Autonomous Sonar Navigation in Indoor, Unknown and Unstructured Environments

A mobile robot operating autonomously in unknown, unstructured environments has to be able to map its environment while at the same time determining its own position accurately within this environment. This paper presents an approach where the bootstrapping problem of concurrent localisation and map building is solved by estimating the respective errors introduced by each of the processes and correcting them accordingly. The success of this approach also hinges on the ability to determine which measurement originates from which feature. A heuristic multiple hypothesis data association framework is developed to deal with this problem. The problems encountered with the implementation of the algorithms on the mobile robot ROAMER are discussed. Real experiments in typical office environments have shown that the robot is able to navigate autonomously in such indoor environments. >