A multi-target dynamic path planning method and system for a tunnel environment

Abstract

The application belongs to the field of tunnel construction automation and robot navigation technology, and provides a multi-target dynamic path planning method and system for tunnel environment, dynamically adjusts node expansion step length according to tunnel curvature of the current node position, adds a safety penalty term related to the tunnel wall distance in the cost function, preferentially reuses the pre-stored planning path segment when planning a new path, and takes the maximum value of the Euclidean distance and the topological distance as the heuristic estimation of the hybrid heuristic function; compared with the classical search algorithm, the number of node expansions and the calculation time are reduced, the wall distance safety cost is taken as a continuous penalty term and integrated into the cost function, the quantitative trade-off between safety and path length is realized, the collision risk is effectively reduced, the maximum value of the distance and the topological distance is taken as the heuristic estimation, the adoptability is maintained and the information is better than the pure Euclidean distance, and the invalid node expansion is reduced.

Description

Technical Field

[0001] This invention belongs to the field of tunnel construction automation and robot navigation technology, and in particular relates to a multi-objective dynamic path planning method and system for tunnel environments. Background Technology

[0002] The tunnel's interior space is long and narrow with a complex structure, presenting a unique environment with both static and dynamic obstacles. This places stringent demands on the robot's real-time performance, safety, and multi-target adaptability in path planning.

[0003] Current graph search-based path planning algorithms employ a uniform node sampling strategy, resulting in a large number of redundant nodes in straight tunnel sections and reduced path smoothness in curved sections due to insufficient resolution. They only use path length as the optimization objective and do not incorporate safety distance into the cost function. Furthermore, classic search algorithms use Euclidean distance as a heuristic function, which severely underestimates the actual cost when crossing tunnel sections, leading to the expansion of a large number of invalid nodes. As a result, current search algorithms are not suitable for path planning of robots in long, narrow, and structurally complex tunnels. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a multi-objective dynamic path planning method and system for tunnel environments. This invention reduces the number of node expansions and computation time, incorporates the safety cost of wall distance as a continuous penalty term into the cost function, achieves a quantitative trade-off between safety and path length, effectively reduces collision risk, takes the maximum value of distance and topological distance as a heuristic estimate, maintains admissibility and has better informativeness than pure Euclidean distance, reduces invalid node expansion, and adapts to tunnel environments with narrow spaces and complex structures.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: In a first aspect, the present invention provides a multi-objective dynamic path planning method for tunnel environments, comprising: Get the tunnel map; Based on the tunnel map, a two-level graph structure including intersection nodes and tunnel section nodes is obtained; according to the principle of real-time adjustment based on the urgency of the task, dynamic priority weights are set for each target point in the two-level graph structure. The target points are sorted based on the current node position, the priority weight of unvisited target points, and geometric angle conditions; According to the sorting and search algorithm, a path is generated between adjacent sorted target points; wherein, in the search algorithm, the node expansion step size is dynamically adjusted according to the tunnel curvature at the current node location, a safety penalty term related to the distance to the tunnel wall is added to the cost function, the pre-stored planned path segment is reused first when planning a new path, and the maximum value of Euclidean distance and topological distance is taken as the heuristic estimate of the hybrid heuristic function.

[0006] Furthermore, when the target points are sorted, if a new task is inserted, the insertion interval of the new task in the current sort is located, the geometric angle conditions of the target points within the insertion interval are redefined, the affected local nodes are reordered, the order of the remaining target points remains unchanged, and the original order of the unaffected parts is retained.

[0007] Furthermore, the heuristic function is constructed based on the intersection node information; an emergency path library is constructed by determining the evacuation path from each intersection node to the nearest safe exit.

[0008] Furthermore, the two-level graph structure includes a topology layer and a refinement layer; the nodes of the topology layer are intersections, and the edges are tunnel segments, and the topology layer includes information on the length, average curvature, and width of the tunnel segments; the refinement layer adaptively generates refinement nodes between the topology nodes according to the degree of tunnel curvature, resulting in a mesh graph for path search.

[0009] Furthermore, the priority weights include urgent tasks with higher weights and ordinary tasks with lower weights, and the target point is accessed within a specified time window; the geometric angle conditions include the angle between the current node and the candidate target point relative to the tunnel centerline, the angle between the current node and the candidate target point relative to the starting point, and the angle between the current node and the candidate target point relative to the temporary target node; the dynamic priority weights... for: ; in, For the first i Initial priority weights for each testing device; λ represents the warning time threshold; λ is the time urgency coefficient. This represents the remaining time.

[0010] Furthermore, the sorting of target points includes: sorting candidate target points from smallest to largest according to their comprehensive cost; the comprehensive cost is the weighted sum of distance cost, angle cost, and priority cost.

[0011] Furthermore, the dynamic adjustment of node expansion step size includes: dynamically adjusting the node expansion step size according to the tunnel curvature at the current node's location, using a larger step size when the curvature is less than a first threshold, and using a smaller step size when the curvature is greater than a second threshold; the safety penalty term monotonically increases as the distance decreases, and the penalty is infinite when the distance is less than the robot's safe radius; when planning a new path, planning is performed in the priority order of direct path reuse, prefix path reuse, sub-path reuse, and no reuse; the heuristic function is a hybrid heuristic function constructed using intersection node information, and the topological distance is the sum of the distance from the current node to the nearest intersection, the topological distance between intersections, and the distance from the intersection to the target point.

[0012] Furthermore, the aforementioned security penalty items for: ; in, This represents the distance from the node to the tunnel wall. The robot's safety radius; The desired safe distance; Penalty coefficient.

[0013] Furthermore, the topological distance for: ; in, For the set of intersection nodes; From the current node n to the intersection The Euclidean distance; Fork in the road To the target point The topological distance.

[0014] Secondly, the present invention also provides a multi-objective dynamic path planning system for tunnel environments, comprising: The data acquisition module is configured to: acquire tunnel maps; The two-level graph structure construction module is configured to: obtain a two-level graph structure including intersection nodes and tunnel section nodes based on the tunnel map; and set dynamic priority weights for each target point in the two-level graph structure based on the principle of real-time adjustment according to the urgency of the task. The target point sorting module is configured to sort target points based on the current node position, the priority weight of unvisited target points, and geometric angle conditions. The path planning module is configured to generate paths between adjacent sorted target points according to sorting and search algorithms; wherein, in the search algorithm, the node expansion step size is dynamically adjusted according to the tunnel curvature at the current node location, a safety penalty term related to the distance to the tunnel wall is added to the cost function, pre-stored planned path segments are reused first when planning new paths, and the maximum value of Euclidean distance and topological distance is taken as the heuristic estimate of the hybrid heuristic function.

[0015] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-objective dynamic path planning method for tunnel environments described in the first aspect.

[0016] Fourthly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the program to implement the steps of the multi-objective dynamic path planning method for tunnel environments described in the first aspect.

