A path planning system for complex land environments based on three-dimensional voxel grid

Abstract

This invention discloses a path planning system for complex land environments based on three-dimensional voxel lattices. It is characterized by comprising a land complex environment mobility assessment and prediction module, a three-dimensional voxel lattice establishment and optimization module, and a path planning module based on the three-dimensional voxel lattice. The technical solution disclosed in this invention mainly addresses the problem of vehicle-ground mechanical interaction through regional mobility assessment and prediction, obtaining an assessment of the vehicle's mobility in regional maneuvering after considering factors such as terrain, landforms, soil, soil quality, and vehicle dynamics; then, three-dimensional voxel data for land path planning is formed through the three-dimensional voxel lattice representation of multi-dimensional environmental factors; finally, the path planning task in complex environments is realized based on the three-dimensional voxel lattice.

Description

Technical Field

[0001] This invention relates to a novel path planning technology. Background Technology

[0002] Commercial navigation software is already quite mature, such as Baidu Maps and Gaode Maps. However, these commercial navigation applications are primarily geared towards civilian use, relying heavily on road network data and basic traffic information to support user navigation needs. They cannot address the navigation challenges of vehicles in complex off-road environments. Currently, domestic research on off-road vehicle navigation largely focuses on evaluating vehicle maneuverability and obstacle avoidance algorithms, with less consideration given to vehicle path planning and navigation capabilities in complex, variable, and roadless off-road environments.

[0003] The demand for navigation in special environments and scenarios is gradually increasing. Whether it's the NATO Reference Mobility Model (NRMM) developed and validated by NATO in the 1960s and 70s, or the Next-Generation NRMM (NG-NRMM) developed by NATO's AVT-248 committee in 2016, these are all multifunctional path planning technologies developed for special environments and complex scenarios. my country is also gradually recognizing the importance of this area and is investing heavily in the development of this new navigation technology.

[0004] In terms of off-road mobility performance evaluation, considering that one-third of my country's land area is plateau and mountainous, and some special plateau and mountainous areas have complex terrain, the Beidou Navigation Innovation Research Institute of Shanghai Jiao Tong University has been working in the field of Beidou navigation for a long time and has established a speed calculation model for off-road vehicles under different road conditions. This calculation model contains vehicle structural parameters, dynamic parameters and road condition parameters. Southeast University, in cooperation with the General Vehicle Department, has conducted research on the off-road vehicle's ability to pass over ridges through multibody dynamics modeling and simulation. Ning Junshuai et al. of the Armored Forces Engineering Academy proposed a comprehensive evaluation process based on the "mission model" and "driving model" methods.

[0005] In terms of path planning under off-road conditions, domestic scholars' research is relatively weak, and is relatively focused on obstacle avoidance algorithms and improving shortest path planning algorithms applicable to road conditions to meet the needs of path planning under off-road conditions. For example, Tian Xiping et al. optimized the distance and direction of the classic Dijkstra algorithm, and proposed an obstacle avoidance algorithm based on direction for the analysis range of off-road maneuver routes; Wang Qisheng et al. proposed to use an optimized maze algorithm for shortest path analysis for off-road path planning, and proposed an optimization algorithm based on the path obtained by the maze algorithm by using the idea of ​​substituting curves with straight lines and spline models; Wu Tianyi et al. analyzed the path planning problem under the combined influence of terrain slope and surface attributes for off-road path planning problems of different types of vehicles, constructed an evaluation function for the improved A* algorithm and an evaluation function for the ant colony algorithm, and designed a path optimization algorithm that considers slope and roughness constraints; Fan Linlin et al. comprehensively considered the influence of terrain elements and landforms on the maneuverability of off-road vehicles, and based on this, proposed an optimization scheme for the heuristic A* algorithm. First, the complex terrain environment is expressed by terrain quantification, and then the evaluation function of the A* algorithm is optimized to explore the shortest path planning problem under the premise that the starting point and the ending point are determined.

[0006] Yan Xingyu et al. addressed the need for large-scale off-road path planning in complex off-road environments by researching and designing a hierarchical off-road path planning method based on accessibility analysis. This method focuses on the efficiency and feasibility of large-scale path planning, proposing a hierarchical planning approach. It achieves hierarchical path planning by constructing two map grids with different resolutions and combines surface cover and surface roughness for regional accessibility analysis, specifically constructing an A* algorithm evaluation function. Tian Hongqing et al. proposed a probabilistic graphical model (AFP-PRM) algorithm based on a potential energy field model. This algorithm uses an artificial potential energy field to model the off-road environment, assess vehicle access risks, and uses a probabilistic graphical algorithm to optimize multi-dimensional access costs between nodes for path planning. Considering vehicle dynamics, a dynamic curvature smoothing method is used to optimize the vehicle trajectory. Li Kunwei et al. analyzed the elements of path planning in dynamic environments and proposed a bidirectional path planning solution. Addressing the specific characteristics of off-road path planning in dynamic environments, they set the search method and search step size for bidirectional path planning, constructed a cost estimation function, and finally proposed a bidirectional A* search algorithm suitable for dynamic path planning. Liang Xiaohui et al. from Beijing University of Aeronautics and Astronautics applied genetic algorithms for trajectory planning, but the algorithm converged very slowly, and the generated paths did not fully consider terrain information such as rivers, resulting in paths that did not meet the requirements of real-world complex off-road environments. Chen Qijie categorized existing path calculation methods into traditional methods and intelligent methods.

