A fast path planning method considering vehicle model and terrain attribute information
By using layered geographic information data and an improved RTT algorithm, combined with vehicle model and terrain attribute information, a vehicle speed map is generated, which solves the problem of insufficient comprehensiveness of path planning algorithms and achieves fast and accurate path planning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA ACAD OF LAUNCH VEHICLE TECH
- Filing Date
- 2022-11-24
- Publication Date
- 2026-05-26
AI Technical Summary
Existing path planning algorithms are not comprehensive enough in considering vehicle models and terrain attributes, resulting in low path planning accuracy and high computational complexity, making it difficult to quickly generate collision-free paths on large maps.
Using layered geographic information data and an improved RTT algorithm, vehicle speed maps are generated. Combined with vehicle model and terrain attribute information, path planning is accelerated and optimized using CUDA. Vehicle trajectories are generated using terrain, obstacle, and vector data layers.
It improves the speed and accuracy of path planning, reduces the time spent on map data retrieval, lowers computational complexity, and enables the rapid generation of near-optimal collision-free paths.
Smart Images
Figure CN116295472B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a rapid path planning method that considers vehicle model and terrain attribute information, belonging to the field of aerospace simulation. Background Technology
[0002] To create a high-level intelligent AI in terrestrial scenarios, it is necessary to provide motion design path planning algorithms for some terrestrial elements to obtain a near-optimal collision-free path from the starting point to the ending point. The following four influencing factors need to be considered:
[0003] 1) Geographic information features such as topographic elevation and topographic features (e.g., slope and aspect) should be taken into account;
[0004] 2) The algorithm should be able to perform real-time path planning, minimize the impact on the simulation speed, and the planning time should be as short as possible. The time complexity T(n) should be as small as possible, and should satisfy T(n)≦O(n2).
[0005] 3) The kinematic constraints of various land combat elements under different equipment conditions (such as the degree of damage suffered) and under different terrain features (such as the constraints on changes in movement speed) should be given priority consideration.
[0006] 4) Use geographic information three-dimensional spatial data.
[0007] To enable a vehicle to move from its starting position to its target position, a path planning algorithm is needed to find a collision-free path that satisfies both kinematic and environmental constraints. Both state-space based and sample-based path planning algorithms can be used to solve path planning problems.
[0008] The state-space-based A* algorithm can find the optimal path and is suitable for path planning on low-complexity road networks. However, due to the combinatorial explosion problem in state spaces, the computational complexity of the A* algorithm increases rapidly with the size of the task, making it unsuitable for path planning on large maps.
[0009] Sampling-based path planning algorithms sample the entire workspace, enabling them to quickly and efficiently find near-optimal paths. Probabilistic route graphs (PRM) and fast exploratory random trees (RRT), along with their derivatives, are all sampling-based path planning algorithms that can solve kinematically constrained path planning problems while satisfying kinematic constraints.
[0010] Most sampling-based path planning algorithms possess probabilistic completeness, meaning that if the problem has a solution, the probability of the algorithm finding the solution approaches 1 infinitely. If there is no reachable path from the given starting point to the destination point, sampling-based path planning algorithms will not converge.
[0011] To address the issue of excessive computational cost in A* as the problem size increases, an improved RRT algorithm can be used to achieve collision-free path planning in 3D space. In vehicle-specific software implementations, the path planning task encompasses several extended and optimized path planning algorithms.
[0012] Currently, various studies are being conducted both domestically and internationally, and path planning algorithms for land elements have been developed in e-sports games. However, the factors considered are not comprehensive enough. For example, key factors such as land slope, aspect, and various types of terrain are not taken into account, and the kinematic constraints of path planning are not adequately considered. Summary of the Invention
[0013] The purpose of this invention is to overcome the above-mentioned defects and provide a fast path planning method that considers vehicle model and terrain attribute information. This solves the technical problem that path planning algorithms do not consider all the factors, resulting in low path planning accuracy. This invention considers multiple terrain factors and can improve the speed of path planning.