[0017] Fifthly, the present invention also provides a computer program product, the computer program product comprising a computer program, which, when executed by a processor, implements the steps of the multi-objective dynamic path planning method for tunnel environments described in the first aspect.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a two-level graph, combines curvature adaptive step-size expansion, and adjusts the priority weight of each target point in the two-level graph structure in real time according to the urgency of the task. In the search algorithm, the node expansion step-size is dynamically adjusted according to the tunnel curvature at the current node's location. A safety penalty term related to the tunnel wall distance is added to the cost function. Pre-stored planned path segments are reused first when planning new paths. The maximum value of Euclidean distance and topological distance is used as the heuristic estimate for the hybrid heuristic function. Compared with classic search algorithms, this invention reduces the number of node expansions and computation time. The wall distance safety cost is incorporated as a continuous penalty term into the cost function, achieving a quantitative trade-off between safety and path length, effectively reducing collision risk. Using the maximum value of distance and topological distance as the heuristic estimate maintains acceptability and is more informative than pure Euclidean distance, reducing invalid node expansions and adapting to tunnel environments with narrow spaces and complex structures. Attached Figure Description

[0019] The accompanying drawings, which form part of this embodiment, are used to provide a further understanding of this embodiment. The illustrative embodiments and their descriptions are used to explain this embodiment and do not constitute an improper limitation of this embodiment.

[0020] Figure 1This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of a tunnel binarized traffic map according to Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of tunnel skeleton extraction and intersection nodes in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the two-level diagram structure of Embodiment 1 of the present invention; Figure 5 The comprehensive cost sorting and initial access order of Embodiment 1 of the present invention; Figure 6 This is a schematic diagram illustrating the safety cost of wall distance in Embodiment 1 of the present invention; Figure 7 This is a schematic diagram of the path cache pool reuse mechanism in Embodiment 1 of the present invention. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0023] Example 1: With the continuous expansion of the scale of tunnel and underground engineering construction, tunnel construction robots have gradually become the core operating equipment of modern tunnel engineering, and their autonomous navigation capabilities directly affect construction efficiency and operational safety.

[0024] The tunnel's interior space is long and narrow with a complex structure, presenting a unique environment with both static and dynamic obstacles. This places stringent demands on the robot's real-time performance, safety, and multi-target adaptability in path planning.

[0025] Current mainstream path planning methods suffer from the following problems when applied to tunnel environments: The graph search-based A* algorithm employs a uniform node sampling strategy, generating numerous redundant nodes in straight tunnel sections and causing decreased path smoothness in curved sections due to insufficient resolution. Existing methods lack safety constraints on tunnel walls; the classic A* algorithm uses only path length as the optimization objective, failing to incorporate safety distance into the cost function. In scenarios with multiple targets accessing the same location consecutively, existing methods execute planning independently for each task, making it impossible to reuse already planned path segments. When new tasks are dynamically inserted, a complete replanning is required, hindering real-time response. The classic A* algorithm uses Euclidean distance as a heuristic function, severely underestimating the actual cost when crossing tunnel segments, leading to the expansion of numerous invalid nodes.

[0026] To solve the above problems, such as Figure 1As shown, this embodiment provides a multi-objective dynamic path planning method for tunnel environments. It constructs a two-level graph by extracting the tunnel topology, establishes a dynamic priority ranking mechanism, and makes four targeted improvements to the A* algorithm. This effectively solves the technical problems in existing technologies, such as computational redundancy, insufficient security, inability to adapt to dynamic task insertion, and low efficiency of heuristic functions in tunnel environment path planning. The method includes: S1. Obtain the tunnel map, extract the tunnel topology, and generate a two-level graph structure containing intersection nodes and tunnel segment nodes.

[0027] S1.1. Use a 3D laser scanner to acquire point cloud data of the tunnel environment.

[0028] A 3D laser scanner is mounted on the robot body. Its working principle involves emitting a laser beam and measuring the reflection time to calculate the distance from the scanner to the surface of the object being measured. Let the laser emission time be... t 0, the time of receiving the reflected signal is t 1. The speed of light is c The distance from the scanner to the measured point is... d satisfy d = c ·( t 1- t 0) / 2. Combined with the horizontal rotation angle recorded by the scanner's internal angle encoder. i and vertical rotation angle f It can calculate the three-dimensional coordinates of the measured point in the scanner's local coordinate system. x L , y L , z L ), specifically x L = d cos f cos i , y L = d cos f ·sin i , z L = d ·sin f In this embodiment, the 3D laser scanner has a horizontal field of view of 360°, a vertical field of view of 30°, and a scanning frequency of 10Hz, acquiring 3D coordinate data of approximately 300,000 points per frame. The 3D coordinate data of all measured points constitute a point cloud.

[0029] S1.2 Project the point cloud data onto a two-dimensional plane to generate a two-dimensional map of the tunnel.

[0030] Since construction robots move on the tunnel surface, their mobility is mainly limited in the horizontal direction. x Along the length of the tunnel y Obstacle distribution in the tunnel width direction and in the height direction. z Obstacles (axis) can be compressed by determining whether they are within the robot's travel height. Let the robot height be... H robot In this embodiment, H robot =1.5 meters. For any point in the point cloud ( x , y , z The passability flag after projection is defined as follows: if z∈[0, H robot Then the projection of that point is the obstacle point, if z [0, H robot Then this point is ignored. Specifically, the tunnel along... x and y Orientation by resolution d Discretize, d =0.1 meters. Assume the tunnel length is... L =500 meters, width W =20 meters, then the discretized grid size is N x = L / d =5000, N y = W / d =200. For each grid cell ( i , j ), i =1,..., N x , j =1,..., N y Its corresponding actual location is ( x i , y j )=(i· d ,j· d ). Count the existence of all projected points within the grid cell. If at least one projected point exists, then F ( i , j )=0 indicates an obstacle; otherwise F ( i , j A value of 1 indicates that passage is permitted. All... F (i , j Save as a grayscale image, where F =0 corresponds to a black pixel. F =1 corresponds to white pixels, resulting in a two-dimensional grayscale map of the tunnel.

[0031] S1.3. Perform binarization processing on the two-dimensional grayscale map of the tunnel.

[0032] Let the pixel value of the grayscale image be... I ( x , y The value range is 0 to 255. Set the grayscale threshold. In this embodiment, =128. For any pixel ( x , y Define the binarized passability flag. F binary ( x , y )for: ; Where 1 represents passable (white) and 0 represents an obstacle (black). Binarization converts a grayscale image into a black-and-white binary image, as shown below. Figure 2 As shown.

[0033] S1.4 Perform morphological closing operation on the binarized image. Morphological closing operation is defined as dilation followed by erosion.