[0007] Ji Qing summarized the corresponding path calculation methods based on different modeling approaches. The most widely used classification method is based on the degree of environmental information mastery, dividing them into global path calculation methods and local path calculation methods. Global path calculation methods, when the environmental map information is completely known, involve preprocessing and modeling, then selecting an appropriate path calculation method. Its efficiency depends entirely on the algorithm selection, balancing optimality and reachability; however, this method, from a global perspective, may not consider details. Local path calculation methods deal with unknown environmental information, so modeling and searching occur simultaneously. This method is often used in mobile robots and unmanned vehicles with autonomous navigation information. It requires real-time processing of environmental information, using rapid computing power to respond promptly to various emergencies. However, the paths searched based on local information may not be very accurate, or even fail to find a path. The two models are essentially the same, but local path calculation is more complex than the global model. With improvements, the two methods can be converted to each other. Wei Guanwei et al. established a penalty function neural network to quantify collisions between paths and obstacles, providing path guidance for robots. Neural network algorithms involve numerous parameters, but currently, there is no clear theoretical basis for the design of network models; they remain in a black box state. Duguleana and Mogan et al. proposed combining neural networks and reinforcement learning to improve the ability of moving bodies such as mobile robots to handle static and dynamic obstacles. In the widely used deep reinforcement learning method, neural networks can handle pathfinding tasks well on their own. Research shows that moving bodies using deep Q-learning algorithms can robustly learn to achieve goals on maps and exhibit good performance on maps that the moving bodies have never encountered before.

[0008] my country has seen a significant increase in navigation needs in complex, special environments outside of civilian applications, and work is underway in this area. This invention addresses these needs by developing and designing a path planning technology based on a 3D voxel lattice for complex land environments. This path planning technology can generate a multi-attribute, integrated 3D voxel lattice and support path planning capabilities in complex land environments.

[0009] The main technical problems in off-road vehicle navigation in China can be summarized as follows:

[0010] (1) Limited Element Assessment. Mobility assessment relies mainly on commander experience, lacking sufficient quantitative decision-making support and considering only a few factors, such as terrain, without taking into account the combined effects of weather, soil quality, natural disasters, and emergencies. The increasing variety and functions of ground vehicles, as well as the complex environment with diverse geological types and geometrical complexity, bring high uncertainty to vehicle model parameters and performance assessments, necessitating adaptive and extended improvements to the system's mechanical model.

[0011] (2) Insufficient consideration of real-time situational information. Current research mainly focuses on static information such as roads and terrain, without comprehensively considering the impact of real-time situational information such as weather, meteorology, and emergencies on off-road navigation.

[0012] (3) Insufficient modeling and prediction capabilities. The algorithm and model optimization are mainly based on basic geographic information systems (GIS). There is a lack of cross-domain joint modeling and analysis capabilities, such as vehicle dynamics, vehicle-soil interaction mechanics modeling, and the ability to estimate maneuver speed, fuel consumption, and travel time in off-road environments.

[0013] (4) The algorithms in current path planning and navigation software are not very practical. They are mainly based on existing road network data to perform simple shortest path analysis, which is far from meeting practical needs. They do not adequately consider off-road mobility and complex off-road environments, such as rain, snow, ice, strong winds, fog and the resulting soil moisture and slippage. Summary of the Invention

[0014] The technical problem this invention aims to solve is that my country's research on vehicle mobility performance evaluation and navigation in complex off-road environments started relatively late. While there is a certain theoretical research foundation, the focus and systematic nature are weak. Most of the work is based on foreign simulation software and platforms for algorithm verification. There is a lack of comprehensive mobility evaluation products, insufficient practical testing, and the level of independent control needs to be improved. The acquisition of ground mechanical parameters is still at the follower level, which restricts the progress of my country's off-road mobility assessment technology.

[0015] To address the aforementioned technical problems, the present invention discloses a path planning system for complex land environments based on a three-dimensional voxel lattice. The system is characterized by comprising a mobility assessment and prediction module for complex land environments, a three-dimensional voxel lattice establishment and optimization module, and a path planning module based on the three-dimensional voxel lattice, wherein:

[0016] The Land Complex Environment Mobility Assessment and Prediction Module further includes:

[0017] The intelligent obstacle recognition unit for complex land environments constructs an intelligent obstacle recognition model based on scene scale and DCNN transfer learning, which is used to quickly identify obstacles in complex land environments.

[0018] The ground mechanical parameter interpretation unit for complex land environments provides a data foundation for realizing coupled analysis of vehicle and ground mechanics, classifies and identifies soil, obtains soil moisture, and then constructs ground mechanical parameters;

[0019] A precise inversion unit for ground mechanical parameters of coupled vehicles is established to create a knowledge base of the correspondence between known soil types and mechanical properties. This allows for the acquisition of a coupled quantitative calculation method that satisfies the ground mechanical property parameters of different vehicles and standard classifications, thereby enabling the inversion of ground mechanical property parameters in unknown areas and the assessment of vehicle traffic performance.

[0020] The vehicle dynamics model based on vehicle-ground interaction is determined by the type of vehicle and ground. The vehicle type is determined by the running gear. Based on the interaction mechanism between different running gears and the ground, a multi-degree-of-freedom mechanical model of the vehicle in a three-dimensional unstructured environment is established to provide a model basis for the evaluation of vehicle passability and maneuverability.

[0021] A vehicle passability assessment unit based on multivariate analysis is used to assess a vehicle's ability to pass through a designated area.

[0022] A speed prediction unit combining vehicle performance parameters and dynamic models is used to evaluate vehicle operating speed. Furthermore, for vehicle operating safety, a deep learning-based driving stability classification is obtained by combining vehicle test data, ground comprehensive characterization model and vehicle speed. At the same time, combining vehicle stability and speed, a rapid maneuvering rule study is carried out to form a rapid maneuvering area evaluation model to obtain a judgment of rapid maneuvering areas.

[0023] The ground mechanical parameter interpretation semantic map construction unit adopts the form of semantic map to characterize the ground mechanical parameter characteristics inversion results of multi-source data. By adopting the semantic map construction method, the process of estimating ground bearing and shear characteristics based on vision is integrated into the ground modeling process to construct a ground mechanical parameter characteristic inversion semantic map that integrates ground geometry and mechanical characteristics, thereby providing terrain geometry and force for vehicle traffic simulation, path planning and mobility control.

