A robot positioning and mapping method based on visual cortex and hippocampus spatial encoding

Abstract

The application provides a robot positioning mapping method based on a visual cortex and hippocampus space coding, and the method comprises the following steps: acquiring linear velocity and angular velocity through a wheel encoder and an inertial measurement unit, and obtaining robot pose information through path integration; collecting scene images by using a camera and extracting GIST features, obtaining a winning cell in a view cell network through Euclidean distance matching, updating weights, and introducing a posteriori entropy adaptive threshold to perform robust scene matching; storing the pose information and the successfully matched weight vector into a hippocampus CA1 area location cell, eliminating cumulative errors through iterative optimization, controlling the map size through space density redundancy pruning, and adopting a hierarchical switching strategy to perform scale stretching optimization, so that fast mapping is realized. Through simulating a biological space coding mechanism, the application improves scene recognition robustness, positioning consistency and mapping efficiency, and is suitable for intelligent navigation of a mobile robot.

Description

Technical Field

[0001] This invention relates to the field of intelligent navigation for mobile robots, and in particular to a robot localization and mapping method based on spatial coding of the visual cortex and hippocampus. Background Technology

[0002] Fast and accurate target-oriented navigation is one of the key technologies for mobile robots. A prerequisite for achieving this capability is that the robot can synchronously complete SLAM tasks in unknown environments. Traditional SLAM methods mainly rely on filters or graph optimization frameworks. While they have achieved significant results in structured environments, they still face the following challenges in large-scale, dynamically changing, or visually degraded scenarios: 1. High computational complexity: The probabilistic method based on Extended Kalman Filter (EKF) has a large computational cost in high-dimensional state spaces, and particle filtering is even more significant as the dimension increases; 2. Accumulated drift is difficult to eliminate: Noise in the sensor signals causes the positioning error to accumulate continuously when the robot runs over long distances; 3. Insufficient robustness of loop closure detection: Feature matching methods that rely on fixed thresholds are difficult to adapt to scenes with different lighting and viewing angles; 4. Uncontrollable map size: As the exploration area expands, the number of stored map nodes continues to increase, affecting system efficiency. Summary of the Invention

[0003] In view of the above situation, the main objective of this invention is to propose a robot localization and mapping method based on spatial coding of the visual cortex and hippocampus, so as to solve the above-mentioned technical problems.

[0004] This invention proposes a robot localization and mapping method based on spatial coding of the visual cortex and hippocampus, the method comprising the following steps: Step 1: Obtain linear velocity and angular velocity using the wheel encoder and inertial measurement unit mounted on the mobile robot. Integrate the angular velocity to obtain the heading angle. Based on the heading angle, perform path integration with the linear velocity to obtain the two-dimensional position coordinates in the world coordinate system. Integrate the linear velocity, angular velocity, and the robot's two-dimensional position coordinates in the world coordinate system to obtain the robot's pose information. Step 2: Use the camera mounted on the mobile robot to capture scene images and extract the feature vector of each frame of the scene image; based on the feature vector of the image, match the most similar cell in the view cell network by Euclidean distance as the winning cell; update the weight of the winning cell and the weight of the neighboring cells to obtain the updated view cell network; calculate the expected activity level of the winning cell in the updated cell network and compare it with the adaptive threshold to obtain the weight vector of the successfully matched winning cell. Step 3: Store the robot's pose information and the weight vector of the winning cell that was successfully matched into the CA1 region of the hippocampus to obtain the set of CA1 region location cells; iteratively optimize the set of CA1 region location cells to obtain the optimized set of CA1 region location cells; perform redundancy pruning on the optimized set of CA1 region location cells to obtain the redundancy-pruned set of CA1 region location cells; construct a hierarchical switching strategy using the redundancy-pruned set of CA1 region location cells, and perform scale scaling optimization using the hierarchical switching strategy until the preset conditions are met, complete the localization and mapping, and obtain the localization and mapping results.

[0005] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention learns the visual features of the environment through a visual cortex-based visual cell network. The network takes GIST features as input, employs a competitive learning mechanism to adaptively increase its size, and introduces a posterior entropy adaptive threshold to achieve robust scene matching, thereby improving the robustness of scene recognition and providing reliable visual evidence for subsequent loop closure detection and cognitive map updates.

[0006] 2. This invention establishes contextual connections and similarity connections between location cells based on temporal adjacency relationships and visual representation similarity relationships, and constructs a cognitive map using a map learning algorithm. This mechanism can transform the cumulative error generated by simple path integration into a global optimization problem under graph structure constraints, thereby improving the consistency of localization results and the stability of the cognitive map.

[0007] 3. This invention introduces a redundant cell pruning algorithm based on spatial density. By merging cells in the cognitive map that exceed the maximum spatial density, it effectively suppresses disordered cell growth and ensures that the size of the cognitive map is controllable. This mechanism makes the size of the cognitive map mainly constrained by the size of the exploration space, rather than continuously growing linearly with exploration time.

[0008] 4. This invention adopts a neural scale scaling optimization algorithm for fast mapping. It adaptively adjusts the map encoding resolution through dynamic hierarchical scheduling and cell fusion-separation mechanism. When fast convergence is required, it switches to a high-level low-resolution mode, and when details need to be improved, it switches to a low-level high-resolution mode. This can improve the mapping speed while maintaining the integrity of pose information and map accuracy. Attached Figure Description

[0009] Figure 1 This is a flowchart of a robot localization and mapping method based on visual cortex and hippocampal spatial coding proposed in this invention; Figure 2 This is a general framework diagram of a robot localization and mapping method based on spatial coding of the visual cortex and hippocampus proposed in this invention. Figure 3 This is a schematic diagram of the process of a mobile robot pose estimation method based on visual cortex and hippocampal spatial coding proposed in this invention. Figure 4 This is a schematic diagram of the visual cell network encoding process for robot localization and mapping based on spatial encoding of the visual cortex and hippocampus proposed in this invention. Figure 5 This is a schematic diagram of the hierarchical switching mechanism for robot localization and mapping based on spatial coding of the visual cortex and hippocampus proposed in this invention. Detailed Implementation

[0010] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0011] These and other aspects of the embodiments of the present invention will become clear from the following description and accompanying drawings. In these descriptions and drawings, some specific embodiments of the present invention are specifically disclosed to illustrate some ways of implementing the principles of the embodiments of the present invention; however, it should be understood that the scope of the embodiments of the present invention is not limited thereto.