[0034] The dilation operation slides the structuring element B across the image. If foreground pixels exist within the area covered by the structuring element, the corresponding position in the output image is considered foreground. The erosion operation slides the structuring element B across the image; only when all pixels within the area covered by the structuring element are foreground pixels is the corresponding position in the output image considered foreground. In this embodiment, the structuring element B is a 3×3 square kernel. The dilation operation fills in small holes in the foreground region, and the erosion operation removes isolated noise and restores areas that have been excessively dilated. The image after the closing operation yields a clean, binarized tunnel pass-through map.

[0035] S1.5 Extract the skeleton from the binarized access map.

[0036] The Zhang-Suen thinning algorithm is employed, iteratively deleting boundary pixels that meet specific conditions until a skeleton with a single pixel width is obtained. Each pixel in the binarized image is defined as either 1 (foreground, passable) or 0 (background, obstacle). The 8-neighborhood of pixel p1 is defined as follows, where p1 is the center pixel, and p2 to p9 are arranged clockwise: p9p2p3, p8p1p4, p7p6p5.

[0037] Define function A(p1) as the number of times the pixel value changes from 0 to 1 in the sequence p2 to p9. Define function B(p1) as the number of foreground pixels in p2 to p9, i.e., B(p1) = p2 + p3 + ... + p9.

[0038] Each iteration consists of two sub-loops. In the first sub-loop, delete pixels p1 that satisfy all of the following conditions: B(p1) is between 2 and 6; A(p1) = 1; p2 × p4 × p6 = 0; p4 × p6 × p8 = 0.

[0039] In the second sub-loop, delete the pixel p1 that satisfies all of the following conditions: B(p1) is between 2 and 6; A(p1) = 1; p2 × p4 × p8 = 0; p2 × p6 × p8 = 0.

[0040] Repeat the two sub-loops above until no pixels can be deleted, at which point you obtain a tunnel centerline skeleton with a width of one pixel. Each pixel in this skeleton represents a point on the tunnel centerline.

[0041] S1.6 Identify intersection nodes from the skeleton.

[0042] For any pixel p on the skeleton, define the number of skeleton pixels in its 8-neighborhood as N8(p). If N8(p) > 2, then pixel p is a fork node. Denote all fork nodes as the set J = {J1, J2, ..., Jm}. The skeleton line segment between adjacent fork nodes is defined as a tunnel segment. For any two adjacent fork nodes Ja and Jb, the tunnel segment between them is denoted as Sab. Record the length Lab, average curvature κab, and width Wab of each tunnel segment Sab.

[0043] The length Lab is calculated by multiplying the number of skeleton pixels by the pixel resolution δ, i.e., Lab = count(Sab) × δ, where count(Sab) is the number of skeleton pixels on Sab.

[0044] The average curvature κab is calculated by fitting the centerline curve of the tunnel segment. Let the sequence of centerline points of tunnel segment Sab be {( x 1, y 1), ( x 2, y 2), ..., ( xK , yK The least squares method is used to fit the quadratic curve. y = ax^ 2 +bx+c That is, minimizing the error function After solving for the coefficients a, b, and c, the curvature... mean curvature .

[0045] The width Wab is obtained by averaging the cross-sectional widths perpendicular to the centerline within the tunnel segment. For each point on the centerline ( xk , yk ), calculate the normal direction of the centerline at the point, search for the nearest black pixels (tunnel walls) on the left and right sides in the binarized map along the normal direction, record the sum of the distances on both sides as the tunnel width at the point, and take the average of the widths of all centerline points to obtain Wab.

[0046] In this embodiment, three intersection nodes are extracted and labeled J1, J2, and J3, respectively. Figure 3 The tunnel consists of four sections, labeled S1 (entrance to J1), S2 (J1 to J2), S3 (J1 to J3), and S4 (J2 to exit). Calculations show: L1 = 150 meters, κ1 = 0.02; L2 = 80 meters, κ2 = 0.15; L3 = 50 meters. k 3=0.01; L4=220 meters, k 4 = 0.03. The tunnel width Wab is between 3.5 meters and 4 meters.

[0047] S1.7 Construct a two-level graph structure, including a topology layer and a refinement layer, such as... Figure 3 and Figure 4 The topology layer constructs a graph structure with intersection nodes as vertices and tunnel sections as edges. G topo =( V topo , E topo ),in V topo =J∪{entrance, exit} E topo ={( u , v )| u , v ∈ V topo There are tunnel sections connected u and v Each topological edge ( u , v Store its attribute: length L uv Mean curvature k uv ,width W uv In this embodiment, V topo ={Entrance, J1, J2, J3, Exit} E topoThe edge (entrance, J1) corresponds to S1, the edge (J1, J2) corresponds to S2, the edge (J1, J3) corresponds to S3, and the edge (J2, exit) corresponds to S4.

[0048] The refinement layer adaptively generates refinement nodes between topology nodes based on the tunnel curvature. A curvature threshold is defined. k low =0.05 and k high =0.10. For the tunnel section S Its mean curvature is k S Refine the node generation step size for: ; In this embodiment, take =2.0 meters, =1.0 meter, =0.5 meters. For tunnel segment S, its starting coordinates P_start and ending coordinates P_end are known. The distance is calculated along the centerline of the tunnel segment in steps of 0.5 meters. Sampling generates refined nodes. Each refined node... n Store its global coordinates ( x n , y n ) and pass signs F binary ( x n , y n All refinement nodes constitute the refinement layer graph structure. G refine The refined nodes are connected according to the sampling order, and the edge weight between adjacent refined nodes is the Euclidean distance between the two points. In this embodiment, S1 is a straight segment. k 1 = 0.02 k low Using Δ s =2.0 meters, generating approximately 75 refined nodes; S2 is a bend. k 2=0.15> k high Using Δ s =0.5 meters, generating approximately 160 refined nodes; S3 is a straight segment, k 3 = 0.01 k low Using Δ s =2.0 meters, generating approximately 25 refined nodes; S4 is a straight segment, k 4 = 0.03 k low Using Δ s=2.0 meters, generating approximately 110 refined nodes. The total number of refined nodes is 370.

[0049] S2. Set dynamic priority weights for each target point and adjust them in real time according to the urgency of the task.

[0050] Let G = {G1, G2, ..., GN} be the set of monitoring points that the robot needs to visit, and N be the total number of monitoring points. In this embodiment, N = 5. Each monitoring point... G i For a specific location within the tunnel, such as a ventilation equipment monitoring point or a gas concentration monitoring point, its coordinates are ( xi , y i (Known)

[0051] Each monitoring point G i Initial priority weight The value range is [0, 1], depending on the task type. Task types are divided into three levels based on construction safety regulations and work procedures: emergency tasks (such as gas leak detection, collapse early warning) are set to... =1.0, for important tasks (such as support structure monitoring, ventilation equipment inspection) take... =0.8, for routine tasks (such as lighting equipment inspection and ambient temperature recording) =0.5. In this embodiment, the initial weights of the 5 monitoring points are set as follows: =0.5 (Ventilation equipment monitoring) =0.5 (Lighting equipment monitoring) =0.8 (gas concentration monitoring) =0.8 (support structure monitoring) =0.5 (Emergency equipment inspection).