[0024] The 3D voxel lattice creation and optimization module further includes:

[0025] Based on the multi-element voxel lattice, a three-dimensional navigable domain map construction unit is established. For the core elements of complex off-road environments related to the mobility characteristics of ground vehicles and the interaction between vehicles and the environment, an abstract representation model is established. Based on the voxel model, a spatial topological map structure is used for the digital representation of unstructured complex terrain and complex off-road environments.

[0026] Abstract representation unit for complex off-road environment elements, used to realize the abstract representation of complex field environment elements;

[0027] A map data structure unit based on voxel model spatial topology is used to establish spatial topology based on voxel model. Utilizing the spatial continuity between different instance objects, spatial information based on semantic point clouds is clustered, with each cluster considered an independent object in space. Semantic map objects are created for different categories based on the number of semantic segmentation categories. In the map data structure, the semantic segmentation map is combined with a global voxel grid map. Voxel grid objects are managed by a global voxel grid set, organized according to object instances, and used for predicting impassable areas for vehicles in subsequent path planning. Furthermore, to achieve [further details regarding the classification of each class / category], [further details are needed]. To perceive object instances in the semantic map, an incremental density clustering algorithm based on voxel grid elements was designed for each category of complex off-road environment spatial semantic map. Based on the voxel grid coordinates, all voxel grids that meet the conditions are retrieved in a certain neighborhood. The inherent relationship between the voxel grid instance information after semantic segmentation is used to generate their topological relationship. The voxel model reflects the topological relationship between spatial objects through the adjacency relationship between cubes. The adjacency relationship describes the topological relationship between two adjacent voxels. For the topological relationship between multiple consecutive voxels, path and connectivity relationships are used to describe it, thereby generating a complex terrain map data structure for complex off-road environments.

[0028] Path planning module based on 3D voxel lattice:

[0029] Utilizing extensive empirical data from ground vehicle passability tests, a data-driven global path planning algorithm is obtained based on a multi-element voxel grid 3D passable domain map. According to differentiated application requirements and ground vehicle mobility requirements, a multi-state constrained optimization objective function is established. A hierarchical search is performed on the complex voxel grid off-road environment map, and a heuristic search algorithm is used to iteratively find passable nodes. Combined with a deep neural network, empirical passable path results with different rewards are obtained, thus achieving the optimal set of path points that satisfies the objective function.

[0030] Preferably, in the land complex environment ground mechanical parameter interpretation unit, a soil and rock classification system for ground mechanical properties is constructed, rapid survey and identification of ground soil and rock characteristic types in land complex off-road environments is performed, a soil moisture inversion method based on remote sensing data is used, and ground soil mechanical property parameters for land complex environments are constructed.

[0031] Preferably, in the vehicle dynamics model, the vehicle is divided into two types of running gear: tracked and wheeled.

[0032] Preferably, in the vehicle passability assessment unit: the ground is parametrically characterized, and road surface information, terrain information, landform information and climate information are integrated. Based on a combination of subjective and objective methods, features of various types of information are extracted and selected to obtain feature types that are highly correlated with vehicle passability, thereby constructing a ground comprehensive characterization parameter model covering geometric and geological characteristics.

[0033] Regarding the vehicle's own characteristics, the theoretical design or experimental experience of structural dimensions, power characteristics, and performance indicators are combined to conduct a deconstructive analysis of the vehicle's own characteristics. Combined with the ground comprehensive characterization parameterization model, the factors affecting passability are studied to reveal the interaction mechanism between vehicle parameters, ground characterization, and passability, establish quantitative analysis relationships, and form a multivariate judgment evaluation model.

[0034] Preferably, the maneuver speed prediction unit, based on the passability assessment, studies a maneuver speed prediction method that obtains a power flow model based on the vehicle's power transmission characteristics and the vehicle-ground coupled dynamics model, targeting different vehicle and ground types.

[0035] Preferably, in the abstract representation unit of the complex off-road environment elements, for the terrain and soil voxels, it is necessary to extend the solid voxels, add the element attributes of soil elements, and use the octree data structure to compress the terrain with the same type of attributes. Then, multi-resolution terrain contour surface extraction is achieved through indirect visualization.

[0036] Preferably, the map data structure unit utilizes the characteristics of hash mapping and uses the std::map object provided by the C++STL library to construct key-value pairs of voxel grid IDs and voxel grid objects. Voxel grids do not need to be pre-created. Once the sensor observations of the system platform are transformed to the reference coordinate system, if an observation falls into a voxel grid object that has already been created in the complex off-road environment space, the voxel grid at that location will be dynamically created.

[0037] Preferably, the path planning module based on three-dimensional voxel grids also provides a neural network path planning method based on LSTM, which reorganizes the imported data into a three-dimensional voxel grid with multiple element attributes based on an octree structure, and finally forms a three-dimensional passable area map.

[0038] The generated 3D passable area map serves as the base map for the path planning component. When path planning is needed, users input corresponding commands into the system based on their starting point, destination, and highest priority application requirements. After reading the code corresponding to the command, the system then initiates path planning.

[0039] After receiving user input, the subsystem first establishes a specific path optimization function f(x), which is related to application requirements, navigation unit model, and navigable area map. Based on the generated 3D navigable area map and f(x), the vehicle's off-road maneuvering area map is divided into several voxel grids of different sizes. i The constructed 3D feasible region map has each voxel grid with a unique number and rich attribute parameters. The specific path optimization process is as follows:

[0040] a) The subsystem first searches for the trajectory on a high-resolution three-dimensional feasible domain map to determine a coarse path trajectory;

[0041] b) Based on the coarser path trajectory, perform accurate path optimization again on a higher resolution map according to the optimization function;

[0042] c) After repeating the above two steps multiple times, a precise path trajectory P is finally determined. ab and its coordinates.

