Automatic generation method for city simulation test scene of intelligent unmanned system

Abstract

The invention discloses an automatic generation method of an intelligent unmanned system city simulation test scene, and relates to the field of simulation scene construction and intelligent unmanned system test.The method comprises the steps that based on terrain boundary parameters, a plurality of rectangular areas with function labels are generated through Poisson disc sampling and a distance gradient algorithm; constructing a navigation grid representing the passable area; arranging a plurality of road intersections by adopting a constrained Poisson sampling algorithm; calculating the heart weight based on each intersection, and calculating the connection strength between the intersections through a gravity model; generating a global road network topology structure by adopting an improved minimum spanning tree algorithm; performing path search by adopting an algorithm A to obtain an optimal path, and automatically generating a road model and ancillary facilities; and extracting road network, road parameters and subsidiary facility information in the generated urban scene, and generating a standardized scene description file according to an OpenDRIVE standard format. And the construction efficiency of the simulation test scene, the test effectiveness and the cross-platform compatibility are improved.

Description

Technical Field

[0001] This invention relates to the fields of simulation scene construction and intelligent unmanned system testing technology, and in particular to an automated method for generating urban simulation test scenes for intelligent unmanned systems. Background Technology

[0002] The large-scale application of intelligent unmanned systems, such as autonomous vehicles, mobile robots, and various unmanned equipment, in typical urban scenarios relies on comprehensive, efficient, and high-coverage testing and verification. Simulation testing, capable of verifying the perception, decision-making, planning, and control capabilities of intelligent unmanned systems in a virtual environment across all elements, processes, and with high concurrency, has become a key means of evaluating their robustness, safety, and effectiveness in complex urban scenarios. Constructing a diverse, highly realistic simulation test scenario library that conforms to the realities of urban environments is a crucial foundation for expanding the capabilities of intelligent unmanned systems and enhancing their intelligence level.

[0003] However, existing methods for constructing urban simulation test scenarios for intelligent unmanned systems still have significant shortcomings, mainly in the following aspects: First, scene creation is inefficient and lacks automation. Currently, most scene generation relies on manual modeling, requiring the drawing of roads, the placement of buildings, and the setting of traffic facilities one by one. This is time-consuming, labor-intensive, and costly, making it difficult to quickly generate large-scale, diverse test scenes. Although some template-based generation methods exist, their flexibility is poor, making it difficult to adapt to different testing needs and hindering the automated batch construction of scenes.

[0004] Second, the scene layout and structure lack realism and rationality. Existing methods often fail to simulate the development patterns and spatial structures of real cities. For example, the distribution, density gradient, and topology of urban functional areas (such as commercial, residential, and industrial zones) are often generated randomly or using simple rules, resulting in unreasonable scene layouts and unnatural road connections. These methods fail to reflect typical characteristics of real cities, such as central agglomeration, functional zoning, and road network hierarchy, thus reducing the credibility and effectiveness of the test results.

[0005] Third, the parametric control and standardized output capabilities are weak. During the scene generation process, key parameters (such as road geometry parameters, lane attributes, facility layout, etc.) lack a systematic and standardized control mechanism, making it difficult to ensure compliance with industry standards (such as the "Highway Engineering Technical Standards"). Furthermore, the generated results often rely on the proprietary formats of specific simulation platforms, lacking the automatic output capability of industry-standardized scene description files such as OpenDRIVE. This results in poor reusability and compatibility of scenes across different simulation tools, increasing the fragmentation of the testing ecosystem and integration costs.

[0006] Fourth, the spatial feasibility and rationality of the generated scenes are not adequately guaranteed. During the automated generation process, spatial conflicts can easily arise between elements such as buildings, roads, and intersections (e.g., overlapping buildings and roads, excessively close intersections, and impassable paths). Existing methods often lack effective spatial constraints and collision detection mechanisms, requiring significant manual intervention for correction after generation, thus failing to achieve true "what you see is what you get."