[0012] Please see Figure 1 and Figure 2 This invention proposes a robot localization and mapping method based on spatial coding of the visual cortex and hippocampus, which includes the following steps: Step 1: Obtain linear velocity and angular velocity using the wheel encoder and inertial measurement unit mounted on the mobile robot. Integrate the angular velocity to obtain the heading angle. Based on the heading angle, perform path integration with the linear velocity to obtain the two-dimensional position coordinates in the world coordinate system. Integrate the linear velocity, angular velocity, and the robot's two-dimensional position coordinates in the world coordinate system to obtain the robot's pose information. In step 1, linear velocity and angular velocity are acquired using the wheel encoder and inertial measurement unit mounted on the mobile robot. The angular velocity is integrated to obtain the heading angle. Based on the heading angle, path integration is performed using the linear velocity to obtain the two-dimensional position coordinates in the world coordinate system. The linear velocity, angular velocity, and the robot's two-dimensional position coordinates in the world coordinate system are integrated to obtain the robot's pose information. The specific steps are as follows: The linear velocity and angular velocity of the mobile robot are acquired in real time by a wheel encoder and an inertial measurement unit mounted on the robot; the heading angle is calculated by integrating the angular velocity over time. By using the heading angle and combining it with the linear velocity, the path integral is performed to obtain the robot's two-dimensional position coordinates in the world coordinate system; Linear velocity, angular velocity, and the robot's two-dimensional position coordinates in the world coordinate system are integrated to obtain the robot's pose information.

[0013] For details, please refer to Figure 3 In the process of calculating the heading angle based on angular velocity by integrating over time, the following relationship exists: ; in, Indicates the initial heading angle. Indicates the time step. This represents the heading angle at time t. Represents the angular velocity at time t; In the process of using the heading angle and linear velocity to perform path integration to obtain the robot's two-dimensional position coordinates in the world coordinate system, the following relationship exists: ; in, This represents the two-dimensional position coordinates of the robot in the world coordinate system at time t. This represents the robot's initial two-dimensional position coordinates in the world coordinate system. Represents the linear velocity at time t. This represents the cosine value of the heading angle at time t. This represents the sine value of the heading angle at time t.

[0014] This step performs path integration based on the self-motion information to obtain the robot's pose prior at the current moment. The pose prior includes two-dimensional position coordinates and orientation angle.

[0015] Step 2: Use the camera mounted on the mobile robot to capture scene images and extract the feature vector of each frame of the scene image; based on the feature vector of the image, match the most similar cell in the view cell network by Euclidean distance as the winning cell; update the weight of the winning cell and the weight of the neighboring cells to obtain the updated view cell network; calculate the expected activity level of the winning cell in the updated cell network and compare it with the adaptive threshold to obtain the weight vector of the successfully matched winning cell. In step 2, scene images are captured using a camera mounted on the mobile robot, and feature vectors are extracted from each frame of the scene images. Based on the feature vectors, the most similar cell in the view cell network is matched as the winning cell using Euclidean distance. The weights of the winning cell and its neighboring cells are updated to obtain the updated view cell network. The expected activity level of the winning cell is calculated in the updated cell network and compared with an adaptive threshold to obtain the weight vector of the successfully matched winning cell. The specific steps are as follows: Scene images are captured by a camera mounted on a mobile robot; feature vectors are obtained for each frame of the scene image using the GIST feature extraction algorithm; in the view cell network, the similarity between the image feature vector and the weight vector of each view cell is calculated using Euclidean distance; the cell with the highest similarity between the image feature vector and the weight vector of each view cell is selected as the winning cell. After the winning cell is determined, the weight vector of the winning cell is updated, and the weight vectors of the other cells in the neighborhood of the winning cell are adjusted to complete the update of the weight vector of the winning cell and the weight vectors of the other cells in the neighborhood of the winning cell. After updating the weight vector of the winning cell and the weight vectors of the remaining cells in the neighborhood of the winning cell, calculate the cumulative sum of squared errors of the winning cell at the current time. When the sum of the squared errors of the winning cell at the current time exceeds a preset threshold, or when the current image feature vector is outside the neighborhood of the winning cell, a view cell is added, and an updated view cell network is obtained. By utilizing the prior probabilities in the updated view cell network, the expected activity level of the winning cell at the current moment is obtained through prediction. Posterior entropy is defined based on the feature vectors of an image; After normalizing the posterior entropy, the mean, cumulative sum of squares, and standard deviation of the historical sequence are obtained by recursive calculation. The original fixed threshold is adaptively scaled using the mean, cumulative sum of squares, and standard deviation of the historical sequence to obtain an adaptive threshold. The expected activity level of the winning cell at the current moment is compared with an adaptive threshold. If the expected activity level of the winning cell at the current moment is greater than the adaptive threshold, the winning cell is determined to be a successful match, and the weight vector of the successfully matched winning cell is obtained; otherwise, the match fails.

[0016] Please see Figure 4Visual cortex cells are the neural basis for spatial memory and contextual retrieval, capable of adaptively learning and representing visual structures in the environment through competitive activation and synaptic plasticity. Based on this, this invention proposes a contextual memory network inspired by visual cells to achieve continuous learning and memory retrieval of visual input features. Specifically, in the process of calculating the similarity between the feature vector of the image and the weight vector of each view cell using Euclidean distance, and selecting the cell with the highest similarity between the image's feature vector and the weight vector of each view cell as the winning cell, the following relationship exists: ; in, Indicates the number of the winning view cell. Indicates the index of the view cell. This represents the index operation that minimizes the distance function. This represents the feature vector of the image at time t. This represents the weight vector of the u-th view cell at time t; After determining the winning cell, the weight vector of the winning cell is updated, and the weight vectors of the remaining cells in the winning cell's neighborhood are adjusted simultaneously. The following relationship governs the process of updating both the winning cell's weight vector and the weight vectors of the remaining cells in the winning cell's neighborhood: ; in, This represents the weight vector of the u-th view cell in the neighborhood. This represents the square of the Euclidean distance. This represents the weight vector of the winning view cell at time t. This represents the initial learning rate of the view cell. This represents the total winning frequency of the winning cell between time 0 and t-1. This represents the influence coefficient of the winning cell on its neighboring cells. Let represent the updated weight vector of the u-th view cell in the neighborhood at time t+1.