[0043] To address the problems mentioned in the background art, the technical solution disclosed in this invention mainly focuses on three parts: First, it solves the problem of interaction between the vehicle and the ground mechanics by regional mobility assessment and prediction, and obtains the vehicle's ability to pass through regional mobility after considering factors such as terrain, landform, soil, soil quality, and vehicle power; then, it forms three-dimensional voxel data for land path planning by three-dimensional voxel lattice representation of multi-dimensional environmental factors; finally, it realizes the path planning task in complex environments based on the three-dimensional voxel lattice.

[0044] To address the challenges of complex unstructured off-road environments, insufficient real-time situational awareness, and weak autonomous decision-making capabilities during maneuver navigation in complex environments, this invention proposes a task-driven 3D voxel lattice real-time path planning technology. This novel path planning technology offers the following two advantages:

[0045] (1) Diverse evaluation elements and strong real-time performance. Based on the accurate evaluation results of vehicle mobility in complex environments, this technology uses dynamic events such as real-time weather changes, natural disasters, road traffic, electromagnetic countermeasures and threat situations as important planning parameters for path planning. The parameter information is stored and represented in the form of a three-dimensional voxel grid. During the path planning process, the task information and dynamic events are comprehensively evaluated to achieve real-time response to the maneuvering task and dynamic emergencies.

[0046] (2) Excellent modeling and prediction capabilities, and superior practicality. By constructing a multi-element voxel grid, unstructured ground environments can be represented, and a dynamic octree approach is used to manage the 3D navigable domain map. Each voxel grid point can store navigable / impassable information generated by the mobility assessment module, extended information such as terrain, obstacle semantics, and fuel consumption, task attributes, navigation performance assessment results, and other dynamic event elements. The 3D voxel grid integrates mobility readiness assessment results and navigation performance assessment results, while simultaneously accessing dynamic information in real time, making the path planning results more reasonable and accurate. Attached Figure Description

[0047] Figure 1 This illustrates the overall technical concept of the present invention;

[0048] Figure 2 This illustrates the overall approach to mobility assessment and prediction in complex land environments;

[0049] Figure 3 The study of the ground integrated characterization parameter model is illustrated;

[0050] Figure 4 The diagram illustrates the predicted speed of maneuver;

[0051] Figure 5 This illustrates the abstraction process of elements in a complex off-road environment;

[0052] Figure 6 A schematic diagram illustrating terrain volume modeling using basic elements;

[0053] Figure 7 This illustrates real-time environment understanding based on voxel grid maps;

[0054] Figure 8 This is a voxel lattice topological diagram. Detailed Implementation

[0055] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0056] like Figure 1The present invention, after acquiring a large amount of geographic information data, performs mobility and navigation performance assessments on the current complex environment based on the characteristics of the navigation vehicle, obtaining vehicle traversability data and navigation positioning performance data in the current environment, respectively. Then, based on an octree structure, the data is reorganized into a three-dimensional voxel grid with multiple element attributes, ultimately forming a three-dimensional traversable area map. This generated three-dimensional traversable area map serves as the base map for the path planning component. Given a starting and ending point for maneuvering, path planning can be performed, and the corresponding planning results can be returned.

[0057] Based on the above ideas, the present invention discloses a specific scheme for a path planning system for complex terrestrial environments based on three-dimensional voxel lattices, including:

[0058] 1. Land-based complex environment mobility assessment and prediction module

[0059] The overall approach to the mobility assessment and prediction module in complex land environments is as follows: Figure 2 As shown, the research content is mainly divided into the following five parts:

[0060] 1) Intelligent obstacle recognition unit for complex land environments

[0061] To address the need for rapid obstacle recognition in complex terrestrial environments, this paper researches intelligent obstacle recognition methods based on scene scale and DCNN transfer learning. The challenges lie in: the high semantic level and well-defined physical boundaries making intelligent interpretation difficult; limited sample sizes; and the need for an appropriate scale to represent the overall characteristics. Therefore, to address the issues of low intelligence and accuracy in obstacle recognition, this invention constructs an intelligent obstacle recognition model based on scene scale and DCNN transfer learning. This invention utilizes publicly available datasets and DCNN transfer learning strategies; it uses scenes as primitives to represent the overall characteristics of obstacles and constructs methods for complex semantic patterns.

[0062] 2) Terrestrial mechanical parameter interpretation unit for complex terrestrial environments

[0063] Interpreting ground mechanical parameters in complex terrestrial environments provides the data foundation for coupled analysis of vehicle and ground mechanics. Because the ground mechanical properties in complex terrestrial environments are significantly influenced by soil type and soil moisture, interpreting these parameters requires soil classification and identification, obtaining soil moisture data, and subsequently constructing the ground mechanical parameters. This includes: constructing a soil and rock classification system for ground mechanical properties; rapid surveying and identification of soil and rock characteristics in complex off-road environments; a soil moisture inversion method based on remote sensing data; and the construction of ground soil mechanical property parameters for complex terrestrial environments.

[0064] 3) Precise inversion unit of ground mechanical parameters of coupled vehicle

[0065] In order to quantitatively describe the relationship between ground soil and rock types and their corresponding mechanical property parameters, as well as the mechanical characteristics of different vehicle types, this study investigates the mechanical characteristics of soil types that affect vehicle traffic, establishes a knowledge base of the correspondence between known soil types and mechanical properties, and obtains a coupled quantitative calculation method that satisfies the mechanical property parameters of different passing vehicles and standard classifications of ground. This enables the inversion of ground mechanical property parameters in unknown areas and the assessment of vehicle traffic performance.

