Intelligent path planning method in digital twin environment

Abstract

The invention relates to an intelligent path planning method in a digital twinborn environment. The method comprises the steps of determining a needed target area, obtaining satellite data of the target area and performing preprocessing, constructing a digital twinborn model of the target area, performing modeling by adopting a block modeling thought, and finally importing a deep learning model to obtain the digital twinborn model of the target area. Projecting a target area simulated by the digital twin model to a two-dimensional grid area; and projecting a real obstacle in the target area to the two-dimensional grid area, performing expansion operation on the obstacle, and if the projection of the obstacle crosses a plurality of grids, expanding the projection of the obstacle to all the crossed grids. An intelligent path planning plane algorithm A * is adopted, A * is embedded into the digital twinborn model, and a path corresponding to the minimum global cost F from the starting point to the target point is selected. Experiments prove that the path planned by the method provided by the invention is shortest in time consumption, so that the response efficiency of emergency departments for handling emergencies is improved.

Description

Technical Field

[0001] The present invention relates to intelligent path planning, and in particular to an intelligent path planning method in a digital twin environment. Background Art

[0002] In recent years, a number of emergencies have exposed the inefficiency of the emergency response mechanisms of relevant departments. If relevant departments could respond quickly and reach the emergency response site promptly, not only would the emergency be quelled quickly, but harm to the public would also be minimized. Against this backdrop, optimal path planning, a key tool for improving emergency response efficiency, has garnered widespread attention. Traditional path planning technologies, which mostly rely on flat maps, have significant limitations when dealing with complex and changing real-world situations. Therefore, more intelligent and efficient solutions are urgently needed.

[0003] Digital twin technology, one of the most anticipated emerging technologies in recent years, transforms real-world physical entities into digital models, offering a new approach to problem analysis, optimization, and decision-making during real-world missions. In the field of route planning, the application of digital twin technology is expected to enable personalized and precise route planning through simulation of real-world scenarios and real-time feedback, thereby better adapting to the needs of various emergency situations.

[0004] Furthermore, current emergency route selection methods for conflict management by relevant departments are limited, primarily relying on charts and maps. However, route selection has a crucial impact on the emergency safety management system of relevant units. Traditional emergency route selection methods, limited to emergency route diagrams and emergency drills, are ineffective and lack practical application value, making them ineffective for route planning in emergencies. Summary of the Invention

[0005] In view of the above problems existing in the prior art, the technical problem to be solved by the present invention is: how to plan a path with practical application value in response to emergency situations.

[0006] To solve the above technical problems, the present invention adopts the following technical solution: an intelligent path planning method in a digital twin environment, comprising the following steps:

[0007] S1: Determine the target area and obtain satellite data of the target area.

[0008] S2: Preprocess the satellite data to remove noise from the satellite data, and then perform cleaning, noise reduction and unification preprocessing steps on the satellite data. In view of the missing buildings in the data acquisition process, rasterization and vector data processing are used to extract linear data to make the representation complete.

[0009] S3: Build a digital twin model of the target area. Use a block modeling approach to develop the model. Save each block model and convert it into an FBX file. Then integrate and process it in Unreal Engine. Finally, import the deep learning model to obtain the digital twin model of the target area.

[0010] S4: Project the target area simulated by the digital twin model onto a two-dimensional grid. Each grid cell represents a real-world distance of one meter. Black grid areas represent blocked spaces, while white areas represent accessible areas. Actual obstacles in the target area are projected onto the two-dimensional grid and expanded. If the obstacle projection spans multiple grid cells, the obstacle projection is expanded to all the grid cells it spans.

[0011] S5: Use the intelligent path planning planar algorithm A*, embed A* into the digital twin model, and select the route corresponding to the minimum global cost F from the starting point to the target point.

[0012] Preferably, in the process of block modeling, when the S3 does not fit the actual longitude and latitude of the satellite image, in order to import the twin model into the twin scene, the actual data alignment digital twin model method is adopted; the satellite image map is imported into QGIS; the base project is adjusted to the projection coordinate WGS84-32650; in QGIS, the local satellite image map of the twin model is compared with the satellite image map, and three diagonal points are set to align the projection coordinates; then it is converted into FBX format and imported into the C4D modeling software; the twin model is fitted with the local satellite image map of the twin model, and the X, Y, and Z coordinates are adjusted in the C4D modeling software.

