Service Robot Navigation In Large,Dynamic And Complex Indoor Environments

Robot navigation is a fundamental research in robotics,the ability of moving freely in environments is the prerequisite to accomplish other complex tasks for robots.In recent decades,with the advancement of robotic technologies and the growing demand for service robots in society,academia and industry have invested a lot of resources on the reasearch and application of robot navigation,which makes the indoor robot navigation technology much more mature.Indoor robot navigation technology can be divided into three stages:industrial phase,domestic service phase and public sevice phrase.The public service scenarios are more complex than factory scenarios or home scenarios,therefore they bring new problems and challenges to the current indoor robot navigation technology,which is also the background of this study.At present,the mainstream framework of indoor robot navigation includes three major components:map-building,robot localization and navigation control.Although these techniques have been developed and progressed in home scenarios,when be ap-plied to large-scale,complex and dynamic public service scenarios,they are still faced with many engineering difficulties and theoretical problems.This thesis attemps to make a step on these issues,the main work and contribution are summarized as fol-lows:1.Robot navigation requires detailed environmental model,therefore to develop a map-mapping algorithm suitable for large-scale environment is the basic work of robot navigation.In most cases,the sizes of the public service scenarios are much more larger than ordinary home scenarios,they often can reach tens of thousands square meters or even larger scale.The existing map-building algorithms in large scale scenarios are inefficient and resource-consuming.In this work,an efficient representation of quadtree map is proposed and a coding mechanisum is adopt to accelerate the speed of accessing node data in quadtree.Besides,a map-building algorithm based on quadtree map representation is designed for large-scale envi-ronments,the experiment shows that the proposed method could reduce 50%-70%memory consumption.2.The public service scenarios are complex and highly dynamic,especially with the flow of crowds.Robot navigation in crowded environments depends on re-liable robot localization.However,robot localization in dynamic environments is quite challenging.This work first analyzes the causes of the localization fail-ures in dynamic environments,and then proposed two improvements based on odometry calibration and self-perception of dynamics in environments.The new proposed method greatly reduces the impact of erroneous observations and thus significantly improves the robustness of localization in dynamic environments.3.The traditional methods of navigation control lack consideration of pedestrian factors in environments,thus the navigation performance can't meet the needs of human-aware navigation in public scenarios.Aiming to solve the problems existing in such case,pedestrians are identified and tracked in our system,and their movement status is estimated and predicted.the pedestrian prediction is used to improve the local path planning algorithm and make the robot navigation more nature.In general,this work improves the current robot navigation tenniques from three aspects:map-building,robot localization and navigation control.The proposed meth-ods and improvements are realized and verified in the project of KeJia shopping mall guide,which is deployed in a real shopping mall for about 40 days.