[0007] Therefore, there is an urgent need for a method to generate urban simulation test scenarios for intelligent unmanned systems that can be fully automated, have a layout that conforms to the laws of real cities, allow flexible parameter configuration, standardize output, and ensure spatial feasibility. This method would improve the efficiency of scenario construction, the realism of testing, and the compatibility of tools, and meet the urgent needs of intelligent unmanned systems for large-scale, diverse, and high-standard test scenarios. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention provides an automated method for generating urban simulation test scenarios for intelligent unmanned systems, comprising the following steps: S1. Based on the input terrain boundary parameters, a Poisson disk sampling and distance gradient algorithm is used to generate multiple rectangular regions with functional labels; for each rectangular region, a building density parameter is set according to the functional label, spatial sampling is performed inside the region and spatial constraint verification is executed, and a three-dimensional building model is generated based on the verified sampling points and the matched building model. S2. Based on the rectangular region and the three-dimensional building model, construct a navigation grid representing the passable area; within the passable area of ​​the navigation grid, use a constrained Poisson sampling algorithm to set up multiple road intersections; S3. Calculate the centrality weight of each intersection based on its location. Based on the centrality weight and the distance between intersections, calculate the connection strength between intersections using a gravity model. Based on the connection strength, generate the global road network topology using an improved minimum spanning tree algorithm. The improved minimum spanning tree algorithm introduces a limit on the maximum number of connections between intersections. S4. Based on the navigation grid, for each pair of connecting intersections in the road network topology determined in S3, use A... The algorithm performs path search to obtain the optimal path and smooths the path point sequence to generate a three-dimensional road spline; based on the three-dimensional road spline and preset road grade parameters, it automatically generates a road model and ancillary facilities. S5. Extract the road network, road parameters, and ancillary facility information from the urban scene generated in S1 to S4, and generate a standardized scene description file according to the OpenDRIVE standard format.

[0009] Further, in S1, the generation of multiple rectangular regions with functional labels using the Poisson disk sampling and distance gradient algorithm includes: The sampling point spacing is dynamically adjusted based on a density gradient function at a preset distance from the city center; the sampling spacing in the city center area is smaller than that in the edge area; the center point of the area is generated based on the adjusted spacing, and a rectangular area of ​​corresponding size is generated with each center point; Based on the normalized distance from the center point of each rectangular area to the city center, functional labels are assigned to each area according to a preset probability distribution. The functional labels include commercial areas, residential areas, and industrial areas.

[0010] Furthermore, the spatial constraint verification includes at least: setback constraints between sampling points and region boundaries, minimum spacing constraints between sampling points, terrain slope constraints at sampling points, and conflict detection between sampling points and planned road areas.

[0011] Further, in S2, constructing a navigation grid representing a passable area includes: The terrain is divided into uniform grid units; Traverse each grid cell. If it is located within the functional area boundary or the collision body of the building model, mark it as a blocked cell. If its distance to the functional area boundary or the building model is less than or equal to a preset threshold, mark it as a high-cost cell. The remaining grid cells are marked as passable cells.

[0012] Furthermore, in S3, the improved minimum spanning tree algorithm specifically includes: (a) Initialize each intersection as an independent connected component; (b) Traverse the intersection connection pairs in descending order of connection strength; (c) For the current pair of connections, if they belong to different connected components and the current number of connections at both intersections has not reached the maximum number of connections threshold, then establish a connection between the pair of intersections and merge the corresponding connected components. (d) For the current connection pair, if they belong to the same connected component, their connection strength is within the previous predetermined percentage, and the current connection number of the two intersections has not reached the maximum connection number threshold, then a connection is established between the two intersections to form a loop.

[0013] Furthermore, the improved minimum spanning tree algorithm also includes: When the number of connections at an intersection reaches the maximum connection threshold, it is marked as a specific intersection type.

