From L0 to L5: Building and Testing an Autonomous Driving System

With the rapid development of autonomous driving technology, the process of converting a new car from a basic prototype to a road‑ready vehicle has become more systematic. The author shares personal experience on how each subsystem of an autonomous vehicle is gradually installed, debugged, and integrated, turning a "novice" car into a "senior driver".

According to the SAE J3016 standard, autonomous driving is classified into six levels (L0–L5). The article focuses on upgrading an L0 vehicle to L4/L5 capabilities.

1. Line‑Control Retrofit – Replacing the mechanical steering linkage with electronic actuation, similar to using a game controller for a racing game. Vehicles equipped with advanced driver‑assist features (ACC, LKA, valet parking) already have the necessary low‑level control interfaces, making them suitable candidates for line‑control conversion.

After retrofit, functional and performance tests verify the ability to switch between manual and autonomous modes and measure response/feedback latency, which is critical for safety and varies across vehicle types.

2. Mapping and Computing Platform – High‑precision maps define the drivable area; the first step after vehicle selection is map acquisition, analogous to familiarizing oneself with a driving course. A robust on‑board computing platform (power, communication, storage, CPU/GPU) provides the foundation for all higher‑level functions.

3. Sensors and Localization – Selection of LiDAR, millimeter‑wave radar, and cameras depends on cost, power, performance, and field‑of‑view. Proper mounting and calibration (intrinsic and extrinsic) are essential for accurate perception and localization.

4. Perception System – Collected sensor data is used to train models for obstacle and traffic‑sign detection. Iterative data collection, model training, and on‑vehicle testing gradually improve perception capabilities.

5. Control System – Longitudinal control (throttle/brake) and lateral control (steering) are tuned first separately on a closed track, then integrated. Successful integration enables the vehicle to follow recorded trajectories, comparable to passing a driving‑test course.

6. Planning System – The planner decides the vehicle’s future path based on the perceived environment, akin to an instructor directing lane changes or turns during a driving exam. More advanced planning yields higher SAE levels (L3‑L5).

7. Skill Iteration – New features undergo extensive simulation testing, followed by staged real‑world road tests. Data from each test cycle feeds back into simulation, creating a continuous improvement loop that gradually upgrades the vehicle from basic straight‑line driving to complex maneuvers such as overtaking.

After mastering these steps, a fleet of autonomous vehicles can be deployed, enabling faster data collection, parallel testing, and scaling the technology to larger vehicle classes such as buses and trucks.

— End of article.