Path planning method and system integrating multi-mode terrain perception and vehicle dynamics

Abstract

The invention belongs to the technical field of path planning, and discloses a path planning method and system fusing multi-modal terrain perception and vehicle dynamics, and the method comprises the steps: obtaining and fusing multi-modal geographic data to generate a terrain physical representation, and calculating the terrain complexity based on the terrain physical representation; predicting an overturning index and a stability margin of the vehicle based on the terrain physical representation, and generating a trafficability cost function; taking the trafficability cost function as a constraint condition, selecting a motion element according to terrain complexity to perform trajectory fitting, and generating a vehicle path; executing the vehicle path to obtain a vehicle state, correcting the trafficability cost function by using the vehicle state, and optimizing the vehicle path; and constructing a loss function, training the path planning model, and performing path planning by using the trained model. According to the invention, by establishing a terrain vehicle bidirectional coupling mechanism, safe and adaptive local path planning under a complex unstructured terrain is realized, and the traffic capacity and safety of the system in a dynamic environment are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of path planning technology, and in particular to a path planning method and system integrating multimodal terrain perception and vehicle dynamics. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Traditional local path planning algorithms mostly rely on geometric traversability models for path planning. This makes them more adaptable on hard surfaces. However, in complex terrain, they ignore key mechanical parameters such as surface stiffness, adhesion coefficient, and subsidence risk, leading to serious deviations in the assessment of the "true traversability" of the planned path. For example, in soft mud or steep slopes, a path generated solely based on geometric obstacle avoidance principles may cause dangerous situations such as vehicle sinking or skidding due to insufficient consideration of the ground's bearing capacity.

[0004] Existing algorithms generally fail to consider vehicle dynamics constraints. Key parameters such as gradeability, roll stability thresholds, and tire slip ratios are not incorporated into the path optimization objective system. This makes the planned paths susceptible to safety accidents such as rollovers, skidding, or chassis seizures during actual execution in complex environments, seriously threatening driving safety. Furthermore, the path generation mechanism of current fixed-order motion model algorithms is rigid and struggles to adapt to the high curvature and continuous perturbations of complex terrain. This results in poorly smoothed generated trajectories and frequent sudden changes in control commands, which in turn exacerbates vehicle jitters, increases actuator wear, and shortens their service life.

[0005] In particular, existing traditional methods lack environmental adaptability, and the perception-planning-control processes are separated from each other. They cannot build a dynamic closed-loop system from terrain perception to vehicle dynamics response to real-time path updates, and cannot optimize planning strategies by accumulating historical experience. As a result, the system responds laggingly when facing dynamic obstacles or sudden terrain changes, and cannot make effective adjustments in time, resulting in frequent planning failures in new scenarios. Summary of the Invention

[0006] To address the above issues, the present invention proposes a path planning method and system that integrates multimodal terrain perception and vehicle dynamics. By establishing a terrain-vehicle bidirectional coupling mechanism and combining variable-order motion primitive optimization and architecture, safe and adaptive local path planning is achieved in complex unstructured terrain, significantly improving the system's traffic capacity and safety in dynamic environments.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present application provides a path planning method fusing multi-modal terrain perception and vehicle dynamics, comprising the following steps: acquiring multi-modal geographic data, fusing the multi-modal geographic data, generating terrain physical representation, calculating terrain complexity based on the terrain physical representation; inputting the terrain physical representation into a vehicle dynamics model, predicting the overturning index and stability margin of the vehicle, and generating a passability cost function based on the overturning index and stability margin; using the passability cost function as a constraint condition, selecting motion primitives of different orders according to the terrain complexity to perform trajectory fitting, and generating a vehicle path; executing the vehicle path to obtain the vehicle state, and using the vehicle state to correct the passability cost function and optimize the vehicle path; constructing a hybrid loss function to train the path planning model, and using the trained model to perform path planning.

[0008] As an optional implementation, the path planning method further comprises uploading the trained local model parameters to a central server, and the central server aggregates the model parameters to form global shared parameters.

[0009] As an optional implementation, acquiring multi-modal geographic data specifically comprises: cameras provide texture entropy and infrared reflectivity, laser radar reconstructs terrain elevation curvature through three-dimensional point cloud coordinates, uses point cloud density to deduce subsidence probability, millimeter wave radar soil penetration detection obtains soil dielectric constant, and combines Bayesian subsidence probability model to dynamically correct ground stiffness coefficient and multi-spectral vegetation analysis to quantify the attachment coefficient of coverage and water distribution.

[0010] As an optional implementation, the overturning index is calculated based on the terrain physical representation: ; wherein, is the terrain curvature, is the ground stiffness, is the vehicle speed.

[0011] As an optional implementation, the stability margin is calculated based on the terrain physical representation: ; wherein, is the roll moment, is the pitch moment, is the lateral roll moment critical threshold.

[0012] As an optional implementation, the passability cost function is: ; wherein, a ground stiffness, a minimum safety stiffness threshold, , , a weight parameter, a rollover risk index, a terrain gradient, a Heaviside step function.

[0013] In a second aspect, the present application provides a path planning system fusing multi-modal terrain perception and vehicle dynamics, comprising: a first module configured to: acquire multi-modal geographic data, fuse the multi-modal geographic data, generate a terrain physical representation, and calculate terrain complexity based on the terrain physical representation; a second module configured to: input the terrain physical representation into a vehicle dynamics model, predict a rollover index and a stability margin of the vehicle, and generate a passability cost function based on the rollover index and the stability margin; a third module configured to: take the passability cost function as a constraint condition, select motion primitives of different orders according to the terrain complexity to perform trajectory fitting, and generate a vehicle path; a fourth module configured to: execute the vehicle path, obtain a vehicle state, correct the passability cost function using the vehicle state, and optimize the vehicle path; a fifth module configured to: construct a hybrid loss function, train a path planning model, and perform path planning using the trained model.

