Security inspection robot dog autonomous path planning method, device and medium

Abstract

This invention discloses an autonomous path planning method, device, and medium for a security patrol robot dog. The method includes: fusing LiDAR point cloud data, visual images, and plantar pressure multimodal sensor data to jointly construct a three-dimensional geographic map containing the spatial distribution of obstacles and the mechanical properties of ground softness and hardness, achieving comprehensive and accurate perception of geometric obstacles and hidden terrain risks. Simultaneously, a terrain-adaptive grid optimization strategy is adopted, dynamically adjusting the grid size according to the complexity of the terrain, effectively reducing computational overhead while ensuring planning accuracy in complex terrains, thus balancing algorithm accuracy and operational efficiency.

Description

Technical Field

[0001] This invention relates to the field of robot dog path technology, and in particular to a method, device and medium for autonomous path planning of a security patrol robot dog. Background Technology

[0002] With the rapid development of robotics and artificial intelligence, security patrol robots, as a new type of intelligent mobile platform, have shown broad application prospects in autonomous inspection tasks in industrial parks, substations, chemical plants, underground utility tunnels, and complex outdoor environments. Compared to traditional wheeled or tracked mobile platforms, robot dogs have superior terrain adaptability and obstacle-crossing capabilities, enabling them to move stably on unstructured terrains such as stairs, gravel, soft soil, and gullies, thus becoming an important vehicle for security patrol tasks in complex environments. However, the complex geographical environment and varied ground conditions also place higher demands on the robot dog's autonomous navigation and path planning capabilities.

[0003] Autonomous path planning is one of the core technologies for intelligent mobile robots to achieve fully autonomous inspection. Its goal is to find an optimal or near-optimal feasible path from the starting position to the target position within a known or partially known environmental map, while simultaneously satisfying multiple constraints such as safety, smoothness, and optimal energy consumption. Existing path planning methods can be mainly divided into graph search-based methods (such as A* algorithm and Dijkstra's algorithm), random sampling-based methods (such as RRT and PRM), and biomimetic intelligent optimization-based methods (such as genetic algorithms and particle swarm optimization). Among these, the A* algorithm has been widely used in practical engineering due to its simplicity, high search efficiency, and ease of incorporating heuristic information. However, most of the above methods are geared towards structured environments, heavily reliant on terrain geometry and obstacle distribution information, while insufficiently considering mechanical properties such as ground hardness and bearing capacity, making them difficult to directly apply to complex scenarios with soft soil, slippery surfaces, or alternating soft and hard terrain.

[0004] In existing technologies, robot dogs typically rely on 3D point cloud maps constructed by LiDAR or environmental images acquired by visual sensors for path planning, generating passable areas by identifying obstacle locations and geometric dimensions. However, the complexity of the environment in security patrol scenarios is not only reflected in the spatial distribution of static obstacles, but also in the spatial heterogeneity of dynamic physical characteristics such as terrain bearing capacity, ground friction, and foot sinking risk. For example, when a robot dog walks on soft soil, even if there are no obvious obstacles in the area, its feet are prone to sinking, slipping, or even tipping over, leading to patrol mission failure or equipment damage. In traditional methods, ground hardness information is often recorded as independent attribute data, lacking fusion processing with obstacle spatial distribution information, and even more so lacking the technical means to systematically integrate it into the path planning cost function and grid passability judgment criteria, making it difficult to achieve a comprehensive balance between terrain geometric risks and mechanical risks during path search.

[0005] Furthermore, existing raster map-based path planning methods typically employ a fixed-size raster partitioning strategy, discretizing the geographic region globally using uniform length and width dimensions. While this strategy is applicable to flat and uniform terrain, it struggles to balance map accuracy and computational efficiency in areas with hard and soft ground boundaries, dramatic terrain undulations, and significant differences in obstacle density. Too small a raster size leads to a rapid increase in map size, increasing the computational burden and storage overhead of path searching; too large a raster size results in the loss of terrain details, particularly the excessive smoothing of key terrain features at abrupt boundaries, thus reducing the safety and reliability of path planning. Currently, there is a lack of an optimization scheme that can adaptively adjust the raster size based on local terrain complexity to effectively control map size and computational complexity while maintaining path accuracy.

[0006] Furthermore, most existing robot dog path planning methods rely on static map data for one-time path generation. This means that inspection tasks are only performed after path planning is complete, lacking real-time perception and response capabilities to changes in terrain and environment during the inspection process. However, in actual security inspection scenarios, the environmental conditions of the target geographical area are not static: ground hardness may change significantly after rain, temporarily placed items may create new obstacles, and inspected areas may become impassable due to disturbances. During the robot dog's movement along the initial path, when a pre-marked passable grid becomes impassable due to environmental changes, traditional static planning methods lack the ability to dynamically evaluate the environmental state of the current grid and its surrounding area and replan the path. They often can only perform global replanning or interrupt the task; the former is inefficient, and the latter cannot guarantee task continuity.

[0007] In summary, how to construct an autonomous path planning method for security inspection robot dogs that can comprehensively consider the spatial distribution of obstacles and the mechanical properties of terrain, adaptively adjust the discretization resolution of the map, and dynamically respond to environmental changes during the inspection process to adjust the path has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] This invention provides a method and system for autonomous path planning for a security patrol robot dog, which at least solves some of the technical problems existing in the background art as described above.

[0009] According to a first aspect of the present invention, an autonomous path planning method for a security patrol robot dog is provided, comprising the following steps: S1. The target security patrol robot dog sends the first, second and third collected data within the target geographical area to the target cloud, and the target cloud generates a target three-dimensional geographical area map of the target geographical area. S2. The target cloud generates a set of target geographic grids within the target geographic area based on the target three-dimensional geographic area map; the set of target geographic grids includes several target geographic grids, and the target geographic grids are geographic grid areas obtained after optimizing the initial geographic grids. S3. The target cloud generates the initial path of the target inspection robot dog based on all the target geographic grids, and sends the initial path to the target inspection robot dog; S4. The target inspection robot dog receives the initial path and performs inspection along the initial path. When the target inspection robot dog reaches any target geographic grid on the initial path, it collects the environmental data of the geographic grid and regenerates the subsequent path as a new path based on the environmental data of the geographic grid.

