Method for planning global path of inspection robot in structured road environment

Abstract

The invention relates to a method for performing global path planning in a structured road environment based on an inspection robot, which comprises the following steps of: firstly, acquiring elevation information and map information of the environment by adopting different software, and constructing a three-dimensional elevation topological map of a road in the environment to be inspected; then, comprehensively considering two topographic factors of path length and path fluctuation and energy consumption requirements of the robot, and establishing an optimization function model of global path planning; on the basis, model parameters of a global path planning optimization function are determined through inverse reinforcement learning; and finally, solving a global path planning problem by means of a Floyd algorithm and a particle swarm algorithm, and giving a global path planning scheme. The trajectory planned by the method is closer to an ideal specified trajectory, and can still maintain high adaptability and robustness in the face of complex and changeable environmental conditions, thereby greatly improving the navigation performance and operation convenience of the robot in practical application.

Description

Technical Field

[0001] The invention belongs to the technical field of outdoor global path planning of inspection robots, and in particular relates to a global path planning method for inspection robots in a structured road environment. Background Art

[0002] With the rapid development of social economy, inspection robots have been widely used in many places such as exhibition centers, production plants, logistics warehouses and campuses. Usually, when performing tasks, inspection robots are required to start from the starting point, pass through all roads in the inspection environment, and ensure that each road is walked at least once, and finally return to the starting point for charging, while ensuring energy consumption. This problem is equivalent to a full coverage multi-constraint global path planning problem in a structured road environment. The effective solution to this problem is an important factor in determining whether the inspection robot can smoothly and efficiently perform inspection tasks.

[0003] The global path planning algorithm generally includes environmental modeling and path search, in which the acquisition of the global optimal path requires comprehensive factors such as path length, navigation time or energy minimization. The current main path planning methods are mainly divided into traditional methods, intelligent bionic optimization methods and deep learning methods. Traditional global path planning methods include artificial potential field method, simulated annealing algorithm, A* algorithm and RRT algorithm, among which the most representative ones are A* algorithm and RRT algorithm. However, such methods can quickly obtain the optimal path under effective and small environmental scales and have a good effect. However, with the expansion of the environmental scale, such algorithms have problems such as low algorithm efficiency, easy to fall into local optimality and unable to solve environmental uncertainty. Intelligent bionic optimization algorithms such as ant colony algorithm, genetic algorithm and particle swarm algorithm are widely used to solve path optimization problems due to their intelligence and robustness, and have achieved good results, but there are also serious premature phenomena and easy to fall into local optimal solutions in the later stage. In order to solve the above problems, scholars have further improved the algorithm to avoid the occurrence of such problems.

[0004] The key to achieving path planning based on reinforcement learning and inverse reinforcement learning in deep learning methods is the setting of reward functions and action mechanisms. For example, the reward function of inverse reinforcement learning is learned by a method based on maximum entropy calculation. By simulating the driving behavior of a large number of users, the suboptimality problem of random expert data is solved, but the results are uncertain in unfamiliar environments. Alternatively, inverse reinforcement learning is used to create an approximate cost function generator for visual navigation challenges to obtain an implicit objective function based on visual navigation, which can achieve end-to-end path planning, but has poor adaptability to complex environments. The above algorithms are mainly aimed at improving the problem of two-dimensional space or seeking the shortest path between two vertices. However, in the face of the multi-constrained full coverage problem of the inspection robot inspecting all edges in a three-dimensional specified environment and returning to the starting point, the existing technical methods highlight a series of limitations, such as low algorithm efficiency, easy to fall into local optimal solutions, and difficulty in dealing with environmental uncertainties. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art, the present invention proposes a global path planning method for an inspection robot in a structured road environment based on inverse reinforcement learning and particle swarm algorithm: first, different software are used to obtain the elevation information and map information of the environment, and a complete three-dimensional elevation topological map of the road is constructed; then, the two terrain factors of path length and path undulation and the energy consumption requirements of the robot are comprehensively considered to establish an optimization function model for global path planning; on this basis, the model parameters of the global path planning optimization function are determined through inverse reinforcement learning; finally, with the help of Floyd algorithm and particle swarm algorithm, the global path planning problem is solved, and a global path planning solution is given.

