Steel structure roof climbing robot path planning method, system, medium and equipment

Abstract

The present invention relates to the technical field of steel structure monitoring. The present invention discloses a path planning method, system, medium and equipment for a steel structure roof climbing robot, including: dividing the surface of the steel structure roof into a number of triangular blocks, and based on the visibility of each observation point to all triangular blocks, with the minimum number of observation points as the goal, after optimizing a number of observation points under the steel structure attachment constraint and the grid full coverage constraint, edge weight assignment is performed to form a spatial observation graph, and the shortest adjacency matrix is ​​obtained by introducing a virtual node and a Freud algorithm, and each observation point in the spatial observation graph is used as a node, the shortest path distance in the shortest adjacency matrix is ​​used as the edge weight, and the virtual node is used as the starting point and the end point to construct a traveling salesman problem expansion graph, and the optimal path of the climbing robot is obtained through a reinforcement learning strategy network based on a graph neural network. The full coverage path planning of the climbing inspection robot for the high-altitude steel structure roof is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of steel structure monitoring, and in particular to a path planning method, system, medium and equipment for a steel structure roof climbing robot. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Existing robot path planning methods often focus on the coverage of the robot's path over a planar area, whereas the path of a climbing robot is along a spatial steel structure and requires full coverage of the target space by the robot's visual space. Existing path planning methods cannot achieve this function. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a path planning method, system, medium and equipment for a steel structure roof climbing robot, establishes a mathematical model with the goal of minimizing the number of observation points, and constructs steel structure attachment and grid full coverage as constraints, so as to achieve efficient visual full coverage of the steel structure roof by the observation points, and on this basis, based on the optimal Hamiltonian path theory, combines virtual nodes and Floyd algorithm, constructs an extended graph of the traveling salesman problem, establishes an equivalent relationship between the traveling salesman problem path and the Hamiltonian path, and realizes the full coverage path planning of the climbing inspection robot for the high-altitude steel structure roof.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A first aspect of the present invention provides a path planning method for a steel structure roof climbing robot, which comprises:

[0007] Obtain the steel structure model, divide the steel structure roof surface into several triangular blocks, and determine the visibility of each observation point to all triangular blocks;

[0008] Based on the visibility of each observation point to all triangle blocks, with the goal of minimizing the number of observation points, under the constraints of steel structure attachment and full grid coverage, after optimizing several observation points, edge weights are assigned according to path distance, passage width and obstacle density to form a spatial observation graph. On the basis of the spatial observation graph, the shortest adjacency matrix is ​​obtained by introducing virtual nodes and Floyd algorithm, and each observation point in the spatial observation graph is taken as a node, the shortest path distance in the shortest adjacency matrix is ​​taken as the edge weight, and the virtual nodes are taken as the starting point and end point to construct the extended graph of the traveling salesman problem.

[0009] Based on the extended graph of the traveling salesman problem, the optimal path for the climbing robot is obtained through a reinforcement learning strategy network based on a graph neural network.

[0010] Furthermore, the visibility of each observation point to all triangle blocks is determined by the following constraints:

[0011] Visual sensor detection angle constraints: ;in, is the normal vector of the kth triangle block, is the centroid of the kth triangle block, is the i-th observation point, for With T k The connection between The angle formed;

[0012] Detection depth constraints of visual sensors: ;in, Indicates the minimum visual depth of the visual sensor, Indicates the maximum visual depth of the visual sensor;

[0013] No occlusion constraint: At the observation point P i and the centroid T of the triangle block k There is no obstruction in between.

[0014] Furthermore, the steel structure attachment constraint is: ; Among them, each support on the surface of the climbable steel structure roof is abstracted into a curve, l i Represents the spatial equation of the i-th curve; Represents the spatial coordinates of the climbing robot; Indicates that the climbing robot is located on the i-th line; represents the spatial distance between the climbing robot and the i-th line, that is: .