[0014] In a third aspect, the present application provides an electronic device comprising a memory and a processor, and computer instructions stored on the memory and running on the processor, wherein when the computer instructions are run by the processor, the method of the first aspect is completed.

[0015] In a fourth aspect, the present application provides a computer readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method of the first aspect is completed.

[0016] In a fifth aspect, the present application provides a computer program product comprising a computer program, wherein when the computer program is executed by a processor, the method of the first aspect is completed.

[0017] Compared with the prior art, the present application has the following beneficial effects: The present invention proposes a path planning method that integrates multimodal terrain perception and vehicle dynamics. It integrates multimodal terrain perception, vehicle dynamics constraints and machine learning strategies, breaks through the limitations of traditional path planning geometric models, and establishes a closed-loop link of terrain dynamic properties → vehicle dynamic response → path feasibility assessment → path optimization generation. By establishing a bidirectional coupling mechanism between terrain dynamic properties and vehicle dynamic response, combined with variable-order motion primitive optimization and U-shaped segmentation learning architecture, safe and adaptive local path planning under complex unstructured terrain is achieved. It solves key problems in traditional methods such as the lack of terrain physical property modeling, neglect of vehicle dynamics constraints, and rigid path generation, significantly improving the system's traffic capacity and safety in dynamic environments. It is suitable for complex natural environments such as mountains, jungles, soft surfaces, and trenches. It can perceive terrain characteristics, adapt to vehicle driving capabilities, and dynamically generate safe and feasible local paths.

[0018] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0020] Figure 1 The decision-making framework and technical elements of the present invention; Figure 2 This is a framework diagram of the path planning method integrating multimodal terrain perception and vehicle dynamics of the present invention; Figure 3 This is a framework diagram of the multimodal geographic data fusion of the present invention; Figure 4 A flow chart of adaptive path generation driven by mechanical constraints of the present invention; Figure 5 A flowchart of data collaborative optimization of the present invention; Figure 6 Flowchart of the U-shaped segmentation learning evolutionary layer in multi-vehicle path planning. DETAILED DESCRIPTION

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

[0022] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0023] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the example embodiments of the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0024] The embodiments in the present application and the features in the embodiments can be combined with each other in the case of no conflict.

[0025] Embodiment 1 As Figure 2 shown, the embodiment provides a path planning method fusing multimodal terrain perception and vehicle dynamics, including the following steps: Obtaining multimodal geographic data, fusing the multimodal geographic data, generating terrain physical characterization, and calculating terrain complexity based on the terrain physical characterization; Inputting the terrain physical characterization into a vehicle dynamics model, predicting the overturning index and stability margin of the vehicle, and generating a passability cost function based on the overturning index and stability margin; Taking the passability cost function as a constraint condition, selecting motion primitives of different orders according to the terrain complexity to perform trajectory fitting, and generating a vehicle path; Executing the vehicle path, obtaining the vehicle state, correcting the passability cost function by using the vehicle state, and optimizing the vehicle path; Constructing a hybrid loss function, training the path planning model, and performing path planning by using the trained model.

[0026] The specific scheme of the present application is as follows: The path planning method fusing multimodal terrain perception and vehicle dynamics of the present application is a new type of multimodal coupled local path planning algorithm (MMC-LPPA), a bidirectional coupling mechanism of terrain vehicle, a variable order motion primitive optimization model, a U-shaped segmentation learning-mechanics model hybrid architecture are proposed, a closed loop link of "terrain mechanical properties → vehicle dynamic response → path feasibility evaluation → path optimization generation" is established, and the adaptability of the planning system to complex natural environment and the driving safety are significantly improved.

[0027] The terrain-vehicle bidirectional coupling mechanism uses multimodal sensors to perceive the terrain's physical properties in real time, including ground stiffness, adhesion coefficient, subsidence probability, soil moisture content, and surface roughness. These terrain physical responses are quantified and input into the vehicle's dynamic model, constructing a nonlinear response function that dynamically predicts the vehicle's stability margin and rollover index at specific terrain locations. This is then embedded into the path cost function in the form of a unified path passability score, forming a closed-loop decision chain from "terrain mechanical properties → vehicle dynamic response → path feasibility." This overcomes the limitations of traditional path planning algorithms, which rely solely on geometric features such as elevation, obstacles, and slope for passability assessments. It achieves a quantitative mapping of terrain status to path risk and supports triggering path replanning when the terrain changes or when the vehicle's status approaches a boundary. This significantly improves the system's path reliability and operational safety in slopes, soft soil, trenches, and other abrupt landforms.

[0028] The variable-order motion primitive optimization model is a variable-order polynomial motion primitive optimization unit driven by mechanical constraints. This optimization unit dynamically selects the order of trajectory expression based on the local terrain complexity index. In low-complexity areas, third-order trajectories are used to improve computational efficiency. In medium-disturbance areas, fifth-order curves are used to ensure smooth transitions. In high-risk complex terrain, seventh-order trajectories are enabled to meet acceleration continuity requirements, thereby achieving structural adaptability and dynamic adjustability of path generation. At the same time, vehicle stability constraints are added to the trajectory optimization process, and trajectory smoothness, mechanical consumption cost, and path safety function are integrated to form a joint optimization target. Quasi-Newton gradient optimization is used for real-time solution to ensure that the path achieves the optimal balance between smoothness, controllability, and mechanical stability. This solves the problems of poor smoothness, sudden steering changes, and control discontinuity in complex terrain caused by fixed-order trajectory structures in traditional path planning.

