第一章 自动驾驶的起源与发展脉络 / Chapter 1 Origin and Development Context of Autonomous Driving

自动驾驶（Autonomous Driving）是指搭载先进传感器、控制器、执行器等装置的车辆，能够在部分或全部驾驶场景下替代人类驾驶员完成对车辆的感知、决策、控制和执行操作，实现安全、高效、合规的行驶目标。作为智能交通系统的核心组成部分，其技术范畴覆盖从低级的驾驶辅助到高级的完全无人驾驶全谱系，核心价值在于降低人为失误导致的交通事故、提升交通运行效率、重塑未来出行生态。> Autonomous Driving refers to vehicles equipped with advanced sensors, controllers, actuators and other devices, which can replace human drivers in some or all driving scenarios to complete perception, decision-making, control and execution operations of vehicles, and achieve safe, efficient and compliant driving goals. As the core component of intelligent transportation systems, its technical scope covers the full spectrum from low-level driving assistance to high-level fully driverless driving, with core values of reducing traffic accidents caused by human error, improving traffic operation efficiency, and reshaping the future mobility ecosystem.

国际汽车工程师学会（SAE International）在2021年发布的《J3016驾驶自动化分级标准》是全球通用的自动驾驶技术分级依据，将驾驶自动化划分为L0至L5共6个等级，不同等级的核心差异在于驾驶任务的承担主体、系统的作用边界以及驾驶员的责任范围。下表清晰展示了各等级的核心特征： | 驾驶自动化等级 | 驾驶任务承担主体 | 驾驶员责任 | 系统责任边界 | 典型应用场景 | 商业化落地时间 | |----------------|------------------|------------|--------------|--------------|----------------| | L0 无自动化 | 人类驾驶员 | 全部驾驶任务 | 仅提供预警 | 车道偏离预警、碰撞预警 | 已普及 | | L1 驾驶辅助 | 人类驾驶员+系统 | 全部驾驶任务 | 单一维度控制（转向/加减速） | 自适应巡航、车道保持 | 已普及 | | L2 组合驾驶辅助 | 人类驾驶员+系统 | 驾驶员全程监控 | 多维度协同控制 | 特斯拉Autopilot、小鹏NGP | 2015年后普及 | | L3 条件自动驾驶 | 系统为主 | 系统请求时接管 | 特定场景下完全负责 | 高速公路自动驾驶、拥堵跟车 | 2023年奔驰Drive Pilot率先落地 | | L4 高度自动驾驶 | 系统为主 | 无需接管 | 特定区域/场景下完全负责 | Robotaxi（自动驾驶出租车）、无人配送 | 2018年Waymo率先试点 | | L5 完全自动驾驶 | 系统 | 无责任 | 全场景全工况完全负责 | 所有乘用车、商用车场景 | 预计2030年后落地 |

> The SAE International's *J3016 Levels of Driving Automation Standard* released in 2021 is the globally accepted basis for autonomous driving classification, dividing driving automation into 6 levels from L0 to L5. The core differences between levels lie in the subject responsible for driving tasks, the boundary of system functions, and the responsibility scope of drivers. The table above clearly shows the core characteristics of each level.

> 注：L3与L2的核心差异在于责任主体转移：L2阶段若发生事故责任主体为驾驶员，L3阶段在系统激活的特定场景下责任主体为汽车厂商/系统开发者。L4与L3的核心差异在于接管需求：L3阶段驾驶员需随时准备接管车辆，L4阶段系统可完全自主运行，无需人类介入。 > Note: The core difference between L3 and L2 is the transfer of responsibility: in L2, the driver is responsible for accidents, while in L3, the car manufacturer/system developer is responsible in specific activated scenarios. The core difference between L4 and L3 is the takeover requirement: in L3, drivers need to be ready to take over the vehicle at any time, while in L4, the system can operate completely autonomously without human intervention.

一、自动驾驶的起源：从科幻设想到技术雏形

> ### 1.1 Origin of Autonomous Driving: From Sci-Fi Concepts to Technical Prototypes 自动驾驶的概念并非诞生于21世纪的AI热潮，其雏形可追溯至20世纪初的无线电控制技术探索，甚至更早的科幻文学想象。1925年，美国发明家Francis Houdina团队改造了一辆 Chandler 汽车，通过车顶安装的无线电接收装置，由后方跟随的车辆发送控制信号，实现了车辆的前进、转向和刹车操作，这辆被称为“Phantom Car（幻影车）”的装置在纽约街头公开行驶，吸引了大量民众围观，被认为是全球首个公开演示的自动驾驶原型，虽然其本质是远程遥控而非自主决策，但首次验证了“无需人类驾驶员操控即可行驶”的可行性。> In 1925, a team led by American inventor Francis Houdina modified a Chandler car, which was equipped with a radio receiver on the roof. Control signals sent by a following vehicle enabled the car to move forward, turn and brake. This vehicle, known as the "Phantom Car", was publicly driven on the streets of New York, attracting large crowds of onlookers. It is considered the first publicly demonstrated autonomous driving prototype. Although it is essentially remote-controlled rather than autonomous decision-making, it first verified the feasibility of "driving without a human driver".