[0014] Furthermore, in S4, the automated generation of the road model includes: The three-dimensional road spline is sampled in segments, and the number of lanes, lane width, shoulder width and slope parameters are configured for each segment according to the road grade, and lane lines and traffic markings are generated.

[0015] Furthermore, in S4, the automated generation of auxiliary facilities includes: Streetlights and green plants are laid out along the three-dimensional road spline according to preset rules, and corresponding special facilities are laid out according to the type of surrounding functional area.

[0016] Furthermore, in step S5, generating a standardized scene description file according to the OpenDRIVE standard format includes: An XML file structure conforming to the OpenDRIVE specification is constructed, and road geometry data, lane attributes, intersection information, and ancillary facility data are written into the corresponding XML nodes, ultimately generating an xodr format file.

[0017] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: This invention automates the entire process from functional zoning, building layout, road network generation to standardized file export. It generates functional zones that conform to real-world urban density patterns using Poisson disk sampling and distance gradient algorithms, constructs a structurally sound road topology using a gravity model and an improved minimum spanning tree algorithm, and integrates navigation meshes with A... The algorithm ensures spatial feasibility and ultimately outputs scene files that conform to the OpenDRIVE standard. This significantly improves the efficiency of simulation test scene construction, test effectiveness, and cross-platform compatibility while ensuring the realism of scene layout, spatial rationality, and parameter standardization. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the process steps of the method for automatically generating urban simulation scenes of intelligent unmanned systems disclosed in this invention; Figure 2 This is a schematic diagram of the functional areas and building automation generation process disclosed in this invention; Figure 3 This is a schematic diagram illustrating the implementation process of road network generation based on a gravity model disclosed in this invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention aims to provide an automated method for generating urban simulation test scenarios for intelligent unmanned systems. This method features a fully automated process, layout that conforms to real-world urban patterns, flexible parameter configuration, standardized output, and guaranteed spatial feasibility. It should be noted that the implementation of this invention can rely on game engines such as Unreal Engine 5 (hereinafter referred to as UE5) or professional simulation development platforms, utilizing their provided physics engines, geometric modeling, and scripting systems. Some specific parameters involved in the embodiments (such as distance units, thresholds, and probability values) can be adjusted according to the scale and accuracy requirements of the actual simulation and should not be construed as limiting the invention.

[0022] Please see Figure 1 This document illustrates the overall process of the intelligent unmanned system's automated urban simulation scene generation method of the present invention. The method begins with inputting basic terrain data and sequentially goes through five core stages: functional area and building generation, navigation grid and intersection layout, road network topology generation, road geometry and facility modeling, and finally outputs a standardized scene file. This achieves fully automated urban scene construction from scratch, which will be described in detail below.

[0023] S1. Based on the input terrain boundary parameters, a Poisson disk sampling and distance gradient algorithm is used to generate multiple rectangular regions with functional labels. For each rectangular region, a building density parameter is set according to its functional label. Spatial sampling is performed within the region and spatial constraint verification is executed. A three-dimensional building model is generated based on the verified sampling points and the matched building model.

[0024] Those skilled in the art will understand that the goal of this step is to automatically generate multiple zones within a given terrain boundary that conform to the distribution patterns of real urban functional areas (such as dense central commerce and dispersed peripheral industries), and to generate 3D architectural models with matching style and density within each zone. The specific process is as follows: Figure 2 As shown: S11. Parameter Input and Parsing The system receives initial parameters input by the user through a graphical user interface or configuration file. These parameters include, but are not limited to: terrain boundaries, range of the number of regions, range of region sizes, and density gradient parameters. The terrain boundaries define the world coordinates of the scene area, typically a rectangular region determined by the coordinates of its lower left and upper right corners; the range of the number of regions is the expected number of functional regions to be generated, for example, 20-50; the range of region sizes is the length and width dimensions of each rectangular region, for example, (500, 1500) units; and the density gradient parameter is a coefficient used to control the density difference between the city center and the periphery.