[0010] In a specific embodiment, in step S1, the first data collected within the target geographical area is the raw point cloud dataset of the target geographical area collected by the target security patrol robot dog through its own lidar device; the second data collected within the target geographical area is the set of obstacle images within the target geographical area collected by the target security patrol robot dog through its own camera device; and the third data collected within the target geographical area is the ground pressure dataset of the target geographical area collected by the target security patrol robot dog through its own foot pressure sensor device.

[0011] In one specific embodiment, step S1 further includes the following steps: S11. Generate an initial three-dimensional point cloud map based on the first collected data within the target geographical area, wherein the initial three-dimensional geographical area map is generated based on the original point cloud dataset of the target geographical area collected by the target security patrol robot dog through its own lidar device. S12. Based on the second collected data within the target geographical area, generate an initial obstacle dataset, wherein the initial obstacle dataset includes several initial obstacle data, wherein any initial obstacle data includes: an initial obstacle ID, an obstacle type label corresponding to the initial obstacle ID, and an obstacle geographical location corresponding to the initial obstacle ID; S13. Based on the third collected data within the target geographical area, generate a set of ground hardness for the initial geographic grid within the target geographical area; divide the target geographical area into several initial geographic grids; obtain the ground hardness corresponding to the initial geographic grid based on the ground pressure data corresponding to any initial geographic grid in the third collected data. S14. Generate a target three-dimensional geographic region map of the target geographic region based on the initial three-dimensional point cloud map, the initial obstacle dataset, and the ground hardness set of the initial geographic grid.

[0012] In one specific embodiment, the optimization process in step S2 further includes the following steps: S21. Based on the initial obstacle information and initial ground hardness within any initial geographic grid, determine the intermediate geographic grid; based on the initial obstacle information and initial ground hardness corresponding to any initial geographic grid, generate the passage risk priority corresponding to the initial geographic grid; when the passage risk priority corresponding to the initial geographic grid is greater than a preset risk upper limit threshold, determine the initial geographic grid as an impassable geographic grid and insert it into the impassable grid set; when the passage risk priority corresponding to the initial geographic grid is not greater than the preset risk upper limit threshold, determine the initial geographic grid as a passable geographic grid and insert it into the passable grid set. S22. Determine the target geogrid based on the initial ground hardness of the intermediate geogrid; determine the length and width of the target geogrid corresponding to the intermediate geogrid based on the initial ground hardness of the intermediate geogrid, the length of the intermediate geogrid, and the width of the intermediate geogrid.

[0013] In one specific embodiment, step S3 further includes the following step to generate the initial path: S31. When all target geographic grid location information meets the preset path generation conditions, the initial path of the target inspection robot dog is generated; the preset path generation conditions refer to the existence of a connected path consisting of several adjacent target geographic grids between the target geographic grid where the starting position of the target inspection robot dog is located and the target geographic grid where the target position is located, and each target geographic grid on the connected path is a passable geographic grid. S32. When the location information of all target geographic grids does not meet the preset path generation conditions, a new target geographic grid is generated from the set of inaccessible grids as a key geographic grid, so as to generate the initial path of the target inspection robot dog based on all key geographic grids and all target geographic grids.

[0014] In one specific embodiment, step S31 further includes the following steps: S311. Obtain all target geographic grid information, wherein the target geographic grid information includes: the location of the target geographic grid and the traffic risk priority of the target geographic grid; S312. Input all target geographic grid information into the A* path generation model to generate the initial path for the target inspection robot dog.

[0015] In one specific embodiment, step S32 further includes the following steps: S321. Obtain information on all impassable geographic grids in the impassable grid set. The information on the passable geographic grids includes: the passability risk priority of any impassable geographic grid and the coordinates of the center point of the impassable geographic grid corresponding to the impassable geographic grid. S322. Based on the access risk priority of the impassable geographic grid and the preset access risk upper limit threshold, calculate the optimization potential index of the impassable geographic grid. When the optimization potential index of the impassable geographic grid is greater than the preset optimization potential threshold, mark the impassable grid as an impassable grid to be optimized and insert it into the set of impassable grids to be optimized. S323. For each impassable grid in the set of impassable grids to be optimized, calculate the local environment accessibility fusion index of the impassable grid based on the accessible geographical grid information adjacent to the impassable grid; when the local environment accessibility fusion index of the impassable grid is greater than the preset local environment accessibility fusion threshold n, mark the impassable grid as an optimizable accessible grid and insert it into the set of optimizable accessible grids. S324. For each optimizable accessible grid in the optimizable accessible grid set, perform optimization processing to generate a new accessible geographic grid. S325. The newly added passable geographic grids are merged into all the existing target geographic grids to form an updated set of target geographic grids. Then, return to step S31 and re-execute the A* path generation model until the initial path of the target inspection robot dog is generated, or until there are no optimizable impassable grids that meet the optimization conditions. If so, it is determined that there is no connected path in the target 3D geographic area map and the path planning failure information is output.

[0016] In a specific embodiment, step S324 specifically includes the following sub-steps: S3241. Obtain the ground hardness distribution information corresponding to the original initial geographic grid area of ​​the optimizable accessible grid, and estimate the optimized ground hardness at the center point of the optimizable accessible grid by using bilinear interpolation based on the ground hardness values ​​of adjacent accessible geographic grids. S3242. Based on the optimized ground hardness of the optimizable accessible grid and the access risk priority of adjacent accessible geogrids of the optimizable accessible grid, calculate the optimized access risk priority; the optimized access risk priority is calculated based on the number of adjacent accessible geogrids of the optimizable accessible grid, the access risk priority of each adjacent accessible geogrid, the ground hardness of each adjacent accessible geogrid, and the hardness difference attenuation coefficient. S3243. Compare the optimized passage risk priority with the preset passage risk upper limit threshold. If the optimized passage risk priority is not greater than the preset passage risk upper limit threshold, mark the optimizable passable grid as a newly added passable geographic grid and update the corresponding target geographic grid information to the target geographic grid set. If the optimized passage risk priority is greater than the preset passage risk upper limit threshold, maintain its impassable state and do not include it in the newly added passable geographic grid set.

[0017] According to a second aspect of the present invention, a computer device includes a memory and a processor, the memory storing a computer program executable on the processor, the processor executing the program to implement the steps of the above-described method.