[0066] 4) Vehicle dynamics model based on vehicle-ground interaction

[0067] The vehicle-ground interaction model is primarily determined by the type of vehicle and the type of ground. Different types of ground have different characterization parameters, while the type of vehicle is mainly determined by the type of its running gear. Existing vehicles are mainly divided into tracked and wheeled running gears, while the types of ground vary due to their constituent components. Therefore, studying the interaction mechanism between tracked / wheeled vehicles and different types of ground can effectively reflect their interaction relationship, obtain the corresponding types of mechanical parameters for both the vehicle and the ground, and provide guidance for the inversion of ground mechanical parameters. Based on the interaction mechanism between different running gears and the ground, a multi-degree-of-freedom mechanical model of vehicles in a three-dimensional unstructured environment is established, providing a model basis for evaluating vehicle passability and maneuverability.

[0068] 5) Vehicle passability assessment unit based on multivariate analysis

[0069] Vehicle passability primarily refers to a vehicle's ability to traverse a designated area, a GO / NOGO judgment problem. Passability assessment research first requires parametric characterization of the ground surface, integrating road surface, terrain, geomorphological, and climatic information. Based on a combination of subjective and objective methods, features are extracted and selected from various types of information to obtain feature types highly correlated with vehicle passability. This allows for the construction of a comprehensive ground characterization parameter model encompassing geometric and geological properties, such as... Figure 3 As shown.

[0070] Regarding the vehicle's own characteristics, the vehicle's own characteristics are deconstructed and analyzed by combining theoretical design or experimental experience with structural dimensions, power characteristics, and performance indicators. The factors affecting passability are studied by combining the ground comprehensive characterization parameter model, revealing the interaction mechanism between vehicle parameters, ground characterization and passability, establishing quantitative analysis relationships, and forming a multi-dimensional judgment evaluation model.

[0071] 6) A speed prediction unit combining vehicle performance parameters and dynamics models

[0072] Mobility assessment primarily focuses on evaluating vehicle operating speed. Building upon passability assessment, and considering different vehicle and terrain types, research is conducted on mobility speed prediction methods based on power flow models, utilizing vehicle power transmission characteristics and vehicle-ground coupled dynamics models. Figure 4 As shown. Regarding vehicle operational safety, a deep learning-based study on driving stability classification is conducted, combining vehicle test data, a comprehensive ground characterization model, and vehicle speed. The determination of rapid maneuvering zones requires combining vehicle stability and speed characteristics to develop rapid maneuverability rules, forming a rapid maneuverability zone assessment model, thereby enabling the determination of rapid maneuverability zones.

[0073] 7) Ground mechanical parameter interpretation semantic map construction unit

[0074] To better interpret the ground mechanical parameter inversion results from multi-source data, a semantic map approach is adopted to characterize these results. By incorporating semantic map construction methods, the process of estimating ground bearing and shear characteristics based on vision is integrated into the ground modeling process. This results in the construction of a semantic map that integrates ground geometry and mechanical properties, providing terrain geometry and force data for vehicle traffic simulation, path planning, and mobility control.

[0075] 2. 3D Voxel Lattice Creation and Optimization Module

[0076] Path planning and maneuver guidance require maps that reflect terrain, features, strata, and geological characteristics. To this end, this invention first establishes an abstract representation model of the core elements of complex off-road environments, focusing on the maneuver characteristics of ground vehicles and the interaction between vehicles and the environment. Then, it studies a spatial topological map structure based on voxel models for the digital representation of unstructured complex terrain and complex off-road environments.

[0077] Based on prior knowledge of the impact of geographical environment and vehicle performance on vehicle trafficability, this invention studies the use of voxel models to achieve comprehensive modeling and representation of various factors affecting ground vehicle traffic in complex off-road environments.

[0078] Abstract representation of environmental elements in complex off-road environments. This study investigates methods for abstracting and representing landforms, features, and subsurface environmental elements within a specific area. Based on an octree voxel structure, the modeling process for complex off-road terrain is implemented. Regular voxels are used to achieve grid partitioning, discretizing elements such as terrain surface and subsurface geology. Each voxel records the attributes of various elements related to the current complex off-road environment, such as terrain elevation, temperature, humidity, and geological properties.

[0079] Map data structure based on voxel model spatial topology. This study investigates a data structure with fast 3D spatial search and data compression capabilities, establishes spatial topology relationships of voxel models, and constructs a multi-element voxel lattice 3D traversable domain map to support subsequent research on ground vehicle path planning algorithms.

[0080] 1) Three-dimensional walkable domain map construction unit based on multi-element voxel lattice

[0081] Path planning and maneuver guidance require maps that reflect terrain, features, strata, and geological characteristics. To this end, this invention first establishes an abstract representation model of the core elements of complex off-road environments, focusing on the maneuver characteristics of ground vehicles and the interaction between vehicles and the environment. Then, it studies a spatial topological map structure based on voxel models for the digital representation of unstructured complex terrain and complex off-road environments.

[0082] 2) Abstract representation unit of complex off-road environment elements

[0083] Research on vehicle route planning for complex terrain, including the abstract representation process of specific complex outdoor environmental elements, such as... Figure 5 As shown.

[0084] This paper uses an octree voxel structure to model terrain volumes as an example to illustrate the process of modeling complex off-road environments based on voxel models. A terrain volume can include the terrain surface, underground geology, and surface air. First, the "boundary volume" of a specified region is determined, and the resolution of the terrain voxels is specified. Then, the terrain surface, underground geology, and surface air are segmented into voxels, and the volumes are discretized using solid, liquid (groundwater), and gas voxels, respectively.

[0085] like Figure 6 As shown, for topographic soil voxels, it is necessary to expand the solid voxels by adding soil feature attributes, such as soil type, moisture content, permeability, compressibility, and shear strength. An octree data structure is used to compress topographic features with the same type of attributes, and then multi-resolution topographic contour surface extraction is achieved through indirect visualization methods.

