GPS error range reduction method based on linear kinematic model

GPS is an effective tool to localize robots within an absolute coordinate frame. However, depending on the application, even the accuracy provided by the GPS in an ideal environment might be too low for precise control. In this work we propose a method for reducing the error range of GPS measurements based on an assumption of linear motion. While previous research in this area has mainly improved accuracy using pre-processing and Map-matching, the proposed method makes an assumption of one-dimensional motion, since most vehicle moves linearly, for making corrections. Raw measurements are projected onto a line generated using a method of least squares linear regression. Therefore the error range is reduced to the distance of the intersections of the rail line and the ellipsoid body that indicates the measurement error. Through experiments we demonstrate that the error range reduced at most 15 % at locally compared to the raw measurements. It is possible to apply this method also to other kinematic models, Map-matching algorithms, consumer GPS, and in real-time applications.