[0014] To achieve the above-mentioned objectives, the present invention provides the following technical solution:
[0015] A fast path planning method considering vehicle model and terrain attribute information is proposed, characterized by the following steps: generating hierarchical geographic information data, which contains several data layers, each containing the same type of geographic information; obtaining the vehicle speed reduction data for each data layer in the hierarchical geographic information data based on the vehicle model; merging the vehicle speed reduction data from each data layer into a single data layer to obtain a vehicle speed map; and generating the vehicle trajectory using a path planning algorithm based on the vehicle speed map and the vehicle model. The hierarchical geographic information data includes an obstacle layer, a terrain layer, and a vector data layer. The system for generating layered geographic information data includes a base layer, a data layer, a service layer, an application layer, and a presentation layer. The base layer contains an open-source map engine. The service layer provides map display, map editing, and route planning services. The data layer stores vector data. The presentation layer receives externally input terrain, obstacle, and vehicle models and stores them in the data layer. The application layer retrieves the map engine and terrain layer from the base layer, as well as the map display service from the service layer, to load the terrain layer into the map engine. The application layer retrieves the map editing service from the service layer to annotate the geographic information in the terrain layer loaded into the map engine, obtaining a terrain annotation layer. This annotation layer is then integrated with the vector data layer and obstacle layer in a pyramid-like manner to obtain layered geographic information data, which is output to the presentation layer. The presentation layer displays the layered geographic information data. The geographic information in each terrain layer shares the same logic, including mountains and hills, built-up areas, rivers and lakes, deserts and Gobi, forests and farmland, or water systems and green spaces. Geographic information with the same logic results in the same speed reduction for vehicles. The vehicle model includes vehicle type, performance description, operational model, energy consumption model, sensor model, initial speed, and the influence factors or damage levels of different geographical information on vehicle speed. The path planning algorithm is the RTT algorithm, and the specific method is as follows:
[0016] S1 sets the initial point X init With target point X goal and specify the state space M;
[0017] S2 starts from the initial point X init Starting from the state space M, random sampling is performed on the sampling distance step_size and the maximum number of sampling times max_iteration. Based on the vehicle speed map and vehicle model, it is determined whether the movement conditions are met between adjacent sampling points.
[0018] If the movement condition is not met in S3, resampling is performed; if the movement condition is met, the distance between the sampling point Xr and all nodes in the already generated set of sampling points is calculated to obtain the node Xn closest to the sampling point Xr. From Xn, a new node X is generated by moving from Xn to Xr with a set step size, and the node is stored. If there are points between Xn and X that do not meet the movement condition, return to S2 for resampling.
[0019] S4 repeats S2 to S3 to generate a randomly expanded tree. When a node in the randomly expanded tree reaches the target point X... goal When nearby, find a path from the initial point X in the randomly expanded tree. init To target point X goal The path to the target point X goal The nearby standard is the distance between this node and the target point X. goal The distance is less than or equal to the preset value.
[0020] When performing random sampling, add the following judgment: there is a certain probability that the target point X will be directly selected. goal As sampling points. The state space M is defined by the initial point X. init To target point X goal The connecting line spans a square area equal to half the diagonal length. It also includes acceleration using the CUDA algorithm during the generation of vehicle trajectories based on vehicle speed maps and vehicle models. The vector data layer is converted into slope data; other vehicles in the vector data layer are treated as obstacles.
[0021] This invention provides a fast path planning method that considers vehicle model and terrain attribute information, comprising:
[0022] Generate stratified geographic information data, which contains several data layers, with each data layer containing the same type of geographic information.
[0023] Based on the vehicle model, the loss data of vehicle speed for each data layer in the geographic information layered data is obtained;
[0024] By combining the loss data of vehicle speed from each data layer into one data layer, a vehicle speed map is obtained.
[0025] Based on the vehicle speed map and vehicle model, a path planning algorithm is used to generate the vehicle trajectory.
[0026] Furthermore, the geographic information layered data includes a barrier layer, a topographic layer, and a vector data layer.