[0029] The U-shaped split learning-mechanical model hybrid architecture is a hybrid path planning framework that integrates U-shaped split learning with a vehicle dynamics model. This architecture is based on a multi-vehicle collaborative distributed training strategy. Each vehicle continuously optimizes its local policy network based on actual driving experience, and policy gradients are regularly aggregated through a central server to form a global shared policy. Compared to traditional centralized training or single-vehicle training methods, U-shaped split learning in path planning scenarios combines on-device perception, multi-source fusion at the center level, learned planning capabilities, and on-device dynamic constraints with fast and safe control into a trainable, experience-driven closed-loop system. This system improves adaptability in complex unstructured terrain and dynamic obstacles, while also balancing real-time requirements and privacy / bandwidth constraints.

[0030] In the decision-making process, the model introduces a "mechanical model safety layer" as a barrier for policy execution, that is, before the output path / action of the RL policy, a round of feasibility screening is conducted through the vehicle response model, if there is a dangerous action or path segment that violates the risk of kinetic stability or exceeds the standard of subsidence, the system will automatically reject the policy action and fall back to the alternative path or trigger local re-planning. This hybrid mechanism organically combines the flexibility of data-driven and the certainty of physical models, not only improves the adaptability of the strategy in uncharted terrain, but also guarantees the safety and robustness of path planning, solving the problems of frequent environmental changes, difficult experience transfer, poor adaptability to new terrain in complex terrain environment.

[0031] The construction of local path planning includes a four-layer closed-loop algorithm architecture of "perception, coupling, decision-making, and evolution". Through the collaborative interaction of the multi-modal sensing layer, the terrain vehicle coupling layer, the variable trajectory optimization layer, and the U-shaped segmentation learning evolution layer, safe and reliable local path planning is achieved in complex terrain. The decision-making architecture and technical elements of the application are shown in Figure 1 .

[0032] Specifically, in the multi-modal perception layer, heterogeneous sensors such as laser radar, multi-spectral camera, millimeter wave radar, and IMU generate spatiotemporally aligned multi-source data streams through hardware synchronization, generating a terrain physical parameter tensor. The texture entropy quantifies the macroscopic roughness of the ground, the dielectric constant deduces the ground stiffness , the laser radar point cloud density characterizes the terrain curvature subsidence probability and water content parameters. These parameters together form a multi-scale description to characterize the micro-pore structure of the terrain soil. The three achieve all-weather robust perception through a defect compensation mechanism. Through the physical inversion subsystem and the improved YOLOv8seg-DLKA model, the semantic features of the terrain are extracted, and a cross-modal feature tensor is constructed: .

[0033] In the terrain vehicle coupling layer, the tensor is input into the multi-degree-of-freedom dynamics response model to calculate the overturning risk index and the six-dimensional stability margin , which are used as vehicle stability criteria and fused into a weighted cost function. The passability cost function is generated: ; The function adjusts the dynamic weight through the Heaviside step function to amplify the weight in the overturning overrun area, forcing the path to detour. When the overturning index exceeds the safety threshold, Output is 1, otherwise 0. When replanning instruction is output. In response to terrain mutation events.

[0034] In the variable order trajectory optimization layer, based on the complexity of the terrain dynamically switch trajectory order. And adopt L-BFGS to solve the following mixed loss function, through real-time benchmark update and risk adaptive weighting balance safety and efficiency.

[0035] ; Where the meaning of each parameter is described in detail below.

[0036] In the U-shaped segmentation learning evolution layer, a double-buffered experience pool is used to store differentiated samples, and a physical barrier check is used to realize safety policy screening. A distributed-centralized hybrid learning framework is constructed to realize multi-vehicle collaborative experience sharing: ; Where, is the global model parameter at the moment, responsible for storing all vehicles shared terrain traffic knowledge; is the U-shaped segmentation learning rate to control the knowledge fusion speed; is the number of effective participating vehicles; is the local model update amount; is the safety check function; is the regularization strength coefficient; is the gradient calibration operator.

[0037] In the U-shaped segmentation learning evolution layer, a double-buffered experience pool is used to store and resample the diversified samples generated by the perception-planning-control process, and a physical barrier check mechanism is used to screen the uploaded update amount for safety, thereby ensuring the stability and safety of the distributed-centralized hybrid framework in multi-vehicle collaborative path planning.

[0038] Under the constraint of Krum algorithm, the global strategy is updated to form a closed-loop learning link of "perception-coupling-optimization-execution-experience feedback-model evolution". This architecture significantly improves the planning adaptability and safety in complex unstructured environments through cross-layer data bus and physical-learning coupling mechanism.

[0039] Based on this technology layer, a new type of multimodal coupled local path planning algorithm (Multimodal-Mechanics Coupled Local Path Planning, MMC-LPPA) is proposed and designed, which adopts a phased processing mode, and its algorithm framework is shown in Figure 2 .

[0040] The algorithm flow is: Construct the MMC_LPPA(sensor_data, vehicle_state) function to implement the multimodal perception, mechanical coupling, and path optimization process to generate a safe trajectory. The parameter sensor_data contains multimodal sensor data such as the lidar, camera, and IMU; vehicle_state contains the vehicle's current state information, such as speed, acceleration, and load. The return value trajectory is the generated safe trajectory, formatted as spatial coordinates and a time series [x(t), y(t), z(t)].