[0052] Define time window constraints: for each monitoring point G i Must be within the specified time interval [ t i ^ start , t i ^ end Access was completed within [the specified area]. t i ^ start The earliest accessible time. t i ^ end This is the latest time that must be completed. The time window is set according to the requirements of the construction procedure; for example, gas concentration monitoring must be completed within 30 minutes after the blast. Let the current time be... t curr ,butG i Remaining time Δ t i = t i ^ end - t curr When Δ t i A value less than 0 indicates that the time window has expired, the target point is inaccessible, and an alarm needs to be triggered.

[0053] Dynamic priority weight The calculation formula is: ; in, For the first i Initial priority weights for each testing device; This is the warning time threshold, indicating how many minutes in advance to begin increasing priority. In this embodiment, we take... =10 minutes, meaning an alert is triggered when less than 10 minutes remain. λ is the time urgency coefficient, controlling the rate at which priority increases as remaining time decreases. The larger λ is, the more significant the impact of urgency on priority. In this embodiment, λ = 0.5.

[0054] max(0,( T warning -Δ t i ) / T warning ) Item: When Δ t i ≥ T warning When Δ is zero, the priority remains at its initial value; when Δ t i < T warning When the term is positive, it increases with Δ. t i Decrease and increase, the maximum value approaches 1. The min(1,·) term ensures p i No more than 1.

[0055] In this embodiment, the time window for each monitoring point is set as follows: t 3^ end =60 minutes (gas concentration monitoring must be completed within 60 minutes), t^ for other monitoring points end =120 minutes. Initial moment t curr At 0 minutes, Δ t 3 = 60 minutes ≥ T warning=10 minutes, therefore p3=p3^(0)=0.8. If the current time t curr =55 minutes, then G3's Δ t 3 = 5 minutes, substituting into the formula, we get: ; That is, when only 5 minutes remain, G The priority weight of 3 is increased to the highest value of 1.0 to ensure that the robot has priority access to the target point.

[0056] S3. Sort the target points based on the current node position, the priority weight of unvisited target points, and geometric angle conditions; when a new task is dynamically inserted, only the affected local nodes are reordered, while the original order of the unaffected parts is preserved.

[0057] S3.1, Let the robot's current position be... =( x curr , y curr The set of unvisited target points is G unvisite d For any candidate target point G i ∈ G unvisite d Calculate its overall cost The overall cost consists of three parts: distance cost, angle cost, and priority cost, calculated using the following formula: ; in, , , Let be the weighting coefficient, satisfying + + =1. The weighting coefficient reflects the importance of different cost factors in the ranking decision. In a tunnel environment, distance cost reflects path length, angle cost reflects whether the target point is in the direction of travel, and priority cost reflects the urgency of the task. This embodiment takes... =0.4, =0.3, =0.3, meaning that the distance cost has the highest weight, while the angle cost and priority cost have equal weights.

[0058] Distance Cost (P_curr, Gi) is the normalized Euclidean distance, calculated using the following formula: ; in, This represents the maximum Euclidean distance between any two points on the map. In this embodiment, the tunnel is 500 meters long, 20 meters wide, and its diagonal length is √(500^2 + 20^2) ≈ 500.4 meters. =500 meters. The normalized distance cost ranges from [0, 1].

[0059] The angle cost reflects the degree of deviation of the candidate target point relative to the tunnel centerline direction. Let the tangent direction of the tunnel centerline at the robot's current position be... v tunnel →, this direction is achieved by fitting the quadratic curve in step S1. P curr Calculation of derivative at: v tunnel →=(1,2) axe curr + b ).vector P curr G i →=( x i - x curr (y_i-y_curr). Angle α ( P curr , G i ) is defined as: ; Where "·" represents the dot product of vectors, and ||·|| represents the magnitude of the vector. Angle α The range of values ​​for is [0, π]. When α =0 indicates that the target point is located in front of the tunnel centerline. α When the angle is equal to π, it indicates that the target point is directly behind. The cost of normalizing the angle is: ; Priority cost p i This refers to the dynamic priority weight calculated in step S2, with a value range of [0, 1]. Higher priority weights result in higher priority values. p i The larger, (1- p i The smaller the value, the lower the cost. For example, if the priority weight of a certain target point is... p i =0.2, then its priority cost term is 0.3×(1-0.2)=0.24; if p i =0.9, then the priority cost is 0.3×(1-0.9)=0.03, which significantly reduces the overall cost.

[0060] Calculate the comprehensive cost of all candidate target points Then, sort the targets by cost from smallest to largest to obtain the target point visit order O = [G(1), G(2), ..., G(N)]. The target point with the lowest overall cost is visited first.

[0061] In this embodiment, the robot's current position is the tunnel entrance (0, 0), and the coordinates of each target point are as follows: G1 (120, 0), G2 (80, 0), G3 (150, 5), G4 (200, 3), G5 (50, 0). The initial priority weights are all initial values: p1=0.5, p2=0.5, p3=0.8, p4=0.8, p5=0.5. The tangent direction of the tunnel centerline at the entrance is... v tunnel →=(1,0) (tunnel along) x (Positive axis). Calculate the comprehensive cost for each target point: G5: Distance = 50 meters, Normalized distance = 50 / 500 = 0.04; Angle α =0°, =0 / 180=0; Priority cost=0.3×(1-0.5)=0.15; C=0.4×0.04+0.3×0+0.15=0.016+0+0.15=0.166.

[0062] G2: Distance = 80 meters, Normalized distance = 80 / 500 = 0.064; Angle α =0°, =0; Priority cost =0.15; C = 0.4 × 0.064 + 0.15 = 0.0256 + 0.15 = 0.1756.

[0063] G1: Distance = 120 meters, Normalized distance = 120 / 500 = 0.096; Angle α =0°, =0; Priority cost =0.15; C = 0.4 × 0.096 + 0.15 = 0.0384 + 0.15 = 0.1884.

[0064] G3: Distance = √[(150-0)^2 + (5-0)^2] = √(22500 + 25) = √22525 ≈ 150.08 meters, Normalized distance = 150.08 / 500 = 0.3002; Angle α =arctan(5 / 150)≈1.91°, =1.91 / 180≈0.0106; Priority cost =0.3×(1-0.8)=0.3×0.2=0.06; C=0.4×0.3002+0.3×0.0106+0.06=0.1201+0.0032+0.06=0.1833.

[0065] G4: Distance = √[(200-0)^2 + (3-0)^2] = √(40000 + 9) = √40009 ≈ 200.02 meters, Normalized distance = 200.02 / 500 = 0.4000; Angle α = arctan(3 / 200) ≈ 0.86°, =0.86 / 180≈0.0048; Priority cost=0.06; C=0.4×0.4000+0.3×0.0048+0.06=0.16+0.0014+0.06=0.2214.

