Research On Key Technologies Of Robot Localization And Obstacle Avoidance

In practical situations,robots are often in an unknown and unstructured environment.The localization of robot in the real environment depends on simultaneous localization and mapping(SLAM),that is,to implement self localization and building incremental map by evaluating the real environment information base on sensor datas obtained at the same time of movement.However,in the traditional SLAM methods,only X and Y coordinates and Yaw are considered in the mapping and navigation of mobile robot in the process of mapping and navigation,which leads to the lack of information about the flatness of the ground as well as the concave and convex obstacles,which will have an important impact on the pose of the robot in the process of moving.The carpet strips,down stairs and other areas indoor are not taken into account,and the interference of indoor and outdoor uneven road to the robot’s mobile posture will also affect the mapping accuracy.In this paper,the mathematical principle and calculation process of concave and convex obstacle detection for mobile robot with different posture are studied,and the mathematical model of flatness prediction is established.It is the first time to propose the conception of concave and convex obstacle map and realize a real-time pavement flatness prediction.Compared with other methods,this research enables the robot to perceive more obstacle information in complex environment,the information of small and concave obstacles of map is expanded,while directly integrated into the navigation map,so that the mobile robot can rely on the obstacle map to avoid dangerous areas.The main contents of this paper is as follows:(1)Experimental design.Aiming at the problem of SLAM not considering the flatness of the ground and the information of concave and convex obstacles,a method of using lidar sensor combined with mathematical detection model to realize the ground detection is proposed.The selection of experimental sensor is considered from the detection principle and advantages and disadvantages of depth camera and lidar,and the research method of simulation experiment is defined.(2)Environment construction.Using Gazebo simulation software,Turtlebot experimental simulation platform based on indoor simulation environment and Clearpath HUSKY experimental platform based on outdoor simulation environment are built respectively,and the required sensor plugins are added.The tool parameters are modified to create outdoor concave and convex terrain simulation model suitable for Gazebo.(3)Mathematical modeling and simulation.The mathematical principle of road roughness detection with different robot pose is established.The flatness calculation problem is convered into the projection trandormation from the motion pose to the initial pose,and the flatness detection model is established.The position and pose of the robot at different times are obtained by IMU sensor,and the flatness of the laser ranging data under the current position and pose is calculated by combining with the flatness detection model.The concave and convex obstacle map is constructed based on the optimized odometer pose.Compared with the navigation map constructed by other algorithms,this algorithm can provide more information of obstacle.(4)Experimental research.First,Lidar data is preprocessed,the point cloud data is fused and voxel grid filter are used to get point cloud maps with concave and convex ground information.Experiments show that both methods can be used to construct point cloud maps on concave and convex terrain.The single line Lidar is more adaptable and economical.For the application of ground detection,the single line Lidar proposed in this paper shows less data redundancy and computation.