[0017] However, the view cell network not only involves updating the weight vector, but also focuses on the continuous expansion of the network to learn new structures of the input data; therefore, this invention controls the increase of the number of cells in the network by calculating the cumulative sum of squared errors of the winning cells corresponding to the current input vector. Specifically, after updating the weight vector of the winning cell and the weight vectors of the other cells in the winning cell's neighborhood, the following relationship is involved in calculating the cumulative sum of squared errors of the winning cell at the current time step: ; in, Represents the dimension of the weight vector. Indicates a dimension index. This represents the cumulative squared error of the winning cell before the update at time t+1. Let represent the feature value of the visual feature vector input at time t in the m-th dimension. This represents the weight value of the winning cell's weight vector in the m-th dimension at time t. This represents the cumulative squared error of the winning cell before the update at time t; When the sum of the squared errors of the winning cell at the current moment exceeds a preset threshold (set to 0.1 in this invention), or when the current image feature vector is outside the neighborhood of the winning cell, a view cell is added, and the updated view cell network is obtained. The corresponding relationship is as follows: ; in, Represents conditional probability. This represents the initial weight vector of the newly added view cell at time t; In the process of adding a view cell, the accumulated error of the winning cell is attenuated to suppress the generation of too many redundant cells in dense areas. The expression for the attenuation coefficient is: ; in, Indicates the attenuation coefficient. This represents the cumulative squared error of the view cell after decay at time t+1. This represents the cumulative squared error of the view cell after decay at time t.

[0018] When the feature vector of the image to be inspected is input into the view cell network, the firing activity of the view cells reflects the similarity between their stored features and the input features. However, relying solely on a single winning cell for matching scenarios easily increases false positives. Therefore, this invention employs a method of aggregating the activity of multiple highly active cells for matching; Specifically, in the process of using the prior probabilities in the updated view cell network to predict the expected activity level of the winning cell at the current moment, the following relationship exists: ; in, Represents discrete random variables. This represents the time index variable from time 0 to time t. express Activation intensity at all times This indicates the expected activity level of the winning cell at time t. This represents the mathematical expectation operation. This represents the activation intensity level of the view cell corresponding to the random variable. express The time corresponds to the probability that the winning cell will be selected in the current time window. express Activate the cells with the strongest activity at all times. Indicates the activation intensity of the view cell. Represents the posterior probability. This indicates the starting position of the current statistical time window. express Number of view cells at any given time. The window variable representing cell selection, Represents an indexed random variable. This indicates that the maximum posterior probability is selected from the candidate index range. Let represent the posterior probability that the u-th view cell is selected as the matching cell given the current visual features. express Momentary visual characteristics, The normalization constant representing the visual features of the input. This represents the prior probability that the u-th view cell is selected as a matching cell; The prior probability is determined by the historical winning frequency of each cell, and is expressed as: ; in, This represents the learning frequency or matching frequency of the u-th view cell within the k-th statistical window. This represents the number of view cells participating in the statistics within the k-th statistics window. This represents the learning frequency or matching frequency of the v-th view cell within the k-th statistical window.

[0019] However, the concentration of the posterior probability distribution fluctuates with the input scene's confidence level, making it difficult to balance matching performance under different confidence levels with a fixed threshold. In defining the posterior entropy based on image feature vectors, the following relationship exists: ; in, express The posterior entropy at time t, Represents the posterior probability; When the posterior distribution degenerates into a deterministic distribution This indicates that the network is completely confident in the matching results; when the posterior distribution is uniform. hour This indicates that the network is completely uncertain.

[0020] As the size of the view cell network continues to grow during the exploration process, in order to eliminate the scale effect of cell number on the entropy value, the posterior entropy is normalized. The normalized form of the posterior entropy is as follows: Specifically, after normalizing the posterior entropy, the following relationship exists in the process of recursively calculating the mean, cumulative sum of squares, and standard deviation of the historical sequence: ; in, express The mean of the historical time series. express The mean of the historical time series. express The cumulative sum of squares of the historical time sequence. Indicates the deadline The mean of the time-history normalized posterior entropy sequence. express The cumulative sum of squares of the historical time sequence. express Standard deviation of the time-series historical data; In the process of adaptively scaling the original fixed threshold (the range set in this invention is 0.2~0.3) using the mean, cumulative sum of squares, and standard deviation of historical sequences, the following relationship exists: ; in, Indicates an adaptive threshold. This represents the original fixed threshold. when When the threshold is significantly higher than the historical average, the threshold is increased, suppressing false positives; when When the threshold is low, the recall rate is reduced; when... At that time, the set of scenes corresponding to the view cell with the highest current activation intensity is the matching scene.

[0021] This step learns the visual features of the environment through the visual cortex view cell network. This network takes GIST features as input, adopts a competitive learning mechanism to adaptively increase the network size, and introduces a posterior entropy adaptive threshold to achieve robust scene matching.

[0022] Step 3: Store the robot's pose information and the weight vector of the winning cell that was successfully matched into the position cell of the CA1 region of the hippocampus to obtain the set of position cells of the CA1 region of the hippocampus; iteratively optimize the set of position cells of the CA1 region of the hippocampus to obtain the optimized set of position cells of the CA1 region of the hippocampus; perform redundancy pruning on the optimized set of position cells of the CA1 region of the hippocampus to obtain the redundancy-pruned set of position cells of the CA1 region of the hippocampus; construct a hierarchical switching strategy using the redundancy-pruned set of position cells of the CA1 region of the hippocampus, and perform scale scaling optimization using the hierarchical switching strategy until the preset conditions are met, complete the localization and mapping, and obtain the localization and mapping results. In step 3, the robot's pose information and the weight vector of the successfully matched winning cell are stored in the CA1 region of the hippocampus to obtain a set of CA1 region location cells. The set of CA1 region location cells is iteratively optimized to obtain an optimized set. Redundancy is pruned from the optimized set to obtain a redundancy-pruned set. A hierarchical switching strategy is constructed using the redundancy-pruned set, and scale optimization is performed using the hierarchical switching strategy until a preset condition is met, completing the localization and mapping process and obtaining the localization and mapping results. The specific steps are as follows: The robot's pose information and the weight vector of the winning cell that was successfully matched are stored in the position cell of the CA1 region of the hippocampus to create a new position cell and obtain a set of position cells of the CA1 region of the hippocampus. Based on the spatial topology of the discharge field centers of all CA1 region location cells in the hippocampal CA1 region location cell set, the cumulative error of path integral is eliminated through iterative optimization to obtain the optimized hippocampal CA1 region location cell set. Conditional determination is performed on all CA1 region location cells in the optimized set of CA1 region location cells in the hippocampus to divide each CA1 region location cell into the corresponding grid and obtain the cell grid region. If cells already exist in the cell grid region, the cells assigned to the current cell grid region are designated as redundant cells. The weight vectors of the redundant cells are transferred to the cells already existing in the current cell grid region. At the same time, all cells related to the redundant cells are reconnected to the existing cells to obtain the set of cells in the CA1 region of the hippocampus after redundancy trimming. The set of positional cells in the CA1 region of the hippocampus after redundant trimming is used as the basic representation of the lowest level, and the pose adjustment amount of each level is defined. The pose adjustment ratio of each level is calculated by the pose adjustment amount of each level. A layer switching strategy is constructed by using the pose adjustment ratio of each layer; The hippocampal CA1 region location cell set is processed by a hierarchical switching strategy. When the hippocampal CA1 region location cell set is located at the lowest level and the pose adjustment ratio is continuously lower than the downsizing threshold, convergence stops, so as to obtain the hippocampal CA1 region location cell set after scale scaling optimization. The multi-layer pose information and cell connectivity relationships in the location cell set of the CA1 region of the hippocampus after scale scaling optimization are used as the localization and mapping results. The integration mechanism is as follows: When performing a low-level to high-level transition through a hierarchical switching strategy, the first low-level cell in each pair of adjacent cells in the low-level is retained as the cell to be retained, and the other low-level cell is discarded as the cell to be discarded. The visual information of the discarded cell and the retained cell is retained, as well as the pose information of the retained cell. The pose deviation of the discarded cell relative to the retained cell is stored to complete the fusion and obtain the set of cells in the CA1 region of the hippocampus after scale scaling optimization. The separation mechanism is as follows: When a high-level to low-level conversion is performed through a hierarchical switching strategy, the high-level cell is dissociated into two low-level cells. The first low-level cell directly inherits the pose information of the high-level cell, and the pose information of the second low-level cell is compensated and reconstructed through fusion residuals to complete the separation and obtain the set of cells in the CA1 region of the hippocampus after scale scaling optimization.