[0018] According to a third aspect of the present invention, a computer-readable storage medium has a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0019] In this embodiment, a novel autonomous path planning scheme is proposed, addressing the shortcomings of traditional security patrol robot dog path planning, such as neglecting ground mechanics, difficulty in balancing grid accuracy and computing power, static and fixed paths, and poor adaptability to complex environments. This invention integrates multimodal sensing data from LiDAR point clouds, visual images, and plantar pressure sensors to jointly construct a three-dimensional geographic map containing the spatial distribution of obstacles and the mechanical properties of ground softness and hardness, achieving comprehensive and accurate perception of geometric obstacles and hidden terrain risks. Simultaneously, a terrain-adaptive grid optimization strategy is adopted, dynamically adjusting the grid size according to the complexity of the terrain, effectively reducing computational overhead while ensuring planning accuracy in complex terrain, balancing algorithm accuracy and operational efficiency. Furthermore, this invention adds an optimization evaluation and reconstruction mechanism for impassable grids, overcoming the initial path connectivity limitations, and dynamically updating passage risks and triggering local path replanning based on real-time patrol environment data, enabling rapid response to dynamic environmental changes such as ground softening and temporary obstacles. This invention effectively improves the robot dog's safety in navigating complex terrain and the robustness of path planning, avoiding problems such as patrol slippage, sinking, and task interruption.

[0020] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Furthermore, no embodiment in this disclosure is required to achieve all the effects described above. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0022] Figure 1 This is a schematic diagram of the overall architecture of the security patrol robot dog's autonomous path planning provided in the embodiments of this disclosure.

[0023] Figure 2 The flowchart illustrates the autonomous path planning of a security patrol robot dog, as provided in this embodiment of the invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0025] The implementation process of the method described in this application will be described in detail below with reference to specific embodiments. It should be noted that this embodiment is only used to explain this application and is not intended to limit the scope of protection of this application. Conventional adjustments or substitutions of each step by those skilled in the art without departing from the concept of this application should be included in the scope of protection of this application.

[0026] This invention provides an autonomous path planning method for a security patrol robot, such as... Figure 2 As shown, the method includes the following steps: S1. The target security patrol robot dog sends the first, second, and third collected data within the target geographical area to the target cloud, and the target cloud generates a target three-dimensional geographical area map of the target geographical area.

[0027] Specifically, the first data collected within the target geographical area is the original point cloud dataset of the target geographical area collected by the target security patrol robot dog through its own lidar device.

[0028] Specifically, the second data collected within the target geographical area is a set of obstacle images within the target geographical area collected by the target security patrol robot dog through its own camera device.

[0029] Specifically, the third data collected within the target geographical area is the ground pressure dataset within the target geographical area collected by the target security patrol robot dog through its own foot pressure sensor.

[0030] Furthermore, the target geographical area is the geographical area that the target security patrol robot dog needs to patrol.

[0031] In one specific embodiment, step S1 further includes the following steps: S11, an initial three-dimensional point cloud map is generated based on the first collected data within the target geographical area. The initial three-dimensional geographical area map is generated based on the original point cloud dataset of the target geographical area collected by the target security patrol robot dog through its own lidar device. Those skilled in the art know the methods for generating three-dimensional point cloud maps from any three-dimensional point cloud collected by a lidar device in the prior art, and will not be described in detail here.

[0032] S12, Based on the second collected data within the target geographical area, an initial obstacle dataset is generated, wherein the initial obstacle dataset includes several initial obstacle data, and any initial obstacle data includes: an initial obstacle ID, an obstacle type label corresponding to the initial obstacle ID, and an obstacle geographical location corresponding to the initial obstacle ID.

[0033] Furthermore, the initial obstacle ID is a unique identifier of an obstacle identified from any obstacle image within the target geographic area.

[0034] Furthermore, the obstacle type label corresponding to the initial obstacle ID is obtained by inputting an image of any obstacle within the target geographic area into a pre-trained obstacle recognition model. The pre-trained obstacle recognition model is a pre-trained YOLOv5 model. It is known in the art that the YOLOv5 model obtains obstacle type labels in the prior art, and will not be elaborated here.

[0035] Furthermore, the geographical location of the obstacle corresponding to the initial obstacle ID is determined by the obstacle image within the target geographical area. To further understand: using the location coordinates of the target security patrol robot and any obstacle image from the obstacle image set within the target geographical area, the obstacle recognition model outputs the pixel-level bounding box of the obstacle, and the midpoint of the bounding box pixel is calculated. The robot's own global pose coordinates and camera calibration intrinsic and extrinsic parameters are fused to complete the mapping transformation from pixel coordinates to three-dimensional global geographical coordinates. The three-dimensional global coordinates of the midpoint of the transformed obstacle bounding box are used as the geographical location of the obstacle corresponding to the initial obstacle ID.

[0036] S13, Based on the third collected data within the target geographical area, generate a set of ground hardness and softness for the initial geographic grids within the target geographical area; further understood: the target geographical area is divided into m initial geographic grids A1, ..., A1. i , ..., A m A i It is the i-th initial geographic grid, and the length and width of each initial geographic grid are the same; based on any A in the third collected data i The corresponding ground pressure data was obtained in A i Corresponding ground hardness P i .

[0037] Furthermore, P i Meets the following conditions: The ground pressure F within the initial geographic grid corresponding to Ai is collected by the four foot pressure sensors of the target security patrol robot dog. i1 F i2 F i3 F i4 Avg() is the mean function, and Var() is the variance function. It is the first preset minimum positive number.

[0038] Furthermore, the initial geographic grids are rectangular, and adjacent initial geographic grids have overlapping areas.

[0039] S14. Based on the initial 3D point cloud map, the initial obstacle dataset, and the ground hardness set of the initial geographic grid, generate a target 3D geographic region map of the target geographic region.

