High-Order Moment Models of Landmark Distribution for Local Homing Navigation

For local homing navigation, a mobile robot is supposed to return home using snapshots of the surrounding environment. It basically follows the snapshot model, comparing the home snapshot and the current view to determine the homing direction. In this paper, we suggest a high-order moment potential to describe the landmark feature distribution for local homing navigation. The moment potential function calculates the sum of products of the feature and the distance of landmark particles as a holistic view, allowing a high order of the distance. It effectively combines the range sensor values of landmarks in the current view and the visual features. By analogy with the moment in physics, the center of the moment is estimated as the reference point, which is the unique convergence point in the convex moment potential from any view, and using the property, the gradient of moment potential at the current position and home location can derive the homing vector. We provide a proof of convergence for any moment potential with order greater than or equal to one. Also, we demonstrate homing performances with various moment models in real environments to validate our models. The suggested moment models combining both landmark distance and visual feature have better performances than the visual information alone, and high-order moment potentials can be searched to obtain a better description of landmark distribution for a given environment.