Robot Localization with Weak Maps

In this work, we present an approach for indoor localization for a mobile robot based on a weakly-defined prior map. The aim is to estimate the robot's pose where even an incomplete knowledge of the environment is available, and furthermore, improving the information in the prior map according to measurements. We discuss two different approaches to describe the prior map. In the first approach, a complete map of the environment is given to the robot, but the scale of the map is unknown. The map is represented by occupancy grid mapping. We present a method based on Monte Carlo localization that successfully estimates the robot's pose and the scale of the map. In the second approach, the prior map is a 2D sketched map provided by a user, and it does not hold exact metric information of the building. Moreover, some obstacles and features are not fully presented. The aim is to estimate the scale of the map and to modify and correct the prior map knowing the robot's exact pose. The map is represented in the polygonal format in the homogeneous coordinates, and is capable of analyzing the uncertainty of features. We propose two methods to update prior information in the map. One uses a Kalman filter, and the other is based on Geometrical Constraints. Both methods can partially improve the estimate of the dimensions of rooms and locations and the orientation of walls, but they slightly suffer from data association