[0025] S12. Region generation based on Poisson disk sampling and distance gradient The system uses the geometric center of the terrain as the virtual city center and performs the following sub-steps to generate the region center point and rectangular region: (1) Layered density control: The terrain is divided into a core high-density layer, an intermediate transition layer, and an edge low-density layer. A baseline sampling interval, BaseSpacing, is set.

[0026] (2) Dynamic spacing adjustment: Calculate the normalized distance d from the candidate sampling point P to the city center. norm Define a density gradient function F(d) norm ), which satisfies the expression: F(d norm =BaseSpacing (1+α d norm ), where α is the density gradient parameter. The farther away from the city center, the greater F(d) becomes. norm The larger the value, the larger the required sampling interval.

[0027] (3) Poisson disk sampling: A grid-accelerated Poisson disk sampling algorithm is used. First, an attempt is made near the city center (d...). norm Sample points are generated at smaller locations, ensuring that the distance between any two accepted points is not less than their corresponding dynamic spacing F(d). norm Sampling extends to the edge regions to ensure that the distribution of global points becomes less dense from the center to the edge.

[0028] (4) Rectangular region creation: For each successful sampling point (i.e., the region center point C) i ), randomly generate a length L from a preset area size range i and width W i With C i Centered on, L i and W i Create an axis-aligned rectangular region R for the specified size. i And record its boundary coordinates.

[0029] S13. Distance-based probabilistic functional label allocation For each generated rectangular region R i Assign a functional label, such as commercial area, residential area, or industrial area. The specific steps are as follows: (1) First, calculate the center point C of the region. i Normalized distance d to the city center i .

[0030] (2) Then according to d i The distance range into which the object falls is labeled according to a preset probability model: a. Core area (d i <0.35): High probability of being assigned to a commercial area, the rest to a residential area; b. Middle zone (0.35≤d) i ≤0.70): Primarily allocated to residential areas, with a small amount of commercial areas allowed; c. Edge area (d) i >0.70): High probability of being assigned to industrial areas, the rest to residential areas.

[0031] (3) Subsequently, for each region R i Generate a unique identifier that includes its function label and serial number.

[0032] S14. Automated generation of buildings based on spatial constraint verification For each region R with an assigned function label i To generate a building, the specific steps are as follows: (1) Set building density parameters: Set differentiated building density parameters ρ according to the function labels. For example, commercial area > residential area > industrial area.

[0033] (2) Internal raster sampling generates candidate points: in region R i Uniform rasterization sampling is performed internally to generate a set of candidate building base points {B}. j The sampling density is controlled by ρ.

[0034] (3) Spatial constraint verification: For each candidate point B j Perform multi-dimensional verification. Spatial constraint verification includes at least: setback constraints between sampling points and region boundaries, minimum spacing constraints between sampling points, terrain slope constraints at sampling points, and conflict detection between sampling points and planned road areas. Specifically, setback constraints refer to B... j Minimum distance must be maintained from the area boundary; spacing constraint refers to B. j It must maintain a minimum distance from other verified building points (this distance can be dynamically adjusted according to the target building size); slope constraint refers to B. jThe terrain slope at the location must be less than the maximum allowable value; conflict detection refers to B. j It should not fall into the space already reserved for roads (this stage can be judged based on preliminary planning or reserved buffer zones).

[0035] (4) Architectural model instantiation: For candidate points B that pass all validations j Based on the regional function tags, the system matches architectural models of the corresponding style from a pre-set architectural asset library. For example, commercial areas are matched with high-rise buildings with glass curtain walls, residential areas with multi-story residential buildings, and industrial areas with large-span factory buildings. During implementation, random scaling factors (within a certain range), height, and orientation (such as random rotation from 0 to 360 degrees) are applied to increase scene diversity.

[0036] (5) Filling in ancillary facilities: Analyze the remaining open spaces between building groups and selectively generate supporting micro-facilities such as parking lots, small green spaces, and sidewalks.

