Probabilistic Estimation of Incomplete Map Using Gaussian Process

Abstract

Gaussian process is a powerful probabilistic estimation tool which is used widely in engineering fields such as Computer vision, Robotics and sensor networks, etc. This thesis implemented an estimation algorithm of the total map with sparse sensing data using Gaussian Process. In the implemented algorithm, two kinds of kernel functions are applied to the spatial Gaussian Process model

squared exponential kernel and neural network kernel. The performance of the proposed algorithm was verified by the experiments with a simple mobile sensor network. To construct a simple mobile sensor network based on ROS (Robot Operating System) platform, a two wheeled mobile robot (Pioneer3DX) and a two dimensional laser scanner (SICKlms200) are used.