Research On Rat Brain Hippocampal Cognitive Mapping Based On Visual And IMU Information Fusion

With the development of science and technology,robots have appeared in every aspect of our production and life.Autonomous positioning and navigation are the most basic functions of robots.However,the current technology relies on a lot of external resources and computational resources,and its autonomous positioning and navigation ability is far from reaching the level of animals.As biologists continue to study the rat brain and hippocampus,cells related to rat environmental cognition and navigation have been discovered,such as head direction cells,striped cells,grid cells,and place cells,etc.Space cells provide neurophysiological support for humans to learn spatial cognition methods and skills in rats.By simulating the firing characteristics and information transmission mechanism of these cells,human can make the robot cognize the spatial environment like rats.This project is mainly based on the study of the cognitive mechanism of rat brain hippocampal environment,to construct the rat brain hippocampal space cell model,to simulate the firing characteristics and information transfer mechanism of the space cell,and then to construct the robot space environment cognitive map,identify and locate the target location and navigate to the target in the constructed cognitive map.The main research work includes:(1)The rat brain hippocampal space cell model and information transfer neural network model were constructed.The rat brain hippocampal space cell model and information transfer neural network model were constructed according to the information transmission mechanism of rat brain and hippocampus and the firing characteristics of space cells in rat brain and hippocampus.There are four main types of cells associated with spatial navigation in the rat hippocampus: head direction cells,striped cells,grid cells,and place cells.According to the firing characteristics of the four space cells and the information transmission mechanism between their parts in the hippocampus of the rat brain,the angular velocity information was firstly input into the head direction cell,and the angular velocity information of the rats was encoded.The head orientation signal output by the head direction cell is input into the stripe cell together with the linear velocity information,and the information transfer from the subiculum to the parasubiculum is completed.The one-dimensional annular attractor model of the stripe cell is constructed to integrate the current position.The stripe cell firing signal output by the stripe cell is input into the grid cell to complete the information transfer from the parasubiculum to the entorhinal cortex.Finally,information circulates between grid cells and place cells,forming an information transmission system of the entorhorium-ca3 loop,forming a hexagonal grid cell grid field and a single peak place cell field.The simulation results show that the model can carry out good path integration and simulate the firing characteristics of the rat brain hippocampal space cells in physiology.(2)Visual and IMU information fusion to construct cognitive maps.Fusion of visual and IMU information to obtain accurate angular velocity and linear velocity information,and use this as input,constructed a cognitive map based on the rat brain hippocampus information transfer neural network model.IMU information can track the rapid movement of the robot in a short time,visual information can effectively correct the drift error generated by IMU information in the movement,and more precise angular velocity and linear velocity can be obtained by integrating the two.ORB feature extraction and improved o-FAST corner point extraction were carried out on the collected visual image,and the features were matched by RANSAC-ORB feature matching algorithm.The angular velocity and linear velocity information obtained from the visual information were fused with the IMU information and input into the robot to construct the cognitive map,and updating the cognitive map through the closed-loop detection algorithm.(3)Identify and locate the target location and navigate to the target in the constructed cognitive map.When the robot is required to reach a certain place or a person's position in the map,the target position needs to be identified to obtain the position in the cognitive map.In the process of path integration and spatial navigation,angle and velocity information can drive the grid cells.Similarly,in the process of visual recognition,saccades can drive the firing activity of grid cells and encode the features of the image.In this way,the recognition of a target or a face can be realized,and a certain place in the cognitive map can be matched.The current position and target position in the space can be accurately located through the grid cells.After obtaining the position of the target,the shortest path from the current position to the target position is obtained through the improved Dijkstra path planning algorithm,and using segmented navigation to reach the destination.