Global self-localization for actual mobile robots: generating and sharing topographical knowledge using the region-feature neural network

Abstract

This paper presents an approach to global self-localization (GSL) for autonomous mobile robots using the region-feature neural network. Our approach, which is similar to optical character recognition, categorizes discrete regions of space (topographical nodes) using actual mapped sonar data from two different mobile robots. That is, the mapped sonar data assumes the form of a character unique to its respective region, thereby allowing an autonomous vehicle to determine which room it is in without knowing how or when it got there. With a robust exploration routine, this GSL solution can be time-, translation-, and rotation-invariant. The GSL solution can also become independent of the mobile robot used to collect the sensor data. This suggests that a single robot can transfer its knowledge of various learned regions to other mobile robots. This paper describes the exploration routine and the type of neural network used to solve the GSL problem. It also examines the impact and feasibility of two differently configured robots sharing knowledge on various levels.