving at 100 km/h (27.8 m/s) results in a reaction distance of 41.7 meters alone. In contrast, an autonomous driving system with a millisecond-level sensor processing pipeline and pre-programmed response logic has an average reaction time of just 0.1 seconds, resulting in a reaction distance of only 2.78 meters at the same speed: a difference of nearly 15 times. When combined with the physical braking distance (approximately 40 meters for a standard passenger car braking from 100 km/h on a dry asphalt road with good