Autonomous Navigation and Planning Technology for Quad-rotors Unmanned Aerial Vehicle (UAV) System

Yongbo Chen

Abstract: : For a robot to navigate in an unknown environment, it needs to solve simultaneous localization and mapping (SLAM) problem to keep estimating its pose and, simultaneously, building a map of its surrounding environment. This thesis investigates the graph structure of SLAM and applies it in the related problems, including anchor selection and active SLAM, with the applications for Quad-rotors UAV system. First, we explore the relationship between the graphical structure of 2D and 3D pose-graph SLAM and Fisher information matrix, Cramer-Rao lower bounds, and its optimal design metrics based on the assumption of isotropic Langevin noise for rotation and block-isotropic Gaussian noise for translation. Second, we present a high-efficient greedy algorithm, using Cholesky decomposition, approximate minimum degree permutation, order re-use, and rank-1 update technologies, for the problem of choosing a set of anchored poses from a set of possible or potential poses, that minimizes estimated error in pose-graph SLAM, which is proved to be a non-normalized monotone decreasing sub-modular maximization optimization with a cardinality-fixed constraint. Third, as applications of the graph structure results, based on map joining, two active SLAM methods with two different frameworks are presented: one for 2D feature-based SLAM and the other for 3D pose-graph SLAM.