[0041] Phase 1: Multimodal perception fusion, the function is T = MultimodalFusion(sensor_data). The input value is multi-source sensor data, and the output value is the terrain dynamics tensor T = [G(x,y), μ(x,y), P_sink(x,y), ].

[0042] Phase 2: Dynamic constraint generation, using the function constraints = DynamicCoupling(T, vehicle_state). The inputs are the terrain tensor T and the vehicle state, and the output is the dynamic constraint set {Cost_couple, threshold κ}.

[0043] Phase 3: Adaptive path optimization, using the function trajectory = AdaptiveOptimizer(T, constraints) . The inputs are the terrain tensor T and dynamic constraints. The variable optimization unit is integrated with the loss function to solve and output a safe trajectory (spatial coordinates + time series).

[0044] Furthermore, a decision-making framework for coupling multimodal terrain perception with vehicle mechanical performance is proposed, which includes a multimodal terrain perception system and a terrain-vehicle bidirectional coupling mechanism.

[0045] The multimodal terrain perception system constitutes the first module of MMC-LPPA. It generates terrain parameter tensors by spatiotemporally aligning data streams from heterogeneous sensors such as lidar, multispectral camera, and millimeter-wave radar, integrating physical inversion and semantic recognition capabilities. , providing a unified environmental representation for terrain-vehicle coupling decision-making. The specific process is as follows Figure 3 shown.

[0046] First, the camera provides texture entropy and infrared reflectivity. LiDAR through 3D point cloud coordinates Reconstructing terrain elevation curvature , using point cloud density Deducing the probability of subsidence Millimeter wave radar soil penetration detection to obtain soil dielectric constant , and dynamically correct the ground stiffness coefficient by combining the Bayesian subsidence probability model and multispectral vegetation analysis, cover and moisture distribution, quantitative attachment coefficient , integrating to obtain the obstacle probability field , the calculation formula is: ; ; ; ; ； The above parameters are fused to generate terrain physical parameter tensor parameters It provides full-dimensional environmental representation for bidirectional coupling decision-making and calculates terrain complexity through implicit parameter inversion. The calculation formula is: ; ; The multimodal terrain perception system is controlled by two subsystems: the physical inversion subsystem and the semantic recognition subsystem. The physical inversion subsystem integrates multi-source physical signals and quantifies mechanical parameters through an inversion model. The semantic recognition subsystem analyzes visual semantic features to fill in the environmental cognition gaps of the physical model.

[0047] The physical inversion subsystem inverts the subsidence model by combining camera and radar fusion signals. Quantify the roughness of surface particles and accurately capture visual details, but it is limited in foggy and dusty environments because it cannot penetrate obstacles; radar sensors use point cloud density , dielectric constant It analyzes soil porosity and water content, and has all-weather detection capabilities, but lacks recognition of color semantics. The two achieve all-weather robust perception through a defect compensation mechanism: in optically shielded scenarios, the radar's penetration ability compensates for camera failures; in scenarios requiring semantic perception such as mud recognition, the color parsing capabilities of multispectral cameras fill the gaps in radar semantics. This heterogeneous data fusion creates a cross-modal emergence effect that cannot be achieved by single-source sensors, and jointly constructs a multi-dimensional understanding of the physical properties of the surface. By generating inversion information through multispectral images, millimeter-wave radar dielectric constant information, and lidar point cloud density changes, the present invention establishes a Maxwell viscoelastic surface model to predict subsidence behavior under ground unit pressure. The model comprehensively considers key factors such as soil elastic modulus, water content, and contact pressure, and the prediction formula is as follows: ; Where, To predict the amount of subsidence; The surface shear strength parameter characterizes the soil's ability to resist shear deformation. Distribution variance inversion; is the soil plasticity coefficient, which reflects the plastic deformation characteristics of the soil and is obtained by analyzing the infrared reflectivity data of the multispectral camera; The effective tire contact width is calculated based on the spatial distribution characteristics of the lidar point cloud and directly affects the ground pressure distribution. is the local ground pressure of the vehicle; Soil moisture content is obtained by inverting the normalized water index of multispectral bands to quantify the key indicators of soil plasticity and bearing capacity.

[0048] The model output is converted into the subsidence probability as the “subsidence risk factor” in the path accessibility scoring function The subsequent traversability cost function is formally input to ensure that the path planning dynamically responds to changes in surface mechanics.

[0049] The semantic recognition subsystem, in order to solve the recognition problems of unstructured natural terrain, such as blurred boundaries, large scale spans, irregular shapes, and easy loss of small objects, introduces the Deformable Large Kernel Attention (DLKA) mechanism based on the traditional semantic segmentation model, significantly improving the ability to model the semantic features of complex surfaces. In addition, in addition to the traditional IoU loss, the Normalized Wasserstein Distance (NWD) loss function is used to solve the problem of missed small object detection, and the prediction box is measured. With real box The loss function is as follows: ; in, represents the two-dimensional Wasserstein distance, that is, the square of the optimal transmission distance, To calibrate the parameters, the actual measurement improved the F1-score of gully small target detection to 0.87. This accuracy improvement ensures the obstacle probability field in the subsequent terrain parameter tensor. reliability.

[0050] The improved model can achieve high-precision segmentation of key features such as rocks, mud, gullies, etc., and output structured semantic masks which are converted into adhesion coefficients through calculation. , obstacle probability field , as the terrain complexity core input.