[0015] Furthermore, the grid full coverage constraint is ; where k represents the number of triangle blocks and n represents the number of observation points; binary variable Represents the observation point P i Visibility of the j-th triangle block, if , then it means that at the observation point P i The jth triangle block is visible.

[0016] Furthermore, the edge weight assignment is expressed as: ;in, , and All are weights; w ij is the i-th observation point P i To the jth observation point P j Priority;d ij Represents the observation point P from the i-th observation point i To the jth observation point P j Path distance; f ij Represents the observation point P from the i-th observation point i To the jth observation point P j The width of the path; o ij Represents the observation point P from the i-th observation point i To the jth observation point P j The density of obstacles on the path.

[0017] Furthermore, it also includes: acquiring positioning information of ultra-wideband and inertial measurement unit, combining the optimal path of the climbing robot, and obtaining the positioning of the climbing robot through square root unscented Kalman filtering.

[0018] A second aspect of the present invention provides a path planning system for a steel structure roof climbing robot, comprising:

[0019] A steel structure partitioning module is configured to: obtain a steel structure model, partition a steel structure roof surface into a plurality of triangular blocks, and determine the visibility of each observation point to all triangular blocks;

[0020] The graph construction module is configured as follows: based on the visibility of each observation point to all triangle blocks, with the minimum number of observation points as the goal, after optimizing a number of observation points under the constraints of steel structure attachment and full grid coverage, edge weights are assigned according to path distance, pass width and obstacle density to form a spatial observation graph. On the basis of the spatial observation graph, the shortest adjacency matrix is ​​obtained by introducing virtual nodes and the Floyd algorithm, and each observation point in the spatial observation graph is taken as a node, the shortest path distance in the shortest adjacency matrix is ​​taken as the edge weight, and the virtual nodes are taken as the starting point and the end point to construct the extended graph of the traveling salesman problem;

[0021] The path planning module is configured to obtain the optimal path of the climbing robot based on the extended graph of the traveling salesman problem through a reinforcement learning strategy network based on a graph neural network.

[0022] Furthermore, it also includes a positioning module, which is configured to: obtain positioning information of the ultra-wideband and inertial measurement unit, combine the optimal path of the climbing robot, and obtain the positioning of the climbing robot through square root unscented Kalman filtering.

[0023] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor and, when executed by the processor, implements the steps in the path planning method for a steel structure roof climbing robot as described above.

[0024] The fourth aspect of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the program, the steps in the path planning method for a steel structure roof climbing robot as described above are implemented.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The present invention provides a path planning method for a steel structure roof climbing robot, which establishes a mathematical model with steel structure attachment and grid full coverage as constraints with the goal of minimizing the number of observation points, so as to achieve efficient visual full coverage of the steel structure roof by the observation points, and on this basis, based on the optimal Hamiltonian path theory, virtual nodes and Floyd algorithm are combined to construct an extended graph of the traveling salesman problem, and an equivalent relationship between the traveling salesman problem path and the Hamiltonian path is established, so as to achieve full coverage path planning of the high-altitude steel structure roof by the climbing inspection robot.

[0027] The present invention provides a path planning method for a steel structure roof climbing robot, which is based on an IMU and UWB multi-source information combined positioning scheme, introduces a PI-LPM position probability model, and performs positioning information fusion based on SR-UKF to achieve precise positioning of the climbing robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings, which constitute a part of the specification of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but do not constitute limitations of the present invention.

[0029] Figure 1 This is a flow chart of a path planning method for a steel structure roof climbing robot according to Embodiment 1 of the present invention;

[0030] Figure 2 This is a schematic diagram of fusion positioning according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0033] In the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention is further described below with reference to the accompanying drawings and embodiments.

[0034] Embodiment 1

[0035] The purpose of this first embodiment is to provide a path planning method for a steel structure roof climbing robot.

[0036] The steel structure roof climbing robot path planning method provided in this embodiment is as follows: Figure 1 As shown, including:

[0037] Step 1: Study the steel structure surface network division strategy based on the bubble grid method, and transform the continuous visual coverage problem into a discrete grid coverage problem.

