High‑Precision Mapping and Localization Technologies for Autonomous Driving

High‑precision maps are essential for autonomous driving, providing centimeter‑level road network and traffic information that directly affect perception, prediction, planning, and localization modules.

Two main types are covered: high‑precision topological maps, which include detailed lane geometry, lane types, connections, adjacency, and traffic signals; and high‑precision point‑cloud maps, which use dense LiDAR point clouds with labeled special features such as traffic lights and signs.

The generation pipeline involves massive data collection using GNSS, IMU, wheel odometry, LiDAR, and cameras, followed by preprocessing (data extraction, time alignment, image and point‑cloud de‑distortion), semantic extraction via deep learning, pose optimization using SLAM‑based bundle adjustment, and finally point‑cloud merging to produce the high‑precision map.

High‑precision maps benefit perception (reducing false detections of traffic lights, improving lane and obstacle detection), prediction (enhancing vehicle trajectory forecasts based on lane information), planning (enabling speed, lane‑change, and fuel‑efficient decisions), and localization (providing accurate vehicle pose when matched with offline maps).

Three localization approaches are discussed: satellite positioning (with RTK carrier‑phase correction), map‑based positioning (matching online LiDAR data to pre‑built voxelized maps), and fused positioning that combines satellite, map, IMU, and wheel‑odometry data using EKF and error‑state Kalman filters.

Challenges include GNSS signal loss in tunnels, data errors from abrupt GNSS jumps, and difficulties in dense urban environments where trees or buildings obstruct signals, requiring robust sensor fusion and continuous map verification.

Overall, the article provides a comprehensive overview of high‑precision mapping and localization techniques, their integration into autonomous driving stacks, and the practical obstacles that must be addressed for reliable deployment.