Quick and dirty localization for a lost robot

Abstract

The lost robot problem is tackled here. The robot is placed randomly in an environment and, when started up, has to determine where it is. A new method is presented that employs a SOM to provide a shortlist of candidate locations for the robot. A quick and dirty localization method sits on top of the SOM and disambiguates its output by moving the robot a small distance away from the initial position and accumulating evidence. Two studies are presented that evaluate the accuracy and reliability of the method in worlds of different sizes. These yield favorable results and illustrate the tradeoff between accuracy and reliability. The results show that the location of the robot can be computed with a satisfactory degree of reliability and accuracy within a fairly small radius of uncertainty.