| Robot technology is one of the important symbols to measure the development degree of national high-tech and industrial automation.With the development of society,the demand of autonomous intelligent navigation robot is more and more urgent,and giving the robot a higher level of autonomous navigation ability has become one of the main challenges of the development of intelligent robot technology.The traditional robot autonomous navigation technology has poor robustness,scalability and adaptability,and cannot meet the requirements of long-term stable and autonomous work in natural environment.It is the development trend of robot intelligent autonomous navigation to learn from animal navigation patterns in nature and realize autonomous exploration of environment and cognitive map construction,and revealing the computing mechanism of spatial cognition of mammalian brain is the key to brain-inspired intelligent navigation technology.In order to provide computational theory and model for brain-like navigation technology,based on the research progress of spatial cognitive mechanism in brain science,this thesis established a neural computational model around the key neurons and information processing mechanism in the entorhinal cortical-hippocampus neural circuit,and proposed a brain-like spatial cognitive learning algorithm.The main research work of this thesis is as follows:First,a spatial coding model of visual scene was established based on the visual perception mechanism,The visual attention mechanism is used to self-organize feature locations from visual scenes,convert visual features into spatial locations,generate sparse representations of the spatial structure of visual scenes,realize brain-like visual perception and provide preliminary spatial coding for navigation.The results of simulation on the public database show that the model can well encode the spatial location of visual features and reveal the visual perception mechanism during mammalian environmental exploration.Second,a firing rate adaptive neural network model of grid cell was constructed based on the brain space navigation environment representation mechanism,which can form a grid representation of the environment by turning the preliminary spatial coding into a grid coding.The grid representation of the external environment is realized,and the spatial coordinate system is provided for navigation.The head orientation information is locally encoded into modulated position information input,which reproduces the neural coding characteristics recorded in neurobiological experiments and makes up for the deficiency that the existing grid cell models cannot encode head orientation information.Finally,an overall model of entorhinal cortical-hippocampal neural circuit was established based on the information processing mechanism of the entorhinal corticalhippocampus neural circuit,The model includes a grid cell network and a location cell network.The grid cell network completes the representation of the environment and provides input to the place cell network by aligning the grid coding direction through cyclic connections.Place cell networks match environmental representations with cognitive maps.The simulation results show that the model generates sparse location coding for the environment space and provide brain-like representations for constructing cognitive maps.Statistical analysis of the number and area of site fields showed that the model reproduced the firing of site cells with biological characteristics,which proved its effectiveness and mimicry.The biomimetic neural computing model proposed in this thesis explains the brain perception and environmental spatial representation mechanism in mammalian spatial navigation,realizes the brain-like spatial cognitive function,and provides a theoretical basis for the further development of brain-like navigation technology... |
PDF Full Text Request