Research On Construction And Navigation Of Mobile Robot Based On Improved RGB-D SLAM

With the development of science technology and the economic level,robot technology has begun to expand between human production and living.With the continuous integration of robots and artificial intelligence,human-machine collaboration has gradually became a hot research topic in robotics subject.Autonomous mobile robots with cooperative manipulators have gradually become a research hotspot.This article explores the application of robot technology in simultaneous localization and mapping,path planning and navigation with the human-machine collaboration robot as the application background.It builds a mobile robot system combined with mechanical hardware,embedded software application,sensor,map construction and path planning and multi-sensor fusion technology is applied to mobile chassis to achieve high positioning accuracy and navigation efficiency.Firstly,based on the two-wheel differential drive mobile chassis design and build an overall software framework,using Lidar and depth camera to sense the external environment,using MCU to integrate various safety devices such as anti-collision rubber strip,ultrasonic sensor and anti-drop sensor and other safety sensor information to ensure its motion safety.The ROS operating system is used to build the entire software framework to ensure stable operation and data transmission efficiency of the system.Secondly,this paper uses the feature point-based RGBDSLAM framework in map construction.The ORB features commonly used in visual front-end lack the scale and rotation robustness and concentrated distribution.The meshing method combined with the improved feature description algorithm.,is metioned to solve this problem Extracting,while maintaining the real-time performance of the detection algorithm,it ensures the uniform distribution and robustness of the feature points.For the large number of mismatches in the feature matching process,the method of grid filtering combined with PROSAC is used to increase the numbe of the correct matches and reduce erroneous deletion,for the pose estimation error obtained by the traditional method is large,the multi-feature points method is used to reduce the estimation error.For the pose optimization process,the key frame detection and closed-loop detection methods are proposed,IMU is utilized with the odometer data and the pose map optimization to complete the map construction.Then,for the defect that the environment information can not be fully acquired by using the lidar alone,using the Bayesian formula to combine with the 2D plane environment information come from the 3D point cloud detected by the depth camera,a detailed grid map of the indoor environment is established.In path planning,based on the traditional A* algorithm,the 16-direction weighted A* algorithm is improved,which reduces the path length and rotation angle to some extent.The global path planning is improved by the improved A* algorithm,and the local path re-planning is combined with the lidar and depth camera data and dynamic A* algorithm,and finally the global navigation strategy based on A* algorithm is obtained.Finally,using the built mobile robot,the RGBDSLAM experiment of merging IMU and odometer is completed,and the 3D map and global navigation strategy obtained by the fusion of lidar and depth camera are used to carry out physical experiments to verify the feasibility of the algorithm.