[0086] The volumetric modeling methods for various elements in complex off-road environments are essentially the same, all utilizing basic voxels to construct the various elements of the complex off-road environment. When modeling the elements of a complex off-road environment in a specific area, the voxels of each element are extensions based on the basic voxels. Spatially, they are divided into cubes, allowing the attribute values ​​of various elements to be modeled uniformly within a single voxel. Each voxel can record different land, ocean, meteorological, and electromagnetic attribute information, thus achieving unified modeling of various elements in the complex off-road environment, as shown in the figure below. When analyzing or representing any single element, attribute values ​​of the same element type can be extracted from the entire voxel model.

[0087] Theoretically, this approach enables unified modeling of complex off-road environments, constructing a unified voxel model. However, in practice, data compression, storage, and indexing are crucial for modeling. Since voxel data is already enormous, improving the efficiency of data storage and indexing is paramount.

[0088] 3) Map data structure unit based on voxel model spatial topology

[0089] To more efficiently store and retrieve traversable domain maps, it is necessary to establish voxel-based spatial topological relationships. Unlike the TSDF Volume space used by traditional algorithms, the size of voxel grids in real-world complex off-road environments is often difficult to determine, and traditional algorithms struggle to allocate sufficient memory. This invention utilizes the characteristics of hash mapping, using std::map objects provided by the C++ STL library to construct key-value pairs between voxel grid IDs and voxel grid objects. Voxel grids do not need to be pre-created; they are dynamically created only after the system platform's sensor observations are transformed to the reference coordinate system. If an observation falls within a voxel grid object already created in the complex off-road environment space, the voxel grid at that location will be dynamically created. Leveraging the O(lgn) fast retrieval characteristic of hash mapping, this method saves memory and computation time, enabling mapping of larger scenes and meeting the real-time requirements of the Z-platform.

[0090] Commonly used path planning maps only model nodes or drivable roads, failing to represent targets in the ZC environment by label. This means that the map representation cannot distinguish between vehicles, environmental objects, etc., in complex off-road environments. While instance segmentation algorithms can mitigate this deficiency to some extent, they remain ineffective for segmenting objects like trees, shrubs, and unique features of complex off-road environments such as ruins and walls. This invention utilizes the spatial continuity between different instance objects to cluster spatial information based on semantic point clouds. Each cluster is considered an independent object in space. Semantic map objects are established for different categories based on the number of semantic segmentation categories.

[0091] In this data structure, the map of the semantic segmentation part is combined with the global voxel grid map. Since the label of the voxel grid may jump from one category to another as the number of observations increases, in order to avoid wasting time deconstructing and reconstructing voxel grids when storing voxel grid information independently for each category, the voxel grid objects in this invention will be managed by the global voxel grid set and organized according to object instances, which will be used to predict the areas where vehicles cannot pass through in subsequent path planning.

[0092] To achieve object instance perception in the semantic map for each category, the system references the DBSCAN algorithm and designs an incremental density clustering algorithm based on voxel grid elements in the semantic map of complex off-road environments for each category. This invention retrieves all voxel grids that meet the criteria within a certain neighborhood based on voxel grid coordinates, significantly reducing computation time. First, the semantic information obtained from the overall perception is clustered into several clusters. Then, for each new cluster, its adjacency relationship with existing clusters is determined, and the clustering results of existing clusters are merged or new clusters are added.

[0093] After incremental clustering terminates, a comprehensive semantic topological voxel map of the complex off-road environment can be obtained, which can accurately represent objects such as grasslands, large buildings, ruins, and forests that are difficult to represent using instance information. It can also be converted into only local object instances, realizing situational awareness covering a wide range of object sizes and types in complex off-road environments.

[0094] To effectively organize the spatial relationships of labeled elements in a voxel model within an off-road environment, topological relationships are generated by utilizing the inherent associations between voxel lattice instances after semantic segmentation. The voxel model primarily reflects the topological relationships between spatial objects through adjacency relationships between cubes. There are three types of shared relationships between voxels—shared faces, shared edges, and shared points, referred to as face adjacency, edge adjacency, and point adjacency, respectively. Each voxel has 26 adjacent voxels, i.e., 6 face adjacencies, 12 edges, and 8 faces. Adjacency relationships describe the topological relationship between two adjacent voxels; for the topological relationship between multiple consecutive voxels, path and connectivity relationships are used, such as... Figure 8 As shown, this leads to the generation of complex terrain map data structures for complex off-road environments.

[0095] Based on the above methods, this service can generate at least four types of route planning maps, including raster maps, Voronoi polygon maps, visualization maps, topological maps, and 3D voxel maps, which meet the indicator requirements.

[0096] 3. Path planning module based on 3D voxel lattice

[0097] Leveraging extensive empirical data from ground vehicle passability tests, this study investigates data-driven global path planning algorithms based on multi-element voxel lattice 3D drivable domain maps. According to differentiated application requirements and ground vehicle mobility needs, a multi-state constrained optimization objective function is established. A hierarchical search is performed on the complex voxel lattice off-road environment map, using a heuristic search algorithm to iteratively find drivable nodes. Combined with deep neural networks and other methods, empirically drivable path results with different rewards are explored, ultimately achieving the optimal set of path points that satisfies the objective function.

[0098] Based on the diverse application requirements in complex off-road environments, the system performs rapid, real-time global path planning and search based on the provided navigation unit mobility assessment results and navigation performance status assessment results, providing a basis for path selection and providing a basis for the maneuvering paths of land vehicles.

[0099] This invention requires real-time dynamic path planning based on different application requirements input by the user. In complex off-road environments, vehicles are subject to various constraints such as prior knowledge of terrain, climate, and environmental conditions, while also facing multiple application requirements, which increases the difficulty of trajectory prediction.

[0100] The main data inputs related to different vehicle models are:

[0101] 1) Vehicle mobility assessment and prediction database files;

[0102] 2) Vehicle dimensions, weight, and dynamics-related attributes.