[0051] The specific algorithm steps of multimodal perception fusion are as follows: S1: Construct a MultimodalFusion(lidar,camera,radar) function that receives lidar data, camera data, and millimeter-wave radar data as input.

[0052] S2: Utilize sub-function = LaplacianFilter(lidar) processes the lidar point cloud data and applies a missing fill algorithm. Use the Laplacian filter to process the lidar data and calculate the curvature of the terrain. ; S3: Use the subfunction Psink = SinkProbability(lidar, radar) to combine the data of lidar and millimeter-wave radar and obtain the terrain sink probability Psink through the sink probability Bayesian estimation algorithm.

[0053] S4: Use the subfunction G = StiffnessInverse(radar, ε_r) to interpret the millimeter wave physics. This function adds an inversion algorithm based on the millimeter wave radar data and the relative dielectric constant ε_r, using the stiffness inversion algorithm (formula is ) Calculate the stiffness G of the terrain.

[0054] S: Use the subfunction μ = FrictionCompensation(camera, radar) for multispectral semantic compensation. This uses an enhanced fusion mechanism to combine camera and millimeter-wave radar data and calculate the terrain adhesion coefficient μ using a cross-modal adhesion coefficient compensation algorithm.

[0055] S6: The calculated terrain stiffness G, adhesion coefficient μ, sinking probability Psink and terrain curvature Build and return a space-time aligned tensor.

[0056] The terrain-vehicle bidirectional coupling mechanism constitutes the second module of MMC-LPPA. This mechanism establishes a closed-loop interactive system between the terrain's physical properties and the vehicle's dynamic response. It achieves bidirectional dynamic coupling through a "terrain-to-vehicle" forward action chain and a "vehicle-to-terrain" feedback chain, supporting real-time optimization of the traffic cost function.

[0057] This system uses multi-dimensional physical modeling to achieve real-time evaluation and control of vehicle stability, and uses the terrain tensor generated by the above multi-modal perception , closing the terrain-to-vehicle forward action chain in two dimensions, and realizing the mapping of terrain physical properties to vehicle dynamic responses.

[0058] In terms of real-time stability assessment, the overturning index is calculated in real time based on the terrain parameter tensor: ; Among them, the terrain curvature Determines rolling moment, ground stiffness Reflects the response characteristics of the suspension system, vehicle speed Directly affects the centrifugal force effect.

[0059] In terms of multi-degree-of-freedom stability cone analysis, a six-degree-of-freedom dynamic stability margin model is constructed: ; Where, rolling moment , generated by the coupling of lateral acceleration and center of mass height. , depends on the relationship between longitudinal acceleration and slope, and the risk of subsidence , through the composite calculation of ground contact area and subsidence probability, accurate quantification is achieved, forming a complete evaluation closed loop from micro parameters to macro stability.

[0060] The two are used together as vehicle stability criteria and integrated into a weighted cost function to generate a trafficability cost function: ; This function uses a dynamic weight adjustment mechanism: When the Heaviside step function Apply weight amplification to the overturning limit area and force the path to detour; when Output replanning instructions in response to sudden terrain changes.

[0061] The vehicle-to-terrain feedback chain is mainly based on a coupled decision-making mechanism. It dynamically corrects terrain parameter assessment errors through real-time vehicle response data, solves the perception and decision-making mismatch caused by sudden environmental changes, and realizes reverse regulation from vehicle dynamics state to terrain risk assessment correction.

[0062] The input layer of the feedback chain mainly contains the slip rate and vertical stability margin The wheel speed sensor and IMU work together to calculate the slip rate. , quantify the relative deviation between the circumferential speed of the driving wheel and the center of mass speed of the vehicle body, set The six-axis inertial navigation system is combined with the suspension displacement sensor to dynamically calculate the vertical stability margin. Real-time monitoring of vehicle body sinking trend, when The soil bearing capacity risk alarm is triggered.

[0063] When the tire slip rate exceeds the limit, the real-time compensation mechanism of the adhesion coefficient is triggered: ; in, , indicating that the tread shear stress exceeds the Coulomb friction limit. 0.2 is the road surface adaptive damping coefficient, reflecting the slip energy Attenuation effect on adhesion.

[0064] When vertical stability is on the verge of failure, the emergency weighting mechanism for subsidence risk is triggered: ; in, This indicates that the vehicle has exceeded its static load-bearing capacity limit. The risk probability gradient needs to be adjusted accordingly.

[0065] The revised and re-evaluated terrain model is: ; ; Furthermore, the present invention proposes a mechanical constraint-driven adaptive path generation algorithm, which completes the passability cost function output by the coupling layer. After real-time injection, the system dynamically selects the mathematical expression of motion primitives according to the complexity of the terrain, designs a variable-order optimization unit, and quantifies the abstract mechanical risk into an executable trajectory control point sequence through an adaptive order selection strategy. It also develops a multi-objective optimization solver to dynamically balance the safety and motion smoothness of the generated motion primitives through this solver. This part of the algorithm constitutes the third module of MMC-LPPA - the real-time embedding algorithm of mechanical constraints. The process is as follows: Figure 4 shown.

[0066] The specific algorithm steps are as follows: S1: Construct the function DynamicCoupling(T, state), which implements the dynamic coupling function in mechanical performance coupling. It receives the terrain tensor T and the vehicle state state as input, performs real-time calculation of mechanical risk and dynamic weight adjustment, and finally returns the coupling cost and stability threshold.