[0040] In a specific embodiment, those skilled in the art know a method for generating a target three-dimensional geographic region map of a target geographic region based on the initial three-dimensional point cloud map, the initial obstacle dataset, and the ground hardness set of the initial geographic grid. For example, in S14, the lidar coordinate system of the target security patrol robot dog is used as the global reference coordinate system. The image pixel coordinates corresponding to each obstacle in the initial obstacle dataset are uniformly transformed to the lidar global three-dimensional coordinate system. At the same time, the ground hardness data corresponding to each initial geographic grid is accurately bound to the corresponding grid center three-dimensional coordinate position. This completes the unified registration of the spatiotemporal coordinates of the three types of sensor data: lidar point cloud, visual obstacle image, and foot pressure mechanics. This effectively eliminates the spatial offset and temporal error existing in the multi-sensor acquisition process. Based on this, a basic geometric terrain layer of the target geographic region is constructed based on the initial three-dimensional point cloud map. The initial obstacle dataset, which contains obstacle ID, obstacle type label, obstacle geographical location, and height information, is parsed into an independent obstacle label layer and superimposed on the basic geometric terrain layer to completely mark the area. The 3D boundary, spatial occupancy range, and obstacle characteristic attributes of obstacles are used to form a 3D point cloud map that integrates geometric obstacle information. Subsequently, the 3D point cloud map with integrated geometric obstacle information is uniformly divided into initial geographic grid units. The ground hardness values ​​of each initial geographic grid are pre-acquired as core mechanical attribute labels and assigned to the corresponding grid area. For complex areas such as the boundary between soft and hard ground and areas with severe terrain undulations, additional annotations are added to include derived terrain parameters such as ground hardness gradient, ground bearing risk, and friction characteristics, constructing a complete terrain mechanical attribute layer. Finally, the geometric terrain layer, obstacle annotation layer, and terrain mechanical attribute layer are integrated and coupled into a single layer. This allows each spatial grid unit in the target 3D geographic area map to synchronously integrate the 3D geometric terrain contour, obstacle spatial distribution characteristics, and ground hardness mechanical bearing attributes, generating an integrated target 3D geographic area map that combines geometric obstacle perception and terrain implicit mechanical risk perception. This provides accurate, synchronous, and complete multimodal fusion data support for subsequent geographic grid passage risk assessment, adaptive grid size optimization, and dynamic path replanning during inspections.

[0041] S2, the target cloud generates a target geographic grid set within the target geographic area based on the target three-dimensional geographic area map.

[0042] Specifically, the target geographic grid set includes several target geographic grids, wherein the target geographic grids are obtained by optimizing the initial geographic grids.

[0043] In one specific embodiment, the optimization process in step S2 further includes the following steps: S21, determine the intermediate geographic grid based on the initial obstacle information and initial ground hardness within any initial geographic grid; further understood as: based on A i Based on the corresponding initial obstacle information and initial ground hardness, generate A. i Corresponding traffic risk priority R i When R i Greater than the preset risk upper limit threshold R max At that time, determine A i For impassable geographic grids, insert them into the impassable grid set; when R i Not greater than the preset risk upper limit threshold R max Determine A i For accessible geographic grids, i.e., intermediate geographic grids, insert them into the accessible grid set B1, ..., B g , ..., B W Among them, B g It is the g-th intermediate geographic grid, where g ranges from 1 to W, and W is the number of intermediate geographic grids.

[0044] Furthermore, R i Meets the following conditions: ;in, The risk impact weight of obstacle density, The risk impact weight of ground hardness, It is a collaborative weight.

[0045] Furthermore, , , .

[0046] Furthermore, D i Meets the following conditions: ;in, It is A i The corresponding initial geographic grid area is calculated using the length and width of the initial geographic grid; ni is A i The total number of initial obstacles included; H k It is A i The height of the k-th obstacle within the obstacle is extracted from the obstacle image, H. ref It is the maximum height of the target inspection robot dog's foot clearance from the ground.

[0047] In summary, compared with existing technologies, this technology achieves a dual-coupling weighted average of the number and height of obstacles. Low obstacles only generate basic density risk, while tall obstacles are given a higher density weight. This accurately reflects the risks of numerous low obstacles and few tall obstacles, thus meeting the actual obstacle-crossing safety needs of the target inspection robot dog.

[0048] Furthermore, S i Meets the following conditions: ;in, It is a preset reference value for ground hardness, which indicates that the ground is passable without subsidence; This is a preset ground hardness threshold, which reflects that when the initial ground hardness is greater than the preset ground hardness threshold, the target inspection robot dog's foot settlement will suddenly change and it will be unable to pass.

[0049] In summary, existing technologies typically use a linear relationship to describe the impact of ground hardness on the equipment to be used. Compared to existing technologies, when the ground hardness P... i Approaching the critical threshold At that time, the risk of foot subsidence increases exponentially and rapidly. The nonlinear mechanical characteristics of the robot dog sinking, slipping, and overturning in soft soil areas are accurately matched, and the risk characterization is more in line with the real ground conditions.

[0050] The above scheme couples and links obstacle risks with ground hardness risks. An increase in any risk factor will trigger a superlinear increase in overall risk, accurately reflecting the synergistic hazard effect of severe terrain. At the same time, by simulating the saturation characteristics of foot indentation on extremely soft ground, it avoids excessive risk distortion caused by single exponential amplification and the sudden increase in risk in dual high-risk scenarios. When the initial geographic grid has dense obstacles and soft ground at the same time, it adapts to the passage failure situation of extremely complex terrain.

[0051] S22, determine the target geogrid based on the initial ground hardness of the intermediate geogrid; further understood as: based on the initial ground hardness P of Bg. g B g Corresponding length BL gx and B g Corresponding width BL gy Determine B g The corresponding target geographic grid length L gx and width L gy .

[0052] Furthermore, L gx Meets the following conditions: ,in, It is the length weight on the X-axis in the target 3D geographic region map; It reflects the degree of terrain change along the X-axis in the three-dimensional geographic area map of the target.

[0053] Furthermore, L gy Meets the following conditions: ,in, It is the length weight on the Y-axis in the target 3D geographic region map; It reflects the degree of terrain change along the Y-axis in the three-dimensional geographic area map of the target.

[0054] Furthermore, It is the floor function.

[0055] Furthermore, It is B g The ground hardness gradient vector at the center point is calculated using the finite difference method of the hardness values ​​of adjacent grids.

[0056] Furthermore, it is used to measure B g E is an index representing the complexity of the distribution of ground hardness. g E g Meets the following conditions: Where Q represents the number of ground hardness levels in the intermediate geographic grid, and Q is not less than 3, used to refine the ground hardness distribution characteristics; p q It is the proportion of the number of sampling points in the qth ground hardness level range to the total number of sampling points of the grid, where q ranges from 1 to Q; It is the second preset minimum positive number.

[0057] In summary, by characterizing the complexity of the hardness and softness distribution of the intermediate geographic grid, the size of the target geographic grid can be determined. Based on accurately perceiving the terrain abrupt changes (i.e., complexity) in the grid's neighborhood, the subdivision intensity can be increased and the grid side length reduced in directions of drastic terrain gradient changes, while larger grid sizes can be retained in directions of gentle gradient changes. Ultimately, slender rectangular grids are formed at the boundaries of soft and hard ground and at the boundaries of terrain abrupt changes, accurately depicting the boundaries of irregular terrain. Large-sized grids are retained in flat and uniform terrain, significantly reducing the overall computational load of the algorithm and achieving the optimal balance between accuracy and efficiency.