[0103] The methods related to data input involve constraint functions and optimization processes for path optimization. They will regenerate a 3D navigable domain map, and through global and local path planning, generate the path trajectory of the maneuvering unit according to application requirements.

[0104] Based on the generated path trajectory and the dynamic characteristics of different navigation unit types, different control parameters are generated to guide the maneuvering unit.

[0105] Addressing seven main application requirements—speed priority, distance priority, road priority, safety priority, fuel consumption priority, commander autonomous decision-making, and intelligent mode—this invention, in addition to conducting research on traditional path planning algorithms, will also explore a neural network path planning method based on LSTM.

[0106] This invention uses data from relevant databases within the platform, as well as data provided by the land-based complex environment mobility assessment and prediction subsystem and the complex environment ubiquitous navigation performance prediction and assessment subsystem, as data sources. It imports geographic entity data, navigation electronic map data, vehicle parameter data, and satellite navigation signal monitoring and sensing network data. Furthermore, it utilizes mobility assessment and prediction products to determine the accessibility and vehicle mobility within the target area. Since the terrain, landforms, and geology of the areas traversed by the mobile navigation unit are relatively complex, this invention reorganizes the imported data into a three-dimensional voxel grid form with multiple element attributes based on an octree structure, ultimately forming a three-dimensional accessible area map.

[0107] The generated 3D navigable area map serves as the base map for the route planning component (hereinafter referred to as the "navigable map"). When route planning is needed, the user inputs corresponding commands into the system based on their starting point, destination, and highest priority application requirements. After reading the code corresponding to the command, the system initiates route planning.

[0108] After receiving user input, the subsystem first establishes a specific path optimization function f(x), which is related to application requirements, the navigation unit model, and the navigable area map. Based on the generated 3D navigable area map and f(x), the vehicle's off-road maneuvering area map is divided into several voxel grids of different sizes. i The constructed 3D feasible region map has each voxel grid with a unique number and rich attribute parameters. The specific path optimization process is as follows.

[0109] a) The subsystem first searches for the trajectory on a high-resolution three-dimensional feasible domain map to determine a coarse path trajectory;

[0110] b) Based on the coarser path trajectory, perform accurate path optimization again on a higher resolution map according to the optimization function;

[0111] c) After repeating the above two steps multiple times, a precise path trajectory P is finally determined. ab and its coordinates.

[0112] When the subsystem performs path finding through the above steps, due to different optimization conditions arising from application requirements, more than one optimized path result may be obtained, resulting in multiple paths that meet or similarly meet the application requirements. After generating each path, the subsystem stores information such as path travel time, fuel consumption, distance, safety, and maximum speed as basic path attribute information, forming a corresponding data file.

[0113] Determine path P ab Then, based on the GIS coordinates output by the path, the area that the motor vehicle passes through under the guidance of the path can be determined. The mobility attribute parameters of the area the vehicle passes through can be read, searched and determined. The main parameters of the search process are shown in Table 1 below.

[0114] Table 1 Input and Output of Intelligent Path Planning

[0115]

[0116] Based on the information such as the highest speed in the area from the above data, the overall travel time T of this planned route can be estimated. Then, based on the set arrival time t, the recommended departure time t0 can be calculated for the reference of the command personnel when carrying out maneuvers.

