Study On Simultaneous Localization And Mapping For The Spherical Amphibious Robot

Amphibious robots have flexible mobility and can adapt to complex environments.However,to work in a strange environment,the robot relies heavily on its precise positioning and the perception of the surrounding environment.At present,GPS,which is mainly used in robot positioning,is affected by rf interference and other factors,leading to inaccurate positioning of the robot and unable to provide the surrounding environment information to the robot.In recent years,the rapid development of Simultaneous Localization and Mapping(SLAM)algorithm provides a solution for robot autonomous Localization and Mapping of the surrounding environment.First of all,in order to realize the positioning and mapping of spherical amphibious robot in a strange environment,this paper improved the ORB-SLAM2 algorithm so that it could conduct accurate and stable positioning on spherical amphibious robot.By transplanting ORB-SLAM2 to the robot,the robot has the ability of positioning and mapping.The intensity of light in the external environment,the complexity of texture and the rapid movement of the robot in a short time will affect the visual positioning.While Inertial Measurement Unit(IMU)can accurately record the trajectory when moving fast,but has zero drift when stationary.In this paper,by combining visual and IMU information and using unscented kalman filter,the two information are combined to form a visual inertial odometer so that the robot can get more accurate and stable positioning.Secondly,in the ORB-SLAM2 algorithm,the map constructed around is only a collection of image feature points,which cannot provide effective navigation information.At the same time,the long time to build the map will result in a huge map file,which will be difficult for the robot to process.In this paper,octree map is used instead of sparse point cloud map to reduce computation and facilitate robot navigation.Then,the noise caused by sensor error in octree map is removed by statistical filtering method,and the map file size is reduced while the environmental information is retained,so that the robot can build maps for a long time.Finally,in order to verify the feasibility and effectiveness of the improved algorithm,a validation experiment is carried out.Firstly,simulation experiments are carried out on data set TUM-VI to verify the effectiveness of the algorithm.In the experiment,the NDI optical tracking system is used to record the actual moving trajectory of the robot,and then the trajectory recorded by the robot is compared and verified.Finally,the mapping performance is compared from the file size,precision and other indicators of map file.Experiments verify the effectiveness and feasibility of the visual-inertial positioning algorithm in positioning and mapping.