An GPS/DR navigation system using neural network for mobile robot

Abstract

Dead reckoning (DR) is frequently used for mobile robot navigation as it can provide precise short term navigation information but the errors of a DR system can accumulate over time. A global positioning system (GPS) can be used for outside navigation and localization but the error of a single GPS receiver is still big even though an error intentionally introduced into the system called the selective availability policy (SA) was already removed. Standard differential GPS (DGPS) can provide an accuracy of less than one meter but it is too expensive for the mass market aside from the need of having a base station to provide differential data. This paper proposes a new GPS/DR fusion method based on the data characteristics of a cheap single GPS receiver and use neural network to estimate the output of the GPS receiver to provide precise navigation information to the mobile robot. Simulation results validated the performance of the proposed method and showed its potential use in outdoor mobile robot navigation.