[0006] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is:

[0007] A global path planning method for an inspection robot in a structured road environment comprises the following steps:

[0008] S1: Obtain elevation information and map information of the environment to be inspected, and construct a three-dimensional elevation topological map of the roads in the environment to be inspected;

[0009] S2: Based on the three-dimensional elevation topological map of the environment to be inspected, a global path planning problem model for the inspection robot is established;

[0010] S3: Adjust the model parameters of the global path planning problem in the inspection environment through inverse reinforcement learning;

[0011] S4: Find the optimal path between all vertices in the topological map through Floyd's algorithm;

[0012] S5: Global optimal path planning is achieved through particle swarm algorithm.

[0013] Preferably, S1 comprises the following sub-steps:

[0014] S11: Based on Amap, the image segmentation technology is used to extract road distribution information and build a two-dimensional road topology map;

[0015] S12: Using Tuxin Earth software, collect the height of the road profile relative to the sea level in the patrol area every k meters, i.e., elevation data;

[0016] S13: The road elevation data is integrated with the two-dimensional road topological map to construct a three-dimensional elevation topological map of the road.

[0017] Preferably, S2 comprises the following sub-steps:

[0018] S21: The three-dimensional road topology map established for the inspection environment has N vertices and T edges. The set of T edges is denoted as H. The length of the edge between any two vertices i and j is calculated as l. i,j , the fluctuation is g i,j , energy consumption is c i,j ;

[0019] If the extraction interval of path elevation data is k meters, then the path length l i,j That is, the number of elevation data multiplied by k meters, as shown in the following formula:

[0020] l i,j =Num*k

[0021] Path undulation g i,j , calculated by the elevation data difference between two adjacent points on the road, as follows:

[0022]

[0023] Where: Indicates the elevation value of the mth point on the path;

[0024] Path energy consumption i,j , obtained according to the COPERT vehicle energy consumption prediction model, the specific formula is as follows:

[0025]

[0026] Where: is the road slope of the mth segment in the path between vertices i and j;

[0027] S22: Considering the two terrain factors of path length and undulation and the energy consumption requirements of the robot, the description function of the global path planning problem is constructed - the global path fitness F is as follows:

[0028]

[0029] α1+α2+α3=1

[0030] Where: H is the set of edges between vertices in global path planning, F(i,j) is the path fitness function between vertex i and vertex j, and α1, α2, and α3 are the corresponding parameter weights.

[0031] Preferably, S3 includes the following sub-steps:

[0032] S31: Take the local environment path and perform path planning according to human preference as the expert trajectory;

[0033] S32: According to the global path description model, the Floyd algorithm is used to obtain the optimal path between each vertex in the local environment path, and this is used as the strategy trajectory;

[0034] S33: Compare the strategy trajectory and the expert trajectory of the starting point and the end point in the local environment path, obtain the trajectory difference, and determine whether the trajectory difference is less than the expected threshold. If it is less than the expected threshold, keep the current weight parameter; if not, proceed to the next step;

[0035] S34: Adjust the weight parameters to obtain new weight parameters;

[0036] S35: When the termination condition is met, the iteration is stopped, otherwise, the process returns to S32 for parameter tuning.

[0037] Preferably, S4 includes the following sub-steps:

[0038] S41: for the three-dimensional elevation topological map of the environment to be inspected, establish an optimal path weight matrix PM between vertices and initialize all its matrix elements;

[0039] S42: Determine the optimal path between any two vertices.

[0040] As a preferred embodiment, S41 is specifically as follows: the three-dimensional elevation topological map of the environment to be inspected contains N vertices, and the PM matrix is ​​an N*N square matrix, wherein the matrix element p ij Initialize according to the following rules: If there is an edge between two vertices, calculate the fitness value F of the edge between the two vertices i,j , let p ij＝ F i,j ; If there is no direct edge between two vertices, then p ij＝ +∞.