[0037] S2. Based on the rectangular region and the 3D building model, construct a navigation grid representing the passable area; within the passable area of ​​the navigation grid, use the constrained Poisson sampling algorithm to set up multiple road intersections.

[0038] Those skilled in the art will understand that the navigation mesh provides drivability constraints for subsequent path planning, and intersections are the core hubs of the road network. This step, based on UE5's Navigation Mesh and PCG modules, achieves the automated generation of the navigation mesh and intersections. This step can be further subdivided into: S21, Navigation Mesh Generation (1) Input the grid cell size and obstacle detection radius parameters.

[0039] (2) Based on the world boundary of the terrain, the entire terrain area is evenly divided into multiple regular grid units of uniform size.

[0040] (3) Traverse each grid cell and mark its accessibility based on the functional areas and building models generated in step S1: If the center point of a mesh cell is located within the functional area boundary, or intersects with the collision body range of any generated building model, it is marked as a blocking cell.

[0041] If the distance from the center point of a grid cell to the nearest functional area boundary or building is less than or equal to the preset obstacle detection radius, it is marked as a high-cost cell and assigned a higher movement cost.

[0042] Otherwise, mark the grid cell as a passable cell and set the movement cost to the normal value.

[0043] (4) Output a data structure containing all cell types, world coordinates and movement costs, i.e., the navigation grid.

[0044] S22 & S23: Intersection Layout Based on Constrained Poisson Sampling (1) Parameter setting: Preset number of target intersections N target and the minimum spacing D between intersections min .

[0045] (2) Candidate area selection: The world coordinate regions corresponding to all “passable” units in the navigation grid are used as candidate areas for intersection generation.

[0046] (3) Constrained Poisson sampling to generate intersection nodes: First, forcibly initialize the first intersection J0 in the passable area near the city center. Repeat this process until N is generated. targe At any intersection or when the maximum number of attempts is reached: randomly select a candidate point P within the candidate area. candidate Calculate P candidate 2D planar distances to all generated intersection points. If distance exists... <D min If not, then this candidate point is abandoned. Confirm P. candidate The current grid cell is in a "passable" state. If the verification passes, then in P... candidate Create a crossroads node instance (e.g., create an Actor with a sphere visual component in UE5) and add a uniform label to it, such as "Crossroad".

[0047] (4) Record all generated intersections {J k The world coordinates of}.

[0048] S3. Calculate the centrality weight of each intersection based on its location. Based on the centrality weight and the distance between intersections, calculate the connection strength between intersections using a gravity model. Based on the connection strength, use an improved minimum spanning tree algorithm to generate the global road network topology. The improved minimum spanning tree algorithm introduces a limit on the maximum number of connections between intersections.

[0049] Those skilled in the art will understand that the goal of this step is to determine how the intersections should be connected to form the structural framework of the global road network. The process is as follows: Figure 3 As shown: S31. Calculate the intersection centrality weight. Taking the geometric center of the terrain as the city center, for each intersection J k Calculate its Euclidean distance from the city center. k Find the maximum distance among all distances, dist. max Calculate the normalized distance nd k = dist k / distmax The centrality weight w is calculated using linear interpolation. k It satisfies the formula: w k = W max -(W max -W min ) nd k W max and W min Preset weights are assigned to intersections in the city center and on the outskirts. In one specific embodiment, the weight of intersections in the city center is set to 1.0, representing the highest importance; the weight of intersections on the outermost edge is set to 0.3, maintaining the lowest but non-zero importance. That is, the closer to the city center, the higher the weight.

[0050] S32. Calculate connection strength based on gravity model For any two intersections J a and J b (Their weights are w respectively) a ,w b The straight-line distance is d ab ), calculate its connection strength I ab A variant of the gravitational model formula:

[0051] Where G is the gravitational constant and γ is the distance decay exponent (usually taken as 1 or 2). It can be seen that the larger the weight and the closer the distance, the stronger the connection.