Claims

1. A path planning system for complex terrestrial environments based on three-dimensional voxel lattices, characterized in that, It includes a module for maneuverability assessment and prediction in complex land environments, a module for establishing and optimizing three-dimensional voxel lattices, and a module for path planning based on three-dimensional voxel lattices, wherein: The Land Complex Environment Mobility Assessment and Prediction Module further includes: The intelligent obstacle recognition unit for complex land environments constructs an intelligent obstacle recognition model based on scene scale and DCNN transfer learning, which is used to quickly identify obstacles in complex land environments. The ground mechanical parameter interpretation unit for complex land environments provides a data foundation for realizing coupled analysis of vehicle and ground mechanics, classifies and identifies soil, obtains soil moisture, and then constructs ground mechanical parameters; A precise inversion unit for ground mechanical parameters of coupled vehicles is established to create a knowledge base of the correspondence between known soil types and mechanical properties. This allows for the acquisition of a coupled quantitative calculation method that satisfies the ground mechanical property parameters of different vehicles and standard classifications, thereby enabling the inversion of ground mechanical property parameters in unknown areas and the assessment of vehicle traffic performance. The vehicle dynamics model based on vehicle-ground interaction is determined by the type of vehicle and ground. The vehicle type is determined by the running gear. Based on the interaction mechanism between different running gears and the ground, a multi-degree-of-freedom mechanical model of the vehicle in a three-dimensional unstructured environment is established to provide a model basis for the evaluation of vehicle passability and maneuverability. A vehicle passability assessment unit based on multivariate analysis is used to assess a vehicle's ability to pass through a designated area. A speed prediction unit combining vehicle performance parameters and dynamic models is used to evaluate vehicle operating speed. Furthermore, for vehicle operating safety, a deep learning-based driving stability classification is obtained by combining vehicle test data, ground comprehensive characterization model and vehicle speed. At the same time, combining vehicle stability and speed, a rapid maneuvering rule study is carried out to form a rapid maneuvering area evaluation model to obtain a judgment of rapid maneuvering areas. The ground mechanical parameter interpretation semantic map construction unit adopts the form of semantic map to characterize the ground mechanical parameter characteristics inversion results of multi-source data. By adopting the semantic map construction method, the process of estimating ground bearing and shear characteristics based on vision is integrated into the ground modeling process to construct a ground mechanical parameter characteristic inversion semantic map that integrates ground geometry and mechanical characteristics, thereby providing terrain geometry and force for vehicle traffic simulation, path planning and mobility control. The 3D voxel lattice creation and optimization module further includes: Based on the multi-element voxel lattice, a three-dimensional navigable domain map construction unit is established. For the core elements of complex off-road environments related to the mobility characteristics of ground vehicles and the interaction between vehicles and the environment, an abstract representation model is established. Based on the voxel model, a spatial topological map structure is used for the digital representation of unstructured complex terrain and complex off-road environments. Abstract representation unit for complex off-road environment elements, used to realize the abstract representation of complex field environment elements; A map data structure unit based on voxel model spatial topology is used to establish spatial topology based on voxel model. Utilizing the spatial continuity between different instance objects, spatial information based on semantic point clouds is clustered, with each cluster considered an independent object in space. Semantic map objects are created for different categories based on the number of semantic segmentation categories. In the map data structure, the semantic segmentation map is combined with a global voxel grid map. Voxel grid objects are managed by a global voxel grid set, organized according to object instances, and used for predicting impassable areas for vehicles in subsequent path planning. Furthermore, to achieve [further details regarding the classification of each class / category], [further details are needed]. To perceive object instances in the semantic map, an incremental density clustering algorithm based on voxel grid elements was designed for each category of complex off-road environment spatial semantic map. Based on the voxel grid coordinates, all voxel grids that meet the conditions are retrieved in a certain neighborhood. The inherent relationship between the voxel grid instance information after semantic segmentation is used to generate their topological relationship. The voxel model reflects the topological relationship between spatial objects through the adjacency relationship between cubes. The adjacency relationship describes the topological relationship between two adjacent voxels. For the topological relationship between multiple consecutive voxels, path and connectivity relationships are used to describe it, thereby generating a complex terrain map data structure for complex off-road environments. Path planning module based on 3D voxel lattice: Utilizing extensive empirical data from ground vehicle passability tests, a data-driven global path planning algorithm is obtained based on a multi-element voxel grid three-dimensional passable domain map. According to differentiated application requirements and ground vehicle mobility requirements, a multi-state constrained optimization objective function is established. A hierarchical search is performed on the complex voxel grid off-road environment map. A heuristic search algorithm is used to iteratively find passable nodes. Combined with a deep neural network, empirical passable path results with different rewards are obtained, thus achieving the optimal set of path points that satisfies the objective function. In the vehicle passability assessment unit: the ground is parametrically characterized, and road surface information, terrain information, landform information and climate information are integrated. Based on a combination of subjective and objective methods, features of various types of information are extracted and selected to obtain feature types that are highly correlated with vehicle passability, thereby constructing a ground comprehensive characterization parameter model covering geometric and geological characteristics. Regarding the vehicle's inherent characteristics, based on theoretical design or experimental experience in structural dimensions, dynamic characteristics, and performance indicators, a deconstructive analysis of the vehicle's inherent characteristics is conducted. Combined with a ground-based comprehensive characterization parameterized model, the factors influencing passability are studied, revealing the interaction mechanism between vehicle parameters, ground characterization, and passability. A quantitative analysis relationship is established, forming a multi-dimensional assessment model. In the abstract representation unit of the complex off-road environment elements, for the terrain and soil voxels, it is necessary to extend the solid voxels, add the feature attributes of soil elements, and use the octree data structure to compress the terrain with the same type of attributes. Then, multi-resolution terrain isosurface extraction is achieved through indirect visualization methods. The map data structure unit utilizes the characteristics of hash mapping and uses the std::map object provided by the C++ STL library to construct key-value pairs of voxel IDs and voxel objects. Voxels do not need to be created in advance. After the sensor observations of the system platform are transformed into the reference coordinate system, if an observation falls into a voxel object that has been created in the complex off-road environment space, the voxel at that location will be dynamically created.

2. The path planning system for complex terrestrial environments based on a three-dimensional voxel lattice as described in claim 1, characterized in that, In the ground mechanical parameter interpretation unit for complex terrestrial environments, a soil and rock classification system for ground mechanical properties is constructed, rapid survey and identification of ground soil and rock characteristic types in complex off-road environments is performed, a soil moisture inversion method based on remote sensing data is used, and ground soil mechanical property parameters for complex terrestrial environments are constructed.

3. The path planning system for complex terrestrial environments based on a three-dimensional voxel lattice as described in claim 1, characterized in that, In the vehicle dynamics model, the vehicle is divided into two types of running gear: tracked and wheeled.

4. The path planning system for complex terrestrial environments based on three-dimensional voxel lattice as described in claim 1, characterized in that, Based on the passability assessment, the proposed speed prediction unit studies a speed prediction method that uses a power flow model to obtain a speed prediction method for different vehicle and ground types, based on vehicle power transmission characteristics and vehicle-ground coupled dynamics model.

5. The path planning system for complex terrestrial environments based on a three-dimensional voxel lattice as described in claim 1, characterized in that, The path planning module based on three-dimensional voxel grids also provides a neural network path planning method based on LSTM, which reorganizes the imported data into a three-dimensional voxel grid with multiple element attributes based on an octree structure, and finally forms a three-dimensional passable area map. The generated 3D passable area map serves as the base map for the path planning component. When path planning is needed, users input corresponding commands into the system based on their starting point, destination, and highest priority application requirements. After reading the code corresponding to the command, the system then initiates path planning. After receiving user input, the subsystem first establishes a specific path optimization function f(x), which is related to application requirements, navigation unit model, and navigable area map. Based on the generated 3D navigable area map and f(x), the vehicle's off-road maneuvering area map is divided into several voxel grids of different sizes. i The constructed 3D feasible region map has each voxel grid with a unique number and rich attribute parameters. The specific path optimization process is as follows: a) The subsystem first searches for the trajectory on a high-resolution three-dimensional feasible domain map to determine a coarse path trajectory; b) Based on the coarser path trajectory, perform accurate path optimization again on a higher resolution map according to the optimization function; c) After repeating the above two steps multiple times, a precise path trajectory P is finally determined. ab and its coordinates.

Images