Research On Autonomous Localization Of Indoor Mobile Robot Based On Graph Optimization SLAM

Simultaneous Localization and Mapping(SLAM)is a critical technology for mobile robots to achieve autonomous localization and navigation.Li DAR is the most used sensor in SLAM technology.However,due to the limited Li DAR data and weak scene recognition ability,the traditional laser SLAM system is prone to the problem of significant positioning error or even positioning loss in the environment with a single feature.In addition,due to the influence of both the Li DAR itself and external factors,noise points are prone to appear during scanning,which can also reduce the positioning accuracy of the system.To improve the positioning accuracy of the system under the above circumstances,this thesis uses the graph optimization SLAM algorithm conducts in-depth research on the indoor mobile robot from the two perspectives of the detection of the physical environment with a single feature and the noise reduction of the Li DAR data and proposes two effective improvement methods.The main research contents of this thesis are as follows:(1)A long corridor detection method is proposed to improve positioning accuracy.Scan-tosubmap is the most widely used scan-matching method in laser SLAM systems.To improve the positioning accuracy of the laser SLAM system in a long corridor environment,this thesis proposes a scan-to-submap-based long corridor detection method suitable for structured environments.The covariance matrix is constructed according to the pose transformation matrix generated in the scanto-submap process and the corresponding score.The ratio(k)of diagonal elements of the covariance matrix represents the uncertainty difference in the x and y directions during scan matching.Finally,the pose of the odometer,IMU,and scan matching are fused by the parameter k,and a more accurate robot pose is obtained.(2)A clustering denoising method based on keyframe extraction is proposed.The accuracy of Li DAR data directly affects the positioning accuracy of robots.Therefore,this thesis proposes a clustering noise reduction method based on keyframe extraction suitable for unstructured environments.Referring to the concept of keyframe in visual SLAM,the histogram is used to reduce the dimension of the data,and then the similarity between the histograms is compared.The scan with low similarity is the keyframe.The scan without new environment features is discarded through keyframe extraction,and only the extracted keyframes are denoised.To reduce the time complexity of the clustering process,a region segmentation method suitable for 2D Li DAR is proposed in this thesis.When clustering,only the points in the same region are traversed,significantly reducing the traversal dimension.The proposed method can reduce the overall calculation pressure of the system and improve the positioning accuracy of the system.The thesis uses Gazebo simulation,MIT data set and ROS mobile robot to verify and analyze the proposed algorithm.When experimenting in a corridor environment,the proposed corridor detection method can reduce environmental feature mapping errors by about 20.2%.When experimenting in an indoor environment,the proposed clustering noise reduction method can improve the accuracy of feature mapping by 17.5% while reducing the average CPU usage by 10.6%.