Research On Bionic Positioning And Navigation Methods For Indoor Robots

With the increasing application of indoor robots,the need for autonomous positioning and navigation technology becomes more urgent.At present,domestic and foreign research on indoor robot positioning and navigation technology is still based on accurate measurement and complex mathematical calculations for map-building.Such methods lead to complex algorithms and poor real-time performance,and are not the same as human behavior in positioning navigation.Therefore,positioning and navigation as the core technology for indoor robots to complete service work is still a research focus for indoor robots.Based on the existing visual positioning methods,this paper focuses on the navigation mechanism of human visual positioning from the physiological aspect.On the one hand,photoreceptor cells in the human visual system are divided into cone cells and rod cells,in which the cone cells acquire the color information and edge information of the image in the bright scene,and the rod cells acquire the edge information in the dark scene.Therefore,characteristics can be extracted from the color and edge information of the object,and the extracted features are remembered by the brain.In addition,because of the binocular vision,people can also acquire depth images in space.On the other hand,there are three kinds of cells,including place cells,grid cells and head-direction cells in the human brain that make people have the ability to locate and navigate.Place cells are sensitive to a specific location and can remember the location of the marker.Grid cells can correlate and memorize the locations of cell memories at each location,which is similar to topological maps in life.The head-direction cells have the function of sensing azimuth orientation.Therefore,human positioning navigation behavior is due to the memory and identification of indoor markers,memory of place cells,and memory of grid cells.Therefore,the bionic positioning and navigation method proposed in this paper mainly includes the following four parts: bionic object representation and recognition method research,bionic map-building method,using map bionic positioning method and using map bionic navigation method.Bionic object characterization and identification select the color,shape,symmetry and color texture for item characterization,and store the characteristics of the marker in the item database.In recognition,the items acquired from the natural scene are matched with the characteristics of the markers stored in the database to determine whether they are landmarks for positioning.Bionic map-building uses markers as the nodes of the topological map.The distance and direction information obtained from the depth image is used as the edge of the map.The database is used to store the topological map database.When using maps to positioning,the items collected by the robot are first identified through the item database.After being confirmed as a marker,the robot is searched in the topological map database to determine its relative position with the adjacent nodes in the environment and complete the positioning.When using a map to navigate,on the basis of the completion of the positioning task,the topological map database is retrieved to obtain the nodes,directions,and distances required to reach the destination to form a navigation route.According to the route,the robot starts from the current position and arrives in turn to complete the navigation.In order to verify the positioning navigation method proposed in this paper,a robot system which uses Kinect bionic eye function to obtain color images and depth images is built.This system uses the bionic brain of the IPC to process visual information,store items database and topology map database and completes the recognition,positioning and navigation algorithms.In addition,it uses MPU9250 sensor bionic head-direction cells to obtain orientation and assist navigation.In the validation experiment,some items in the laboratory was used as markers,and each experiment was conducted 100 times.The results showed that the average recognition rate of the algorithm is 98.8%,and the calculation complexity is low and the real-time performance is strong.In the real-life scene navigation experiments in the laboratory,the robot runs 26.33 m in a single run in the room and sets 10 positioning points.The average recognition time of each positioning point is 3.82 ms,and the navigation time is 83 s,achieving the whole process of positioning and navigation.