Chapter 6 Localization Algorithms

When people refer to localization, they typically mean obtaining a rough estimate of a target's position on a map. For example, Professor X might be located in his office on the 15th floor, or my current GPS coordinates might be 39.649548, -75.789969. For humans, precise positioning isn't always necessary for route planning. However, for autonomous vehicles, localization accuracy is critical. The system must determine not just a general location—like "somewhere on the Street"—but precise information such as the specific lane, the distance from the curb, and the vehicle's orientation. This level of detail is essential for the onboard computing system to ensure safe and reliable operation.

In a typical autonomous driving system, the localization module performs two primary tasks:

• High-frequency pose tracking: Delivers real-time updates on the vehicle's position to support navigation and object detection.

• Low-frequency global re-localization: Estimates a coarse position to reinitialize the tracking algorithm when pose tracking fails.

The implementation of the localization module is closely tied to the types of sensors installed on the vehicle. In this chapter, we begin by providing an overview of the localization problem, then discuss localization using four different sensor modalities, and finally present the latest research developments in the field.