[0027] Furthermore, the system used to generate layered geographic information data includes a base layer, a data layer, a service layer, an application layer, and a presentation layer;
[0028] The base layer includes an open-source map engine;
[0029] The data layer stores the vector data layer input from the outside and the terrain and obstacle layers input from the application layer;
[0030] The service layer provides map display and map annotation services;
[0031] The application layer calls the map annotation service in the service layer to annotate the vector data layer in the data layer, generate the terrain and landform layer and the obstacle layer, and store them in the data layer;
[0032] The application layer, based on the map engine of the base layer, calls the terrain layer, obstacle layer, and vector data layer. Relying on the map display service of the service layer, the application layer displays the terrain layer, obstacle layer, and vector data layer in layers in the presentation layer.
[0033] Furthermore, for the obstacle layer, the same geographic information type means that the geographic information has the same obstacle attribute;
[0034] For the topographic and geomorphological layer, the same type of geographic information means that the geographic information has the same characteristics;
[0035] For the topographic layer, geographical information with the same characteristics includes mountains and hills, built-up areas, rivers and lakes, deserts and Gobi, forests and farmlands or water systems and green spaces; geographical information with the same characteristics will cause the same loss of vehicle speed.
[0036] Furthermore, the vehicle model includes vehicle type, performance description, operational model, energy consumption model, sensor model, or damage level.
[0037] Furthermore, the path planning algorithm is an improved RTT algorithm based on vehicle speed maps, and the specific method is as follows:
[0038] S1 sets the initial point X init With target point X goal and specify the state space M;
[0039] S2 starts from the initial point X init Starting from the state space M, random sampling is performed on the sampling distance step_size and the maximum number of sampling times max_iteration. Based on the vehicle speed map and vehicle model, it is determined whether the movement conditions are met between adjacent sampling points.
[0040] If the movement condition is not met, S3 will resample; if the movement condition is met, the distance between the sampling point Xr and all nodes in the already generated set of sampling points will be calculated to obtain the node Xn closest to the sampling point Xr. From Xn, the specific speed value in the vehicle speed map will be multiplied by the set time step to move to Xr, a new node X will be generated, and the node will be stored.
[0041] If there is a point with zero speed in the vehicle speed map on the line connecting Xn and X, then return to S2 for resampling; otherwise, generate an edge of the random extended tree.
[0042] S4 repeats S2 to S3, generating a randomly expanded tree containing several edges. When a node in the randomly expanded tree reaches the target point X... goal When nearby, find a path from the initial point X in the randomly expanded tree. init To target point X goal The path to the target point X goal The nearby standard is the distance between this node and the target point X. goal The distance is less than or equal to the preset value.
[0043] Furthermore, the following criteria are added when performing random sampling:
[0044] There is a certain probability that the target point X will be directly... goal As sampling points.
[0045] Furthermore, the state space M is defined by the initial point X. init To target point X goal The area of a square is defined by a line whose length is half the length of the diagonal.
[0046] Furthermore, it also includes using the CUDA algorithm to accelerate the process of generating vehicle trajectories based on vehicle speed maps and vehicle models.
[0047] Furthermore, the vector data layer is converted into slope and gradient data, and the slope and gradient data are used to generate vehicle speed loss data to generate a vehicle speed map.
[0048] Other vehicles are treated as obstacles to generate vehicle speed maps.
[0049] Compared with the prior art, the present invention has the following advantages:
[0050] (1) This invention reduces the time consumed by map data retrieval by merging geographic information data layers in the preprocessing, and effectively improves the efficiency of route planning while ensuring the accuracy of route planning.
[0051] (2) The present invention uses GPU acceleration optimization design to improve the computation time of path planning, thereby further improving the efficiency of path planning. Attached Figure Description
[0052] Figure 1 Flowchart of the path planning algorithm of this invention;
[0053] Figure 2 This is a flowchart of the CUDA acceleration process of the present invention. Detailed Implementation
[0054] The features and advantages of the present invention will become clearer and more apparent from the following detailed description.
[0055] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments. Although various aspects of embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless specifically indicated otherwise.
[0056] First, unlike traditional path planning methods based on simulation, this method needs to consider both geographical and topographical constraints and kinematic constraints under different damage states to obtain an optimal solution in the solution space. Second, this method has high computational speed requirements to reduce the impact on simulation speed. The smaller the time complexity T(n), the better. It should satisfy T(n) ≦ O(n2).