[0058] S3, the target cloud generates the initial path of the target inspection robot dog based on all target geographic grids, and sends the initial path to the target inspection robot dog.

[0059] In one specific embodiment, S3 further includes the following step to generate the initial path: S31, when all target geographic grid location information meets the preset path generation conditions, generate the initial path of the target inspection robot dog.

[0060] Specifically, the preset path generation bar refers to a connected path consisting of several adjacent target geographic grids between the target geographic grid at the starting position of the target inspection robot dog and the target geographic grid at the target location, and each target geographic grid on the connected path is a passable geographic grid.

[0061] Furthermore, step S31 also includes the following steps: S311, Obtain all target geographic grid information G={G1, ..., G...} j , ..., G m}, G j This refers to the information of the j-th target geographic grid; wherein, the target geographic grid information includes: the location of the target geographic grid and the traffic risk priority of the target geographic grid; the value of j ranges from 1 to M, where M is the number of target geographic grids.

[0062] S312, input G into the A* path generation model to generate the initial path of the target inspection robot dog.

[0063] In one specific embodiment, step S312 further includes the following steps: S3121, from all the target geographic grids, determine the target geographic grid where the starting inspection node of the target inspection robot dog is located is marked as C1, and determine the target geographic grid where the ending inspection node of the target inspection robot dog is located is marked as C2.

[0064] S3122, Construct an initial open list and an initial closed list. The initial open list is used to store path nodes that have been generated but not yet expanded (i.e., target geographic grids), and the initial closed list is used to store path nodes that have been expanded (i.e., target geographic grids).

[0065] S3123, add the starting inspection node C1 to the initial open list, set g(C1) = 0, and calculate F(C1).

[0066] S3124, calculate the total cost F(r) from C1 to any current path node r, where F(r) satisfies the following condition: F(r) = g(r) + H(r). Select the node r with the smallest F(r) value from the initial open list as the current extended node. If the current extended node r is the end inspection node C2, then jump to S3127; otherwise, move the current extended node r from the initial open list to the initial closed list.

[0067] Furthermore, the calculation method for F(C1) is the same as that for F(r).

[0068] Furthermore, g(r) satisfies the following condition: Where T is the total number of target geographic grids between the starting inspection node C1 and the current path node r; when t=1, the starting inspection node of the path is C1, and when t=T, the ending node of the path is the current path node r; U(t-1, t) is the Euclidean distance between the center points of two adjacent path nodes. Rref is the passage risk priority of the t-th target geographic grid between the starting inspection node C1 and the current path node r; Rref is the preset passage risk priority threshold. It is a risk cost influencing factor, used to adjust the degree of influence of travel risk on route cost; It is a risk adjustment index and The risk adjustment index is used to differentiate the cost characteristics of different risk regions: Further explanation: When When the marginal cost of low-risk areas decreases, the A* path generation model utilizes the target geographic grid; when When the marginal cost of high-risk areas increases, the A* path generation model abandons the high-risk target geographic grid; △θ is the change angle of the path movement direction, that is, the orientation offset of the current path segment relative to the previous path segment; the first path segment defaults to △θ=0, with no turning cost; Is the shift affecting the weight and .

[0069] Furthermore, H(r) satisfies the following condition: Where U(r, C2) is the straight-line Euclidean distance from the center point of the current path node r to the center point of the end inspection node C2; ​​R 0 (r, C2) is the average passage risk priority of all target geogrids covered by the straight path from the current path node r to the end inspection node C2. It is the risk prediction weight and This is used to adjust the algorithm's sensitivity to predicting future path risks; It is a distance correction factor and This is used to optimize the heuristic value for distant nodes, accelerating the algorithm's convergence speed; U max It is the maximum distance between diagonal geographic grids in the target 3D geographic region map.

[0070] S3125, Obtain the set of all adjacent target geographic grids B of the current extended node r. 0 r Among them, B 0 rcomprising: target geographic grids in a total of 8 directions related to the current expansion node r, including up, down, left, right, upper left, upper right, lower left, and lower right, and taking the target geographic grid in any one direction as an adjacent path node of the current expansion node r, wherein, for B 0 r any one adjacent path node J in B, the following judgment conditions are executed: (1) If the adjacent path node J is inserted into the initialized closed list, or the traffic risk priority ΔR of the adjacent path node J is greater than the preset traffic risk upper limit threshold R max , then the said node is skipped and not processed; (2) If the adjacent path node J is not inserted into the initialized open list, insert the adjacent path node J into the initialized open list, record the parent node of the adjacent path node J as r, and calculate the actual cumulative traffic cost g(J) of the adjacent path node J according to the formula in S3124, wherein g(J) meets the following condition: g(J)=g(r)+Cost(r→J), wherein Cost(r→J) is the corresponding term in the general term of the sum of g(r) in S3124; (3) If the adjacent path node J is inserted into the initialized open list, calculate the new actual cumulative traffic cost g'(J)=g(r)+Cost(r→J) that reaches the adjacent path node J via the current path node r. If g'(J) < g(J), update the parent node of the adjacent path node J to r, update g(J) to g'(J), and recalculate F(J)=g'(J)+H(J) at the same time.

[0071] Further, the heuristic estimate H(J) and the total cost value F(J) are obtained by referring to the aforementioned steps.

[0072] S3126, repeatedly execute S3123 to S3125 until the initialized open list is empty or the end patrol node C2 is found.

[0073] S3127, starting from the end patrol node C2, backtrack the parent node pointer in sequence until reaching the start patrol node C1, connect all the passed target geographic grids in order, and generate the initial path of the target patrol robot dog.

[0074] S32, when all the position information of the target geographic grids does not meet the preset path generation condition, generate a new target geographic grid from the impassable grid set as a key geographic grid, so as to generate the initial path of the target patrol robot dog according to all the key geographic grids and all the target geographic grids.