[0038] For the inspection task of steel structure, obtain the steel structure model, divide the surface of the steel structure roof into several triangle blocks with similar differences, and start from the observation point P i Observe the target and construct a triangle block k on the surface of the steel structure. If the following constraints are met, it means that the measurement point P i A high-quality image of triangle block k is obtained, that is, the triangle block is visible.

[0039] (1) Detection angle constraint of visual sensor: If the observation point is known P i The maximum cone angle of view is , assuming that the centroid of triangle block k is T k . P i With the centroid of the triangle block T k The angle between the line connecting the triangles and the normal vector of the triangle block is β k,i , the conditions that need to be met are as follows: , ;in, is the normal vector of the triangle block; is the centroid of the triangle block; is a waypoint; for With the centroid of the triangle block T k The line between them and the normal vector of the triangle block The angle between them; for The maximum cone angle of view.

[0040] (2) Detection depth constraint of the visual sensor: The triangle block k must be within the depth range of the visual sensor, that is, the observation point P i To the centroid T k The projection distance on the normal vector of the triangle block k needs to be within the depth range of the visual sensor. That is, the conditions that need to be met are as follows: . Indicates the minimum visual depth of the visual sensor; Indicates the maximum visual depth of the vision sensor.

[0041] (3) No occlusion constraint: At the observation point P i and the centroid T k There is no obstruction in between.

[0042] represents the climbing robot at the observation point P i The observable triangle blocks introduce the concept of visibility matrix. i , for all triangle blocks T = {T 1 ,T 2 ,…T k}, introduce binary variables Indicates the visibility of the kth triangle block, if , then it means that at the observation point P i The kth triangle block is visible, that is, it satisfies the detection angle constraint of the visual sensor, the detection depth constraint of the visual sensor, and the no-occlusion constraint; otherwise, it is invisible. Therefore, the visibility matrix is ​​introduced to represent the observation point P i The visibility of all triangle blocks T can be obtained, namely: .

[0043] Step 2: Establish a mathematical model for optimizing the full coverage observation points of the climbing robot, with the minimum number of observation points n as the objective function and component (steel structure) attachment and full grid coverage as constraints.

[0044] Among them, the objective function is expressed as: .

[0045] Among them, the component attachment constraint is expressed as: ;set up ;in, Indicates that the climbing robot is located on the i-th line, Represents the spatial coordinates of the climbing robot , each support on the surface of the climbable steel structure roof is abstracted into a curve, i Represents the spatial equation of the i-th curve; represents the spatial distance between the climbing robot and the i-th line, that is: .

[0046] Among them, the grid full coverage constraint is expressed as: ; Set the inspection path coverage matrix ; k represents the number of triangle blocks corresponding to the observation point Pi, n represents the number of observation points; for all triangle blocks , introduce binary variables Represents the observation point P i Visibility to the jth triangle block. , then it means that at the observation point Pi The jth triangle block is visible; otherwise, it is invisible. Therefore, the visibility matrix is ​​introduced to represent the observation point P i All triangle blocks can be obtained T Visibility, that is: By setting the visual coverage constraint, the climbing robot passes through a series of observation points. Finally, visual space coverage is achieved. The visibility matrix of each observation point on the entire path constitutes the inspection path coverage matrix. If full visual space coverage is to be achieved, each triangle block must be detected at least once. Therefore, the visual space coverage constraint (grid full coverage constraint) can be expressed as: .

[0047] Step 3: A large-scale parallel particle swarm algorithm based on GPU (graphics processing unit) is proposed to solve the problem (the mathematical model of optimizing the full coverage observation points of the climbing robot) and obtain a number (m) of observation points V = {v 1 ,v 2 ,……,v m}, achieving full visual coverage with the least number of observation points; after obtaining a number of observation points, the connection criteria of the observation points are studied and a spatial observation map is formed, and then the edge weights are assigned according to the path distance, passage width, and obstacle density: ;in, , and They represent the weights of the actual path distance, passage width and obstacle density factor, respectively, and their values ​​are adjusted according to actual application requirements; w ij Used to calculate the observation point P i To other observation points P j The priority is used in the Floyd-Warshall algorithm to obtain the optimal path that traverses all observation points; d ij Represents the path distance, that is, from the i-th observation point P i To the jth observation point P j Path distance; f ij It means passage width, which refers to the width of the path or the difficulty of passing the path; o ij Represents the obstacle density, from the observation point P i To observation point P j The density of obstacles on the path.