[0057] This invention provides a rapid path planning method that considers equipment status and terrain attribute information, comprising the following steps:
[0058] Generate layered geographic information data (including obstacle layer, topographic and geomorphological labeling layer, vector data layer, etc.);
[0059] Based on the vehicle model, the hierarchical geographic information data is converted into data affecting vehicle speed, and then merged to generate a speed map;
[0060] By combining the speed map and considering the current equipment status information, the speed map and the influence of the current equipment status on the speed are superimposed, and the vehicle trajectory is generated based on the sampling method.
[0061] During trajectory generation, the matrix operations such as Mercator projection transformation and inverse transformation in the path planning algorithm are accelerated and optimized based on the CUDA acceleration method.
[0062] Furthermore, geographic information layered data needs to include obstacle layers, topographic and geomorphic labeling layers, vector data layers, etc.
[0063] Furthermore, the vehicle model should include information such as vehicle type, performance description, operational model, energy consumption model, sensor model, and damage level.
[0064] Furthermore, the sampling-based method for generating vehicle trajectories using the RRT algorithm follows the following process:
[0065] Algorithm input:
[0066] 1) The terrain elevation data stored in the software is imported into the elevation layer generated by the software, that is, the elevation map in TIFF format is imported into the elevation layer generated by the software.
[0067] 2) Velocity map, which is a polygonal region annotation stored in the software. This annotation needs to carry the region's attribute information.
[0068] 3) Land movement element information, which refers to the land movement element objects that need to be path planned. These objects should have information such as initial movement speed, damage level, and the influence factor of different terrains on movement speed.
[0069] 4) Starting point X of land movement elements init and the expected target point X goal Latitude and longitude information.
[0070] 5) Path planning related parameters. The parameters that need to be specified are the sampling distance step_size and the maximum number of samplings max_iteration.
[0071] Calculation process:
[0072] 1) Set the initial point X init With target point X goal and specify the state space M;
[0073] 2) Randomly sample the sampling distance step_size and the maximum number of sampling times max_iteration in the state space M. Determine whether the movement conditions are met between adjacent sampling points based on the velocity map and land movement element information. If not, resample.
[0074] 3) If the movement condition is met, calculate the distance between the sampling point Xr and all nodes in the already generated sampling point set based on the elevation information data, obtain the nearest node Xn, move from Xn to Xr with a set step size, and generate a new node X. If there are points between Xn and X that do not meet the movement condition, return to 2) resampling.
[0075] 4) After repeated iterations, a random expansion tree is generated. When a node in the random expansion tree reaches the vicinity of the destination location, a path from the initial point to the target point can be found in the random expansion tree.
[0076] Algorithm output: A path consisting of the coordinates of multiple nodes.
[0077] Furthermore, considering the determination of different vehicle types, based on the vehicle element object input during initialization, the movement speed information of the vehicle element under different terrains is obtained, thereby determining whether a certain terrain can be regarded as an obstacle for the vehicle element.
[0078] Furthermore, considering the judgment of other vehicle elements, treating them as obstacles can prevent collisions between two vehicle elements. If other vehicle elements are also in the path of a vehicle element, it will seek an alternative route to avoid them.
[0079] To address the issue of slope data, the imported elevation data is converted into slope data. This helps avoid situations where steep slopes prevent vehicles from passing through, yet the path planning shows that the route is passable.
[0080] Furthermore, when performing random sampling, an additional criterion is added: there is a certain probability that the endpoint will be directly used as the sampling point. Using this method can fix the direction of path tree growth, which is beneficial to speeding up path planning.
[0081] The random sampling area is no longer the entire velocity map size, but a square area with the starting point and ending point as the diagonal length of 1 / 2. This avoids situations where the starting point and ending point are located on one side of the map, but random sampling is frequently performed on the other side, causing paths to fail to survive for a long time or generating paths with unnecessary turns.
