Research On Indoor Hierarchical Mapping Of Service Robots

The perspection and understanding of indoor environments for service robots is the foundation and long-range goal of Artificial Intellegence(AI),since only service robots understanding the human environment,can human-robot interaction be achieved.Generally,the human-robot interaction is accomplished by human sending tasks to service robots,directly,and then,service robots perform these tasks.Those tasks,which are given by the human language,are semantic relations hiding many complicated information.Those information includes locations of objects,descriptions of objects and relations of objects,corresponding to metric maps,feature maps and semantic maps,respectively.Single type of map hardly satisfies tasks given by the human language,thus,it is significant for the basic research of service robots to construct a comprehensive indoor map.However,different types of maps can not be merged,easily,but hierarchical map is an effective way to achieve maps merging,thus,hierarchical mapping is the research emphasis in this work.In order to achieve the hierarchical map ping for service robots,researches are focused on the metric mapping,the feature mappin g and semantic mapping,respectively:The metric map is the foundation of the hierarchical mapping,and the pose estimation is the foundation of the metric mapping.A Norm(PMWN,Polar Metric-Weighted Norm)is proposed to achieve the scan matching of 2D scans,which can control influences caused by the rotation and the translation of service robots and reduce iterations to improve the efficiency.The rotation angles are calculated by residual error estimation,and the translation are calculated by the geometric property of PMWN.Finally,the metric map is constructed by fusing poses given by the odometer and PMWN,respectively,based on KF(Kalman filter).The feature map is middle layer of the hierarchical map.A division of 2D scans provided by the metric map based on hypothesis testings is proposed to extract corner features,segment features and arc features.Feature matching methods based on hypothesis testings are adopted to match similar features,which are then fused to one stable feature by their weights.Meanwhile,those features are used to achieve the loop closure detection to improve the accuracy of the metric map.After the feature extraction,the feature matching and the feature fusion,the feature map is constructed,finally.The semantic map is the top layer of the hierarchical map.A semantic region is proposed to express the semantic map.Semantic regions are formed by semantic symbols,which contain the local contour of indoor environment,and semantic symbols are collected by matching the local contours of the service robot with semantic symbols.Weights of semantic symbols are updated by Particle filter,and then,the current semantic region the service robot entering is estimated by Viterbi algorithm.A semantic navigation algorithm is achieved by Bayesian Network to navigate the service robot to the specified region.The hierarchical map is consisted of the metric map,the feature map and the semantic map.Tasks of service robots are wrapped by the cascade of each layer in the hierarchical map to achieve the description of tasks in human langu age.If small changes are occured in indoor environment,the relations of each layer in the hierarchical map can be changed,flexibly,to fit into the unchanged tasks.Firstly,experiments about the metric map,the feature map and semantic map are conducted,and then,comprehensive experiments for the hierarchical map are achieved.The experiments for each map are different.First,for the metric map realized by PMWN,we compare the PMWN method with PSM and Mb-ICP to evaluate the accuracy,efficiency and loop closure of PMWN.Second,for the feature map achieved by features,the accuracy and the stability of features are testified by features extraction experiments in the complex environment and large-scale environments.Third,for the semantic map implemented by semantic regions,the long-term abilities are verified by the long-term experiment and the robot kidnapping experiment.Finally,for the hierarchical map,its flexibility can be testified by experiments of changing the indoor environment without changing the task,and the stability of the hierarchical map for the robot executing tasks can be proved by experiments of the service robot perform a repeated task in the large-scale environment.