Research On The Episodic Cognitive Mapping And Robot Navigation Method Based On The Cognitive Mechanism Of Rat Hippocampus

In recent years,more and more robots have appeared in our daily life.This provides an updated requirement for the ability of the robot: not only to perceive the external environment and the state of itself accurately through its own sensor,but also to be more intelligent to solve the robot "‘where am I?' ‘Where do you want to go?' ‘How to go'" these three basic questions.It is the biggest challenge to make the robot have a deeper cognitive ability,to build a cognitive map with more intelligentized and deeper meaning and to make a rapid and accurate navigation for the intended target.Mammals(mice,humans,etc.)can quickly and accurately navigate to a predetermined target location,such as homing and foraging.A large part of the perception of the outside world in mammals(mice,people,etc.)is transmitted from the eyeball to the brain through visual information where the visual area of the brain and transferred to the two brain regions("what loop" and "where loop").After encoding the information separately,the information between the entorhinal cortex and hippocampus is gathered to complete the information input part of cognitive activity.At the same time,biological cognitive neuroscience research shows that the hippocampus in rat brain plays a commanding role in spatial cognition.Based on the theory of biological visual pathway and the theory of hippocampal activation related to declarative episodic memory in biological cognitive neurology,combining the background knowledge of the neural network model of the hippocampal space cells,this paper proposes a method of constructing the episodic cognitive map based on the hippocampus of the rat brain,which satisfies Meet the requirement of robot to build accurate cognitive map.Based on this,a robot oriented bionic navigation model based on episodic cognitive memory is studied.The main research work includes as follows:(1)An episodic cognition model based on rat hippocampus is proposed.Based on the biological visual cognitive environment process,the spatial cell firing characteristics and spatial mapping encoding in the hippocampal structure,the whole network model of rat hippocampus episodic Cognition is established.The technology of image feature extraction and feature matching in the field of computer vision is used to complete the construction of the "where loop" of the biological visual information,and obtain the input information of the precise posture perception of the robot.Next,based on the continuous attractor model,the path integration characteristics and the firing characteristics for the space complete expression are completed.From the grid cells to the place cells to compete with the firing characteristics;the place cell cluster expresses the spatial environment characteristics,and finally obtains the robot's precise posture perception.The experimental results show that the cognitive model can better simulate the cognitive process of biological vision and space cells to the environment,and can obtain the local precise position and pose information of the robot.(2)A method of environmental cognitive method based on closed loop detection of BoW model and deep neural network is proposed.The method uses RGBD camera sensor to collect information from external environment,and constructs BoW model as the expression elements of environmental information and closed loop detection method.At the same time,using the method based on the deep neural network to identify the specific object recognition of the environment,complete the construction of declarative memory,and get the episodic information of the environment,forming the "what loop" of the biological visual information with the BoW model.The environmental cognitive method has good robustness under different illumination conditions,which lays the foundation for building accurate and intelligent cognitive maps.(3)A cognitive map building method based on rat brain hippocampus cognitive model is proposed.The relative position of the mouse in the current environment is obtained by integrating the information of the outside perceived speed and direction from the visual pathway "where loop" with the head-direction cells and band cells inside the organism.The expression of the current environment from the place cell is integrated with the recognition of the external environment objects from the "what loop" of the visual pathway,completing the episodic cognition of the current environment,and constructing the model of the episodic memory unit,and integrating the information of the current position.The whole visual loop-the entorhinal cortex-hippocampal circuit provides a basis for the dynamic expression of spatial memory.When the system detects a closed-loop,the episodic information in the episodic memory unit is corrected by using the episodic information in the "what loop",and the place cell cluster is distributed to the previously firing state.The eventual episodic memory unit on the cognitive map includes the cell firing coding information of the place cell cluster,the location information in the corresponding environment,the rotation information of the position and posture of the robot at the next moment and the previous moment,the relative topological information of the cognitive map and the episodic information of the environment.Physical experiments show that the method can construct accurate and intelligent episodic cognition map in large space of laboratory.Finally,we use the episodic cognitive map to complete the navigation of scene oriented targets in the condition of changing state space and complicated obstacles.This subjects received the national natural science fund project “No.61573029” and the BeiJing natural science fund project “No.4162012”,the research results of this subjects have great significance to psycholigy research,medical treatment,robot navigation and so on.