[0041] As a preferred embodiment, S42 is specifically as follows: calculate the fitness values ​​of all paths from vertex i to j, where the path with the smallest fitness value is the optimal path between the two vertices. Let p ij＝ min F i,j .

[0042] Preferably, S5 includes the following sub-steps:

[0043] S51: Initialize the particle swarm algorithm parameters, number the vertices and edges of the three-dimensional elevation topology map, initialize the particles, and generate a global path;

[0044] S52: Calculate the fitness of the global path of each particle according to the optimal path weight matrix between vertices, and determine the global optimal path;

[0045] S53: Determine whether the fitness value of the current global optimal path is less than the historical global optimal path, and whether the termination condition is met;

[0046] S54: If the termination condition is not met, the global path corresponding to the particle is updated and the process returns to S52;

[0047] S55: If the termination condition is met, the currently saved global optimal path is output as a solution to the global path planning problem.

[0048] The present invention firstly adopts different software to obtain the elevation value and map information of the road and constructs a complete three-dimensional elevation topological map; then, the description function of the global path planning is established by comprehensively considering the two terrain factors of path length and path undulation and the energy consumption requirements of the robot; next, expert data with human preference is set, and the parameters of the global path planning function are adjusted through inverse reinforcement learning; finally, the vertex particle swarm algorithm combining the Floyd algorithm and the particle swarm algorithm is used to realize efficient solution of the global path planning problem.

[0049] Compared with the prior art, the beneficial effects and advantages of the present invention are:

[0050] 1. The present invention effectively reduces the complexity of global path planning of inspection robots in structured road environments, and has made significant breakthroughs in modeling global planning problems and parameter setting for different inspection environments. By using inverse reinforcement learning technology, the automatic optimization and setting of global path planning model parameters are realized, thereby improving the adaptability and accuracy of parameter configuration. The trajectory planned by the present invention is not only closer to the ideal specified trajectory, but also can maintain a high degree of adaptability and robustness when facing complex and changeable environmental conditions, thereby greatly improving the navigation performance and ease of operation of the robot in practical applications.

[0051] 2. The vertex particle swarm algorithm proposed in the present invention shows good performance in many aspects and has significant advantages in finding the optimal solution for global path planning. It can converge to a high-quality solution more effectively with higher efficiency, and provides a new and efficient solution to the global path planning problem of inspection robots in structured road environments. It also provides valuable reference and basis for subsequent related research. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is the overall flow chart of the present invention;

[0053] Figure 2 This is an example of establishing an elevation topology map of a structured road; (a) is a real scene map, (b) is a topology map of a structured road extracted from the background; (c) is a three-dimensional road information map, which contains information on a two-dimensional topology map and an elevation map;

[0054] Figure 3 This is a schematic diagram of the optimal path weight matrix between vertices based on Floyd's algorithm;

[0055] Figure 4 It is the global planning path diagram of inverse reinforcement parameters;

[0056] Figure 5 It is a global path planning graph with the starting point "P" and the optimal F value;

[0057] Figure 6 It is a real-life map of a large campus area;

[0058] Figure 7 It is a large-area campus road topology map. DETAILED DESCRIPTION

[0059] The technical scheme of the present invention is further specifically described below through examples, which are provided for the purpose of illustrating the present invention and are not intended to limit the present invention. Based on the examples in this application, all other examples obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0060] Example 1

[0061] Reference Figure 1 , a global path planning method for an inspection robot in a structured road environment, comprising the following steps:

[0062] S1: Obtain elevation information and map information of the environment to be inspected, and construct a three-dimensional elevation topological map of the roads in the environment to be inspected;

[0063] S2: Based on the three-dimensional elevation topological map of the environment to be inspected, a global path planning problem model for the inspection robot is established;

[0064] S3: Adjust the model parameters of the global path planning problem in the inspection environment through inverse reinforcement learning;

[0065] S4: Find the optimal path between all vertices in the topological map through Floyd's algorithm;

[0066] S5: Global optimal path planning is achieved through particle swarm algorithm.

