Research On Simultaneous Localization And Mapping Of Indoor Robots

With the development of science and technology,indoor service robots have entered people's life.In recent years,simultaneous localization and mapping technology has become a research hotspot in the field of robots,and it is considered as the key technology to realize the autonomous navigation for indoor robots.In this paper,simultaneous localization and mapping technologies based on lidar and vision sensor are analyzed,and a simultaneous localization and mapping method by combing both lidar and vision is proposed,which realizes efficient and stable indoor robot localization and mapping by fusing multiple information.The main research contents include:(1)The motion model of indoor service robot is modeled,the odometry model is given,and the lidar and camera are modeled respectively.At the same time,the grid map model suitable for navigation and the specific calculation method of grid map are given;(2)A method for joint optimization of laser vision is proposed.The initial value and laser constraint are given by laser correlative scan matching,visual ORB feature points matching results and re-projection errors are taken as visual constraints,and the random sampling consensus algorithm is used to eliminate visual mismatching information to ensure the accuracy of visual constraints.A joint error function is constructed for laser vision constraints.The localization of robot is optimized by the method of non-linear optimization,so as to obtain more accurate localization results;(3)The loop closure information is detected by visual bag of words model and the gird map constructed for laser data closes the loop.The key frame is constructed and the word information of the key frame is extracted from the bag of words model to identify whether the scene is closed.After identifying the loop closure information,the new constraints provided by the loop closure information are added to the graph optimization framework and all poses of robot is optimized.Finally,the closed loop of the grid map is realized by updating the map with the optimized robot pose and laser data,which solves the problem that laser mapping is difficult to be closed the loop;(4)A loop closure detection algorithm based on multi-information fusion is proposed.The search range of loop closure is limited by the current position of robot.The loop closure information is initially detected by visual bag of words.Finally,the real loop closure is verified by the result of laser scan matching,which ensures the accuracy of the loop closure detection.Finally,the proposed method is implemented on the actual robot development platform,and experiments are carried out in real indoor scenes.The experimental results show that the laser vision fusion method in this paper has higher localization accuracy than the pure laser method.At the same time the proposed method can effectively distinguish similar scenes,detect the loop closure information and complete loop closure grid map construction.