Dynamic Mapping For Domestic Service Robot

Mobile service robots are gradually becoming close to people's lives,The transi-tion of robots,from research to industrial processe,is challenging robotics researches.Simultaneous localization and mapping in dynamic environments is a typical problem.Most of the previous researches assume that the environment is static;When serving people,mobile service robots have to face the movement of people,the change of fur-nituresfurniture and,so on.Existing SLAM algorithm cannot meet the needs of this kind application.This paper assessed the characteristics of dynamic environments and proposed a graph based SLAM algorithm for indoor dynamic environment.This work mainly fo-cused on problem formalization,uncertainty analysis,error handling and data associa-tion,related content are as follows:This article firstly introduces the development and progress of the research of the SLAM problem,then describes the graph-optimization based SLAM algorithm,a state of art technology of the SLAM problem.The principle and technical details are presented,in order to analyze new challenges.According to the motion characteristics of obstacles,they are divided into two categories:dynamic obstacles and semi-dynamic obstacles.A framework which combines the extracting geometric features of environments filtering dynamic ob-stacles and the detection of semi-dynamic obstacles by local sub-maps is then proposed solving the SLAM problem in dynamic environments.As to the feature extraction of laser data problem.Based on the mechanism of laser sensors and characteristics of laser data,an adaptive threshold segmentation algorithm is proposed.Laser data are split according to the distance between the real pose and the predicted pose,which is calculated by its neighbor points,of an end point.The geometric feature extraction method and the data fusing strategy all presented in this chapter.As for the data association problem of nodes in a pose-graph,an innovation al-gorithm which fuses odomery data and laser data is given to generate accurate constraints between nodes in a pose-graph.The algorithm can not only reduce the accumulated error of odomery,but also generate an initial pose for the laser match procedure.As for the feature indigent laser data-matching problems,a Fisher information matrix,whose eigenvalues and eigenvectors can depict information contained in laser data,is used to evaluation the probability of a bad match.In order to modeling the movement of semi-dynamic obstacles,Information pose-graph models are defined by adding tags in the laser data and node information,which can reduce the probability of error caused by false loop closure constraints.