[0013] Preferably, the calculation process of the global cost F in S5 is as follows: F=G+H, where G represents the total number of grids required to move from the starting point a to the target point grid along the path of the search grid, and H represents the estimated cost of moving from a node in any direction in the grid to the target point grid, calculated by Euclidean distance or Manhattan distance.

[0014] Compared with the prior art, the present invention has at least the following advantages:

[0015] The method of the present invention performs digital twin modeling on the target area, projects the target area simulated by the digital twin model into a two-dimensional grid area, then projects the obstacles into the two-dimensional grid area and expands it, and finally uses a global optimization algorithm to find the minimum cost route from the starting point to the target position.

[0016] Furthermore, by modeling actual application scenarios, the feasibility of embedding intelligent path planning algorithms within a digital twin environment was verified. The experiment used both local and global optimal algorithms to plan and explore road options, enabling the agent to proactively find its path. This method demonstrated that the path planned by this method minimized time, thereby improving the efficiency of emergency response departments in handling emergencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Exploration area for the "A*" star algorithm.

[0018] Figure 2 Pathfinding diagram for the "A*" star.

[0019] Figure 3 These are twin scene pictures of Jintang Prison.

[0020] Figure 4 Model a panoramic view of the Golden Hall.

[0021] Figure 5 This is a first-person perspective image of handling tasks in motion.

[0022] Figure 6 A multi-point pathfinding map.

[0023] Figure 7 Path graph for Dijkstra algorithm pathfinding.

[0024] Figure 8 This is a time comparison chart between the A* algorithm and the Dijkstra algorithm.

[0025] Figure 9 Obstacle expansion processing diagram. DETAILED DESCRIPTION

[0026] The present invention is described in further detail below.

[0027] An intelligent path planning method in a digital twin environment includes the following steps:

[0028] S1: Determine the target area and obtain satellite data of the target area.

[0029] S2: Preprocess the satellite data to remove noise from the satellite data, and then perform cleaning, noise reduction and unification preprocessing steps on the satellite data. In view of the missing buildings in the data acquisition process, rasterization and vector data processing are used to extract linear data to make the representation complete.

[0030] S3: Build a digital twin model of the target area using a block-based modeling approach. [Model the target area's buildings, lawns, trees, roads, sidewalks, and other areas in a block-based manner.] Each block model is saved and converted into an FBX file. This file is then integrated and processed in Unreal Engine. Finally, it is imported into a deep learning model to create a digital twin model of the target area.

[0031] S4: Project the target area simulated by the digital twin model onto a two-dimensional grid. Each grid cell represents a real-world distance of one meter. Black grid areas represent blocked spaces, while white areas represent accessible areas. Actual obstacles in the target area are projected onto the two-dimensional grid and expanded. If the obstacle projection spans multiple grid cells, the obstacle projection is expanded to all the grid cells it spans.

[0032] like Figure 9 Display situation. Figure 9 (a) depicts the original form of the obstacle in the real scene. Figure 9 (b) shows the changes in two-dimensional space after the grid is expanded. Each grid cell represents a distance of one meter in the real world. The black grid area represents blocked space, that is, the area that cannot be reached; the white area represents the accessible area, that is, the range that can be reached. The grid coordinate values ​​will gradually increase from left to right and from bottom to top. Figure 9 It shows a 20x20 grid environment map of a real scene after obstacle expansion. The lowest grid point is at (1,1) and the highest grid point is at (20,20).

[0033] S5: Use the intelligent path planning planar algorithm A*, embed A* into the digital twin model, and select the route corresponding to the minimum global cost F from the starting point to the target point.

[0034] The A* algorithm can be understood as combining breadth-first search with heuristic search. It begins at the starting position of a grid path graph, performing a breadth-first search followed by a heuristic search using an evaluation function. After a round of searching the graph, it evaluates each position traversed in that round, selects the node with the lowest cost and best position, and then begins a new round of heuristic search from this best node until the target position is reached. The search process of embedding the A* algorithm in a twin scenario mainly includes:

[0035] (1) Search area:

[0036] Set up a single exploration grid, editable in blueprints as a square or hexagon, for easier calculation and observation.

[0037] Now edit it into a square grid. Suppose the agent intends to move from point a to another point b next door, such as Figure 1 As shown, the white Agent's current position represents the starting point a, the fire effect represents the ending point b, and the red square represents the wall in the middle.