[0048] Step 4. On this basis, the optimal Hamiltonian path is used to model the visual full coverage problem, and the TSP (traveling salesman problem) expansion graph is constructed through virtual nodes and the Floyd-Warshall algorithm.

[0049] Step 401: Initial graph construction.

[0050] A complete graph (spatial observation graph) G is initially established, which contains m observation points, i.e., the m observation points optimized in step 4, and the path distance adjacency matrix D of each observation point is constructed.

[0051] Input: A set of observation points V = {v 1 ,v 2 ,……,v m}, adjacency matrix D = { w ij}.

[0052] The output is: the initial graph G=(V,E), and the edge set E consists of the paths between all observation points.

[0053] Step 402: Introduce a virtual node.

[0054] In order to form the optimal Hamiltonian path, a virtual node v is constructed. 0 , as the starting point and end point of the path, connect the virtual node and all the observation points in the initial graph to generate the expanded observation point set V ’ , transforming the problem into a closed loop problem to facilitate path planning.

[0055] Input: initial graph G=(V,E) and adjacency matrix D.

[0056] Output: Expanded graph G ‘ =(V ’ ,E ‘ ) and the expanded adjacency matrix D ’ .

[0057] Step 403: Use the Floyd-Warshall algorithm to calculate the shortest path.

[0058] Take each observation point as a node and use the Floyd-Warshall algorithm to calculate the shortest path distance between all nodes. ’ Convert to a complete graph.

[0059] Input: Expanded adjacency matrix D ’ .

[0060] Output: Updated shortest adjacency matrix D min .

[0061] Step 404: construct a TSP expansion graph.

[0062] Take each observation point as a node, and use D minThe shortest path distance in 0 As the starting point and end point, ensure that the path is closed at the start and end of the virtual node, so that the problem is transformed into a TSP problem.

[0063] Input: shortest path adjacency matrix D min , virtual node v 0 .

[0064] Output: TSP extended graph G TSP .

[0065] Step 5. Establish the equivalence relationship between the TSP path and the Hamiltonian path, adopt a reinforcement learning strategy network based on graph neural network (GNN), effectively learn and extract the geometric features of steel structures of different scales to improve the generalization ability, and construct a small data set to effectively initialize the strategy network using imitation learning, so as to efficiently solve the large-scale TSP problem and provide the optimal path planning for full coverage of the steel structure roof.

[0066] Among them, the reinforcement learning strategy network based on graph neural network is the existing network.

[0067] For the reinforcement learning policy network based on graph neural network, the input is the TSP extended graph (GTSP), and the output is the optimal Hamiltonian path, which represents the optimal path with full coverage of observation points, that is, the optimal path for the climbing robot.

[0068] Step 6. To address the problems of reflection interference, signal blocking and unavailable GPS (Global Positioning System) signals inside steel structures, a multi-source information combined positioning solution based on UWB (ultra-wideband) and IMU (inertial measurement unit) sensors is established. Combined with the path constraint that the climbing robot can only be located on the surface of the steel structure, a position probability (PI-LPM) model is established, and the kernel probability density model is used to calculate the constrained probability density map near the structure.

[0069] Assumptions x is the possible position of the robot on the climbable path (optimal path with full coverage), sample X={x 1 , x 2 , …, x N} is the position of N independent and identically distributed path sampling points, and the path sampling point x i The robot appears at the sampling point x after planning the path. i The probability of nearby can be expressed as follows: ;in, is the kernel function; H is the bandwidth matrix, also known as the window; is the scaling kernel function.