[0023] Please see Figure 5 During the robot's exploration of the environment, the pose information output by the entorhinal-hippocampal CA3 structure and the contextual information output by the visual cortex are stored in the position cells of the hippocampal CA1 region. Specifically, in the process of storing the robot's pose information and the weight vector of the successfully matched winning cell into the position cell of the CA1 region of the hippocampus to create a new position cell and obtain the set of position cells in the CA1 region of the hippocampus, the following relationship exists: ; in, This represents the location of the i-th cell in the CA1 region; The visual information corresponding to the i-th CA1 region cell is provided by the visual cortex's visual cell network; This represents the pose information corresponding to the i-th CA1 position table, output by the siphon-hippocampal path integral model. This represents the coordinates of the discharge field center of the i-th CA1 region cell in the world coordinate system, and its initial value is consistent with the position output by the path integral. This represents the heading angle corresponding to the i-th cell in the CA1 region.

[0024] Spatial relationships exist between position cells in different CA1 regions. These related position cells will influence each other's pose information. Specifically, during the creation of a new position cell, the cells associated with the new position cell are defined using the following formula: ; in, This represents the pose adjustment amount from the i-th CA1 region cell to the j-th CA1 region cell. This represents the cell located in the j-th CA1 region. This represents the visual information corresponding to the cell at the j-th CA1 region location.

[0025] Spatial association mainly includes the following two cases: contextual connection and similar contextual connection; among them, contextual connection represents the connection established between adjacent cells, while similar contextual connection represents the connection established between two cells with similar visual information; this association mechanism enables all CA1 region location cells to exhibit a complex topological structure in space. Specifically, in the process of iteratively optimizing the spatial topology of the firing field centers of all CA1 region location cells in the hippocampal CA1 region location cell set to eliminate the cumulative error of path integrals and obtain the optimized hippocampal CA1 region location cell set, the following relationship exists: ; in, This represents the change in pose of the cell at the i-th CA1 region. This represents the number of cells connected to the cell at the i-th CA1 region location. This represents the number of cells connected to the cell at position i. Indicates the index of the connected CA1 region location cell; This represents the pose correction rate, with a value range of [value range missing]. A value that is too large will cause oscillations in the map learning process, while a value that is too small will cause the map learning speed to be slow. Therefore, it is usually set to 0.5. This represents the pose information of the p-th CA1 region cell that is spatially associated with the i-th CA1 region cell. This represents the relative pose transition adjustment amount from the p-th CA1 region cell to the j-th CA1 region cell; By iteratively calculating the above formula, the cumulative error caused by path integrals can be largely eliminated, and an accurate cognitive map corresponding to the current environment can be constructed.

[0026] As the robot explores its environment, the number of cells in the CA1 region will continuously increase. This inevitably leads to redundancy in experience nodes, affecting the learning efficiency of the cognitive map. To effectively address the information overload caused by the increase in the number of cells, this invention proposes a redundant cell pruning method based on spatial density. Specifically, in the process of conditionally determining all CA1 region cells in the optimized set of CA1 region cells in the hippocampus to divide each CA1 region cell into its corresponding grid and obtain the cell grid region, the following relationship exists: ; in, This represents the set of all cells located in the CA1 region at the current moment. This represents the grid region of the existing cell at the current CA1 region location. This represents the coordinates of the center of the discharge field of the b-th CA1 region cell in the world coordinate system. This represents the coordinates of the top-left vertex of the g-th grid region. This represents the coordinates of the bottom right vertex of the g-th grid region. This represents the cell located in the b-th CA1 region. This represents the two-dimensional position coordinates of any point within the grid area; In the process of transferring the weight vectors of redundant cells to existing cells in the current cell grid region, and simultaneously reconnecting all cells related to redundant cells to existing cells to obtain the set of cells in the CA1 region of the hippocampus after redundancy trimming, the following relationship exists: ; in, This represents the cell located in the a-th CA1 region. This represents all spatial associations from other CA1 region location cells to the a-th CA1 region location cell. This represents all spatial associations from the a-th cell in the CA1 region to other cells in the CA1 region. This represents the spatial association from cells in other CA1 regions to the b-th CA1 region. This represents the spatial association between the b-th cell in the CA1 region and other cells in the CA1 region. This represents the number of spatial associations from other CA1 region cells to the b-th CA1 region cell. This represents the number of spatial associations from the b-th cell in the C1 region to other cells in the CA1 region. This indicates the corresponding operation to be performed when the preconditions are met. This represents the index variable indicating the spatial association from other CA1 region cells to the b-th CA1 region cell. The index variable represents the spatial association from the b-th CA1 region cell to other CA1 region cells; The above method ensures that the number of cells in the CA1 region is directly proportional to the size of the exploration space and independent of the exploration time, thus ensuring that the size of the cognitive map remains controllable.