[0052] S33: Join pair filtering and sorting First, set the maximum connection distance threshold D. max and minimum gravitational threshold I min For all possible (J) a J b Yes, if d ab >D max or I ab min If the connection pair fails, it is filtered out; subsequently, all valid connection pairs that passed the filter are processed by I. ab Sort the values ​​from largest to smallest to obtain a priority list L. connections .

[0053] S34. Improved minimum spanning tree algorithm for generating topology. (1) J at each intersection k Consider it as an independent connected component C k Initialize the current connection count (count) for each intersection. k = 0. Set the maximum number of connections (MaxLinks) for a single intersection. Initialize an empty result edge set E. result .​

[0054] (2) Traverse the priority list to establish connections.

[0055] Those skilled in the art can understand that the first traversal is to ensure connectivity. Specifically, traverse the connection pairs (J connections , J a , J b ) in the order of L: If J a and J b belong to different connected components C a and C b , and count a < MaxLinks and count b < MaxLinks; then add the edge (J a , J b ) to E result , count a ++, count b ++, and merge the connected components C a and C b .

[0056] Those skilled in the art can understand that the second traversal is to increase loops and enrich the network. Specifically, traverse the connection pairs (J connections , J a , J b ) in the order of L: If J a and J b already belong to the same connected component, and count a < MaxLinks and count b < MaxLinks; then add the edge (J a , J b ) to E result to form a loop, count a ++, count b ++.

[0057] (3) During the connection process, if the count k of an intersection reaches MaxLinks, mark it as a special type (such as "XCrossRoad"). Finally, E result defines the topological structure of the global road network, that is, which intersections need to be connected by roads.

[0058] S4. Based on the navigation grid, for each pair of connected intersections in the road network topological structure determined in S3, use A The algorithm performs path search to obtain the optimal path and smooths the path point sequence to generate a 3D road spline. Based on the 3D road spline and preset road grade parameters, it automatically generates a road model and ancillary facilities.

[0059] Those skilled in the art will understand that this step transforms the topological connection into a specific three-dimensional road geometry and adds supporting facilities. Specifically, it includes the following steps: S41. Path search based on navigation grid For E result Each edge in the equation (i.e., a pair of intersections J that need to be connected) s and J e ): with J s and J e As the starting point and the ending point; on the navigation mesh generated in step S2, run A The algorithm performs pathfinding. A The algorithm's cost function is f(n) = g(n) + h(n), where g(n) is the actual cost from the starting point to node n (cumulative grid cell movement cost), and h(n) is the heuristic cost (such as Manhattan distance or Euclidean distance); A The algorithm avoids blocking cells but can traverse high-cost cells (which have even higher costs); the output originates from J. s To J e Optimal path point sequence Path {se} = {P0, P1, ..., P m}, where P0 = J s ,P m = J e .

[0060] S42, Path Smoothing and 3D Spline Generation For Path {se} To simplify the process, intermediate points with minimal angle changes are removed to reduce redundancy. For the filtered path points, a cubic B-spline curve is used for interpolation fitting to generate a smooth, continuously curvatured three-dimensional spline curve. {se} The fitting process must ensure that the curve meets the minimum turning radius constraint; the spline is resampled to ensure that the sampling points are uniformly distributed along the curve length, resulting in the final path point sequence Q used for modeling. {se} .

[0061] S43. Generated based on standard road models First, define the road's classification (e.g., arterial road, secondary road, local road). Based on the *Highway Engineering Technical Standards*, configure parameters for each classification: number of lanes, single-lane width, shoulder width, median strip (if applicable), maximum longitudinal slope, cross slope, etc. Then, follow the spline... {se}, with Q {se} Using points as a reference, and based on the configured road width, a 3D geometric model (road surface) of the road is generated through extrusion or loft operations. Finally, based on the road model, lane lines (solid / dashed) and edge lines are automatically generated. Guide arrows and lane markings are generated at a certain distance (50-100 meters) before intersections. Deceleration markings and pedestrian caution markings are generated near pre-set important facilities such as schools and hospitals.