[0067] S2: Real-time computation of mechanical risks and strengthening of closed-loop algorithms. Obtaining curvature from the terrain tensor T and stiffness G, obtain the velocity v from the vehicle state, and calculate the vehicle rollover risk index κ, κ = 1 / (1 + exp(-(0.5 * + 0.3 * (G ** -1) + 0.2 * (v ** -1)))) .

[0068] S3: Use the six-degree-of-freedom stability cone algorithm and obtain the terrain tensor T and vehicle state state to calculate the stability DSM = StabilityConeModel(T, state).

[0069] S4: Dynamic weight adjustment algorithm (new decision logic). Check whether the calculated capsizing index κ is greater than 0.35. If κ is greater than 0.35, multiply the weight parameter β by 1.5 to achieve risk-driven weight amplification.

[0070] S5: Returns the coupling cost Cost_couple(α,β,γ)+DSM_threshold for subsequent path optimization solution.

[0071] By verifying these two types of mechanical constraints, the trajectory generation of the vehicle based on the vehicle dynamics boundary is directly or indirectly embedded in the path generation process, which improves the vehicle's adaptability in dynamic environments and optimizes driving safety and motion smoothness.

[0072] The variable-order optimization unit is a path optimization method proposed by the present invention that integrates a variable-order polynomial trajectory structure with a mechanical response-driven loss function. This method dynamically selects the order of motion primitives and introduces a mechanical penalty term to improve the path's physical adaptability and execution stability in complex terrain. This method can achieve a path generation result that satisfies vehicle dynamics constraints while also exhibiting high passability and control smoothness in complex unstructured terrain. Specifically described as follows: According to the terrain complexity index The real-time calculation results are used to select motion primitives of different orders for trajectory fitting: ; Polynomial trajectories of different orders are suitable for different levels of terrain structures. The specific corresponding relationships are shown in Table 1.

[0073] Table 1 Correspondence between terrain structure and polynomial order;

[0074] When the terrain is assessed as flat, a third-order polynomial trajectory is selected to meet the requirements for displacement and velocity continuity while maintaining computational efficiency. When the path segment contains moderate obstacles or localized undulations, a fifth-order polynomial trajectory is selected, introducing acceleration continuity constraints to ensure smooth dynamic response. When the path traverses complex terrain such as high-curvature fractures, slope-to-sag, and soft interface areas, a seventh-order polynomial trajectory is selected with additional acceleration continuity constraints to ensure smooth controller execution and vehicle posture stability. This variable-order optimization unit mechanism not only improves the adaptability of the path generation structure but also significantly alleviates the control discontinuity issues of traditional fixed-order paths in complex terrain, enhancing the overall dynamic response performance of the system.

[0075] In order to make the solution more accurate, the present invention designs a multi-objective optimization solver, which uses the limited memory BFGS algorithm (L-BFGS) to solve the trajectory parameters and constructs the following hybrid loss function to achieve multi-objective optimization: ; Variable definition: : Polynomial control point vector (decision variable), which determines the spatial shape of the trajectory, : Initial reference path control point. The description of the loss function is shown in Table 2.

[0076] Table 2 Description of the loss function;

[0077] Furthermore, in order to achieve a dynamic balance between safety and efficiency, the present invention introduces two core technical features: First, when the mechanical cost function value exceeds the risk threshold ( > ), automatically increase the mechanical cost weight To 1.5, give priority to avoiding dynamic safety risks such as sinking and overturning; second, the initial reference path control point Refreshing every 200 milliseconds based on the dynamic terrain environment ensures that the optimization target is consistent with the real-time scenario, preventing trajectory invalidation due to sudden environmental changes. By dynamically adjusting weights and updating path benchmarks in real time, the efficiency of trajectory optimization is maintained while ensuring safety, resolving the conflict between safety and efficiency in traditional methods. This part of the algorithm constitutes the fourth module of MMC-LPPA - the path optimization solution algorithm. The specific algorithm steps are as follows: S1: Construct the function AdaptiveOptimizer(T, constraints). This function implements adaptive optimization, dynamically adjusting the decision order based on terrain complexity, generating motion primitives, and performing path optimization using a multi-objective solution algorithm. The parameter T is a terrain-related information object from which information such as terrain complexity C_terrain can be obtained, and constraints is a list of constraints.

[0078] S2: Variable-order decision algorithm. Obtain terrain complexity C_terrain from terrain information T using T.get_terrain_complexity(). Dynamically determine the polynomial order n based on the terrain complexity C_terrain. If C_terrain is less than 0.3, n is 3; if C_terrain is between 0.3 and 0.7, n is 5; and if C_terrain is greater than 0.7, n is 7.

[0079] S3: Obtain the initial polynomial path p0. Use the motion primitive generation algorithm to perform polynomial fitting based on the global path and the determined polynomial order n, and obtain p0 = PolynomialFitting(global_path, n).

[0080] S4: Multi-objective solution algorithm. Define the loss function loss_function(p), which contains the mechanical cost term ,in is the weight coefficient, Cost_mech(p) is the mechanical cost function; the jerk integral term is: , is the weight coefficient, which measures the smoothness of the path; the deviation term from the initial path , is the weight coefficient, which is used to measure the degree of deviation between the current path p and the initial path p0.

[0081] S5: Use the L-BFGS optimizer LBFGS_Optimizer as the solver, set the loss function and hard constraints constraints=[κ<0.35,P_sink<0.4], including the overturning index κ less than 0.35 and the sinking probability P_sink less than 0.4.

