Using the Global Positioning System for Relative Navigation of Mobile Robots

Mobile robots rely on many sensors including the Global Positioning System (GPS) for navigation. It is important to consider the particulars of the application when determining how best to incorporate GPS observables into a state estimation pipeline. In this thesis, we show how GPS can be fit into a relative navigation paradigm. We begin by developing a single-receiver GPS odometry pipeline that achieves a relative drift rate of 0.6%. We then combine GPS with visual localization and obtain few-centimetre-level path-tracking accuracy on a joint indoor-outdoor route with large appearance change. Finally, we implement GPS odometry as an alternative to visual odometry (VO) in Visual Teach Repeat 3 providing robust navigation in the case of sensor failure. Our methods were validated through both offline evaluations on datasets we collected and live closed-loop robot experiments.M.A.S