Research And Application Of Cognitive Map Building Based On Brain Spatial Cognition Mechanism

Multiple related regions in brain are evolved for supporting spatial cognition,helping animals move freely around different environments and achieve effective navigation between different spatial locations.It is one of the indispensable basic capabilities to implement much more sophisticated cognitive functions.Neuroscience experiments suggested that animals could,according to their spatial experience,build cognitive maps that reflect topological structures of environments and readily find flexible routes for accomplishing specific daily tasks.The cognitive map theory directly contributed to generating large amounts of instructive researches about the spatial cognition mechanism in brain,further setting off a new upsurge of exploring the inner spatial representation and navigation mechanism in related brain regions and developing bioinspired spatial cognitive systems based on mobile robot platforms.Researchers linked firing patterns of cell population with specific individual behaviors and discovered brain circuits and various kind of spatial cells related to the inner spatial cognition mechanism.Studies achieving computational modeling about these circuits and spatial cells at the cell level facilitate positive interaction and long-term coevolution between neuroscience and computer science.As a burgeoning interdisciplinary field,the spatial cognition research aims to build up a systemic understanding of animal's spatial cognition mechanism based on neuroscientific exploration to related brain regions and circuits.And then modeling works are done by the use of computational scientific tools and in turn speculate possible working flows and neuronal representation mechanism,and achieve a more deep and rounded view of spatial cognition in brain.Based on neuroscience experimental results about the brain circuits related to spatial cognition,this dissertation studied the modeling problem of bio-inspired cognitive map,developed a cognitive map building framework that links spatial cells in hippocampus and entorhinal cortex and then achieve the environmental map building on mobile robot platforms.The research findings of this dissertation are as follows:1.The environmental exploration in cognitive map building is studied and an automatic object search framework is achieved,which not only can be applied to the robotic local navigation in indoor environments,but also help robot automatically explore and collect environmental information and do preparation for cognitive map building.Automatic object searching is one of the essential skills for domestic robots to operate in unstructured human environments,including object identification,obstacle avoidance,path planning,and navigation.Although object searching and its counterparts,such as object recognition,and detection,are active research areas in the fields of robotics andcomputer vision,its realization is seldom explored in robotic systems that enables a robot not only to identify but also to approach the object of interest with obstacle avoidance.Moreover,it is a challengeable attempt to use a single RGB-D sensor to perform obstacle avoidance task.To help robot automatically explore in environment,collect environmental data and prepare for cognitive map building,an automatic object searching framework was proposed for a mobile robot equipped with a single RGB-D camera.The obstacle avoidance is achieved by a behavior learning algorithm based on deep belief networks(DBN).The target object is recognized using scale-invariant feature transform(SIFT)descriptors and the relative position between the target and mobile robot is estimated from the RGB-D data.Subsequently,the mobile robot makes a path planning to the target location using an improved bug-based algorithm.The framework was tested in indoor environments and requires the robot to perform obstacle avoidance and automatically search and approach the target object.The results indicated that the system is collision free and reliable in performing searching tasks.This system's functions make itself have the potential of being used for local navigation in real environments,environmental exploration and environmental data collection.2.The bio-inspired cognitive map building mechanism is studied,in which the improved dynamics for path integration makes the grid cell model's multi-scale extension be more adaptable to different environments,the hierarchical vision processing enhances the system time performance and the cognitive map building for mobile robot in real environments is achieved.Neuroscience research shows that,by relying on internal spatial representations provided by the hippocampus and entorhinal cortex(EC),mammals are able to build topological maps of environments and navigate.Taking inspiration from mammals' spatial cognition mechanism,entorhinal-hippocampal cognitive systems have been proposed for mobile robots,generally including grid cell modeling for path integration,place cell modeling for spatial representation,vision processing and cognitive map building.However,path integration and vision processing are time-consuming,and the existing model of grid cells is hard to achieve in terms of adaptive multi-scale extension for different environments,resulting in the lack of viability for real environments.An optimized dynamical model of grid cells is built for path integration in which recurrent weight connections between grid cells are parameterized in a more optimized way.Grid firing patterns with specific spatial scales can thus be accurately achieved for the multi-scale extension of grid cells.In addition,a novel hierarchical visual template organization mechanism is proposed for improving the time performance of vision processing and speeding up loop closure detection.Experiment results on the robotic platform demonstrate that our proposed entorhinal-hippocampal model can successfully build cognitive maps,reflecting the robot's spatial experience and environmental topological structures.3.The path integration mechanism based on grid cell modeling is studied,in which a general path integration mechanism is proposed based on spatial mapping representation and the path integration in 1D,2D and 3D space is achieved.The discoveries of spatial sensitive neurons,such as place cells and grid cells,show us spatial cognition may arise from neural activity of these neurons.Exploring how they represent Euclidean space is the key to knowing the brain's representation and encoding for space-related tasks.Grid cells have been kept in focus in spatial cognition research field,which are considered as path integrators in brain.As we all know,a considerable proportion of animals perform daily activities in 3D space but most models have been done in 2D space.A grid cell modeling mechanism was proposed by mapping the agent's self-motion in Euclidean space to the neuronal activity of grid cells,which can achieve multi-scale hexagonal patterns of grid cells from recurrent neural network(RNN)and enables path integration for 1D,2D and 3D spaces.Compared with currently existing models,the proposed modeling mechanism will benefit from the combination of singlecell based and network based grid cell modeling.Different from oscillatory interference(OI)models,recurrent connections between grid cells are maintained and it avoids the connection matrix's symmetry limitation and spatial representation in continuous attractor network(CAN).Different from the existing works which need to learn weights of RNN to get the vector representation of grid cells,the proposed mechanism can obtain network connection weights by direct matrix operations.Moreover,it is a general mechanism that can achieve multi-scale grid cell models for path integration in 1D,2D and 3D spaces.4.As an extension of the bio-inspired spatial cognition research,the grid cells' coding advantage in motion planning and distance metric mechanism are studied.Because of unique distributed and hexagonal spatial firing patterns and multiple spatial scales,grid cells have gained great attention in the subject of bio-inspired cognitive map research.Researchers has given top priority to finding the distance measuring mechanism embedded in grid cells and broaden its scope into non-spatial information's organization and processing.In this paper,the coding advantage of grid cells in motion planning is demonstrated by integrating grid cell activity into rapidly-exploring random tree(RRT)and achieving grid-cell-based multi-scale motion planning.We also underline the necessity of grid cells' distance metric ability and work out the spatial position decoding and distance metric mechanism embedded in grid cells' population activity by analyzingconnection weights' distribution between grid cells and place cells.With the help of artificial intelligence technique,we provide an instructive way of thinking about how grid cells function as a metric for coding space.