[0066] Sort by total cost from smallest to largest: G5(0.166), G2(0.1756), G3(0.1833), G1(0.1884), G4(0.2214), resulting in the initial access order: G5, G2, G3, G1, G4.

[0067] S3.2, When a new task G new During dynamic insertion, let the current sorting order be O = [G(1), G(2), ..., G(k), ..., G(N)], and the robot has completed the previous steps. m The current node is a visit to a target point. P curr The specific method for locating the insertion interval is as follows: calculate... G new The combined cost of the remaining target points G(m+1), G(m+2), ..., G(N) is determined in ascending order of combined cost. G new Insertion position k Set the insertion position. k satisfy C ( G new Between C ( G ( k -1)) and C ( G ( k If the insertion occurs between )), the new order after insertion will be:

[0068] in, From the original order k The initial target point, After insertion The target points and their subsequent target points are shifted one position to the right, while the remaining unaffected target points (i.e., G(1) to G(m) and...) arrive The order remains unchanged.

[0069] In this embodiment, the robot has completed visits to G5 and G2, and its current location is... P curr =(82, 3). The remaining target points in the current sorting order are G3, G1, G4 (the original order was G5, G2, G3, G1, G4, the first two have been completed). The new task G6 has coordinates (95, 3) and a priority weight p6 = 1.0 (urgent task). Calculate the combined cost of G6 and the remaining target points: G6: Current distance = √[(95-82)^2+(3-3)^2] = 13 meters, normalized distance = 13 / 500 = 0.026; included angle α =0°, =0; Priority cost = 0.3 × (1 - 1.0) = 0; C(G6) = 0.4 × 0.026 + 0 = 0.0104.

[0070] G3: Current distance = Meters, normalized distance = 68.03 / 500 = 0.1361; included angle α =arctan(2 / 68)≈1.68°, =1.68 / 180≈0.0093; Priority cost = ;C(G3)=0.4×0.1361+0.3×0.0093+0.06=0.0544+0.0028+0.06=0.1172.

[0071] G1: Current distance = √[(120-82)^2+(0-3)^2] = √(38^2+(-3)^2) = √(1444+9) = √1453 ≈ 38.12 meters, normalized distance = 38.12 / 500 = 0.0762; included angle α =arctan(3 / 38)≈4.51°, =4.51 / 180≈0.0251; Priority cost =0.3×(1-0.5)=0.15; C(G1)=0.4×0.0762+0.3×0.0251+0.15=0.0305+0.0075+0.15=0.1880.

[0072] G4: Current distance = √[(200-82)^2+(3-3)^2] = 118 meters, normalized distance = 118 / 500 = 0.236; included angle α =0°, =0; Priority cost =0.15; C(G4) = 0.4 × 0.236 + 0.15 = 0.0944 + 0.15 = 0.2444.

[0073] Sorted by overall cost: G6 (0.0104), G3 (0.1172), G1 (0.1880), G4 (0.2444). Therefore, before G6 is inserted into G3, the new order is G6, G3, G1, G4. The relative order of G3, G1, and G4 remains unchanged.

[0074] S4. Between adjacent sorted target points, a micro-path is generated using an improved A* algorithm. The improvements include: dynamically adjusting the node expansion step size based on the tunnel curvature at the current node's location; adding a safety penalty term related to the distance to the tunnel wall to the cost function; establishing a path cache pool to store planned path segments, which are reused preferentially when planning new paths; and constructing a hybrid heuristic function using the information of intersection nodes, taking the maximum value of the Euclidean distance and the topological distance as the heuristic estimate.

[0075] S4.1 and A* algorithms in refining layer graphs G refine Run on. G refine The nodes are refined nodes, and the set of nodes is denoted as . V refine , | V refine |=370. G refine The edges are connections between adjacent refined nodes. For each refined node... n Its neighboring nodes are one node before and one node along the tunnel centerline, as well as nodes in the possible left and right branch directions. Edge weight Defined as the Euclidean distance between two nodes: ; The following is from the current position P curr Taking the path planning from G6=(82,3) to G6=(95,3) as an example, the four core modules of the improved A* algorithm are explained in detail.

[0076] S4.2 The specific implementation of curvature adaptive step-size expansion is as follows: The traditional A* algorithm uses a fixed step size Δs_fixed when expanding nodes, that is, the distance from all neighboring nodes to the current node is fixed. This algorithm dynamically adjusts the expansion step size Δs(P) according to the tunnel curvature κ(P) at the current node's location.

[0077] curvature The calculation is performed using the quadratic curve of the tunnel centerline fitted in step S1. Let the slope of the tangent at the current position P be...k =2 axe + b The curvature is: ; The step size Δs(P) is expanded according to the curvature. choose: ; in, =0.05 is the first curvature threshold; anything below this value is considered a straight segment. =0.10 is the second curvature threshold; values ​​higher than this are considered bends. =2.0 meters is a large step size, used for straight segments to reduce the number of nodes; =1.0 meter is a medium step size, used for gentle curves; A small step size of 0.5 meters is used for curved sections to ensure path accuracy. The current position (82, 3) is located within tunnel segment S1, which has a curvature of [missing information]. k 1 = 0.02 =0.05, determined to be a straight segment, using a large step size Δ s =2.0 meters.

[0078] When expanding a node, use the current node as the starting point. n =( x , y Using a central point as the center, calculate the coordinates of candidate nodes in eight directions. (Set of direction vectors) for: ; For direction vector d →=( dx , day )∈ The coordinates of the candidate node are: ; Each candidate node must pass a accessibility check. The corresponding coordinate in the binary accessibility map generated in step S1 is then queried. F binary ( x , y ) value, if F binary ( x , y If )=1, the node is passable and is added to the open list; if F binary ( x , y If )=0, the node is impassable (located in the tunnel wall or obstacle), and the candidate node is skipped.

[0079] After using a large step size from (82, 3) to (95, 3), the node expansion sequence is as follows: (84, 3) is expanded from (82, 3); (86, 3) is expanded from (84, 3); (88, 3) is expanded from (86, 3); (90, 3) is expanded from (88, 3); (92, 3) is expanded from (90, 3); (94, 3) is expanded from (92, 3); (95, 3) is expanded from (94, 3) (the last step has a step size of 1 meter, which is less than Δs). The path point sequence has a total of 8 nodes.

[0080] If a traditional fixed step size Δs_fixed = 0.5 meters is used, the path point sequence needs to start from (82, 3) and move 0.5 meters each step: (82, 3), (82.5, 3), (83, 3), ..., (95, 3), requiring approximately 26 nodes. The large step size strategy reduces the number of nodes by approximately 69%, significantly reducing the computational complexity of the A* algorithm.

[0081] S4.3 The specific implementation of the wall distance safety cost is as follows: add a safety penalty term related to the tunnel wall distance to the cost function of A* to ensure that the robot path maintains a safe distance from the tunnel wall.

