Mobile Robot Active Localization Based On Heuristic Search

Mobile robot localization is the foundation of realizing robot’s autonomous navigation.However, most current localization approaches are passive, i.e., they do not exploit theopportunity to control the robot’s actuators during localization. These approaches are apt to beinefficient in localization since the robot wanders aimlessly. Active localization that has beenpaid more and more attentions in recent years, not only considers the problem of poseestimation but also researches how to control the robot’s behavior on basis of the results oflocalization.However, most existing active localization approaches need to calculate entropy foreach possible path, which leads to a relatively high algorithm complexity and poor real-timeperformance. We use Monte Carlo localization (MCL) as the basic localization algorithmand several aspects of research are studied in the thesis focused on the mobile robot activelocalization algorithm.(1) The various models associated with robot localization are established, includingcoordinate system, map, odometry, motion and observation model; The Bayesian Filter andMarkov localization are analyzed. The Particle Filter localization algorithm is emphaticallyresearched, and the sample impoverishment is discussed. Meantime, we introduce the basicactive localization and mainly analyze several improved algorithms and point out theircharacteristics and shortcomings.(2) A multiple hypothesis active removal algorithm based on heuristic search is proposed.The samples often aggregate into some clusters after the global localization begins. Eachcluster is considered to be a hypothesis of where the robot might be located. Firstly aclustering algorithm is used to separate all the particles into diferent clusters, then constructthe solution space trees and determine the weight of each node in the tree. Finally, ourapproach applies priority queue-type branch and bound algorithm to solve the optimal pathsearch problem. The aim is to distinguish the different clusters through actively controllingthe robot’s motion, and to converge to the true pose as soon as possible.(3) The localization accuracy could gradually reduce in some environments after theparticles converge into the single cluster. Therefore, based on the above work, we present thelocalization accuracy active improvement algorithm based on heuristic search. The maindistinction between both proposed active localizations is the determination of the node weight.This novel approach is to further improve the localization accuracy and keep it in high level.The conclusions and directions for future research work are discussed in the last of thisthesis.