[0070] The standard multivariate Gaussian kernel function with dimension d is: .

[0071] The position of the climbing robot is composed of three-dimensional position information. If d=3, it can be expressed as: .

[0072] According to the kernel density estimation method, the Gaussian sum function of N independent and identically distributed sampling points is linearly superimposed and normalized to obtain the probability density of the PI-LPM (position probability map based on path constraints) model under the 3D Gaussian kernel density: .

[0073] This embodiment proposes an information fusion method based on PL-PLM position probability and UWB / IMU combined positioning based on square root unscented Kalman filter (SR-UKF) to achieve high-precision spatial positioning of the robot that meets the component path constraints, such as Figure 2 As shown, specifically including:

[0074] Step 601: Obtain UWB / IMU positioning information (observation data) and the optimal path for full coverage of observation points.

[0075] Step 602: prediction stage.

[0076] The matrix constructed by Sigma point is: ;in, : The square root matrix of the state covariance matrix; : The dimension of the state vector; : Extended parameters of UKF; Represents the Sigma point set at time k-1, the first point is the state estimation mean, and the remaining points are generated by weighting the square root of the covariance matrix; Represents the mean of the state estimate at time k-1.

[0077] The Sigma point generated by the above formula is processed by the state equation, and the state quantity and error variance matrix at time k are predicted: ; Among them, the number of sigma points in the Sigma point set is 2n+1; Represents the actual value of the i-th sigma point at the k-1th moment; : The predicted value of the i-th sigma point at time k-1 after the state transition model; : state transfer matrix, used to map the state at the previous moment k-1 to the current moment k; : Control input matrix, used to map control input to state space; : Control input at time k-1 (IMU positioning information, including acceleration and angular velocity, etc.); : The predicted state mean at time k, which is obtained by weighted summation of all sigma points; : The mean weight of the i-th sigma point, which is used to calculate the predicted state mean; : The prediction error covariance matrix at time k, which is the covariance of the sigma point and the process noise covariance Q k Calculated; : The covariance weights of the i-th sigma point, these weights are used to calculate the prediction error covariance matrix; : Process noise covariance matrix, representing the uncertainty of the model; : The final value of the prediction error covariance matrix at time k; : A function that updates the covariance matrix , to include the covariance between the initial sigma point and the predicted state mean.

[0078] Step 603: Update phase.

[0079] UWB data observation model: ;in, :UWB observation value, the absolute position of the robot obtained through UWB; : UWB observation function; : Observation noise of UWB.

[0080] PI-LPM data observation model: ; : The observed value of PI-LPM represents the probability value corresponding to the robot position; : The observation function of PI-LPM, that is , maps the location to the probability value on the path probability distribution; : Observation noise of PI-LPM.

[0081] calculate ;in, : Joint observations, including observation data of UWB and PI-LPM; : The predicted observed mean.

[0082] Calculate the variance matrix of the output variable at time k: ;in, : Used to represent the intermediate calculation results of the observation variance matrix; Represents the QR decomposition operation, which is used for matrix decomposition; : Observation noise covariance matrix, representing the measurement error.

[0083] calculate ;in, : The observation covariance matrix after Cholesky (square root method) update.

[0084] Calculate the covariance of the state quantity and the observation quantity at time k: ;in, : The joint covariance matrix of state and observation; : The state mean at the prediction time; : The predicted observed mean.

[0085] Calculate the Kalman filter gain: ;in, : Kalman gain matrix, used to update the state estimate, indicating the weight of the observation on the state estimate; : The joint covariance matrix of state and observation, indicating the correlation between state and observation; : The covariance matrix of the observations, indicating the uncertainty of the observations.

[0086] Update the state quantity and the square root of the error variance matrix: ;in, : Update the intermediate values ​​of the prediction error covariance matrix.