[0082] For nodes n Its distance to the nearest tunnel wall d wall ( n ) is calculated by querying a binarized access map. The specific method is as follows: n Centered on the 8 directions (the same as the direction set in step S4) D (Same) with step size d s The search increments by 0.1 meters until a black pixel (obstacle) is encountered or the maximum search distance is reached. D search_max =5 meters. Assume the search direction. d →The distance at which the first black pixel is encountered is r d ,but d wall ( n )=min{ d ∈ D} r d .

[0083] Security penalty function Defined as: ; in, The robot's safety radius is the minimum distance from the robot's center to its outer contour. In this embodiment, we take... =0.5 meters. When d < When the robot collides with the obstacle, the penalty is infinite, and the node is not scalable. The desired safe distance is the minimum distance that the robot path is expected to maintain between the robot and the tunnel wall. In this embodiment, we take... =1.0 meters. When d≥ At this point, safety is considered to be without additional penalty. α is the penalty coefficient, controlling the intensity of the safety penalty. In this embodiment, α = 10. Figure 6 .

[0084] Improved cost function Defined as: ; in, From the starting point to the node n The cumulative distance cost; n 0, n 1, ..., n { Kn} is from the starting point to n The path node sequence; For nodes n.k. arrive n { k The Euclidean distance of +1}; p (d wall ( n k ))· Represents path segment n k → n { k A penalty of +1 is imposed, with the penalty value proportional to the path length. The A* algorithm uses g'(n) as the actual cost when expanding nodes, along with the heuristic function. Summing yields the evaluation function : ； choose Expand the smallest node.

[0085] In this embodiment, the tunnel width is 3.5 meters, and the robot's safe radius is... d min =0.5 meters. The tunnel centerline is located in the middle of the tunnel, approximately 1.75 meters from each of the two tunnel walls. If a node is located in the exact center of the tunnel, d wal l =1.75 meters ≥ d safe =1.0 meter, then p=0, no additional penalty. If a node deviates 0.4 meters from the centerline, then d_wall = 1.75 - 0.4 = 1.35 meters, still ≥ d_safe, p = 0. If a node deviates 1.0 meter from the centerline, then d wal l =0.75 meters, d min ≤0.75< d safe , p =10×(1.0-0.75)^2=10×0.0625=0.625. If a node is offset from the centerline by 1.2 meters, then... d wal l =0.55 meters, p=10×(1.0-0.55)^2=10×0.2025=2.025. If a node is offset from the centerline by 1.3 meters, then d wal l =0.45 meters< d min =0.5 meters, p =+∞, this node is not expandable. This design makes the A* algorithm tend to select nodes that are far from the tunnel wall, while prohibiting paths from crossing the tunnel wall.

[0086] S4.4 The specific implementation of path segment cache reuse is as follows: In multi-target continuous access scenarios, adjacent path segments often overlap. A path cache pool C is established to store planned path segments in key-value pair format. The key is a node pair (…). u , v ), the value is from u arrive v The path node sequence. The cache pool is initially empty and grows dynamically during the path planning process.

[0087] The priority order of path retrieval is as follows: First priority – direct matching: If the key (A, B) exists in the cache pool, return C(A, B) directly.

[0088] Second priority – prefix matching: Traverse all keys in the cache pool ( u , v ),like u =A and B are in the node sequence of C(A, v). Let C(A, v) = [A = n0, n1, n2, ..., nL = v], and there exists an index k such that n.k. If the value is B, then the prefix is ​​extracted and returned as [A=n0, n1, ..., nk=B].

[0089] Third priority - sub-path matching: Traverse all paths in the cache pool. If a path contains both A and B and A appears before B. Given a path P = [p0, p1,..., pM], there exist indices i < j such that pi = A and pj = B, then extract the sub-path and return [pi = A, p{i+1},..., pj = B].

[0090] Fourth priority - no match: Perform A* search and store the new path in the cache pool C(A, B).

[0091] In this embodiment, there already exists a path from (82, 3) to (120, 3) in the cache pool. This path was calculated and stored during the previous planning from G2(80, 3) to G1(120, 3). The path sequence from (82, 3) to (120, 3) is: (82, 3), (84, 3), (86, 3), (88, 3), (90, 3), (92, 3), (94, 3), (96, 3), (98, 3), (100, 3),..., (120, 3). This sequence passes through (95, 3) (which is between (94, 3) and (96, 3) and requires interpolation in the actual sequence). Through prefix matching, directly intercept the path prefix from (82, 3) to (95, 3) without performing A* search, saving approximately 0.3 seconds of calculation time.

[0092] S4.5. The specific implementation of the topology-aware heuristic function is as follows: The classic A* algorithm uses the Euclidean distance as the heuristic function =‖ n - g ‖, which seriously underestimates the actual cost when crossing tunnel segments. This algorithm constructs a hybrid heuristic function using the set of fork nodes J extracted in step S1: ; where is the Euclidean distance, ; is the topological distance Calculate the topological distance using fork node information The calculation formula for ; where is the Euclidean distance from the current node n to the fork j , ; is the topological distance from the fork j to the target point g , pre-computed by running the Dijkstra algorithm on the topological graph G_topo. The topological graph G topo has nodes that are the set of fork nodesJ Including the entrance and exit, the edge weight is the length of the tunnel segment. L u v The pre-calculated results are stored in a | J |×| J In the distance matrix |.

[0093] Proof of the admissibility of hybrid heuristics: for any node n Actual shortest path cost SP ( n , g )satisfy: SP ( n , g )≥ h euclid ( n This is because the Euclidean distance is the shortest geometric distance between two points (triangle inequality). SP ( n , g )≥ h euclid ( n This is because topological distance is a lower bound estimate of the path after ignoring obstacles. therefore SP ( n , g )≥max( h euclid ( n ), h topo ( n ))= h hybrid ( n The heuristic function satisfies the acceptability condition (the heuristic function is not greater than the actual cost).

[0094] In this embodiment, from (82, 3) to G6(95, 3), n and g Located within the same tunnel segment S1. Calculation h euclid =√[(95-82)^2+(3-3)^2]=13 meters. Calculate. h topo The nearest intersection is J1, located at x =150 meters away, d path ( n J1) = √[(150-82)^2 + (0-3)^2] = √(68^2 + 9) = √(4624 + 9) = √4633 ≈ 68.06 meters. dtopo (J1, g = The topological distance from J1 to G6. Since G6 is located on S1 and precedes J1, traveling from J1 to G6 requires traveling in the reverse direction along S1. d topo (J1, g = |150-95| = ​​55 meters. h topo =68.06 + 55 = 123.06 meters. max(13, 123.06) = 123.06 meters. However, in reality, it only takes 13 meters to get from (82, 3) to G6 because G6 is located between n and J1. Therefore, the topological distance is overestimated. The correct topological distance calculation should take into account... n and g The situation within the same tunnel section. Therefore, correction is needed: when n and g Within the same tunnel segment, topological distance degenerates into Euclidean distance. This algorithm adds a condition in its implementation: if... n and g If they are located in the same tunnel section, then h topo ( n )= h euclid ( n ).