[0082] S6: Call the solve method of the solver to perform optimization and return the solution result.

[0083] The fifth module of MMC-LPPA constructs a hybrid loss function, trains the path planning model, and uses the trained model for path planning.

[0084] Furthermore, in order to solve the problems of frequent environmental changes, difficulty in experience transfer, and poor adaptability to new terrain in complex terrain environments, the present invention proposes a U-shaped split learning-mechanical safety co-evolution mechanism, including a U-shaped split learning architecture enhanced by physical constraints and dual-buffer training driven by mechanical experience.

[0085] like Figure 5 As shown in the figure, the physically constrained U-shaped split learning architecture adopts a dual-channel design of "physical constraint layer-learning layer" to achieve a deep integration of data-driven flexibility and physical safety determinism: The architecture achieves the coordinated optimization of data-driven and physical security through a dual-channel decision-making mechanism. In terms of physical security barriers, the system verifies the overturning index in real time at a high frequency of 200Hz. and slip rate , based on which the security boundary is clearly defined ; When the verification passes, the local gradient It can participate in U-shaped split learning aggregation, and if verification fails, it will immediately switch to a rule-based backup strategy within 10ms to ensure instantaneous response capabilities under dangerous working conditions.

[0086] U-shaped split learning aggregation uses a constrained weighted security protocol to achieve the collaborative evolution of the global model within the physical boundary. Its core formula is: ; Where, is the time step The global U-shaped segmentation model parameters (including encoder and decoder) maintained by the central server; is the learning rate, which controls the global model update step size; is the number of clients participating in the aggregation; For the The local U-shaped segmentation model gradient update amount of each client; This is a security indicator function that returns 1 for clients that pass physical security verification, and 0 otherwise. is the regularization coefficient (default value is 0.05), balancing optimization efficiency and model stability; Harmonize the function for policy consistency to suppress the tendency of local model overfitting; is the time step Time The feature extraction gradient of the local model. This formula is obtained by the current global parameter Add correction items to achieve iterative updates; This is a security aggregation item, which means that only client gradients that have passed security verification are aggregated to avoid malicious updates; To regularize the harmonic term, the local gradient direction is adjusted through the harmonic function to prevent overfitting of the local model.

[0087] like Figure 6 As shown in the figure, in the U-shaped split learning evolution layer, the diversified samples generated by the entire perception-planning-control process are stored and resampled through a double-buffered experience pool, and the uploaded updates are screened for security in combination with a physical barrier verification mechanism, thereby ensuring the stability and security of the distributed-centralized hybrid framework in multi-vehicle collaborative path planning.

[0088] The architecture simultaneously supports three dynamic update modes: periodic aggregation, event-driven response, and incremental update, and integrates the dual protection of Byzantine fault tolerance algorithm and mechanical feature watermark: Krum algorithm: Can accurately filter out malicious nodes, embedded resonant frequency Isomechanical fingerprints enable reliable tracking and verification of model integrity.

[0089] The described dual-buffer training architecture, driven by mechanical experience, uses an environmental perception module to collect real-time terrain and vehicle status data. When a risk probability greater than 0.27 is detected, path execution results are prioritized for storage in the risk buffer pool. The remaining data is pre-processed through image and laser sampling and then entered into the stable buffer pool. Regarding the buffer pool management strategy, the stable buffer pool utilizes a FIFO principle combined with an energy screening mechanism, retaining only the 20% of path samples with the lowest energy consumption. The risk buffer pool, based on a LRU algorithm combined with risk level sorting, prioritizes the 30% of event samples with the highest survival probability.

[0090] The dual buffer pools complement each other in terms of label dimension design. The stability buffer pool labels terrain types and physical parameters to support basic strategy optimization; the risk buffer pool focuses on failure cause classification to drive robust adversarial training. The training objective function is designed to be differentiated as follows: Stabilize buffer pool training to minimize movement deviation and energy consumption ; Risk buffer pool training to maximize survival probability and align with backup strategy The two buffer pools implement atomic data migration through pointer exchange, ensuring zero interruption in the training process. When the stable buffer pool is full, the system automatically triggers the buffer exchange and global aggregation process, and dynamically adjusts the loss weight based on the real-time risk event count: ; The two buffer pools form a closed-loop mechanism, which includes forward and reverse bidirectional enhancement flows. The forward flow is the terrain parameter tensor generated by the perception layer. Directly input the coupling layer to calculate the passability cost The reverse flow is the evolution layer that aggregates multiple vehicle experiences through U-shaped split learning and dynamically updates the perception layer weights. The global strategy update formula of the evolution layer is: in The verification results are fed back to the perception layer to adjust the feature extraction model.

[0091] This architecture achieves the co-evolution of security and efficiency through differentiated sample management, complementary labeling system and dynamic weight mechanism, providing a highly reliable training foundation for U-shaped segmentation learning in complex terrain.

[0092] The present invention can comprehensively consider the various characteristics of complex terrain and the mechanical properties of the vehicle itself, allowing the local path planning algorithm to better perceive different types of complex natural environments and make reasonable path planning decisions based on the actual terrain conditions and vehicle status, thereby enhancing its adaptability in complex natural environments.