[0082] Example:
[0083] The geographic information layered data described in this invention includes obstacle layers, terrain and landform labeling layers, vector data layers, etc. Each layer of data has the same type, and the layers are relatively independent of each other. The connection between them needs to be established based on the speed information of the vehicle model.
[0084] This invention, based on a vehicle model, converts layered geographic information data into vehicle speed information data and merges them to generate a speed map. The layered geographic information data is divided into different layers according to different types of geographic information. Different types of layers have different effects on vehicles. For example, rivers and lakes can hinder vehicle movement, while mountains, hills, deserts, and Gobi will affect vehicle speed to varying degrees. Obstacles in the obstacle layer can prevent vehicles from moving. Vector layers provide information on slope and gradient, which in turn affect vehicle speed. The speed map is generated by combining the speed reduction data of different regions into a single layer. The vehicle model must include information such as vehicle type, performance description, operational model, energy consumption model, sensor model, and damage level. Different vehicle types and attributes are also affected differently by the speed map.
[0085] This invention combines velocity maps and equipment status information to generate vehicle trajectories based on a sampling method. The motion state of any moving target can be determined based on its initial position, initial velocity, and subsequent acceleration at any time. All subsequent calculations and processing of the moving target's motion state are based on this. The idea behind RTT (Random Time To Path) is to rapidly expand a set of tree-like paths to explore most of the space and find feasible paths. The RTT algorithm is a randomized algorithm for the state space. By performing collision detection on the sampled points, it avoids the large computational burden caused by precise modeling of the space, effectively solving path planning problems in high-dimensional spaces and with complex constraints. Figure 1 The process is as follows:
[0086] Algorithm Input: Terrain elevation data stored in the software, i.e., importing TIFF format elevation maps into the elevation layer generated by the software. Velocity map, i.e., polygonal region labels stored in the software; these labels need to carry region attribute information. Land movement element information, i.e., objects of land movement element classes that need to be used for path planning; these objects should have information such as initial movement speed, damage level, and the influence factors of different terrain features on movement speed. Starting point X of the land movement element. init and the expected target point X goal Latitude and longitude information. Path planning related parameters. Parameters that need to be specified include the sampling distance `step_size` and the maximum number of samples `max_iteration`.
[0087] Calculation process: Step 1, set the initial point X init With target point X goal and specify the state space M;
[0088] The second step involves random sampling of the sampling distance step_size and the maximum number of sampling times max_iteration in the state space M. Based on the velocity map and land movement element information, it is determined whether the movement conditions between adjacent sampling points are met. If not, resampling is performed. The third step involves calculating the distance between the sampling point Xr and all nodes in the already generated set of sampling points based on the elevation information data, obtaining the nearest node Xn, and moving from Xn to Xr with a set step size to generate a new node X. If there are points between Xn and X that do not meet the movement conditions, return to step 2) for resampling. The fourth step involves generating a random expansion tree after repeated iterations. When a node in the random expansion tree reaches the vicinity of the destination location, a path from the initial point to the target point can be found in the random expansion tree.
[0089] Algorithm output: A path consisting of the coordinates of multiple nodes.
[0090] By employing the RTT algorithm for path planning based on random selection, an approximate optimal route can be obtained while satisfying kinematic constraints, thus quickly solving path planning problems with motion constraints.
[0091] 1) During trajectory generation, CUDA acceleration methods are used to accelerate and optimize matrix operations such as Mercator projection transformation and inverse transformation in the path planning algorithm. For example... Figure 2 Specifically:
[0092] The first step is to filter out targets that have already been filtered out;
[0093] The second step is to load the target location;
[0094] The third step is to run the kernel function and enter GPU mode.
[0095] Step 4: Mercator projection transformation;
[0096] Fifth step: Update the target's spatial acceleration, velocity, and position;
[0097] Step 6: Target state relocation;
[0098] Step 7: Update the target's direction and attitude;
[0099] Step 8: Update the target constraint attributes;
[0100] Step 9: Mercator projection inverse transformation;
[0101] Step 10: The kernel function finishes running and exits GPU mode.