[0067] Specifically, refer to Figure 2 , S1 includes the following sub-steps:

[0068] S11: Based on Amap, we use image segmentation technology to extract real-life images, such as Figure 2 (a) Extract road distribution information and build a two-dimensional road topological map, such as Figure 2 (b)

[0069] S12: Using Tuxin Earth software, collect the height of the road profile relative to the sea level in the patrol area every k meters, i.e., elevation data;

[0070] S13: The road elevation data is integrated with the two-dimensional road topological map to construct a three-dimensional elevation topological map of the road, such as Figure 2 (c) as shown.

[0071] S2 includes the following sub-steps:

[0072] S21: The three-dimensional road topology map established for the inspection environment has N vertices and T edges. The set of T edges is denoted as H. The length of the edge between any two vertices i and j is calculated as l. i,j , the fluctuation is g i,j , energy consumption is c i,j ;

[0073] If the extraction interval of path elevation data is k meters, then the path length l i,j That is, the number of elevation data multiplied by k meters, as shown in the following formula:

[0074] l i,j =Num*k

[0075] Path undulation g i,j , calculated by the elevation data difference between two adjacent points on the road, as follows:

[0076]

[0077] Where: Indicates the elevation value of the mth point on the path;

[0078] Path energy consumption i,j , obtained according to the COPERT vehicle energy consumption prediction model, the specific formula is as follows:

[0079]

[0080] Where: is the road slope of the mth segment in the path between vertices i and j;

[0081] S22: Considering the two terrain factors of path length and undulation and the energy consumption requirements of the robot, the description function of the global path planning problem - the global path fitness F is constructed as follows:

[0082]

[0083] α1+α2+α3=1

[0084] Where: H is the set of edges between vertices in global path planning, F(i,j) is the path fitness function between vertex i and vertex j, and α1, α2, and α3 are the corresponding parameter weights.

[0085] S3 includes the following sub-steps:

[0086] S31: Take the local environment path and perform path planning according to human preference as the expert trajectory;

[0087] S32: Based on the global path description model, use the Floyd algorithm to obtain the optimal path between vertices in the local environment path, and use this as the strategy trajectory; the result process of the optimal path between vertices based on the Floyd algorithm is referred to Figure 3 , where ABCDE represents the vertices in the topological structure, AB represents an edge in the topological structure graph, corresponding to an actual section of road; each element in the PM matrix represents the fitness value of the corresponding path;

[0088] S33: Compare the strategy trajectory and the expert trajectory of the starting point and the end point in the local environment path, obtain the trajectory difference, and determine whether the trajectory difference is less than the expected threshold. If it is less than the expected threshold, keep the current weight parameter; if not, proceed to the next step;

[0089] S34: Adjust the weight parameters to obtain new weight parameters;

[0090] S35: When the termination condition is met, the iteration is stopped, otherwise, the process returns to S32 for parameter tuning.

[0091] S4 includes the following sub-steps:

[0092] S41: for the three-dimensional elevation topological map of the environment to be inspected, establish an optimal path weight matrix PM between vertices and initialize all its matrix elements;

[0093] The 3D elevation topological map of the environment to be inspected contains N vertices, so the PM matrix is ​​an N*N square matrix, where the matrix element p ij Initialize according to the following rules: If there is an edge between two vertices, calculate the fitness value F of the edge between the two vertices i,j , let p ij＝ F i,j; If there is no direct edge between two vertices, then p ij =+∞;

[0094] S42: Determine the optimal path between any two vertices;

[0095] Calculate the fitness values ​​of all paths from vertex i to j. The path with the smallest fitness value is the optimal path between the two vertices. Let p ij＝ min F i,j .