[0027] However, even with the pruning mechanism, the learning speed of cognitive maps remains slow, especially when correcting large loop closures. To address this, this invention proposes a neural scale scaling optimization algorithm for rapid mapping, simulating the mechanism by which the hippocampus adjusts learning speed and accuracy by changing the scale of neural encoding. This mechanism is manifested in dynamically scheduling the scale of location cells at different levels: during the phase requiring rapid convergence, the algorithm switches to a higher level, reducing the number of location cells involved in map encoding through scale contraction to achieve rapid learning at low resolution; while when map details need to be improved, it switches back to a lower level, restoring high-resolution representation through scale expansion. Specifically, when defining the pose adjustment amount for each level by using the redundantly pruned set of cells in the CA1 region of the hippocampus as the lowest-level basic representation, the following relationship exists: ; in, This represents the total number of cells in the CA1 region of level l. This represents the pose adjustment of the cell located in the i-th CA1 region of the l-th layer at time t. This represents the pose information of the i-th CA1 region cell in the l-th layer at time t-1. This represents the pose information of the i-th CA1 region cell in the l-th layer at time t; The following relationship exists in the process of calculating the ratio of pose adjustment amounts for each level: ; in, This represents the ratio of the pose adjustment at level l at time t to that at time t-1. This represents the pose adjustment of the cell located in the i-th CA1 region of the l-th layer at time t-1; To suppress the false triggering of level switching caused by transient fluctuations due to sensor noise or newly added local correlations, the maximum and minimum values ​​of the pose adjustment ratio of three consecutive times are taken as the basis for judging level switching. Subsequently, the calculation method of level switching strategy can be derived. Specifically, in the process of constructing the level switching strategy using the pose adjustment ratios of each level, the following relationship exists: ; in, Indicates a hierarchical index. Indicates the level-up threshold. Indicates the descent threshold. Indicates the total number of levels. Indicates the time when the last level switch occurred. Indicates the number of cooling steps. This indicates logical AND operation processing. This represents the minimum value of the ratio of the pose adjustment amount of the l-th level within three consecutive steps set at time t and time t-1. This represents the maximum value of the ratio of the pose adjustment amount of the l-th level within three consecutive steps set at time t and time t-1. Cooling period constraints It can effectively prevent the sudden changes in the optimized structure after upgrading the floor level. The oscillating phenomenon of sudden change and immediate stratification; when When this occurs, it indicates that a large loop is being closed, and a switch to a higher level is needed to speed up map learning; when When this occurs, it indicates that the current level has been sufficiently learned, and the level can be moved down to refine local details; if and When the map converges, it is considered that the map has converged. At this point, the map stops learning and continues learning only when a new CA1 region cell is added.

[0028] During hierarchical switching, the fusion and separation of place cells are crucial for achieving neural scale scaling. Specifically, when performing a transition from a lower to a higher level using a hierarchical switching strategy, the first low-level cell in each pair of adjacent place cells in the lower level is retained as the preserved cell, and the other low-level cell is discarded as the discarded cell. The visual information of both the discarded and retained cells is preserved, along with the pose information of the retained cell. The pose deviation of the discarded cell relative to the retained cell is also stored to complete the fusion and obtain the scale-scaling optimized set of place cells in the CA1 region of the hippocampus. The following relationship exists in the process: ; in, This represents the set of visual information stored in the j-th CA1 region location cell (high-level CA1 region location cell) in the (l+1)-th level. This represents the set of visual information stored in the (2j-1)th CA1 region location cell (high-level CA1 region location cell) in the (l+1)th level. This represents the set of visual information stored in the 2j-th CA1 region location cell (high-level CA1 region location cell) in the l-th level. This represents the pose information of the j-th CA1 region location cell in the (l+1)-th level (high-level CA1 region location cell). This represents the pose information of the (2j-1)th CA1 region location cell in the l-th level (a higher-level CA1 region location cell). This represents the relative pose transformation adjustment between the j-th CA1 region cell (higher-level CA1 region cell) and the (j+1)-th CA1 region cell (higher-level CA1 region cell) in the (l+1)-th level. This represents the relative pose transition adjustment between the (2j-1)th CA1 region location cell (higher-level CA1 region location cell) and the 2jth CA1 region location cell (higher-level CA1 region location cell) in the (l+1)th level. This represents the relative pose transition adjustment between the 2jth CA1 region location cell (higher-level CA1 region location cell) and the 2j+1th CA1 region location cell (higher-level CA1 region location cell) in the (l+1)th level. This represents the pose deviation generated by the j-th CA1 region cell (lower-level CA1 region cell) during the fusion process. This represents the pose information of the 2jth CA1 region location cell in the l-th level (high-level CA1 region location cell); When performing a high-level to low-level transition using a hierarchical switching strategy, the high-level cell is dissociated into two low-level cells. The first low-level cell directly inherits the pose information of the high-level cell, and the pose information of the second low-level cell is reconstructed through fusion residual compensation to complete the separation and obtain the set of cells in the CA1 region of the hippocampus after scale-scaling optimization. The following relationship exists in the corresponding process: ; in, This represents the latest pose of the j-th CA1 region position cell in level l (high-level CA1 region position cell) after high-level optimization. This represents the displacement adjustment amount generated by the j-th CA1 region position cell (high-level CA1 region position cell) after high-level optimization. This represents the initial pose information recorded by the j-th CA1 region location cell (high-level CA1 region location cell) in the (l+1)-th level when it merges from a lower level to a higher level; This hierarchical switching mechanism can accurately restore the pose information of the fused cells on the curve trajectory, effectively suppressing the cumulative interpolation error in the continuous "fusion-separation" cycle, thereby ensuring the robustness of the hierarchical switching process and the integrity of the pose information, while taking into account the speed and accuracy of map learning.