[0087] Step 604: Determine the result of IMU and UWB fusion positioning ( The position information contained in the robot is within the geometric range of a component. If the constraint conditions of the component geometric range are met, the component number and the relative position of the robot relative to the component are output to obtain the fusion positioning result of the component where the robot is located and the relative position; otherwise, the positioning failure information is returned, and a prompt is given to recalibrate the sensor or increase the positioning data sampling. Among them, the relative position and the rod number are matched with the positioning result and the steel structure to determine which rod the robot is on and where it is on this rod.

[0088] In view of the need for visual inspection of steel structure roofs, this embodiment establishes a field of view model of panoramic visual inspection equipment and a mathematical description of the effective perception space of fixed observation points.

[0089] This embodiment establishes a mathematical model with the goal of minimizing the number of observation points and building attachment and full grid coverage as constraints to achieve efficient visual full coverage of the steel structure roof.

[0090] This embodiment is based on the optimal Hamiltonian path theory, combines virtual nodes with the Floyd-Warshall algorithm, constructs a TSP expansion graph, and establishes an equivalent relationship between the TSP path and the Hamiltonian path, so as to provide an effective path planning solution for the visual full coverage problem.

[0091] This embodiment is based on the IMU and UWB multi-source information combined positioning solution, introduces the PI-LPM position probability model, and performs positioning information fusion based on SR-UKF to achieve precise positioning of the climbing robot.

[0092] In general, this embodiment realizes full coverage path planning of the climbing inspection robot for the high-altitude steel structure roof.

[0093] Embodiment 2

[0094] The purpose of the second embodiment is to provide a path planning system for a steel structure roof climbing robot, including:

[0095] A steel structure partitioning module is configured to: obtain a steel structure model, partition a steel structure roof surface into a plurality of triangular blocks, and determine the visibility of each observation point to all triangular blocks;

[0096] The graph construction module is configured as follows: based on the visibility of each observation point to all triangle blocks, with the minimum number of observation points as the goal, after optimizing a number of observation points under the constraints of steel structure attachment and full grid coverage, edge weights are assigned according to path distance, pass width and obstacle density to form a spatial observation graph. On the basis of the spatial observation graph, the shortest adjacency matrix is ​​obtained by introducing virtual nodes and the Floyd algorithm, and each observation point in the spatial observation graph is taken as a node, the shortest path distance in the shortest adjacency matrix is ​​taken as the edge weight, and the virtual nodes are taken as the starting point and the end point to construct the extended graph of the traveling salesman problem;

[0097] A path planning module is configured to: obtain the optimal path of the climbing robot based on the extended graph of the traveling salesman problem through a reinforcement learning strategy network based on a graph neural network;

[0098] The positioning module is configured to obtain positioning information of the ultra-wideband and inertial measurement unit, combine the optimal path of the climbing robot, and obtain the positioning of the climbing robot through square root unscented Kalman filtering.

[0099] It should be noted here that each module in this embodiment corresponds to each step in Example 1 one by one, and the specific implementation process is the same, which will not be repeated here.

[0100] Embodiment 3

[0101] This embodiment provides a computer-readable storage medium on which a computer program is stored. The program is executed by a processor. When the program is executed by the processor, the steps in the path planning method for a steel structure roof climbing robot as described in the above-mentioned embodiment 1 are implemented.

[0102] Embodiment 4

[0103] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the program, the steps in the path planning method for a steel structure roof climbing robot as described in the above-mentioned embodiment 1 are implemented.

[0104] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0105] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A path planning method for a steel structure roof climbing robot, characterized in that: include: Obtain the steel structure model, divide the steel structure roof surface into several triangular blocks, and determine the visibility of each observation point to all triangular blocks; Based on the visibility of each observation point to all triangle blocks, with the goal of minimizing the number of observation points, under the constraints of steel structure attachment and full grid coverage, after optimizing several observation points, edge weights are assigned according to path distance, passage width and obstacle density to form a spatial observation graph. On the basis of the spatial observation graph, the shortest adjacency matrix is ​​obtained by introducing virtual nodes and Floyd algorithm, and each observation point in the spatial observation graph is taken as a node, the shortest path distance in the shortest adjacency matrix is ​​taken as the edge weight, and the virtual nodes are taken as the starting point and end point to construct the extended graph of the traveling salesman problem. Based on the extended graph of the traveling salesman problem, the optimal path for the climbing robot is obtained through a reinforcement learning strategy network based on a graph neural network.

