Introduction to Autonomous Driving: Definition, Classification and Application Value

To understand what autonomous driving truly means, we must first distinguish it from the driver-assistance features already common in many family cars today. While terms like “adaptive cruise control” or “lane-keep assist” are often advertised as “self-driving” capabilities in marketing materials, the core difference lies in who bears responsibility for the vehicle’s actions: in a driver-assistance system, the human driver is always legally and operationally responsible for monitoring the road and intervening when needed, whereas an autonomous driving system is designed to assume full responsibility for safe operation when activated. This distinction is not just a technical detail—it is the foundation of the global regulatory framework for autonomous vehicles, and it shapes every design choice made by engineers working in this field. For context, consider a common family trip: if a parent uses adaptive cruise control on the highway and fails to notice a stopped car ahead, causing a collision, the parent is legally at fault. But if a fully autonomous vehicle (operating within its approved use case) makes the same error, the vehicle manufacturer or software developer is held responsible. This shift in accountability is what defines true autonomous driving, and it is the key milestone the industry has been working toward for nearly a century.

The history of autonomous driving is far longer than most people realize, with roots stretching back to the early days of the automobile. The first recorded prototype of a self-directed vehicle was built in 1925 by engineer Francis Houdina, who modified a 1920s Chandler sedan with a radio control system operated from a follow-up car. Dubbed the “American Wonder,” this vehicle was demonstrated on the streets of New York City, where it successfully navigated traffic, honked its horn, and even signaled turns without a human driver in the front seat. While onlookers were initially amazed, many were also alarmed, with some reporting they thought the car was “haunted” when they saw it moving with no one behind the wheel. This early demonstration captured the public’s imagination, but it would take another 60 years for the technology to move beyond remote-controlled prototypes to vehicles that could “see” and navigate their environment on their own.

<!-- [Print] 插图位置：1925年American Wonder无线电遥控车卡通示意图，尺寸：10cm×8cm，位于页面右侧，配文字说明：这辆被认为是世界上最早的“自动驾驶”原型车，由工程师Francis Houdina通过无线电遥控驾驶，1925年在纽约街头行驶时引发了当地居民的轰动 --> <!-- [Digital] 超链接：https://www.smithsonianmag.com/innovation/first-self-driving-car-1920s-180977442/ 史密森尼杂志关于American Wonder的历史报道，含历史照片 --> <!-- [Tag] 历史节点：1925 American Wonder, 无线电遥控车 -->

The next major leap came in the 1980s, when research institutions began developing vehicles that could perceive their environment using onboard sensors rather than external remote control. In 1986, Carnegie Mellon University launched the Navlab project, which built the first autonomous vehicle to use computer vision and radar for navigation. Navlab 1 was a modified Chevrolet van with 5 desktop computers stacked in the back, a bulky camera mounted on the roof, and a top speed of just 30 km/h. It could only drive in a straight line on a clear road, requiring a human driver to take over for turns or when obstacles appeared, but it proved that vehicles could process environmental data to control their own movement. By 1995, the project had evolved to Navlab 5, which completed a cross-country trip from Pittsburgh, Pennsylvania to San Diego, California, driving 98% of the 3,100-mile journey autonomously, with human intervention required for only 3% of the route, mostly for highway on-ramps and off-ramps. This trip demonstrated that long-distance autonomous driving was technically feasible, even with the limited computing power of the era.

The field gained massive global momentum in the early 2000s, when the U.S. Defense Advanced Research Projects Agency (DARPA) launched the Grand Challenge, a competition offering a $1 million prize to any team that could build a fully autonomous vehicle capable of navigating a 142-mile desert course in California’s Mojave Desert. The 2004 competition was a dramatic failure: none of the 15 participating teams completed the full course, with the best-performing vehicle traveling just 7.2 miles before getting stuck on a rock. But the event sparked unprecedented investment and innovation in autonomous driving research. In 2005, Stanford University’s team won the second Grand Challenge with their vehicle “Stanley,” which completed the full desert course in 6 hours and 54 minutes, averaging 19.1 km/h. The 2007 DARPA Urban Challenge pushed the technology even further, requiring vehicles to navigate a mock urban environment with other cars, traffic lights, and intersections. Carnegie Mellon’s team won this event with their vehicle “Boss,” a modified Chevrolet Tahoe that could obey traffic laws, merge into traffic, and park autonomously. Many of the engineers who participated in these DARPA challenges went on to found the leading autonomous driving companies operating today, including Waymo (spun out of Google’s self-driving car project) and Aurora.

The Global Standard for Autonomous Driving Classification: SAE J3016

As the technology evolved from research labs to commercial prototypes, the need for a universal standard to describe autonomous driving capabilities became clear. In 2014, SAE International (formerly the Society of Automotive Engineers, the world’s largest automotive engineering standards organization) released the J3016 standard, which classifies autonomous driving into six distinct levels, from L0 (no automation) to L5 (full automation). This standard is now used by regulators, manufacturers, and researchers worldwide, and it is the framework we will use throughout this book to describe autonomous driving capabilities.

The six levels are defined by two key factors: the degree of automation (how many driving tasks the system can handle) and the responsible party (who is legally responsible for safe operation when the system is active). The following Mermaid diagram illustrates the full classification framework, with examples relevant to everyday life for young readers:

```mermaid graph TD A[SAE J3016: Global Autonomous Driving Classification Standard] --> B[L0: No Driving Automation] A --> C[L1: Single-Function Driver Assistance] A --> D[L2: Partial Driving Automation] A --> E[L3: Conditional Driving Automation] A --> F[L4: High Driving Automation] A --> G[L5: Full Driving Automation]

%% L0 Details B --> B1[Definition: No automated driving functions; human driver controls all acceleration, braking, and steering] B --> B2[Everyday Example: Grandparents’ 1990s manual transmission sedan, with no cruise control or driver aids] B --> B3[Responsible Party: Human driver 100% of the time] B --> B4[Key Feature: No automated support for any driving task]

%% L1 Details C --> C1[Definition: One single driving function is automated; human driver must monitor all other tasks] C --> C2[Everyday Example: Basic cruise control in a 2010s family car, which maintains a set speed but requires the driver to steer and adjust for traffic] C --> C3[Responsible Party: Human driver 100% of the time] C --> C4[Key Feature: Only one task (speed control) is automated; driver must handle all others]

%% L2 Details D --> D1[Definition: Multiple driving functions are automated simultaneously (e.g. adaptive cruise control + lane-keep assist); human driver must monitor the road at all times and intervene immediately if needed] D --> D2[Everyday Example: Most new electric vehicles sold in 2024, including Tesla Autopilot, BYD DiPilot, and Xpilot, which can follow other cars and stay in lanes on highways, but require the driver to keep their hands on the wheel and watch the road] D --> D3[Responsible Party: Human driver at all times, even when automat