Practical mobile robot self-localization

Abstract

A map-making robot integrates accumulated sensor data into a data structure that can be used for future localization or planning operations. Localization is the process of determining the robot's location within its environment. This paper describes experiments in which a robot simultaneously makes a map and localizes to that map. The map is a collection of tangent vectors constructed from stored sonar readings localized to a series of estimated poses. The vectors retain sensed surface normal information to improve accuracy. The localization scheme is a Hough transform into a space described by the robot's current sonar scan. The Hough transform finds a best fit in the presence of both sporadic sensor noise and discretization error.