[0096] S5 includes the following sub-steps:

[0097] S51: Initialize the particle swarm algorithm parameters, number the vertices and edges of the three-dimensional elevation topology map, initialize the particles, and generate a global path;

[0098] S52: Calculate the fitness of the global path of each particle according to the optimal path weight matrix between vertices, and determine the global optimal path;

[0099] S53: Determine whether the fitness value of the current global optimal path is less than the historical global optimal path, and whether the termination condition is met;

[0100] S54: If the termination condition is not met, the global path corresponding to the particle is updated and the process returns to S52;

[0101] S55: If the termination condition is met, the currently saved global optimal path is output as a solution to the global path planning problem.

[0102] Example 2

[0103] The experimental scene mainly selects the outdoor standard road environment in the campus. The experimental platform is Windows 11, the computer configuration is 11th Gen Intel (R) Core (TM) i7-11800H @ 2.30GHz 8 cores, and the memory is 16GB. At the same time, in order to better measure the performance of the algorithm, the experiment uses four indicators: F, L, G and C. Among them, F is the global path fitness; L = ∑ (i,j)∈H α1l i,j is the global path length fitness; G = ∑ (i,j)∈H α2g i,j is the global path fluctuation fitness and C = ∑ (i,j)∈H α3c i,j is the global path energy consumption fitness.

[0104] Comparison of Inverse Reinforcement Learning Schemes

[0105] In order to verify the role of inverse reinforcement learning in this method, the expert path of human preference will be introduced below, and the formula The [α1, α2, α3] parameters in are tuned.

[0106] Inverse reinforcement learning is used to automatically set parameters to make the trajectory more consistent with the specified trajectory. "HIKLMP" and "DEFGIK" are introduced as expert paths of human preference. The following are the experimental results of random parameters [0.2, 0.43, 0.37] and parameters adjusted by inverse reinforcement learning [0.25, 0.63, 0.12].

[0107] Table 1 Comparison of inverse reinforcement learning parameters and random parameters

[0108]

[0109] As can be seen from Table 1, among the paths generated by the randomly generated initial parameters and the parameters after inverse reinforcement learning with "H", "P" and "D" and "K" as the starting and ending points respectively, the path generated by the random parameters is quite different from the expert path. The parameters obtained after inverse reinforcement learning make the path obtained by the algorithm more consistent with the expert path. Therefore, the path generated by the parameters after inverse reinforcement learning can better reflect the characteristics of the expert path of human preference.

[0110] Reverse strengthening parameter experiment

[0111] To verify the influence of parameters on the trajectory, the weight parameters of the road section cost function are set to observe the changes in the path trajectory. The experiment selects "O" as the starting point, and [α1, α2, α3] are four groups of parameters: [2 / 3, 1 / 6, 1 / 6], [1 / 6, 2 / 3, 1 / 6], [1 / 6, 1 / 6, 2 / 3] and inverse reinforcement parameters [0.25, 0.63, 0.12]. Each group of parameters is tested 10 times and the average fitness is calculated, as shown in Table 2. Finally, the global planning path achieved by one of the inverse reinforcement parameters is selected for display, as shown in Table 2. Figure 4 As shown:

[0112] Table 2 Parameter fitness comparison table

[0113] <![CDATA[[α1,α2,α3]]]> L G C F [2 / 3,1 / 6,1 / 6] 325.58 425.13 418.01 357.57 [1 / 6,2 / 3,1 / 6] 351.83 363.12 401.18 367.58 [1 / 6,1 / 6,2 / 3] 365.58 414.33 361.76 371.15 [0.25,0.63,0.12] 330.08 371.18 381.93 361.06

[0114] As can be seen from Table 2, when the parameters are [2 / 3, 1 / 6, 1 / 6], the L value is small, and the path sought at this time tends to have a short path length; when the parameters are [1 / 6, 2 / 3, 1 / 6], the G value is small, and the path sought at this time tends to have a small path fluctuation; when the parameters are [1 / 6, 1 / 6, 2 / 3], the C value is small, indicating that the path sought at this time tends to have a smaller energy consumption. Compared with the parameters [2 / 3, 1 / 6, 1 / 6], the parameters after inverse reinforcement learning have the smallest F value, but the path fluctuation fitness value and energy consumption fitness value are both the largest, which does not conform to human preferences. The parameters after inverse reinforcement learning can balance the path length, fluctuation and energy consumption factors while ensuring a small F value, which is more in line with human preferences.