[0038] It is known that one step in marching is about 75 centimeters, and the stride of running is slightly larger than that of marching. Here, in order to be close to the real physical world, in the digital twin environment, the exploration area is first divided into a grid map with 100 cm wide and 100 cm connected square grids.

[0039] refer to Figure 1 As can be seen, the area is composed of 100-centimeter squares. This refinement of the search range reduces computational steps and improves route finding speed as a first step. This strategy allows the exploration process to compress all possible path points into a one-dimensional matrix. Each matrix element represents a small cell within the grid, which is color-coded to indicate whether it is unobstructed or congested. A specific route is converted into a list of grid nodes passing through from point A to point B. Only after finding the optimal route will the agent begin moving from the center of the previous grid to the next grid, stopping only when it reaches the destination.

[0040] (2) Start the search:

[0041] After the search area is converted into a grid area, that is, the pathfinding range is transformed into an easily explorable node area, the exploration process of finding the shortest path begins. In Unreal Engine, you can use its own ray detection box. Starting from the starting point O, you can calculate the numerical value required for the ray from the current box to the surrounding adjacent boxes. The distance calculation formula is used to calculate the numerical value required to move from the exploration grid to the end point. The calculated data is stored in the distance variable and repeated in sequence until the target is found.

[0042] To perform a search in a twin environment, perform the following steps:

[0043] (1) First, start at the starting agent's location and place it in the "pending variable." This "pending variable" is like a virtual shopping basket. Although it currently contains only one box, the number of boxes will gradually increase as the exploration area expands. The path search process may pass through any node in the list, or it may completely avoid them. In general, this is actually a collection of "pending grids."

[0044] (2) Study all feasible paths and accessible areas around the starting point, avoiding touching terrain grids that cannot be crossed, such as rocks and buildings. At the same time, they are added to the list to be processed. The starting position O is the "parent grid" of these points stored in the grid. When we want to describe the path, the information of the parent grid is very important. [i] .

[0045] (3) Remove coordinate O from the pending list and add it to the "processed list", which contains all nodes that do not need to be explored again.

[0046] The grid to be processed is presented in the twin environment as Figure 2 The square with a value of 000 means that the square has been included in the list of solved problems. To facilitate observation of the data, each calculated value is presented above each small square using a numerical variable.

[0047] Next, select the grid that is approximately searchable for the node grid in the start-up list and repeat the previous steps. Finally, select the grid with the lowest global cost f value (the grid with the smallest value), such as Figure 2 shown.

[0048] (3) Path scoring:

[0049] The key to choosing which node square a path passes through is the following formula: F = G + H. G is the total number of squares required to reach the target square along the path from starting point a. H is the estimated cost of moving from any node in the grid to the final location. This can be calculated using either Euclidean distance or Manhattan distance; Manhattan distance is used here.

[0050] Since there may be various obstacles on the road (such as walls, buildings, plants, etc.), there is no way to know the length of the path in advance. When choosing a path to explore, you need to add a Boolean variable to determine whether the exploration grid is passable.

[0051] The optimal route search process involves iterating through the list of problems to solve and selecting the smallest unit block with the lowest global cost F to determine the selected trajectory. G, the distance between the starting point and the current position, measures the cost of the path. To reduce computational complexity, the diagonal cost is set to a fixed value.

[0052]

[0053] After the calculation is completed, add a light module to the map to display the route through color contrast, and search for the square with the minimum cost in reverse order from the target end square. Figure 2 shown

[0054] Specifically, during the block modeling process, when the S3 does not fit the actual longitude and latitude of the satellite image, in order to import the twin model into the twin scene, the actual data alignment digital twin model method is adopted; the satellite image map is imported into QGIS; the base project is adjusted to the projection coordinate WGS84-32650; in QGIS, the local satellite image map of the twin model is compared with the satellite image map, three diagonal points are set, and obvious buildings or landmark buildings are selected to align the projection coordinates; then converted into FBX format and imported into C4D modeling software; the twin model is fitted with the local satellite image map of the twin model, and the X, Y, and Z coordinates are adjusted in the C4D modeling software.

[0055] Specifically, the calculation process of the global cost F in S5 is as follows: F = G + H, where G represents the total number of grids required to move from the starting point a to the target point grid along the path of the search grid, and H represents the estimated cost of moving from a node in any direction in the grid to the target point grid, calculated by Euclidean distance or Manhattan distance.