[0075] Further, the S32 step further comprises the following steps: S321, Obtain information on all impassable geographic grids in the impassable grid set, whereby the information on drivable geographic grids includes: the drivability risk priority RB of any impassable geographic grid W. W The coordinates of the center point of the corresponding impassable geographic grid are (X... W Y W ); S322, according to RB W With R max Calculate the optimizability potential index YQ of the impassable geographical grid W. W And when YQ W >Q 0 When Q is not allowed, the impassable grid W is marked as an impassable grid to be optimized and inserted into the set of impassable grids to be optimized. 0 It is a preset optimizable potential threshold, where YQ W YQ is used to characterize the feasibility of converting the impassable geographic grid into a navigable grid; where YQ W Meets the following conditions: Among them, RB W ≥R max ; It is the distance decay factor and Used to adjust the distance between the impassable grid and the starting and ending inspection nodes; D ref It is a preset distance threshold, and the preset distance threshold is a positive integer multiple of the average side length of the target geographic grid in the target 3D geographic region map; D min (W, C1, C2) is the shortest straight line distance from the center point of the impassable grid W to the starting inspection node C1 and the ending inspection node C2.

[0076] S323, for each impassable grid W0 in the set of impassable grids to be optimized, calculate the local environmental accessibility fusion index M of the impassable grid W0 to be optimized based on the accessible geographical grid information adjacent to W0. W Among them, M W Meets the following conditions: ; Where, N W It is the total number of adjacent accessible geographic grids of the impassable grid W0 to be optimized; It is the passage risk priority of the f-th adjacent passable geographic grid; UW fUS0 is the Euclidean distance between the center point of the impassable geogrid W0 to be optimized and the center point of the f-th adjacent traversable geogrid; US0 is the area covered by the set of adjacent traversable geogrids of the impassable geogrid W0 to be optimized, calculated by multiplying the number of adjacent traversable geogrids by the area of ​​a single geogrid; US max It is the maximum coverage area of ​​the set of adjacent accessible geographic grids among all the inaccessible grids to be optimized; ω is the area coverage weight and 0 < ω < 1; further, when M W The local environment passability fusion threshold M is greater than the preset threshold. min When the time comes, mark the impassable grid W0 to be optimized as an optimizable passable grid and insert it into the set of optimizable passable grids; S324, for each optimizable accessible grid WK in the optimizable accessible grid set, perform optimization processing to generate a new accessible geographic grid W. new Specifically, it includes the following sub-steps: S3241, Obtain the original initial geographic grid region A0 of the optimizable walkable grid W. W The corresponding ground hardness distribution information is used, and based on the ground hardness values ​​of adjacent accessible geogrids, the optimized ground hardness P at the center point of the optimizable accessible geogrid W is estimated using bilinear interpolation. W ; S3242, based on the optimized passable grid W, the optimized ground hardness P W The optimal traffic risk priority R is calculated based on the traffic risk priority of the grid and its adjacent accessible geogrids. W ; Furthermore, R W Meets the following conditions:

[0077] Where, N W It is the number of adjacent accessible geographic grids of the optimizable accessible grid W; RW e It is the passage risk priority of the e-th adjacent passable geographic grid; PW e ξ is the ground hardness of the e-th adjacent passable geogrid; ξ is the hardness difference attenuation coefficient, ξ>0.

[0078] S3243, Compare R W Compared with the preset upper limit threshold R of traffic risk max If R W ≤R max Then, the optimizable traversable grid W is marked as a newly added traversable geographic grid, and the corresponding target geographic grid information is updated to the target geographic grid set; if R W >R maxIf it is not passable, it will remain in an impassable state and will not be included in the newly added passable geographic grid set.

[0079] In step S325, the newly added passable geographic grids are merged into all existing target geographic grids to form an updated set of target geographic grids. Then, the process returns to step S31 and re-executes the A* path generation model until the initial path for the target inspection robot dog is generated, or until no path satisfying YQ is found. W >Q 0 And M W ≥M min If an optimizable but impassable grid is found, it is determined that there is no connected path in the target 3D geographic area map, and a path planning failure message is output.

[0080] In summary, step S32 optimizes the potential index YQ. W The introduction of this method prioritizes the selection of impassable grids that are closer to the starting and ending points and have a lower risk of exceeding limits for optimization, avoiding blind global searches and improving the efficiency of path replanning; at the same time, it integrates the local environmental accessibility index M. W With the optimized traffic risk priority R w The dual verification ensures the feasibility and safety of the newly added passable geographic grid, enabling the path planning to still have the ability to adapt when faced with initial connectivity failure, thus improving the robustness and practicality of the robot dog's path planning in complex terrain.

[0081] S4, the target inspection robot dog receives the initial path and performs inspection along the initial path. When the target inspection robot dog reaches any target geographic grid on the initial path, it collects the environmental data of the geographic grid and regenerates the subsequent path as a new path based on the environmental data of the geographic grid.

[0082] In one specific embodiment, step S4 further includes the following steps: S41. The target inspection robot dog inspects along the initial path, reaching each target geographic grid G. new At the same time, the real-time environmental data of the current grid is collected by airborne sensors to generate a real-time environmental parameter set for the current grid; the real-time environmental data includes: real-time point cloud data collected by lidar, real-time obstacle image data collected by camera device, and real-time ground pressure data collected by foot pressure sensor device; the real-time environmental parameter set includes at least: real-time obstacle distribution parameters, real-time obstacle height parameters, and real-time ground softness and hardness parameters.

[0083] Furthermore, while collecting current grid environment data, the target inspection robot dog records the current timestamp t and the robot dog's pose information, and stores the collected data in association with the timestamp for subsequent time series change analysis.

[0084] S42, According to the current grille G current The real-time environmental parameter set is used to calculate the real-time traffic risk priority of the current grid. and the historical traffic risk priority R of the current grille. old A comparison is made to determine the type of environmental state change at the current grid; the types of environmental state changes include: environmental improvement, environmental deterioration, and environmental stability, among which historical traffic risk priority R old The priority is calculated in step S3.

[0085] Furthermore, Meets the following conditions: ; in, It is the current real-time obstacle density risk indicator for the grid. This is the real-time ground hardness risk index for the current grid. W1 is the risk impact weight of obstacle density, W2 is the risk impact weight of ground hardness, and W3 is the coordination weight. It is an environmental dynamic adjustment factor used to characterize the impact of the change in current grid environmental parameters relative to historical data on risk.

[0086] Furthermore, the environmental dynamic adjustment factor Meets the following conditions: ; in, It is a real-time environmental change detection indicator for the current grille. It is an environmental change influencing factor and It is used to adjust for the nonlinear impact of environmental changes on traffic risks. The adjustment index is fixed.