[0062] S44, Layout of Ancillary Facilities Specifically, streetlight models are alternately deployed along both sides of the road at preset intervals (e.g., 30 meters). The height of the streetlights can be adaptively adjusted according to the road width. Street tree models are deployed within a preset bandwidth on both sides of the road using random sampling. Shrubs and other green landscaping are concentrated at intersection corners. Corresponding facilities are added based on the functional labels of the road area, such as adding billboards in commercial areas and fire hydrants and signs near industrial areas. Traffic signs (such as speed limit signs and directional signs) are automatically matched and deployed according to road function, intersection turning requirements, and speed limits. For example, the speed limit is set at 60 km / h for main roads in commercial areas, 40 km / h for secondary roads, and 30 km / h for side roads. The installation height of the signs is uniformly 2.5-3.0 meters, with an angle of 30-45 degrees to the road centerline to ensure visibility.

[0063] S5. Extract the road network, road parameters, and ancillary facility information from the urban scenes generated in S1 to S4, and generate a standardized scene description file according to the OpenDRIVE standard format.

[0064] Those skilled in the art will understand that this step converts the entire generated scene into the industry-standard OpenDRIVE format, ensuring compatibility. Specifically, it includes the following steps: S51, Extraction of Core Scene Information Traverse the entire generated scene and systematically extract the following information: (a) Road network topology: Intersection list {J k} and its connection relationship E result .

[0065] (b) Road geometry and properties: Spline corresponding to each road {se} The geometric data, length, and all parameters configured in step S43 (number of lanes, width, slope, etc.). (c) Ancillary facilities: the type, ID, precise location (coordinates), orientation, size and other attributes of all facilities such as streetlights, greenery, and traffic signs.

[0066] S52, OpenDRIVE XML file framework construction Using an XML processing library (such as Python's xml.etree.ElementTree), following the OpenDRIVE specification, create the root element of the XML file and then construct the header accordingly. <header>), Road definition <road>), intersection definition ( <junction>Lane attributes <lanes>Core node structures such as )

[0067] S53, Information Writing and File Generation First, the core information extracted in step S51 is written into the corresponding nodes of the XML file constructed in S52. For example, road geometry data is written into... <planview>and <geometry>The node writes the lane information. <lanes>The node writes the information about the ancillary facilities. <objects>Next, the written data undergoes format validation and standardization to ensure all parameters conform to the OpenDRIVE specification. Finally, a standardized scene description file with the .xodr extension is generated. This file can be directly imported and used by mainstream simulation tools for autonomous driving, robotics, and other intelligent unmanned systems that support OpenDRIVE (such as Carsim and Prescan), achieving seamless scene integration.

[0068] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.< / objects> < / lanes> < / geometry> < / planview> < / lanes> < / junction> < / road> < / header>

Claims

1. An automated method for generating urban simulation test scenarios for intelligent unmanned systems, characterized in that, Includes the following steps: S1. Based on the input terrain boundary parameters, a Poisson disk sampling and distance gradient algorithm is used to generate multiple rectangular regions with functional labels; for each rectangular region, a building density parameter is set according to the functional label, spatial sampling is performed inside the region and spatial constraint verification is executed, and a three-dimensional building model is generated based on the verified sampling points and the matched building model. S2. Based on the rectangular area and the three-dimensional building model, construct a navigation grid representing the passable area; Within the passable area of ​​the navigation grid, multiple road intersections are set up using a constrained Poisson sampling algorithm; S3. Calculate the centrality weight of each intersection based on its location. Based on the centrality weight and the distance between intersections, calculate the connection strength between intersections using a gravity model. Based on the connection strength, generate the global road network topology using an improved minimum spanning tree algorithm. The improved minimum spanning tree algorithm introduces a limit on the maximum number of connections between intersections. S4. Based on the navigation grid, for each pair of connecting intersections in the road network topology determined in S3, use A... The algorithm performs path search to obtain the optimal path and smooths the path point sequence to generate a three-dimensional road spline; based on the three-dimensional road spline and preset road grade parameters, it automatically generates a road model and ancillary facilities. S5. Extract the road network, road parameters, and ancillary facility information from the urban scene generated in S1 to S4, and generate a standardized scene description file according to the OpenDRIVE standard format.

