Self-localization based on a short-term memory of bearings and odometry

In this paper we introduce a localization method which is based on a memory of horizontal bearings to landmarks and odometry. The approach is perfectly suited for mobile robots equipped with a camera because bearings can be extracted from images with high accuracy. In contrast to existing approaches, our method does not need any internal representation of the robot's position which is updated by alternating motion and sensor updates. In our approach the location is calculated by applying constraints on the robot's position which are derived from the observations and performed actions that are stored in a short-term memory. We give a detailed description of the method and analyze the properties of different observation selection mechanisms. Results of experiments done in simulation and conducted on a Sony Aibo robot are presented in this paper demonstrating the precision of the method.