[0087] Furthermore, the real-time environmental change detection index Meets the following conditions: ; Where, n new n is the current real-time number of obstacles within the grid. old P represents the number of historical obstacles in the current grid. new It is the real-time ground hardness of the current grid, P old It is the historical ground hardness of the current grating, k 0 1 is the sensitivity coefficient to changes in the number of obstacles, k 0 2 is the sensitivity coefficient to changes in ground hardness, and k 0 1+k 0 2=1, For the third preset minimum positive number, It is the fourth preset minimum positive number.

[0088] S43, According to the current grille G current Real-time traffic risk priority Its historical value The difference △R current Determine the current grid's response level to environmental changes; when △R current The absolute value is greater than the preset change response threshold ΔR 0 When a significant environmental change is detected in the current grid, a neighborhood influence propagation mechanism is triggered; the neighborhood influence propagation mechanism is used to transmit the priority change of the current grid to neighboring geographic grids.

[0089] Furthermore, △R current Meets the following conditions:

[0090] Where, when △R current >△R 0 When △R is reached, it is judged as environmental degradation; when △R is reached... current <-△R 0 When |△R| is considered an environmental improvement; when |△R| is considered an environmental improvement. current |≤△R 0 At that time, the environment was determined to be stable.

[0091] Furthermore, when the neighborhood influence propagation mechanism is triggered, for the current grid G... new Any adjacent grid of order G neighbor Its updated traffic risk priority ; Meets the following conditions: ; in It is the adjacent grille G neighbor Prioritize historical traffic risks. It is the neighborhood influence propagation attenuation coefficient and 0 < <1, This is the Euclidean distance between the current center point of the grid and the center points of the adjacent grids, where d0 is a reference value per unit distance. It is the distance decay exponent and >0, It is the directional influence coefficient, used to characterize the differences in the propagation of environmental changes along different directions.

[0092] S44. Determine the environmental status of the local area where the current grid is located based on the updated traffic risk priority of the current grid and its adjacent grids; when the updated grid traffic risk priority exceeds the preset traffic risk upper limit threshold, mark the grid as an impassable grid and update the impassable grid set; when the updated grid traffic risk priority does not exceed the preset traffic risk upper limit threshold, mark the grid as a passable grid and update the passable grid set.

[0093] S45. Based on the updated passable and impassable grid sets, determine whether there are impassable grids in the uninspected path ahead of the target inspection robot dog; if there are impassable grids, trigger the path replanning mechanism, using the grid G ​​currently located by the target inspection robot dog as the reference point. current As the new starting inspection node, the grid G ​​at the original target location is used. target To conclude the inspection node, the A* path generation model is re-executed using all updated target geographic grid information to generate a new path.

[0094] Furthermore, in the aforementioned path replanning mechanism, dynamic adjustments are made based on the updated grid traffic risk priority, and the A* path generation model obtains a heuristically estimated cost. , Meets the following conditions: ; in, It is the path from the center point of the current path node r to the end inspection node G. target The straight-line Euclidean distance from the center point It is the path from the current node r to the end inspection node G. target The average access risk priority of all target geographic grids covered by a straight path. It is the risk prediction weight and 0 < <1, used to adjust the algorithm's sensitivity to predicting future path risks based on dynamic changes in the environment.

[0095] S46. If the new path generated after path replanning can be obtained from the current grid G current Connect to the end of the inspection grid G target If the new path is successfully replanned, the new path will be sent to the target inspection robot dog, which will then continue its inspection task. If the path cannot be replanned and a connecting path cannot be generated, an emergency handling mechanism will be triggered: the target inspection robot dog will record the location information of the currently impassable area and retreat along the already inspected path to the nearest safety grille. At the same time, it will report the path planning failure information to the target cloud and wait for the cloud to issue a new inspection instruction.

[0096] Furthermore, in the emergency handling mechanism, when the target inspection robot dog retreats in reverse along the already inspected path, it checks the passage risk priority of each grid on the retreat path grid by grid to ensure the safety of the retreat path; if it detects that the environment of the grid behind it has also deteriorated and become impassable during the retreat process, it triggers the standby mode and sends an emergency help signal to the target cloud.

[0097] In summary, step S4 enables the robot dog to instantly perceive and respond to changes in terrain and environment during inspections through real-time environmental data acquisition and dynamic priority updates; the introduction of a neighborhood influence propagation mechanism ensures that environmental changes in a single grid can be reasonably transmitted to surrounding grids, avoiding path planning failures caused by sudden changes in the local environment; and the dual protection of a path replanning mechanism and an emergency handling mechanism enhances the robot dog's autonomous decision-making ability and task execution robustness in dynamic and complex environments.

[0098] This disclosure provides a computer device including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements some or all of the steps in the above-described method.

[0099] This disclosure provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements some or all of the steps in the above-described method. The computer-readable storage medium can be transient or non-transient.

[0100] It should be understood that the phrase "an embodiment" or "one embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this disclosure. Therefore, "in one embodiment" or "one embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this disclosure, the sequence numbers of the above steps / processes do not imply a sequential order of execution; the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this disclosure. The sequence numbers of the above embodiments of this disclosure are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0101] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0102] As can be understood from the several embodiments provided in this disclosure, the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components may be combined, or integrated into another system, or some features may be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0103] The above are merely embodiments of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A security patrol robot dog autonomous path planning method, characterized in that, The method includes the following steps: S1. The target security patrol robot dog sends the first, second and third collected data within the target geographical area to the target cloud, and the target cloud generates a target three-dimensional geographical area map of the target geographical area. S2. The target cloud generates a set of target geographic grids within the target geographic area based on the target three-dimensional geographic area map; The target geographic grid set includes several target geographic grids, and the target geographic grids are geographic grid regions obtained after optimizing the initial geographic grids; S3. The target cloud generates the initial path of the target inspection robot dog based on all the target geographic grids, and sends the initial path to the target inspection robot dog; S4. The target inspection robot dog receives the initial path and performs inspection along the initial path. When the target inspection robot dog reaches any target geographic grid on the initial path, it collects the environmental data of the geographic grid and regenerates the subsequent path as a new path based on the environmental data of the geographic grid.

2. The autonomous path planning method for a security patrol robot dog according to claim 1, characterized in that, In step S1, the first data collected within the target geographical area is the raw point cloud dataset of the target geographical area collected by the target security patrol robot dog through its own lidar device; the second data collected within the target geographical area is the set of obstacle images within the target geographical area collected by the target security patrol robot dog through its own camera device; and the third data collected within the target geographical area is the ground pressure dataset of the target geographical area collected by the target security patrol robot dog through its own foot pressure sensor device.

