Research On LiDAR SLAM/INS/UWB Multisource Information Fusion Positioning Theory And Method

With the rapid development of intelligent unmanned systems such as autonomous vehicles,robots and drones,highly intelligent robots have assisted or replaced humans in some of work areas and are widely used in emergency rescue,industrial automation and a smarter life.Highly real-time,highly accurate,highly reliable and highly available navigation and positioning information is a necessary foundation and important guarantee for the effective environmental perception,path planning and motion control of mobile robots.As work scenarios become increasingly challenging,the limitations and uncertainties of a single sensor are unable to meet the increasingly rich functional requirements of mobile robots,and multisource information fusion technologies that provide continuous,accurate and robust navigation and positioning information are gradually attracting attention.Global Navigation Satellite System(GNSS)/Inertial Navigation System(INS)multisource fusion positioning systems are generally considered to be a reliable solution for achieving high-accuracy outdoor positioning.In GNSS-denied indoor scenarios,Light Detection and Ranging Simultaneous Localization and Mapping(LiDAR SLAM),INS autonomous dead reckoning navigation and Ultra-Wide Band(UWB)wireless positioning technologies have great potential,but also certain limitations.Therefore,this thesis is dedicated to developing a multisource fusion positioning system with high real-time,high accuracy,high reliability and high availability to fully exploit the performance advantages of different sensors and deeply explore the complementary characteristics between heterogeneous sensors for maximum information gain,thereby improving the overall performance of mobile robots.The main research work and contributions are as follows:(1)To improve the performance of UWB ranging and positioning in real scenarios,a LiDAR point cloud map-based UWB None Line of Sight(NLOS)identification algorithm with less environmental dependence,less prior information and strong universality is proposed.Considering that LiDAR SLAM can directly and accurately acquire spatial fingerprint information from the surrounding environment at high frequency,and can produce higher quality environmental maps with lower effort,the LiDAR point cloud map generated using LiDAR SLAM is combined with position information from UWB anchors for NLOS identification.The tedious data collection and training phase are not needed in the early stage and the frequent spatial changes are handled well.Experimental results show that the UWB NLOS identification algorithm can accurately and efficiently identify NLOS measurements,without significantly increasing the computational effort and meeting the real-time requirements.(2)To prevent the effect of a poor geometric distribution on the positioning system after excluding NLOS measurements and improve the utilization of UWB measurements,an NLOS correction algorithm using a grey prediction model is proposed.For poor Line of Sight(LOS)anchor geometric distributions,the grey prediction model fills in the gaps by predicting NLOS measurements based on historical measurements.The addition of the UWB corrected measurements effectively improves the original poor geometric configuration and minimizes the positioning system Dilution of Precision(DOP),thereby improving the positioning accuracy and robustness of the system.A dynamic positioning experiment verifies the rationality and effectiveness of the UWB NLOS correction algorithm.(3)To alleviate LiDAR SLAM error accumulation over long-term runs while reducing its dependence on loop closure detection,a Robust Kalman Filter(RKF)-based LiDAR SLAM/UWB tightly coupled positioning model is proposed using LiDAR SLAM positioning information,UWB LOS measurements and corrected measurements as inputs to the integrated system.The introduction of robust estimation can minimize or mitigate the undesirable effects of UWB gross errors on the state estimation while preventing the filtering parameters from overdiverging,thereby improving the positioning accuracy and positioning performance of the integrated system.A dynamic positioning experiment demonstrates the accuracy and robustness of the RKF-based LiDAR SLAM/UWB multisource fusion positioning system.(4)To eliminate the motion distortion of LiDAR point clouds and the error associated with the linearization of a conventional Kalman filter,an INS-centric Error-State Robust Iterated Kalman Filter(ESRIKF)-based LiDAR SLAM/INS/UWB tightly coupled fusion algorithm is proposed by introducing an Inertial Measurement Unit(IMU)of proprioceptive sensor to achieve more continuous and accurate locally smooth and globally consistent state estimation.The LiDAR and UWB sensor data are separately fused into the system state at their respective reception times,by continuously correcting the IMU navigation results to eliminate drift errors in the state estimation.To strictly control the data quality of the system input,the UWB-Inertial Odometry(UIO)system considers the ranging accuracy and the geometric distribution,while also introducing robust estimation to suppress the adverse effects of abnormal measurements on parameter estimation.Multiple ESRIKF iterations effectively reduce the linearization error of conventional filters,allowing the system state to better converge to the optimal solution.A dynamic positioning experiment validates the superiority of the ESRIKF-based LiDAR SLAM/INS/UWB multisource fusion positioning system and its robustness against sensor failure,with root mean square errors of 0.059 m,0.064 m and 0.087 m in the X,Y and plane directions,respectively,realizing subdecimeter indoor high-precision positioning.In summary,this thesis systematically studies a LiDAR SLAM/INS/UWB multisource fusion positioning method,independently designs the core algorithm and builds a multisensor synchronous acquisition system and algorithm research software and hardware platform.The algorithm and overall system performance are tested and analyzed using measured data.The research results provide a novel and comprehensive feasible solution for high-precision intelligent unmanned system navigation and positioning in GNSS-denied scenarios.The thesis has 74 figures,21 tables and 191 references.