High‑Precision Maps for Autonomous Driving: Production System and Technical Insights

Introduction This article is the second in the #SpringRecruitmentSeries# and summarizes a talk by Gaode’s High‑Precision Map Business GM Xiang Zhe on the role of high‑precision maps in autonomous driving and the current production system.

Industrial‑level Autonomous Driving Classification Two main categories are described: (1) Assisted‑driving vehicles such as Tesla and XPeng, which require drivers to take over at any time; (2) Level‑4 (L4) fully autonomous vehicles like Google’s self‑driving taxis, which are still in validation and may need 4‑5 more years before widespread use. Both rely heavily on high‑precision maps.

High‑Precision Maps and Autonomous Driving High‑precision maps act as the “brain” of an autonomous vehicle, providing essential information for perception, high‑precision positioning, decision‑making, and vehicle control. At least three of these four functions depend on map data.

Key Map Layers The map consists of four critical layers: road layer, lane layer, positioning objects, and dynamic layer. The road layer aligns HD (high‑precision) data with SD (standard) map data for route planning. The lane layer supplies detailed lane geometry for vehicle control. Positioning objects enable both absolute (GPS/IMU) and relative positioning. The dynamic layer will later incorporate real‑time traffic events.

Challenges on Ordinary City Roads While high‑precision mapping of highways is mature, city streets pose difficulties, especially at intersections where road markings are missing. Accurate mapping of these areas is essential for reliable autonomous navigation.

Map Collection and Production Process The production pipeline includes three stages: data collection, map production, and productization. Collection vehicles equipped with LiDAR, inertial navigation, cameras, etc., gather raw data. Gaode has developed its own high‑precision collection vehicles that offer high accuracy, speed, automation, and safety.

Ensuring Map Freshness To keep maps “fresh,” Gaode employs a three‑tier collection system: (1) expensive professional base‑mapping vehicles for nationwide coverage, (2) professional update vehicles for localized changes, and (3) low‑cost crowdsourced devices for rapid updates. Balancing precision and freshness requires iterative refinement.

Technical Challenges and Research Directions Key research topics include designing multi‑cost, multi‑precision sensing hardware; coordinating heterogeneous devices for synchronized collection; developing algorithms to improve absolute and relative positioning accuracy and ensure seamless data alignment; and integrating image and point‑cloud processing for higher automation in map generation.

Business Impact Gaode’s high‑precision maps have secured commercial orders from major car manufacturers, providing services such as high‑precision positioning, beyond‑visual‑range perception, and lane‑level navigation for intelligent driving models.

Additional Information The article also mentions ongoing spring recruitment opportunities at Gaode and a free e‑book titled “Gaode Technology 2020 Collection,” covering front‑end, algorithms, architecture, automotive engineering, and more.