[0029] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0030] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0031] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A robot localization and mapping method based on visual cortex and hippocampal spatial coding, characterized in that, The method includes the following steps: Step 1: Obtain linear velocity and angular velocity using the wheel encoder and inertial measurement unit mounted on the mobile robot. Integrate the angular velocity to obtain the heading angle. Based on the heading angle, perform path integration with the linear velocity to obtain the two-dimensional position coordinates in the world coordinate system. Linear velocity, angular velocity, and the robot's two-dimensional position coordinates in the world coordinate system are integrated to obtain the robot's pose information; Step 2: Use the camera mounted on the mobile robot to capture scene images, and extract the feature vector of each frame of the scene image; based on the feature vector of the image, match the most similar cell in the view cell network by Euclidean distance as the winning cell; Update the weights of the winning cell and its neighboring cells to obtain the updated view cell network; The expected activity level of the winning cell is calculated in the updated view cell network and compared with an adaptive threshold to obtain the weight vector of the winning cell that successfully matches. Step 3: Store the robot's pose information and the weight vector of the winning cell that was successfully matched into the CA1 region of the hippocampus to obtain the set of CA1 region location cells; iteratively optimize the set of CA1 region location cells to obtain the optimized set of CA1 region location cells; perform redundancy pruning on the optimized set of CA1 region location cells to obtain the redundancy-pruned set of CA1 region location cells; construct a hierarchical switching strategy using the redundancy-pruned set of CA1 region location cells, and perform scale scaling optimization using the hierarchical switching strategy until the preset conditions are met, complete the localization and mapping, and obtain the localization and mapping results.

2. The robot localization and mapping method based on visual cortex and hippocampal spatial coding according to claim 1, characterized in that, In step 1, linear velocity and angular velocity are acquired using a wheel encoder and inertial measurement unit mounted on the mobile robot. The angular velocity is integrated to obtain the heading angle. Based on the heading angle, path integration is performed using the linear velocity to obtain the two-dimensional position coordinates in the world coordinate system. The linear velocity, angular velocity, and the robot's two-dimensional position coordinates in the world coordinate system are integrated to obtain the robot's pose information. The specific steps are as follows: The linear velocity and angular velocity of the mobile robot are acquired in real time by a wheel encoder and an inertial measurement unit mounted on the robot; the heading angle is calculated by integrating the angular velocity over time. By using the heading angle and combining it with the linear velocity, the path integral is performed to obtain the robot's two-dimensional position coordinates in the world coordinate system; Linear velocity, angular velocity, and the robot's two-dimensional position coordinates in the world coordinate system are integrated to obtain the robot's pose information.

3. The robot localization and mapping method based on visual cortex and hippocampal spatial coding according to claim 2, characterized in that, In the process of calculating the heading angle based on angular velocity by integrating over time, the following relationship exists: ； in, Indicates the initial heading angle. Indicates the time step. This represents the heading angle at time t. Represents the angular velocity at time t; In the process of using the heading angle and linear velocity to perform path integration to obtain the robot's two-dimensional position coordinates in the world coordinate system, the following relationship exists: ； in, This represents the two-dimensional position coordinates of the robot in the world coordinate system at time t. This represents the robot's initial two-dimensional position coordinates in the world coordinate system. Represents the linear velocity at time t. This represents the cosine value of the heading angle at time t. This represents the sine value of the heading angle at time t.

4. The robot localization and mapping method based on visual cortex and hippocampal spatial coding according to claim 3, characterized in that, In step 2, scene images are captured using a camera mounted on the mobile robot, and feature vectors are extracted from each frame of the scene images. Based on the feature vectors of the images, the most similar cell is matched as the winning cell in the view cell network using Euclidean distance. Update the weights of the winning cell and its neighboring cells to obtain the updated view cell network; calculate the expected activity level of the winning cell in the updated cell network and compare it with an adaptive threshold to obtain the weight vector of the successfully matched winning cell. The specific steps are as follows: Scene images are captured by a camera mounted on a mobile robot; feature vectors are obtained for each frame of the scene image using the GIST feature extraction algorithm; in the view cell network, the similarity between the image feature vector and the weight vector of each view cell is calculated using Euclidean distance; the cell with the highest similarity between the image feature vector and the weight vector of each view cell is selected as the winning cell. After the winning cell is determined, the weight vector of the winning cell is updated, and the weight vectors of the other cells in the neighborhood of the winning cell are adjusted to complete the update of the weight vector of the winning cell and the weight vectors of the other cells in the neighborhood of the winning cell. After updating the weight vector of the winning cell and the weight vectors of the remaining cells in the neighborhood of the winning cell, calculate the cumulative sum of squared errors of the winning cell at the current time. When the sum of the squared errors of the winning cell at the current time exceeds a preset threshold, or when the current image feature vector is outside the neighborhood of the winning cell, a view cell is added, and an updated view cell network is obtained. By utilizing the prior probabilities in the updated view cell network, the expected activity level of the winning cell at the current moment is obtained through prediction. The posterior entropy is defined based on the feature vector of the image. After normalizing the posterior entropy, the mean, cumulative sum of squares and standard deviation of the historical sequence are obtained by recursion. The original fixed threshold is adaptively scaled using the mean, cumulative sum of squares, and standard deviation of the historical sequence to obtain an adaptive threshold. The expected activity level of the winning cell at the current moment is compared with an adaptive threshold. If the expected activity level of the winning cell at the current moment is greater than the adaptive threshold, the winning cell is determined to be a successful match, and the weight vector of the successfully matched winning cell is obtained; otherwise, the match fails.