2. The path planning method for a steel structure roof climbing robot according to claim 1, characterized in that: The visibility of each observation point to all triangle blocks is determined by the following constraints: Visual sensor detection angle constraints: ;in, is the normal vector of the kth triangle block, is the centroid of the kth triangle block, is the i-th observation point, for With T k The connection between The angle formed; Detection depth constraints of visual sensors: ;in, Indicates the minimum visual depth of the visual sensor, Indicates the maximum visual depth of the visual sensor; No occlusion constraint: At the observation point P i and the centroid T of the triangle block k There is no obstruction in between.

3. The path planning method for a steel structure roof climbing robot according to claim 1, characterized in that: The steel structure attachment constraints are: ; Among them, each support on the surface of the climbable steel structure roof is abstracted into a curve, l i Represents the spatial equation of the i-th curve; Represents the spatial coordinates of the climbing robot; Indicates that the climbing robot is located on the i-th line; represents the spatial distance between the climbing robot and the i-th line, that is: .

4. The path planning method for a steel structure roof climbing robot according to claim 1, characterized in that: The grid full coverage constraint is ; where k represents the number of triangle blocks and n represents the number of observation points; binary variable Represents the observation point P i Visibility of the j-th triangle block, if , then it means that at the observation point P i The jth triangle block is visible.

5. The path planning method for a steel structure roof climbing robot according to claim 1, characterized in that: The edge weight assignment is expressed as: ;in, , and All are weights; w ij is the i-th observation point P i To the jth observation point P j Priority; d ij Represents the observation point P from the i-th observation point i To the jth observation point P j Path distance; f ij Represents the observation point P from the i-th observation point i To the jth observation point P j The width of the path; o ij Represents the observation point P from the i-th observation point i To the jth observation point P j The density of obstacles on the path.

6. The path planning method for a steel structure roof climbing robot according to claim 1, characterized in that: Also includes: The positioning information of the ultra-wideband and inertial measurement unit is obtained, combined with the optimal path of the climbing robot, and the positioning of the climbing robot is obtained through the square root unscented Kalman filter.

7. The path planning system for steel structure roof climbing robots is characterized by: include: A steel structure partitioning module is configured to: obtain a steel structure model, partition a steel structure roof surface into a plurality of triangular blocks, and determine visibility of each observation point to all triangular blocks; The graph construction module is configured as follows: based on the visibility of each observation point to all triangle blocks, with the minimum number of observation points as the goal, after optimizing a number of observation points under the constraints of steel structure attachment and full grid coverage, edge weights are assigned according to path distance, passage width and obstacle density to form a spatial observation graph. On the basis of the spatial observation graph, the shortest adjacency matrix is ​​obtained by introducing virtual nodes and the Floyd algorithm, and each observation point in the spatial observation graph is taken as a node, the shortest path distance in the shortest adjacency matrix is ​​taken as the edge weight, and the virtual nodes are taken as the starting point and the end point to construct the extended graph of the traveling salesman problem; The path planning module is configured to obtain the optimal path of the climbing robot based on the extended graph of the traveling salesman problem through a reinforcement learning strategy network based on a graph neural network.

8. The steel structure roof climbing robot path planning system according to claim 7, characterized in that: It also includes a positioning module, which is configured to: obtain positioning information of the ultra-wideband and inertial measurement unit, combine the optimal path of the climbing robot, and obtain the positioning of the climbing robot through square root unscented Kalman filtering.

9. A computer-readable storage medium having a computer program stored thereon, the program being executed by a processor, characterized in that: When the program is executed by a processor, the steps in the path planning method for a steel structure roof climbing robot as described in any one of claims 1 to 6 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the path planning method for a steel structure roof climbing robot are implemented as described in any one of claims 1-6.

Images