[0102] The present invention has been described in detail above with reference to specific embodiments and exemplary examples; however, these descriptions should not be construed as limiting the present invention. Those skilled in the art will understand that various equivalent substitutions, modifications, or improvements can be made to the technical solutions and embodiments of the present invention without departing from the spirit and scope of the invention, and all such modifications and improvements fall within the scope of the present invention. The scope of protection of the present invention is defined by the appended claims.
[0103] The contents not described in detail in this specification are common knowledge to those skilled in the art.
Claims
1. A fast path planning method considering vehicle model and terrain attribute information, characterized in that, include: Generate stratified geographic information data, which contains several data layers, with each data layer containing the same type of geographic information. Based on the vehicle model, the loss data of vehicle speed for each data layer in the geographic information layered data is obtained; By combining the loss data of vehicle speed from each data layer into one data layer, a vehicle speed map is obtained. Based on the vehicle speed map and vehicle model, a path planning algorithm is used to generate the vehicle trajectory; Geographic information layered data includes a barrier layer, a topographic layer, and a vector data layer; The path planning algorithm is an improved RTT algorithm based on vehicle speed maps. The specific method is as follows: S1 sets the initial point X init With target point X goal and specify the state space M; S2 starts from the initial point X init Starting from the state space M, random sampling is performed on the sampling distance step_size and the maximum number of sampling times max_iteration. Based on the vehicle speed map and vehicle model, it is determined whether the movement conditions are met between adjacent sampling points. If the moving condition is not met, S3 will be resampled; If the movement condition is met, calculate the distance between the sampling point Xr and all nodes in the already generated set of sampling points, obtain the node Xn closest to the sampling point Xr, move from Xn to Xr by multiplying the specific speed value in the vehicle speed map by the set time step, generate a new node X, and store the node. If there is a point on the line connecting Xn and X where the vehicle speed is zero in the speed graph, then return to S2 for resampling; Otherwise, generate an edge of a randomly expanded tree; S4 repeats S2~S3 to generate a randomly expanded tree containing several edges. When a node in the randomly expanded tree reaches the target point X... goal When nearby, find a path from the initial point X in the randomly expanded tree. init To target point X goal The path to the target point X goal The nearby standard is the distance between this node and the target point X. goal The distance is less than or equal to the preset value; It also includes using the CUDA algorithm to accelerate the process of generating vehicle trajectories based on vehicle speed maps and vehicle models; The vector data layer is converted into slope and gradient data, and the slope and gradient data are used to generate vehicle speed loss data to generate a vehicle speed map. Other vehicles are treated as obstacles to generate vehicle speed maps; When performing random sampling, the following criteria should be added: There is a certain probability that the target point X will be directly... goal As sampling points; The state space M is the initial point X. init To target point X goal The area of a square is defined by a line whose length is half the length of the diagonal.
2. The fast path planning method considering vehicle model and terrain attribute information according to claim 1, characterized in that, The system used to generate layered geographic information data includes a base layer, a data layer, a service layer, an application layer, and a presentation layer. The base layer includes an open-source map engine; The data layer stores externally input vector data and terrain and obstacle layers input from the application layer; the service layer provides map display and map annotation services. The application layer calls the map annotation service in the service layer to annotate the vector data layer in the data layer, generate the terrain and landform layer and the obstacle layer, and store them in the data layer; The application layer, based on the map engine of the base layer, calls the terrain layer, obstacle layer, and vector data layer. Relying on the map display service of the service layer, the application layer displays the terrain layer, obstacle layer, and vector data layer in layers in the presentation layer.
3. The fast path planning method considering vehicle model and terrain attribute information according to claim 1, characterized in that, For the obstacle layer, geographic information of the same type means that the geographic information has the same obstacle attribute; For the topographic and geomorphological layer, the same type of geographic information means that the geographic information has the same characteristics; For the topographic layer, geographical information with the same characteristics includes mountains and hills, built-up areas, rivers and lakes, deserts and Gobi, forests and farmlands or water systems and green spaces; geographical information with the same characteristics will cause the same loss of vehicle speed.
4. The fast path planning method considering vehicle model and terrain attribute information according to claim 1, characterized in that, The vehicle model includes vehicle type, performance description, operational model, energy consumption model, sensor model, and damage level.