4D imaging automotive radars are high-resolution, long-range sensors that can scan in both the horizontal and vertical planes, while identifying the distance, direction and relative velocity of detected objects. They are an increasingly important tool in the fight to reduce road deaths. With higher resolution than earlier radar sensors, 4D radars can detect many different reflection points from the same object, which by mapping them out, begin to resemble an image. SLAM, or simultaneous localisation and mapping, is the computational process of building and updating  a map of these reflection points from the unknown surrounding area while simultaneously keeping track of the radar's position whitin the plot. It is an esential input for autonomus systems in that it estimates the position and orientation of the sensor while creating a view of the scene in which the system can operate. It is therefore an important input to all AI code or, in the case of radars, detection algorithms. Radar based SLAM is proving increasingly popular as other sensors suffer from detection difficulties in extreme weather conditions.

For all sensors, an essential input for generationg SLAM is the sensor's position or change in position between readings as, if this is unknown, a coherent, accurate SLAM output cannot be formed and safety algorithms cannot be run. While satellite navigation provides a good method for tracking the sensor's position, the project here seeks to develop a method which can be deployed on Provizio's 4D imaging radar sensor and used when PNT data is unavailable (i.e. when traveling in tunnels etc.) or to improve overall accuracy when used in conjunction with PNT data.