这一概念很快进入科幻领域：1936年，通用汽车在纽约世界博览会上展出了名为“Futurama”的未来城市模型，其中就描绘了配备无线电控制系统的汽车实现自动跟车、自动避障的场景；1940年，通用汽车总裁 Alfred Sloan 公开预测“未来汽车将通过无线电控制实现自动驾驶，彻底消除交通事故”。这些科幻设想虽然不具备技术可行性，但为自动驾驶技术的研发埋下了思想种子。

> This concept quickly entered the field of science fiction: In 1936, General Motors exhibited a future city model named "Futurrama" at the New York World's Fair, which depicted cars equipped with radio control systems achieving automatic following and obstacle avoidance. In 1940, Alfred Sloan, president of General Motors, publicly predicted that "future cars will achieve autonomous driving through radio control, completely eliminating traffic accidents." Although these sci-fi predictions were not technically feasible, they planted the seed of thought for the research and development of autonomous driving technology.

真正意义上的自主决策型自动驾驶原型诞生于人工智能技术发展的萌芽期。1961年，斯坦福人工智能实验室（SAIL）研发了全球首台移动机器人“Shakey”，这台身高1.5米、重达270公斤的机器人配备了摄像头、碰撞传感器和无线电通信装置，能够在室内环境中自主识别障碍物、规划路径并完成移动指令，虽然其行驶速度仅为0.03米/秒，且只能在简单的结构化环境中运行，但首次验证了“机器自主感知环境、自主决策、自主控制运动”的技术路径，被认为是现代自动驾驶的技术鼻祖。

> The truly autonomous decision-making autonomous driving prototype was born in the embryonic stage of artificial intelligence technology development. In 1961, the Stanford Artificial Intelligence Laboratory (SAIL) developed the world's first mobile robot "Shakey". This robot, 1.5 meters tall and weighing 270 kilograms, was equipped with cameras, collision sensors and radio communication devices. It could autonomously identify obstacles, plan paths and complete movement instructions in indoor environments. Although its moving speed was only 0.03 meters per second and it could only run in simple structured environments, it first verified the technical path of "machine autonomous perception of environment, autonomous decision-making and autonomous control of movement", and is considered the technical ancestor of modern autonomous driving.

1970年代，日本筑波工程研究实验室研发的“自主车辆”首次在公共道路上完成了10米/小时的自动驾驶测试，该车辆通过摄像头识别道路白线，配合机械转向装置实现车道保持，标志着自动驾驶技术从实验室走向真实道路场景。同一时期，美国卡内基梅隆大学（CMU）启动了“Navlab（导航实验室）”项目，研发了搭载计算机视觉系统的自动驾驶测试平台，1986年发布的Navlab 1搭载了彩色摄像头和专用的图像处理计算机，能够在结构化道路上以30公里/小时的速度行驶，自动识别车道线和障碍物，是首个具备实用价值的自动驾驶测试平台。

> In the 1970s, the "Autonomous Vehicle" developed by Japan's Tsukuba Engineering Research Laboratory completed the first autonomous driving test on public roads at a speed of 10 km/h. The vehicle identified road white lines through cameras and achieved lane keeping with mechanical steering devices, marking that autonomous driving technology moved from the laboratory to real road scenarios. In the same period, Carnegie Mellon University (CMU) launched the "Navlab (Navigation Laboratory)" project and developed an autonomous driving test platform equipped with computer vision systems. Navlab 1 released in 1986 was equipped with color cameras and a dedicated image processing computer, which could drive at a speed of 30 km/h on structured roads, automatically identifying lane lines and obstacles. It was the first autonomous driving test platform with practical value.

> 【侧边栏：术语小百科】 > 【Sidebar: Glossary】 > 激光雷达（LiDAR, Light Detection and Ranging）：通过发射高频激光束，测量激光反射回传感器的时间差计算距离，可生成周围环境的厘米级精度3D点云图，是目前L4及以上自动驾驶的核心传感器，缺点是成本较高、受极端天气影响较大。 > LiDAR (Light Detection and Ranging): By emitting high-frequency laser beams and measuring the time difference of laser reflection back to the sensor to calculate the distance, it can generate a 3D point cloud map of the surrounding environment with centimeter-level accuracy. It is the core sensor for L4 and above autonomous driving, with the disadvantages of high cost and high impact from extreme weather.

二、自动驾驶的发展阶段与技术突破

> ### 2. Development Stages and Technical Breakthroughs of Autonomous Driving 按照技术成熟度、产业参与主体和商业化进程，自动驾驶的发展可划分为四个阶段：概念萌芽期（20世纪初-1980年）、技术探索期（1980-2000年）、科研突破期（2000-2010年）、商业化落地期（2010年至今）。

2.1 概念萌芽期（20世纪初-1980年）：无线电控制与科幻想象的验证

> #### 2.1 Concept Embryonic Period (Early 20th Century - 1980): Verification of Radio Control and Sci-Fi Imagination 这一阶段的核心特征是自动驾驶从纯科幻设想走向初步的技术验证，技术路径以远程无线电控制为主，尚未实现真正的自主决策。除了前文提到的1925年Phantom Car、1961年Shakey机器人外，1977年日本筑波大学的自动驾驶测试车首次搭载了图像处理系统，能够识别道路白线并实现车道保持，虽然行驶速度仅为10公里/小时，且仅能在无其他车辆干扰的封闭场地运行，但首次验证了“视觉感知+自主决策”的技术可行性，为后续的技术研发奠定了基础。

> The core characteristic of this stage is that autonomous driving moved from pure sci-fi imagination to preliminary technical verification, and the technical path was dominate