3. The autonomous path planning method for a security patrol robot dog according to claim 2, characterized in that, Step S1 further includes the following steps: S11. Generate an initial three-dimensional point cloud map based on the first collected data within the target geographical area, wherein the initial three-dimensional geographical area map is generated based on the original point cloud dataset of the target geographical area collected by the target security patrol robot dog through its own lidar device. S12. Based on the second collected data within the target geographical area, generate an initial obstacle dataset, wherein the initial obstacle dataset includes several initial obstacle data, wherein any initial obstacle data includes: an initial obstacle ID, an obstacle type label corresponding to the initial obstacle ID, and an obstacle geographical location corresponding to the initial obstacle ID; S13. Based on the third collected data within the target geographical area, generate a set of ground hardness for the initial geographic grid within the target geographical area; divide the target geographical area into several initial geographic grids; obtain the ground hardness corresponding to the initial geographic grid based on the ground pressure data corresponding to any initial geographic grid in the third collected data. S14. Generate a target three-dimensional geographic region map of the target geographic region based on the initial three-dimensional point cloud map, the initial obstacle dataset, and the ground hardness set of the initial geographic grid.

4. The autonomous path planning method for a security patrol robot dog according to claim 1, characterized in that, The optimization process in step S2 also includes the following steps: S21. Determine intermediate geographic grids based on the initial obstacle information and initial ground hardness within any initial geographic grid; generate the passage risk priority corresponding to the initial geographic grid based on the initial obstacle information and initial ground hardness corresponding to any initial geographic grid; when the passage risk priority corresponding to the initial geographic grid is greater than the preset risk upper limit threshold, determine the initial geographic grid as an impassable geographic grid and insert it into the impassable grid set. If the access risk priority corresponding to the initial geographic grid is not greater than the preset risk upper limit threshold, the initial geographic grid is determined to be a passable geographic grid and inserted into the passable grid set. S22. Determine the target geogrid based on the initial ground hardness of the intermediate geogrid; Based on the initial ground hardness of the intermediate geogrid, the length of the intermediate geogrid, and the width of the intermediate geogrid, the length and width of the target geogrid corresponding to the intermediate geogrid are determined.

5. The autonomous path planning method for a security patrol robot dog according to claim 1, characterized in that, Step S3 further includes the following step to generate the initial path: S31. When all target geographic grid location information meets the preset path generation conditions, the initial path of the target inspection robot dog is generated; the preset path generation conditions refer to the existence of a connected path consisting of several adjacent target geographic grids between the target geographic grid where the starting position of the target inspection robot dog is located and the target geographic grid where the target position is located, and each target geographic grid on the connected path is a passable geographic grid. S32. When the location information of all target geographic grids does not meet the preset path generation conditions, a new target geographic grid is generated from the set of inaccessible grids as a key geographic grid, so as to generate the initial path of the target inspection robot dog based on all key geographic grids and all target geographic grids.

6. The autonomous path planning method for a security patrol robot dog according to claim 5, characterized in that, Step S31 further includes the following steps: S311. Obtain all target geographic grid information, wherein the target geographic grid information includes: the location of the target geographic grid and the traffic risk priority of the target geographic grid; S312. Input all target geographic grid information into the A* path generation model to generate the initial path for the target inspection robot dog.

7. The autonomous path planning method for a security patrol robot dog according to claim 5, characterized in that, Step S32 also includes the following steps: S321. Obtain information on all impassable geographic grids in the impassable grid set. The information on the passable geographic grids includes: the passability risk priority of any impassable geographic grid and the coordinates of the center point of the impassable geographic grid corresponding to the impassable geographic grid. S322. Based on the access risk priority of the impassable geographic grid and the preset access risk upper limit threshold, calculate the optimization potential index of the impassable geographic grid. When the optimization potential index of the impassable geographic grid is greater than the preset optimization potential threshold, mark the impassable grid as an impassable grid to be optimized and insert it into the set of impassable grids to be optimized. S323. For each impassable grid in the set of impassable grids to be optimized, calculate the local environment accessibility fusion index of the impassable grid based on the accessible geographical grid information adjacent to the impassable grid; when the local environment accessibility fusion index of the impassable grid is greater than the preset local environment accessibility fusion threshold, mark the impassable grid as an optimizable accessible grid and insert it into the set of optimizable accessible grids. S324. For each optimizable accessible grid in the optimizable accessible grid set, perform optimization processing to generate a new accessible geographic grid. S325. The newly added passable geographic grids are merged into all the existing target geographic grids to form an updated set of target geographic grids. Then, return to step S31 and re-execute the A* path generation model until the initial path of the target inspection robot dog is generated, or until there are no optimizable impassable grids that meet the optimization conditions. If so, it is determined that there is no connected path in the target 3D geographic area map and the path planning failure information is output.

8. The autonomous path planning method for a security patrol robot dog according to claim 7, characterized in that, Step S324 specifically includes the following sub-steps: S3241. Obtain the ground hardness distribution information corresponding to the original initial geographic grid area of ​​the optimizable accessible grid, and estimate the optimized ground hardness at the center point of the optimizable accessible grid by using bilinear interpolation based on the ground hardness values ​​of adjacent accessible geographic grids. S3242. Calculate the optimized traffic risk priority based on the optimized ground hardness of the optimizable passable grid and the traffic risk priority of the adjacent passable geographic grids of the optimizable passable grid. The optimized access risk priority is calculated based on the number of adjacent accessible geogrids of the optimizable accessible geogrid, the access risk priority of each adjacent accessible geogrid, the ground hardness of each adjacent accessible geogrid, and the hardness difference attenuation coefficient. S3243. Compare the optimized passage risk priority with the preset passage risk upper limit threshold. If the optimized passage risk priority is not greater than the preset passage risk upper limit threshold, mark the optimizable passable grid as a newly added passable geographic grid and update the corresponding target geographic grid information to the target geographic grid set. If the optimized passage risk priority is greater than the preset passage risk upper limit threshold, then its impassable state will be maintained and it will not be included in the newly added passable geographic grid set.

9. A computer device comprising a memory and a processor, the memory storing a computer program executable on the processor, wherein the processor executes the program to implement the method of any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of any one of claims 1 to 8.

Images