Localization

The vehicle’s localization algorithm is responsible for determining the positioning of the car relative to a selected reference system. Localization algorithms can be divided into 2 categories: relative and absolute reference system.

In the case of relative localization, the positioning of the car is determined using a local coordinate system, which can be placed at the vehicle’s starting point or hooked to the car to move with it. The coordinate system hooked to the vehicle enables the identification and tracking of traffic lanes and detection of obstacles relative to the vehicle. Such localization allows the car to move autonomously along a given route and perform basic maneuvers, such as overtaking, adjusting the speed to the vehicle in front, and avoiding collisions.

This type of localization does not allow the car to decide which road to take at an intersection or highway exit. The lack of information on the localization relative to the map prevents the vehicle from driving to a designated destination.

Absolute localization determines the positioning of the vehicle relative to the Earth’s global coordinate system, making it possible to plan a route to a destination, thus enabling the vehicle to perform tasks required by the Level 5 autonomy.

With absolute localization, it is possible to prepare a detailed map of the area in advance, which is later used to determine the positioning. For example, such a map can be created using a lidar. It will be a 3D representation of the environment’s structure.

Determining the exact location of the vehicle is a demanding task which cannot be accomplished using a single sensor, because there is no ideal sensor that would always indicate the exact positioning. That is why it is necessary to apply multiple sensors that complement each other.

When it comes to the localization algorithm, frequently used sensors include speedometer, the GNSS module, cameras, lidar, and the IMU sensor. Data from these sensors differs in accuracy and error characteristics so it is necessary to use data fusion algorithms, such as the Kalman filter.

Go back to technology