Studying sensor fusion for mobile robot localization

Abstract

A discrete-time linear Kalman Filter that recursively combines motion predictions derived from odometry with measurement corrections from GPS, weighting each information source according to its instantaneous uncertainty through the optimal Kalman gain matrix is developed and applied to a numerical case study. Three localization methods: wheel odometry, GPS, and the Kalman Filter are evaluated in a simulation under realistic noise conditions, including a systematic wheel defect and GPS measurement noise. Results demonstrate that optimal sensor fusion fundamentally resolves the individual limitations of each sensor, providing state estimates that are simultaneously smooth like odometry and globally consistent like GPS, with error characteristics superior to either sensor alone.