[0095] like n and g Located in different tunnel sections, for example n Located at S1 (coordinates (100, 0)). g Located at S4 (coordinates (250, 0)), then the Euclidean distance is... h euclid =150 meters. Topological distance: n The distance to J1 is 50 meters, the distance from J1 to J2 is 80 meters (via S2), and the distance from J2 to... g The distance is 50 meters. h topo =50+80+50=180 meters. h hybrid =max(150, 180)=180 meters, which is closer to the actual remaining cost (the actual path needs to detour through S2, and the actual distance is about 50+80+50=180 meters), and can reduce the expansion of invalid nodes in A*.

[0096] S5. Pre-calculate the evacuation path from each intersection node to the nearest safe exit and store it in the emergency path library.

[0097] S5.1. Starting from each intersection node J∈J and ending at the set of safe exits E, refine the layer diagram. Grefine The A* algorithm is then run to calculate the shortest path. The safety exit set E contains two points: the tunnel entrance and the tunnel exit. In this embodiment, the entrance coordinates are (0, 0) and the exit coordinates are (500, 0).

[0098] For intersection J and exit e ∈ E, let the shortest path calculated by the A* algorithm be... P escape (J, e Its length is d escape (J, e) = Σ_{ k =0}^{L-1} w ( n.k. , n { k +1}), where n 0, n 1, ..., nL A sequence of path nodes. w This represents the Euclidean distance between adjacent nodes.

[0099] intersection J The optimal evacuation route is: ; That is, choose the exit with the shorter distance as the evacuation direction. and its length d escape (J) Stored in the emergency path database R middle.

[0100] In this embodiment: J1 has coordinates (150, 0), is 150 meters from the entrance, and 350 meters from the exit. This is the path from J1 to the entry point.

[0101] J2's coordinates are (230, 0), and its distance to the exit is 270 meters. J2 is located on S4, and the length of S4 from J2 to the exit is 220 meters (the actual distance from J2 to the exit should be 220 meters). The distance to the entrance requires passing through J1: J2→J1 distance is 80 meters, J1→entrance is 150 meters, totaling 230 meters. Therefore, the exit is chosen. This is the path from J2 to the exit.

[0102] J3 has coordinates (200, 0), is 200 meters from the entrance (via J1: J3→J1 distance 50 meters, J1→entrance distance 150 meters), and is 300 meters from the exit (J3→J1→J2→exit: 50+80+220=350 meters). Therefore... This is the path from J3 to the entry point.

[0103] S6. After generating the micro-path from the current position to the target point, the robot moves according to the path point sequence; during the movement, the current position is updated and dynamic priority modeling is re-executed; after reaching the target point, it is removed from the list to be visited and the subsequent path planning is continued.

[0104] The robot moves according to the path point sequence generated in step S4. (Path point sequence) P path =[p0, p1, ..., pM], where p0 is the starting point and pM is the target point. The robot's movement control uses a pure tracking algorithm: in each control cycle, the lateral and heading deviations from the robot's current position to the path points are calculated, and the linear velocity is output. v and angular velocity oh This allows the robot to move along the path.

[0105] During the movement, the robot's current position coordinates are updated every 0.5 seconds via the onboard odometer and inertial measurement unit. P curr The odometer calculates the travel distance using a wheel encoder, while the inertial measurement unit provides heading angle information; the fusion of these two provides a more accurate pose estimate. After updating the position, step S2, dynamic priority modeling, is re-executed to check if the priorities of the remaining target points have changed. Specifically, the remaining time Δ for each remaining target point is recalculated. t i = t i ^ end - t curr If Δ t i < T warning Then increase the priority weight. p i For example, if a target point's time window is about to expire, its priority weight may be increased from 0.5 to 0.8 or higher, causing a change in the sorting order.

[0106] When the robot reaches the current target point G targe t After that, i.e. P curr - G target ||< ε , ε To reach the threshold of 0.1 meters, remove the target point from the list of points to be visited. G unvisited Remove from the middle. If G unvisited If not empty, then continue path planning for the next target point according to the current sorting order O'; ifG unvisited If empty, the task is completed and the robot returns to the designated location.

[0107] S7. When an emergency is detected, locate the nearest intersection node, retrieve the corresponding evacuation route from the emergency route database, and switch to evacuation mode.

[0108] During the robot's task execution, gas sensors (such as methane and carbon monoxide sensors) deployed inside the tunnel monitor environmental data in real time. Let the sensor readings be... C sensor The safety threshold is C threshold .when C sensor > C threshold At that time, an emergency signal is triggered.

[0109] The robot immediately obtains the current node position. Calculate down to each fork node J j distance d j =‖ P curr - J j Locate the nearest intersection: ; From the emergency path database R Take out P escape ( J nearest The robot stops its current task, clears the queue for the current path, and... P escape ( J nearest Set the new movement path and switch to evacuation mode.

[0110] In evacuation mode, the robot operates at the highest safe speed. v max Move towards the exit at a speed of 2.0 m / s. During the movement, continuously monitor environmental data. If sensor readings fall below the safety threshold, choose to continue evacuation or resume the original mission (depending on safety procedures). The entire switching process is completed within 0.5 seconds of detecting an emergency signal, including signal detection, intersection location, path retrieval, and path switching.

[0111] This embodiment constructs a two-level graph by extracting the tunnel topology and combines it with curvature adaptive step-size expansion. Large step sizes are used in straight tunnel sections to reduce the number of nodes, while small step sizes are used in curved sections to ensure path accuracy. Compared to the classic A* algorithm, the number of expanded nodes is reduced by 40%-60%, and the overall computation time is reduced by more than 50%.

[0112] This embodiment incorporates the wall distance safety cost as a continuous penalty term into the A* cost function, achieving a quantitative trade-off between safety and path length. By setting a safety radius and a desired safety distance, the average distance between the path and the tunnel wall is significantly increased, effectively reducing the risk of collision.

[0113] This embodiment establishes a path segment caching and reuse mechanism. In scenarios involving continuous access to multiple targets, overlapping portions of adjacent path segments can be directly reused, avoiding redundant calculations. When dynamic tasks are inserted, only affected local nodes are reordered, while unaffected path segments are retained, achieving millisecond-level response times.

[0114] This embodiment utilizes tunnel intersection information to construct a topology-aware hybrid heuristic function, taking the maximum value of the distance and the topological distance as the heuristic estimate. This heuristic function maintains acceptability. (Actual cost), and its information content is better than pure Euclidean distance, reducing the expansion of invalid nodes.

[0115] This embodiment pre-calculates evacuation routes from each intersection to the safety exit, reducing the response time in emergencies from seconds to milliseconds, thus ensuring construction safety.

