Navigation And Mapping System For Multicopter Based On LiDAR And INS In GPS-denied Environments

Sufficient environment perception and precise state estimation are the premise of intelligent applications of multicopter.However,in GPS-denied environments,the navigation of multicopter and the mapping of the environment are still difficult.To this end,this paper design a multicopter system that can realize the navigation and mapping functions based on light detect and ranging(LiDAR)and inertial navigation system(INS)in GPS-denied environment.We also study the registration of three-dimensional(3D)point cloud in detail.The main contributions made by this thesis are listed as follows:(1)In engineering practice,following the task-driven process,we designed and built a hardware platform taking performance and cost into consideration.(2)In algorithm,to avoid the shortcomings of classical point cloud registration method in efficiency and the robustness when registering sparse point cloud,we design our method based on the GP-SLAM(regionalized Gaussian process map reconstruction based simultaneous localization and mapping)in three-dimensional(2D)space.In order to cope with the difficulties of extending the algorithm to 3D space,the data association method is redesigned,which avoids the precision loss cause by the wrong prediction direction in sub-regions,this modification also significantly accelerates the convergence speed.Meanwhile,we analyze the complexity of several modules of the algorithm and reduced them.By this way,we achieve real-time and accurate pose estimation and map building.(3)In software architecture,in order to cope with the high dynamic movement and high-frequency vibration of multicopter,we make use of the information from INS and barometer to improve the robustness of point cloud registration.In order to ensure both efficiency and accuracy,a high-low frequency two-thread architecture is designed.The high-frequency thread registers the point cloud to offer real-time pose estimation,while the low-frequency thread is responsible for refinement of map and pose estimation.(4)We implemented 7 experiments to evaluate the performance of our method and system.The experiments include the verification of 2D GP-SLAM,mapping test of 3D GP-SLAM using spinning 2D Li DAR and the mapping test and large-scale location test of the system.The results show that our method can be applied to different sensor configurations,and performs well in structured or unstructured,indoor or outdoor and large-scale scenes.Our system is capable to process sensor data in realtime,and the drift of pose is under 1%.