5. The robot localization and mapping method based on visual cortex and hippocampal spatial coding according to claim 4, characterized in that, In calculating the similarity between the feature vector of the image and the weight vector of each view cell using Euclidean distance, and selecting the cell with the highest similarity between the image's feature vector and the weight vector of each view cell as the winning cell, the following relationship exists: ； in, Indicates the number of the winning view cell. Indicates the index of the view cell. This represents the index operation that minimizes the distance function. This represents the feature vector of the image at time t. This represents the weight vector of the u-th view cell at time t; After determining the winning cell, the weight vector of the winning cell is updated, and the weight vectors of the remaining cells in the winning cell's neighborhood are adjusted simultaneously. The following relationship governs the process of updating both the winning cell's weight vector and the weight vectors of the remaining cells in the winning cell's neighborhood: ； in, This represents the weight vector of the u-th view cell in the neighborhood. This represents the square of the Euclidean distance. This represents the weight vector of the winning view cell at time t. This represents the initial learning rate of the view cell. This represents the total winning frequency of the winning cell between time 0 and t-1. This represents the influence coefficient of the winning cell on its neighboring cells. This represents the updated weight vector of the u-th view cell in the neighborhood at time t+1; After updating the weight vector of the winning cell and the weight vectors of the remaining cells in the winning cell's neighborhood, the following relationship is obtained when calculating the cumulative sum of squared errors of the winning cell at the current time step: ； in, Represents the dimension of the weight vector. Indicates a dimension index. This represents the cumulative squared error of the winning cell before the update at time t+1. Let represent the feature value of the visual feature vector input at time t in the m-th dimension. This represents the weight value of the winning cell's weight vector in the m-th dimension at time t. This represents the cumulative squared error of the winning cell before the update at time t; When the sum of the squared errors of the winning cell at the current time exceeds a preset threshold, or when the current image feature vector is outside the neighborhood of the winning cell, a view cell is added, and the updated view cell network is obtained. The corresponding relationship is as follows: ； in, Represents conditional probability. This represents the initial weight vector of the newly added view cell at time t; In the process of adding a view cell, the accumulated error of the winning cell is attenuated to suppress the generation of too many redundant cells in dense areas. The expression for the attenuation coefficient is: ； in, Indicates the attenuation coefficient. This represents the cumulative squared error of the view cell after decay at time t+1. This represents the cumulative squared error of the view cell after decay at time t; In the process of using the prior probabilities in the updated view cell network to predict the expected activity level of the winning cell at the current moment, the following relationship exists: ； in, Represents discrete random variables. This represents the time index variable from time 0 to time t. express Activation intensity at all times This indicates the expected activity level of the winning cell at time t. This represents the mathematical expectation operation. This represents the activation intensity level of the view cell corresponding to the random variable. express The time corresponds to the probability that the winning cell will be selected in the current time window. express Activate the cells with the strongest activity at all times. Indicates the activation intensity of the view cell. Represents the posterior probability. This indicates the starting position of the current statistical time window. express Number of view cells at any given time. The window variable representing cell selection, Represents an indexed random variable. This indicates that the maximum posterior probability is selected from the candidate index range. Let represent the posterior probability that the u-th view cell is selected as the matching cell given the current visual features. express Momentary visual characteristics, The normalization constant representing the visual features of the input. This represents the prior probability that the u-th view cell is selected as a matching cell; The prior probability is determined by the historical winning frequency of each cell, and is expressed as: ； in, This represents the learning frequency or matching frequency of the u-th view cell within the k-th statistical window. This represents the number of view cells participating in the statistics within the k-th statistics window. This represents the learning frequency or matching frequency of the v-th view cell within the k-th statistical window; In defining the posterior entropy based on the feature vectors of an image, the following relationship exists: ； in, express The posterior entropy at time t, Represents the posterior probability; After normalizing the posterior entropy, the following relationship exists in the process of recursively calculating the mean, cumulative sum of squares, and standard deviation of the historical sequence: ； in, express The mean of the historical time series. express The mean of the historical time series. express The cumulative sum of squares of the historical time sequence. Indicates the deadline The mean of the time-history normalized posterior entropy sequence. express The cumulative sum of squares of the historical time sequence. express Standard deviation of the time-series historical data; In the process of adaptively scaling the original fixed threshold using the mean, cumulative sum of squares, and standard deviation of the historical series to obtain an adaptive threshold, the following relationship exists: ； in, Indicates an adaptive threshold. This represents the original fixed threshold.

6. The robot localization and mapping method based on visual cortex and hippocampal spatial coding according to claim 5, characterized in that, In step 3, the robot's pose information and the weight vector of the successfully matched winning cell are stored in the CA1 region of the hippocampus to obtain a set of CA1 region location cells. The set of CA1 region location cells is iteratively optimized to obtain an optimized set. Redundancy is pruned from the optimized set to obtain a redundancy-pruned set. A hierarchical switching strategy is constructed using the redundancy-pruned set, and scale optimization is performed using this strategy until a preset condition is met, completing the localization and mapping process and obtaining the localization and mapping results. The specific steps are as follows: The robot's pose information and the weight vector of the winning cell that was successfully matched are stored in the position cell of the CA1 region of the hippocampus to create a new position cell and obtain a set of position cells of the CA1 region of the hippocampus. Based on the spatial topology of the discharge field centers of all CA1 region location cells in the hippocampal CA1 region location cell set, the cumulative error of path integral is eliminated through iterative optimization to obtain the optimized hippocampal CA1 region location cell set. Conditional determination is performed on all CA1 region location cells in the optimized set of CA1 region location cells in the hippocampus to divide each CA1 region location cell into the corresponding grid and obtain the cell grid region. If cells already exist in the cell grid region, the cells assigned to the current cell grid region are designated as redundant cells. The weight vectors of the redundant cells are transferred to the cells already existing in the current cell grid region. At the same time, all cells related to the redundant cells are reconnected to the existing cells to obtain the set of cells in the CA1 region of the hippocampus after redundancy trimming. The set of positional cells in the CA1 region of the hippocampus after redundant trimming is used as the basic representation of the lowest level, and the pose adjustment amount of each level is defined. The pose adjustment ratio of each level is calculated by the pose adjustment amount of each level. A layer switching strategy is constructed by using the pose adjustment ratio of each layer; The hippocampal CA1 region location cell set is processed by a hierarchical switching strategy. When the hippocampal CA1 region location cell set is located at the lowest level and the pose adjustment ratio is continuously lower than the downsizing threshold, convergence stops, so as to obtain the hippocampal CA1 region location cell set after scale scaling optimization. The multi-layer pose information and cell connectivity relationships in the location cell set of the CA1 region of the hippocampus after scale scaling optimization are used as the localization and mapping results.

