Pose correction based on landmarks and barriers

Abstract

The techniques and systems herein enable pose correction based on landmarks and barriers. A vehicle pose, one or more radar-occupancy grid (ROG) landmark locations relative to the vehicle, and one or more map landmark locations are received. Based on determined association probabilities of candidate pairs (e.g., treating the map landmark locations as observations), one of the ROG landmark locations and one of the map landmark locations are selected as corresponding to each other. An ROG barrier location and a map barrier location are identified, and ripple point locations are identified that are along the barriers at a radial distance from the landmarks. Based on the ripple point locations and a cost function, a pose correction for the pose is determined. By using the described techniques, reliable vehicle localization can be performed using radar data and a map in a wide array of environments without necessitating other sensors.