Autonomous Parking Systems have been evolving rapidly since Valeo first introduced them to the market 15 years ago, initially relying solely on ultrasonic sensors to detect the environment. These sensors have a limited range and field of view, allowing only for the creation of an environmental model confined to the vicinity of the ego vehicle.

With the commoditization of mid-range corner RADARs, a logical enhancement is their integration into parking functionality. This integration enables the creation of larger and more accurate maps, which allows for more advanced functionalities such as trained parking.

The challenge in creating larger maps is maintaining accuracy over long trajectories as odometry errors accumulate. Simultaneous Localization and Mapping (SLAM) is the method used to create a consistent map despite these errors. We use the FAST SLAM approach and tailor it to our needs so that it scales well with the size of the map, at the cost of limited accuracy over very large distances. We find it to be a good compromise for embedded parking systems.