Research On SLAM Algorithm Of Wheeled Robot Based On Multi-source Information

Wheeled robot has become an important member of modern intelligent life because of its high speed and high flexibility.Simultaneous Localization and Mapping(SLAM),which is a key technology for simultaneous localization and mapping in unfamiliar environments,refers to the process that a mobile robot uses the onboard sensor to determine its posture and construct the surrounding environment.SLAM with a single sensor has been widely used with good performance,but it is often difficult to operate robustly in complex working environments,for example:the camera fails to track under the condition of insufficient illumination,missing features and interference from moving objects.LiDAR SLAM is prone to loss of location in environments such as long corridors with a single structure and making it difficult to construct loops.Inertial measurement unit(IMU)has bias,and long time operation will result in large cumulative error.This paper studies the theory and existing framework of single sensor and multi-sensor fusion,and built a tightly coupled SLAM framework for multi-source information fusion based on LiDAR,camera,and IMU.The main work is as follows:Research sensor models for LiDAR,Stereo cameras,and IMU and integrate them into wheeled robot systems.Construct a mathematical model for the SLAM problem,and study the mathematical principles and implementation processes of the LOAM algorithm and ORB-SLAM,two single source information SLAM algorithms for LiDAR and camera.Analyze the positioning and mapping effects of the two algorithms.In order to improve the defects of single SLAM algorithm in different working scenes,the LVI-S AM framework based on the fusion of Lidar,camera and IMU,is used to split the whole system into two subsystems,namely vision-inertia and lidar-inertia.The operating principle of the sub-system is studied and improved.In the visual-inertial system,the reverse optical flow method is used for the front-end data preprocessing part to improve the accuracy of corner point extraction.Meanwhile,the advantage of FAST corner point extraction is utilized to compensate for the increased time consumption.In the back-end marginal part,the residual error is separately optimized by establishing the Schur complement matrix to reduce the running time of marginalization.Experiments on open source data set and campus environment show that the running track accuracy of the improved fusion algorithm can be improved by 7.6%at most,the marginalization time can be reduced by 33%at most,and the established map has less invalid points.It is verified that the improved multi-source information SLAM algorithm can not only make up for the defects of a single sensor,but also realize more accurate positioning and mapping.