| Autonomous localization,map building and navigation,as the key technologies for mobile robots to realize autonomous movement in a rapid development stage.Currently,the raster map constructed by the traditional 2D laser SLAM algorithm can basically describe the state of a region,and is suitable for environment modeling and path planning.However,traditional raster maps do not contain any semantic information to identify which obstacles are present in the scene.In this paper,we focus on a 2D laser SLAM algorithm that incorporates semantic information to improve the robot’s understanding of the surrounding environment.And based on the constructed map,a global path planning algorithm based on the improved A-star algorithm is studied to improve the efficiency and safety of path planning.Finally,map-building navigation experiments were conducted on a mobile robot platform to verify the effectiveness of the proposed algorithm.The specific research includes:(1)Aiming at the problem that single-line Li DAR can only detect obstacles in its installation plane and cannot detect low obstacles,a method of fusing two sensor data from single-line Li DAR and depth camera was proposed.The method converts the depth camera sensor data projection into pseudo-LIDAR data and reorganizes the two-sensor data after timestamp alignment according to certain rules.Through ROS simulation experiments,it is verified that the fused Li DAR data increases the obstacle information in the longitudinal direction.(2)A 2D laser SLAM algorithm that incorporates semantic information was proposed to address the problem that the raster map constructed by the conventional 2D laser SLAM algorithm can only represent the state of each region and cannot understand the surrounding environment.First,the Cartographer algorithm and the YOLO v3 algorithm are used to construct a raster map while object recognition of specified obstacles in the environment,respectively.Subsequently,the target objects identified in the depth map are converted into 3D point cloud data,and two adjacent frames of the target point cloud are matched to obtain the complete target point cloud data,and the size of the target object is determined by searching the extreme values of the three dimensions.Finally,the corresponding model is searched in the pre-established 3D model library and created in the raster map in real time.With the guidance of semantic information in the map,the robot can complete the navigation task more accurately.(3)To address the problems of low planning efficiency and insufficient smooth paths in the conventional A-star algorithm for global path planning,a global path planning algorithm based on the improved A-star algorithm was proposed,which used a bidirectional adaptive step search strategy to reduce the search time and the number of search nodes.At the same time,the evaluation function was improved to narrow the path search range.Finally,the planned paths are smoothed and optimized by means of cubic B spline curves.Through the Matlab comparative simulation experiments,it is verified that the improved A-star algorithm path search is more efficient and the generated paths are smoother.(4)In order to verify the effectiveness of the proposed algorithm in practical map construction and path planning,a map construction experiment was conducted using a mobile robot platform for the living room and corridor environments,and the results of a raster map containing semantic information were obtained,which can visually display the location of each specified object in the map.Subsequently,several path planning experiments were conducted using this map,and safe paths of short length and smoothness were obtained to verify the effectiveness and practicality of the algorithm. |
PDF Full Text Request