[0115] Starting point experimental analysis

[0116] Considering the comprehensive performance of the generated trajectory, the experimental setting uses the inverse enhanced parameters: [0.25, 0.63, 0.12]. The experiment selects "A", "O", "D", "K" and "P" points as starting points, and conducts 10 experiments at each starting point and calculates the average L, G, C and value F. The experimental results are shown in Table 3.

[0117] Table 3 Fitness value statistics

[0118]

[0119]

[0120] Table 3 shows the fitness values ​​with "A", "O", "D", "K" and "P" as starting points. The data results of this experiment are all within an acceptable range, which verifies the effectiveness of the algorithm in global path planning with random starting points in different environments. Figure 5 An example of global path planning with the starting point "P" and the optimal F value is given to further verify the feasibility of the algorithm.

[0121] Comparison of Vertex Particle Swarm Solutions

[0122] The vertex particle swarm scheme is mainly composed of the Floyd algorithm and the particle swarm algorithm. To test the effectiveness of this method, the following experiment uses a larger campus map. A total of 23 vertices and 34 paths are established, of which path segment 30 is a one-way segment. The real scene of the large-scale campus map is as follows Figure 6 As shown in the figure, the large-area campus road topology map is as follows Figure 7 shown.

[0123] To further verify the effectiveness of the present invention, the vertex particle swarm algorithm is compared with the particle swarm algorithm and the ant colony algorithm. The following comparison algorithms are all two-layer ant colony algorithms or particle swarm algorithms. Each group of experiments is carried out 10 times and the average fitness and average time consumption are calculated. The weight parameters are set to [0.3, 0.5, 0.2]. The performance comparison results of the methods are shown in Table 4.

[0124] Table 4 Performance comparison

[0125]

[0126] As can be seen from Table 4, compared with the two-layer ant colony algorithm and particle swarm algorithm, the vertex particle swarm algorithm proposed in the present invention has the best average fitness solved at the starting points "L", "G", "C" and "P", and the time required is the shortest, which further verifies the feasibility of the method of the present invention.

[0127] The present invention first obtains global environmental information with the help of a global map constructed based on a topological structure, and constructs a global planning problem model by comprehensively considering key factors such as path length, path undulation, and vehicle energy consumption; then, a global path planning method based on inverse reinforcement learning and vertex particle swarm algorithm is proposed, inverse reinforcement learning is used to adjust the parameters of the global problem model, and the Floyd algorithm is used to optimize the path between vertices. Finally, the global planning path is obtained through the particle swarm algorithm. The present invention verifies the effectiveness of the proposed method through a campus inspection scenario.

[0128] The present invention is described in detail above in conjunction with the embodiments, but the contents described are only preferred embodiments of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of application of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A global path planning method for an inspection robot in a structured road environment, characterized in that The steps include: S1: Obtain elevation information and map information of the environment to be inspected, and construct a three-dimensional elevation topological map of the roads in the environment to be inspected; S2: Based on the three-dimensional elevation topological map of the environment to be inspected, a global path planning problem model for the inspection robot is established; S3: Adjust the model parameters of the global path planning problem in the inspection environment through inverse reinforcement learning; S4: Find the optimal path between all vertices in the topological map through Floyd's algorithm; S5: Global optimal path planning is achieved through particle swarm algorithm.