[0116] Example 2: This embodiment provides a multi-objective dynamic path planning system for tunnel environments, including: The data acquisition module is configured to: acquire tunnel maps; The two-level graph structure construction module is configured to: obtain a two-level graph structure including intersection nodes and tunnel section nodes based on the tunnel map; and set dynamic priority weights for each target point in the two-level graph structure based on the principle of real-time adjustment according to the urgency of the task. The target point sorting module is configured to sort target points based on the current node position, the priority weight of unvisited target points, and geometric angle conditions. The path planning module is configured to generate paths between adjacent sorted target points according to sorting and search algorithms; wherein, in the search algorithm, the node expansion step size is dynamically adjusted according to the tunnel curvature at the current node location, a safety penalty term related to the distance to the tunnel wall is added to the cost function, pre-stored planned path segments are reused first when planning new paths, and the maximum value of Euclidean distance and topological distance is taken as the heuristic estimate of the hybrid heuristic function.

[0117] The working method of the system is the same as that of the multi-objective dynamic path planning method for tunnel environments in Embodiment 1, and will not be repeated here.

[0118] Example 3: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the multi-objective dynamic path planning method for tunnel environments described in Embodiment 1.

[0119] Example 4: This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the program, it implements the steps of the multi-objective dynamic path planning method for tunnel environments described in Embodiment 1.

[0120] Example 5: This embodiment provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the multi-objective dynamic path planning method for tunnel environments described in Embodiment 1.

[0121] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A multi-objective dynamic path planning method for tunnel environments, characterized in that, include: Get the tunnel map; Based on the tunnel map, a two-level graph structure including intersection nodes and tunnel section nodes is obtained; Based on the principle of real-time adjustment according to the urgency of the task, dynamic priority weights are set for each target point in the two-level graph structure. The target points are sorted based on the current node position, the priority weight of unvisited target points, and geometric angle conditions; According to the sorting and search algorithm, a path is generated between adjacent sorted target points; wherein, in the search algorithm, the node expansion step size is dynamically adjusted according to the tunnel curvature of the current node location, a safety penalty term related to the distance to the tunnel wall is added to the cost function, the pre-stored planned path segment is reused first when planning a new path, and the maximum value of Euclidean distance and topological distance is taken as the heuristic estimate of the hybrid heuristic function. The dynamic adjustment of node expansion step size includes: dynamically adjusting the node expansion step size according to the tunnel curvature at the current node's location; using a larger step size when the curvature is less than a first threshold, and a smaller step size when the curvature is greater than a second threshold; the safety penalty term monotonically increases as the distance decreases, and the penalty is infinite when the distance is less than the robot's safe radius; when planning a new path, the search is performed in the priority order of direct path reuse, prefix path reuse, sub-path reuse, and no reuse; the heuristic function is a hybrid heuristic function constructed using intersection node information; the topological distance is the sum of the distance from the current node to the nearest intersection, the topological distance between intersections, and the distance from the intersection to the target point; The security penalty items for: ； in, This represents the distance from the node to the tunnel wall. The robot's safety radius; The desired safe distance; Penalty coefficient; The topological distance for: ； in, For the set of intersection nodes; To start from the current node n At the fork in the road The Euclidean distance; Fork in the road To the target point The topological distance.

2. The multi-objective dynamic path planning method for tunnel environments as described in claim 1, characterized in that, When sorting the target points, if a new task is inserted, the insertion interval of the new task in the current sort is located. Only the geometric angle conditions of the target points within the insertion interval are redefined, the affected local nodes are reordered, the order of the remaining target points remains unchanged, and the original order of the unaffected parts is retained.

3. The multi-objective dynamic path planning method for tunnel environments as described in claim 1, characterized in that, The heuristic function is constructed based on the information of the intersection nodes; an emergency path library is constructed by determining the evacuation path from each intersection node to the nearest safe exit.

4. The multi-objective dynamic path planning method for tunnel environments as described in claim 1, characterized in that, The two-level graph structure includes a topology layer and a refinement layer; the nodes of the topology layer are intersections, and the edges are tunnel segments. The topology layer includes information on the length, average curvature, and width of the tunnel segments; the refinement layer adaptively generates refinement nodes between the topology nodes according to the degree of tunnel curvature, resulting in a mesh graph for path search.

5. The multi-objective dynamic path planning method for tunnel environments as described in claim 1, characterized in that, The priority weights include urgent tasks with higher weights and ordinary tasks with lower weights, and the target point is accessed within a specified time window; the geometric angle conditions include the angle between the current node and the candidate target point relative to the tunnel centerline, the angle between the current node and the candidate target point relative to the starting point, and the angle between the current node and the candidate target point relative to the temporary target node; the dynamic priority weights... for: ； in, For the first i Initial priority weights for each testing device; λ represents the warning time threshold; λ is the time urgency coefficient. This represents the remaining time.

6. The multi-objective dynamic path planning method for tunnel environments as described in claim 1, characterized in that, The sorting of target points includes: sorting candidate target points from smallest to largest according to their comprehensive cost; the comprehensive cost is the weighted sum of distance cost, angle cost, and priority cost.

7. A multi-objective dynamic path planning system for tunnel environments, characterized in that, include: The data acquisition module is configured to: acquire tunnel maps; The two-level graph structure construction module is configured to: obtain a two-level graph structure including intersection nodes and tunnel segment nodes based on the tunnel map; Based on the principle of real-time adjustment according to the urgency of the task, dynamic priority weights are set for each target point in the two-level graph structure. The target point sorting module is configured to sort target points based on the current node position, the priority weight of unvisited target points, and geometric angle conditions. The path planning module is configured to generate paths between adjacent sorted target points according to sorting and search algorithms; wherein, in the search algorithm, the node expansion step size is dynamically adjusted according to the tunnel curvature of the current node location, a safety penalty term related to the distance to the tunnel wall is added to the cost function, the pre-stored planned path segments are reused first when planning new paths, and the maximum value of Euclidean distance and topological distance is taken as the heuristic estimate of the hybrid heuristic function. The dynamic adjustment of node expansion step size includes: dynamically adjusting the node expansion step size according to the tunnel curvature at the current node's location; using a larger step size when the curvature is less than a first threshold, and a smaller step size when the curvature is greater than a second threshold; the safety penalty term monotonically increases as the distance decreases, and the penalty is infinite when the distance is less than the robot's safe radius; when planning a new path, the search is performed in the priority order of direct path reuse, prefix path reuse, sub-path reuse, and no reuse; the heuristic function is a hybrid heuristic function constructed using intersection node information; the topological distance is the sum of the distance from the current node to the nearest intersection, the topological distance between intersections, and the distance from the intersection to the target point; The security penalty items for: ； in, This represents the distance from the node to the tunnel wall. The robot's safety radius; The desired safe distance; Penalty coefficient; The topological distance for: ； in, For the set of intersection nodes; To start from the current node n At the fork in the road The Euclidean distance; Fork in the road To the target point The topological distance.

Images