Research On Active SLAM Algorithm In Structured Scenarios

Li DAR-based active Simultaneous Localization and Mapping(SLAM)has to deal with two tasks simultaneously: Li DAR SLAM and exploration.Among them,Li DAR SLAM is an indispensable part of active SLAM.However,when state-of-the-art approaches face extreme structured scenes,the extracted geometric features are too sparse to perform effective point cloud registration.In view of this,this thesis proposes an improved Li DAR SLAM approach E-LOAM(LiDAR Odometry and Mapping with Expanded Local Structural Information)and integrates state-of-the-art exploration strategies to achieve efficient and robust active SLAM.State-of-the-art Li DAR SLAM approaches usually extract geometric feature points to perform feature-based registration.The thesis argues that in challenging structured scenes,geometric features are not sufficient to guarantee high-accuracy point cloud registration.Therefore,on the basis of geometric features,this thesis utilizes the intensity value of Li DAR to extract the places with high intensity variations as additional feature points,which enriches the number of feature points.Additionally,this thesis also extracts local structural information around feature points,fits the local structure through the normal distribution function,and realizes the distribution-based registration.By extending in two ways,E-LOAM achieves high-precision pose estimation in structured scenes.Experimental results with the KITTI dataset and real robot platform show the efficacy of E-LOAM.Finally,this thesis combines E-LOAM with TARE(state-of-the-art exploration strategy)to realize an efficient and robust active SLAM.E-LOAM performs real-time highprecision pose estimation,and cooperates with TARE to achieve low-power but highefficiency exploration tasks.The combined active SLAM realizes efficient exploration in structured scenarios.Moreover,experimental results with the simulation system and the real robot platform show that the integrated active SLAM has efficient and robust exploration capabilities.