Study On Initial Positioning Of AGV With Laser SLAM Based On Vision And Path Planning

In recent years,with the gradual popularization and wide application of mobile robots,Simultaneous Localization and Mapping(SLAM)has become a current research focus as the core technology as the core technology of mobile robots.It solves the problem that robot builds map and locates with their own sensor during the running in the absence of a priori environmental information.In this paper,based on the in-depth study of SLAM technology theory,the vision provided initial value AMCL localization algorithm was proposed.The JPS path planning algorithm wastes a lot of memory and computation to storage and compute the meaningless nodes.This paper proposes an improved JPS algorithm based on pruning optimization.This paper is aim at conduct a research which carried out to realize the autonomous positioning and navigation of AGV in combination with the mainstream algorithms in current SLAM research.The main research work and achievements of this paper are as follows:(1)This paper analyzes the working principle of the traditional SLAM system.For traditional laser SLAM,information is relatively simple and repetition rate is high.Because of the global location of the robot fails after the robot is restarted at a non-designated location,a localization algorithm based on visual ORB features is proposed.In the early stage of building map by RGB-D camera to extract the ORB feature points,combined with the information of laser position to construct the dictionary,after rebooting,the robot matches the ORB features of the current frame with the ORB features of the historical key frame in the dictionary though the Bag of Words.Then,relative pose estimation is carried out based on inter-frame features,and the initial pose value after AGV restarted is calculated.Finally,the particle filter algorithm is used to disperse particles near the obtained pose,and the AGV accurate pose is obtained after convergence.(2)The implementation principle of JPS algorithm evolved from A* algorithm is introduced.JPS algorithm retains the basic framework on the basis of A* algorithm and proposes "two definitions and three rules" to effectively reduce the amount of computation and memory occupied of meaningless nodes.The path finding process of JPS algorithm is described with examples.Meanwhile,an improved JPS algorithm based on pruning optimization is proposed to optimize the storage calculation of intermediate inflection points in JPS algorithm and further avoid the extended operation of redundant nodes.In view of the incomplete path caused by the absence of intermediate inflection points,this paper proposes a strategy of supplementing the middle inflection point deleted in the path after pathfinding to supplement the complete path.An example is given to illustrate that the improved JPS algorithm is superior to JPS algorithm,which successfully reduces the computation and memory consumption of meaningless nodes in the extended node process and improves the efficiency of the pathfinding algorithm.(3)Design simulation experiments that obtain open source environment data and run AGV in Gazebo.Lidar builds a raster map to provide pose information for vision.Vision sensor collects visual information,extracts feature points,and classifies and builds feature word bag library.AGV was started at different positions to compare the positioning effects of traditional AMCL positioning algorithm and improved AMCL positioning algorithm.The superiority of the improved JPS pathfinding algorithm over the traditional JPS algorithm is proved in different map specifications.A real site experiment was constructed to verify the effectiveness of the improved AMCL positioning algorithm and the improved JPS algorithm again.