7. The robot localization and mapping method based on visual cortex and hippocampal spatial coding according to claim 6, characterized in that, In the process of storing the robot's pose information and the weight vector of the successfully matched winning cell into the hippocampal CA1 region position cell to create a new position cell and obtain the set of hippocampal CA1 region position cells, the following relationship exists: ； in, This represents the cell located in the i-th CA1 region. This represents the visual information corresponding to the cell at the i-th CA1 region location. This represents the pose information corresponding to the i-th CA1 position table. This represents the coordinates of the discharge field center of the i-th CA1 region cell in the world coordinate system. This represents the heading angle corresponding to the i-th cell in the CA1 region; In the process of creating a new location cell, the cell associated with the new location cell is defined by the following formula: ； in, This represents the pose adjustment amount from the i-th CA1 region cell to the j-th CA1 region cell. This represents the cell located in the j-th CA1 region. This represents the visual information corresponding to the cell at the j-th CA1 region location; In the process of iteratively optimizing the spatial topology of the discharge field centers of all CA1 region location cells in the hippocampal CA1 region location cell set to eliminate the cumulative error of path integrals and obtain the optimized hippocampal CA1 region location cell set, the following relationship exists: ； in, This represents the change in pose of the cell at the i-th CA1 region. This represents the number of cells connected to the cell at the i-th CA1 region location. This represents the number of cells connected to the cell at position i. Indicates the index of the connected CA1 region location cell. This represents the pose correction rate. This represents the pose information of the p-th CA1 region cell that is spatially associated with the i-th CA1 region cell. This represents the relative pose transition adjustment amount from the p-th CA1 region cell to the j-th CA1 region cell; In the process of conditionally determining all CA1 region location cells in the optimized set of CA1 region location cells in the hippocampus to divide each CA1 region location cell into the corresponding grid and obtain the cell grid region, the following relationship exists: ； in, This represents the set of all cells located in the CA1 region at the current moment. This represents the grid region of the existing cell at the current CA1 region location. This represents the coordinates of the center of the discharge field of the b-th CA1 region cell in the world coordinate system. This represents the coordinates of the top-left vertex of the g-th grid region. This represents the coordinates of the bottom right vertex of the g-th grid region. This represents the cell located in the b-th CA1 region. Represents the two-dimensional position coordinates of any point within the current grid area; In the process of transferring the weight vectors of redundant cells to existing cells in the current cell grid region, and simultaneously reconnecting all cells related to redundant cells to existing cells to obtain the set of cells in the CA1 region of the hippocampus after redundancy trimming, the following relationship exists: ； in, This represents the cell located in the a-th CA1 region. This represents all spatial associations from other CA1 region location cells to the a-th CA1 region location cell. This represents all spatial associations from the a-th cell in the CA1 region to other cells in the CA1 region. This represents the spatial association from cells in other CA1 regions to the b-th CA1 region. This represents the spatial association between the b-th cell in the CA1 region and other cells in the CA1 region. This represents the number of spatial associations from other CA1 region cells to the b-th CA1 region cell. This represents the number of spatial associations from the b-th cell in the C1 region to other cells in the CA1 region. This indicates the corresponding operation to be performed when the preconditions are met. This represents the index variable indicating the spatial association from other CA1 region cells to the b-th CA1 region cell. The index variable represents the spatial association from the b-th CA1 region cell to other CA1 region cells; When using the redundantly pruned set of cells in the CA1 region of the hippocampus as the lowest-level basic representation, the following relationship exists when defining the pose adjustment amount for each level: ； in, This represents the total number of cells in the CA1 region of level l. This represents the pose adjustment of the cell located in the i-th CA1 region of the l-th layer at time t. This represents the pose information of the i-th CA1 region cell in the l-th layer at time t-1. This represents the pose information of the i-th CA1 region cell in the l-th layer at time t; The following relationship exists in the process of calculating the ratio of pose adjustment amounts for each level: ； in, This represents the ratio of the pose adjustment at level l at time t to that at time t-1. This represents the pose adjustment of the cell located in the i-th CA1 region of the l-th layer at time t-1; In the process of constructing the level switching strategy using the pose adjustment ratios of each level, the following relationship exists: ； in, Indicates a hierarchical index. Indicates the level-up threshold. Indicates the descent threshold. Indicates the total number of levels. Indicates the time when the last level switch occurred. Indicates the number of cooling steps. This indicates logical AND operation processing. This represents the minimum value of the ratio of the pose adjustment amount at level l at time t to that at time t-1. This represents the maximum value of the ratio of the pose adjustment amount at level l at time t to that at time t-1.

8. The robot localization and mapping method based on visual cortex and hippocampal spatial coding according to claim 7, characterized in that, A hierarchical switching strategy was employed to perform hierarchical switching on cell sets located in the CA1 region of the hippocampus. This hierarchical switching process included both fusion and separation mechanisms. Specifically, the fusion mechanism was as follows: When performing a low-level to high-level transition through a hierarchical switching strategy, the first low-level cell in each pair of adjacent cells in the low-level is retained as the cell to be retained, and the other low-level cell is discarded as the cell to be discarded. The visual information of the discarded cell and the retained cell is retained, as well as the pose information of the retained cell. The pose deviation of the discarded cell relative to the retained cell is stored to complete the fusion and obtain the set of cells in the CA1 region of the hippocampus after scale scaling optimization. The separation mechanism is as follows: When a high-level to low-level conversion is performed through a hierarchical switching strategy, the high-level cell is dissociated into two low-level cells. The first low-level cell directly inherits the pose information of the high-level cell, and the pose information of the second low-level cell is compensated and reconstructed through fusion residuals to complete the separation and obtain the set of cells in the CA1 region of the hippocampus after scale scaling optimization.

9. A robot localization and mapping method based on visual cortex and hippocampal spatial coding according to claim 8, characterized in that, When performing a low-level to high-level transition using a hierarchical switching strategy, the first low-level cell in each pair of adjacent cells in the low-level hierarchy is retained as the preserved cell, and the other low-level cell is discarded as the discarded cell. The visual information of both the discarded and retained cells is preserved, along with the pose information of the retained cell. The pose deviation of the discarded cell relative to the retained cell is also stored to complete the fusion and obtain a scale-optimized set of cells in the CA1 region of the hippocampus. The following relationship exists during this process: ； in, This represents the set of visual information stored in the cell at the j-th CA1 region in the (l+1)-th level. This represents the set of visual information stored in the cell at the (2j-1)th CA1 region in the (l+1)th level. This represents the set of visual information stored in the cell at the 2jth CA1 region location in the l-th level. This represents the pose information of the j-th cell in the CA1 region of the (l+1)-th layer. This represents the pose information of the cell located in the (2j-1)th CA1 region of the l-th layer. This represents the relative pose transformation adjustment between the j-th CA1 region cell and the (j+1)-th CA1 region cell in the (l+1)-th level. This represents the relative pose transition adjustment between cells located in the (2j-1)th CA1 region and cells located in the 2jth CA1 region within the (l+1)th layer. This represents the relative pose transition adjustment between the cell at position 2j in CA1 region and the cell at position 2j+1 in CA1 region of level l+1. This represents the pose deviation of the cell at position j in the CA1 region during the fusion process. This represents the pose information of the cell located in the 2jth CA1 region of the l-th layer; When performing a high-level to low-level transition using a hierarchical switching strategy, the high-level cell is dissociated into two low-level cells. The first low-level cell directly inherits the pose information of the high-level cell, and the pose information of the second low-level cell is reconstructed through fusion residual compensation to complete the separation and obtain the set of cells in the CA1 region of the hippocampus after scale-scaling optimization. The following relationship exists in the corresponding process: ； in, This represents the latest pose of the cell at position j in the CA1 region of level l after high-level optimization. This represents the pose integer value generated by high-level optimization for the cell at position j in the CA1 region. This represents the initial pose information recorded by the j-th CA1 region cell in the (l+1)-th layer during the fusion from a lower to a higher level.

Images