2. The global path planning method for an inspection robot in a structured road environment according to claim 1 is characterized in that S1 includes the following sub-steps: S11: Based on Amap, the image segmentation technology is used to extract road distribution information and build a two-dimensional road topology map; S12: Using Tuxin Earth software, collect the height of the road profile relative to the sea level in the patrol area every k meters, i.e., elevation data; S13: The road elevation data is integrated with the two-dimensional road topological map to construct a three-dimensional elevation topological map of the road.

3. The global path planning method for an inspection robot in a structured road environment according to claim 1 is characterized in that S2 includes the following sub-steps: S21: The three-dimensional road topology map established for the inspection environment has N vertices and T edges. The set of T edges is denoted as H. The length of the edge between any two vertices i and j is calculated as l. i,j , the fluctuation is g i,j , energy consumption is c i,j ; If the extraction interval of path elevation data is k meters, the path length l i,j That is, the number of elevation data multiplied by k meters, as shown in the following formula: l i,j =Number*k Path undulation g i,j , calculated by the elevation data difference between two adjacent points on the road, as follows: Where: Indicates the elevation value of the mth point on the path; Path energy consumption i,j , obtained according to the COPERT vehicle energy consumption prediction model, the specific formula is as follows: Where: is the road slope of the mth segment in the path between vertices i and j; S22: Considering the two terrain factors of path length and undulation and the energy consumption requirements of the robot, the global path fitness F is constructed as follows: α1+α2+α3＝1 Where: H is the set of edges between vertices in global path planning, F(i,j) is the path fitness function between vertex i and vertex j, and α1, α2, and α3 are the corresponding parameter weights.

4. The global path planning method for an inspection robot in a structured road environment according to claim 1 is characterized in that S3 includes the following sub-steps: S31: Take the local environment path and perform path planning according to human preference as the expert trajectory; S32: According to the global path description model, the Floyd algorithm is used to obtain the optimal path between each vertex in the local environment path, and this is used as the strategy trajectory; S33: Compare the strategy trajectory and the expert trajectory of the starting point and the end point in the local environment path, obtain the trajectory difference, and determine whether the trajectory difference is less than the expected threshold. If it is less than the expected threshold, keep the current weight parameter; if not, proceed to the next step; S34: Adjust the weight parameters to obtain new weight parameters; S35: When the termination condition is met, the iteration is stopped, otherwise, the process returns to S32 for parameter tuning.

5. The global path planning method for an inspection robot in a structured road environment according to claim 1 is characterized in that S4 includes the following sub-steps: S41: for the three-dimensional elevation topological map of the environment to be inspected, establish an optimal path weight matrix PM between vertices and initialize all its matrix elements; S42: Determine the optimal path between any two vertices.

6. The global path planning method for an inspection robot in a structured road environment according to claim 5 is characterized in that S41 is specifically: The 3D elevation topological map of the environment to be inspected contains N vertices, so the PM matrix is ​​an N*N square matrix, where the matrix element p ij Initialize according to the following rules: If there is an edge between two vertices, calculate the fitness value F of the edge between the two vertices i,j , let p ij =F i,j ; If there is no direct edge between two vertices, then p ij＝ +∞.

7. The global path planning method for an inspection robot in a structured road environment according to claim 6 is characterized in that S42 is specifically: Calculate the fitness values ​​of all paths from vertex i to j. The path with the smallest fitness value is the optimal path between the two vertices. Let p ij＝ min F i,j .

8. The global path planning method for an inspection robot in a structured road environment according to claim 1 is characterized in that S5 includes the following sub-steps: S51: Initialize the particle swarm algorithm parameters, number the vertices and edges of the three-dimensional elevation topology map, initialize the particles, and generate a global path; S52: Calculate the fitness of the global path of each particle according to the optimal path weight matrix between vertices, and determine the global optimal path; S53: Determine whether the fitness value of the current global optimal path is less than the historical global optimal path, and whether the termination condition is met; S54: If the termination condition is not met, the global path corresponding to the particle is updated and the process returns to S52; S55: If the termination condition is met, the currently saved global optimal path is output as a solution to the global path planning problem.

Images