[0093] Example 2 This embodiment provides a path planning system that integrates multimodal terrain perception and vehicle dynamics, including: The first module is configured to: acquire multimodal geographic data, fuse the multimodal geographic data, generate terrain physical representation, and calculate terrain complexity based on the terrain physical representation; The second module is configured to: input the terrain physical representation into the vehicle dynamics model, predict the rollover index and stability margin of the vehicle, and generate a traversability cost function based on the rollover index and stability margin; The third module is configured to: use the passability cost function as a constraint condition, select motion primitives of different orders according to the complexity of the terrain to perform trajectory fitting, and generate a vehicle path; The fourth module is configured to: execute the vehicle path, obtain the vehicle state, modify the trafficability cost function using the vehicle state, and optimize the vehicle path; The fifth module is configured to: construct a hybrid loss function, train the path planning model, and use the trained model to perform path planning.

[0094] It should be noted that the above modules correspond to the steps in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules can be executed in a computer system as part of the system.

[0095] In further embodiments, there is also provided: An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed by the processor, wherein when the computer instructions are executed by the processor, the method in embodiment 1 is performed. For the sake of brevity, no further details are given here.

[0096] It should be understood that in this embodiment, the processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), off-the-shelf field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0097] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method in embodiment 1 is completed.

[0098] The method in Example 1 can be directly executed by a hardware processor, or by a combination of hardware and software modules in the processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, it will not be described in detail here.

[0099] A computer program product includes a computer program, wherein the computer program implements the method in embodiment 1 when executed by a processor.

[0100] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions contained in program modules, which are executed in a device on a real or virtual processor of a target to perform the process / method described above. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided between program modules as needed. The machine-executable instructions for the program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0101] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the computer or other programmable data processing device, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0102] In the context of the present application, the computer program code or related data can be carried by any suitable carrier to enable the device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, and the like. Examples of signals can include electrical, optical, radio, sound or other forms of propagated signals, such as carrier waves, infrared signals, and the like.

[0103] Those skilled in the art can understand that the units and algorithm steps of the examples described in combination with the embodiments can be realized by electronic hardware or a combination of electronic hardware and computer software. Whether the functions are realized in hardware or software manner depends on the specific application and design constraints of the technical solutions. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0104] Although the specific embodiments of the present application are described above in combination with the drawings, it is not a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications or variations made by those skilled in the art without creative labor on the basis of the technical solutions of the present application are still within the scope of protection of the present application.

Claims

1. A path planning method integrating multimodal terrain perception and vehicle dynamics, characterized by: The following steps are involved: Acquire multimodal geographic data, fuse multimodal geographic data, generate terrain physical representation, and calculate terrain complexity based on terrain physical representation; The terrain physical representation is input into the vehicle dynamics model to predict the vehicle's rollover index and stability margin, and a traversability cost function is generated based on the rollover index and stability margin. Taking the trafficability cost function as a constraint, different orders of motion primitives are selected according to the terrain complexity to perform trajectory fitting and generate the vehicle path. Execute vehicle routing, obtain vehicle status, use vehicle status to modify the traffic cost function, and optimize vehicle routing; Construct a hybrid loss function, train the path planning model, and use the trained model for path planning.

2. The path planning method integrating multimodal terrain perception and vehicle dynamics according to claim 1, characterized in that: The path planning method further includes uploading the trained local model parameters to a central server, and the central server aggregates the various model parameters to form global shared parameters.

3. The path planning method integrating multimodal terrain perception and vehicle dynamics according to claim 1, wherein: Get multimodal geographic data, specifically: The camera provides texture entropy and infrared reflectivity, the lidar reconstructs the terrain elevation curvature through three-dimensional point cloud coordinates, and uses the point cloud density to infer the settlement probability. The millimeter-wave radar soil penetration detection obtains the soil dielectric constant, and combines it with the Bayesian settlement probability model to dynamically correct the ground stiffness coefficient and the multispectral vegetation analysis coverage and moisture distribution to quantify the adhesion coefficient.

4. The path planning method integrating multimodal terrain perception and vehicle dynamics according to claim 1, wherein: Calculation of the overturning index based on physical characterization of the terrain: ； in, is the terrain curvature, is the ground stiffness, For vehicle speed.

5. The path planning method integrating multimodal terrain perception and vehicle dynamics according to claim 1, wherein: Calculate the stability margin based on the physical representation of the terrain: ； in, is the rolling moment, is the pitching moment, is the critical threshold of rolling moment.

6. The path planning method integrating multimodal terrain perception and vehicle dynamics according to claim 1, wherein: The accessibility cost function is: ； in, is the ground stiffness, is the minimum safety stiffness threshold, 、 、 is the weight parameter, is the capsizing risk index, is the terrain gradient, is the Heaviside step function.

7. A path planning system that integrates multimodal terrain perception and vehicle dynamics, characterized by: include: The first module is configured to: acquire multimodal geographic data, fuse the multimodal geographic data, generate a terrain physical representation, and calculate terrain complexity based on the terrain physical representation; The second module is configured to: input the terrain physical representation into the vehicle dynamics model, predict the rollover index and stability margin of the vehicle, and generate a traversability cost function based on the rollover index and stability margin; The third module is configured to: use the passability cost function as a constraint condition, select motion primitives of different orders according to the complexity of the terrain to perform trajectory fitting, and generate a vehicle path; The fourth module is configured to: execute the vehicle path, obtain the vehicle state, modify the trafficability cost function using the vehicle state, and optimize the vehicle path; The fifth module is configured to: construct a hybrid loss function, train the path planning model, and use the trained model to perform path planning.

8. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 6 is completed.

9. A computer-readable storage medium, characterized in that Used to store computer instructions, which, when executed by a processor, complete the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The invention comprises a computer program, which is used to implement the method according to any one of claims 1 to 6 when the computer program is executed by a processor.

Images