[0056] Experimental verification

[0057] 1. Path Exploration in Twin Scenario

[0058] (1) Jintang Prison Model Twin Scene

[0059] like Figure 3 The Jintang twin scene simulated by Unreal Engine

[0060] In the scene built by Unreal Engine, in order to make the target eye-catching, the incident site is displayed in the form of flames. How to realize the visualization of the path algorithm? Generally speaking, Unreal Engine can adopt the blueprint method to code and connect lines based on C language or Python. In the process, the path is found by interactively moving the event site and changing the exploration target position. In the twin environment, the construction of the path finding exploration area is as follows: Figure 4 The red area is shown.

[0061] (2) First-person perspective pathfinding application

[0062] In order to facilitate the officers and soldiers involved in the disposal to deal with the changes in the environment, in this design, the perspective change is used to observe the environment from the first perspective of the officers and soldiers involved in the task, such as Figure 5 shown.

[0063] The route shown by the blue light along the way is the route explored by the A* algorithm, and the task personnel handle it from the first-person perspective.

[0064] (3) Multi-point pathfinding in the third perspective in twin scenes

[0065] like Figure 6 As shown, it is the optimal route map for the three participating forces to reach their destination from the third perspective (staff perspective). The three forces explore and find their way from point A of the gate, point B of the wall, and point C in the middle of the road respectively. The three blue glowing paths along the way are the results of the path finding.

[0066] 2. Comparison between A* algorithm and Dijkstra algorithm in twin environment

[0067] In order to verify the gap between the global optimal algorithm A* algorithm and the traditional local optimal algorithm, after the experiment, the Dijkstra algorithm was used to compare with the A* algorithm. In the environment built by Unreal Engine, Figure 7 As shown in the figure, this is the path explored by the Dijkstra algorithm. It can be seen from the figure that for the red obstacle, the local optimal algorithm crosses it and does not effectively avoid it. However, it can be seen from the above process that the global optimal A* algorithm can effectively avoid the obstacle.

[0068] The Dijkstra algorithm is applied to the simulated scene, through experiments, such as Figure 8 As shown, it is not difficult to find that it has the characteristics of small calculation range, low calculation cost, and fast path obtaining. It can quickly give the path finding result. For open and obstacle-free environments, the Dijkstra algorithm can be used for fast path finding.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent path planning method in a digital twin environment, characterized by: The steps include: S1: Determine the target area and obtain satellite data of the target area; S2: Preprocess the satellite data to remove noise from the satellite data. Then perform cleaning, noise reduction, and unification preprocessing steps on the satellite data. In order to solve the missing buildings in the data acquisition process, rasterization and vector data processing are used to extract linear data to make the representation complete. S3: Build a digital twin model of the target area. Use a block-based modeling approach to develop the model. Save each block model and convert it into an FBX file. Then integrate and process it in Unreal Engine. Finally, import the deep learning model to obtain the digital twin model of the target area. S4: The target area simulated by the digital twin model is projected onto a two-dimensional grid area. Each grid cell represents a one-meter distance in the real world, and the black grid area represents the blocked space. The white part is the passable area; The real obstacles in the target area are projected onto the two-dimensional grid area and the obstacles are expanded. If the obstacle projection spans multiple grids, the obstacle projection is expanded to all the grids it spans. S5: Use the intelligent path planning planar algorithm A*, embed A* into the digital twin model, and select the route corresponding to the minimum global cost F from the starting point to the target point.

2. The intelligent path planning method in a digital twin environment according to claim 1, characterized in that: In the process of block modeling, when the latitude and longitude of S3 do not match the actual satellite image, in order to import the twin model into the twin scene, the actual data alignment digital twin model method is adopted; Import satellite images into QGIS; The base project is adjusted to the projection coordinates WGS84-32650; In QGIS, compare the twin model local satellite image with the satellite image, set three diagonal points, and align the projection coordinates; Then convert it into FBX format and import it into C4D modeling software; Fit the twin model to the local satellite image of the twin model, and adjust the X, Y, and Z coordinates in the C4D modeling software.

3. The intelligent path planning method in a digital twin environment according to claim 2, characterized in that: The calculation process of the global cost F in S5 is as follows: F = G + H Among them, G represents the total number of grids required to move from the starting point a to the target point grid along the path of the search grid, and H represents the estimated cost of moving from a node in any direction in the grid to the target point grid, calculated by Euclidean distance or Manhattan distance.

Images