2. The method for automatically generating urban simulation test scenarios for intelligent unmanned systems according to claim 1, characterized in that, In step S1, the generation of multiple rectangular regions with functional labels using the Poisson disk sampling and distance gradient algorithm includes: The sampling point spacing is dynamically adjusted based on a density gradient function at a preset distance from the city center; the sampling spacing in the city center area is smaller than that in the edge area; the center point of the area is generated based on the adjusted spacing, and a rectangular area of ​​corresponding size is generated with each center point; Based on the normalized distance from the center point of each rectangular area to the city center, functional labels are assigned to each area according to a preset probability distribution. The functional labels include commercial areas, residential areas, and industrial areas.

3. The method for automatically generating urban simulation test scenarios for intelligent unmanned systems according to claim 2, characterized in that, The spatial constraint verification includes at least: setback constraints between sampling points and region boundaries, minimum spacing constraints between sampling points, terrain slope constraints at sampling points, and conflict detection between sampling points and planned road areas.

4. The method for automatically generating urban simulation test scenarios for intelligent unmanned systems according to claim 1, characterized in that, In step S2, constructing a navigation grid representing a passable area includes: The terrain is divided into uniform grid units; Traverse each grid cell. If it is located within the functional area boundary or the collision body of the building model, mark it as a blocked cell. If its distance to the functional area boundary or the building model is less than or equal to a preset threshold, mark it as a high-cost cell. The remaining grid cells are marked as passable cells.

5. The method for automatically generating urban simulation test scenarios for intelligent unmanned systems according to claim 1, characterized in that, In S3, the improved minimum spanning tree algorithm specifically includes: (a) Initialize each intersection as an independent connected component; (b) Traverse the intersection connection pairs in descending order of connection strength; (c) For the current pair of connections, if they belong to different connected components and the current number of connections at both intersections has not reached the maximum number of connections threshold, then establish a connection between the pair of intersections and merge the corresponding connected components. (d) For the current connection pair, if they belong to the same connected component, their connection strength is within the previous predetermined percentage, and the current connection number of the two intersections has not reached the maximum connection number threshold, then a connection is established between the two intersections to form a loop.

6. The method for automatically generating urban simulation test scenarios for intelligent unmanned systems according to claim 5, characterized in that, The improved minimum spanning tree algorithm also includes: When the number of connections at an intersection reaches the maximum connection threshold, it is marked as a specific intersection type.

7. The method for automatically generating urban simulation test scenarios for intelligent unmanned systems according to claim 1, characterized in that, In step S4, the automatic generation of road models includes: The three-dimensional road spline is sampled in segments, and the number of lanes, lane width, shoulder width and slope parameters are configured for each segment according to the road grade, and lane lines and traffic markings are generated.

8. The method for automatically generating urban simulation test scenarios for intelligent unmanned systems according to claim 1 or 7, characterized in that, In step S4, the automated generation of auxiliary facilities includes: Streetlights and green plants are laid out along the three-dimensional road spline according to preset rules, and corresponding special facilities are laid out according to the type of surrounding functional area.

9. The method for automatically generating urban simulation test scenarios for intelligent unmanned systems according to claim 1, characterized in that, In step S5, generating a standardized scene description file according to the OpenDRIVE standard format includes: An XML file structure conforming to the OpenDRIVE specification is constructed, and road geometry data, lane attributes, intersection information, and ancillary facility data are written into the corresponding XML nodes, ultimately generating an xodr format file.

Images