Method for motion planning of articulated vehicle based on neural-control hierarchical hybrid planning

Abstract

The embodiment of the application provides a kind of based on neural control hierarchical hybrid planning articulated vehicle motion planning method.Application in vehicle engineering technical field, the method is by obtaining the starting pose, target pose and environmental obstacle information of vehicle;The starting pose, target pose and environmental obstacle information of vehicle are input into trained neural guide planner and are analyzed and handled, obtain first path planning result, neural guide planner includes: environment encoder and planning network;According to the feasibility verification condition of pre-established, the segmented verification result of first path planning result is verified, and the segmented verification result is obtained;According to segmented verification result, the kinematic planning controller constructed is used to optimize and adjust first path planning result, and the final path planning result is obtained, and the final path planning result includes vehicle pose information and corresponding control instruction.The method improves the planning efficiency and path feasibility, enhances the robustness and practicality under complex working conditions.

Description

Technical Field

[0001] This application relates to the field of vehicle engineering technology, and in particular to an articulated vehicle motion planning method based on neural-control hierarchical hybrid planning. Background Technology

[0002] Traditional articulated vehicle path planning mainly relies on geometric algorithms such as Rapidly-exploring Random Tree (RRT) and RRT (Optimal RRT variant), artificial potential field method, traditional geometric programming algorithms rely on pre-constructed accurate environmental geometric models, making it difficult to directly process raw point cloud and image data from sensors such as LiDAR and cameras; their grid-based discretization search mechanism suffers from an exponential explosion in computational complexity when dealing with high-dimensional, continuous configuration spaces such as articulated vehicles, resulting in insufficient real-time performance; at the same time, this type of method is based on the geometric connectivity assumption and does not fully consider the coupling characteristics of vehicle kinematics and dynamics, resulting in the generated path often being unable to be stably tracked by the underlying controller in complex maneuvers such as reversing and sharp turns; in addition, this type of algorithm has weak adaptability to sudden obstacles or dynamic changes in the environment due to the lack of an effective fast replanning mechanism.

[0003] In recent years, learning-based planning methods have gradually emerged, but most are designed for conventional vehicles, lacking specific designs for the multi-body motion characteristics and reversing stability of articulated vehicles. Therefore, there is an urgent need to propose a motion planning method for articulated vehicles to achieve safe and smooth motion planning for articulated vehicles under all operating conditions in complex environments. Summary of the Invention

[0004] This application provides an articulated vehicle motion planning method based on neural-control hierarchical hybrid planning to address the problems of traditional planning methods, such as strong dependence on environment modeling, poor real-time performance in high-dimensional spaces, infeasibility of path kinematics, and weak dynamic replanning capabilities. The method includes:

[0005] Acquire the vehicle's initial pose, target pose, and environmental obstacle information;

[0006] The vehicle's initial pose, target pose, and environmental obstacle information are input into a trained neural guidance planner for analysis and processing to obtain the first path planning result. The neural guidance planner includes an environmental encoder and a planning network.

[0007] The first path planning result is verified in segments according to the preset feasibility verification conditions, and the segment verification results are obtained.

[0008] Based on the segmented verification results, the constructed kinematic planning controller is used to optimize and adjust the first path planning result to obtain the final path planning result. The kinematic planning controller includes a dual-mode controller, which includes a forward mode and a backward mode. The final path planning result includes vehicle pose information and corresponding control commands.

[0009] Optionally, the step of inputting the vehicle's initial pose, target pose, and environmental obstacle information into a trained neural guidance planner for analysis and processing to obtain a first path planning result includes:

[0010] Environmental obstacle information is encoded using an environmental encoder to obtain an environmental feature vector;

[0011] The vehicle's initial pose, target pose, and environmental feature vectors are input into the planning network, and a bidirectional recursive search is used to determine the forward and backward paths.

[0012] When the distance between the forward path and the backward path in the state space is less than a preset threshold, the forward path and the backward path are connected to obtain the first path planning result.

[0013] Optionally, the step of performing segmented verification on the first path planning result according to preset feasibility verification conditions to obtain segmented verification results includes:

[0014] The first path planning result is divided into multiple continuous path segments;

[0015] Determine whether the change in steering angle corresponding to each path segment is less than a preset angle threshold. If so, determine that the path segment satisfies the steering angle constraint condition.

[0016] Determine whether the curvature of each path segment is less than a preset curvature threshold. If so, determine that the path segment satisfies the curvature constraint condition.

[0017] Collision detection is performed on each path segment, and the detection results are output, which are used to indicate whether there is a collision risk.

[0018] The path segment that simultaneously satisfies the steering angle constraint, the curvature constraint, and has no collision risk is considered a feasible path segment.

[0019] Path segments that do not meet the steering angle constraint and / or curvature constraint, or that pose a collision risk, are designated as infeasible path segments.

[0020] Optionally, the step of optimizing and adjusting the first path planning result using the constructed kinematic planning controller based on the segmented verification results to obtain the final path planning result includes:

[0021] For any infeasible path segment, determine the vehicle's driving direction based on the starting pose and target pose corresponding to the infeasible path segment;

[0022] When the vehicle is traveling in the forward direction, the forward mode of the dual-mode controller is invoked to output the corresponding control command, which includes the articulated angular velocity.

[0023] When the vehicle is traveling in the reverse direction, the reverse mode of the dual-mode controller is invoked to output the corresponding control command;

[0024] The motion process of the vehicle in the segment of the infeasible path is simulated using the constructed kinematic model of the articulated vehicle to obtain the corresponding vehicle pose information;

[0025] The infeasible path segments are adjusted according to the control commands and vehicle pose information to obtain the adjusted path segmentation planning results.

[0026] The adjusted path segmentation planning results and feasible path segmentation are integrated to obtain the final path planning result.

[0027] Optionally, when the vehicle's direction of travel is forward, the step of calling the forward mode of the dual-mode controller to output the corresponding control command includes:

[0028] Define the reference point as ;

[0029] The front of the vehicle is in the following condition: ;

[0030] Reference point The error angle between the front vehicle body and the front vehicle body state is: ;

[0031] in, Let x be the x-coordinate of the midpoint of the front axle. Let be the ordinate of the midpoint of the front axle. The x-coordinate of the reference point, The ordinate of the reference point, For the front axle heading angle;

[0032] The desired hinge angle for forward mode is: ;

[0033] in, For articulated steering angle, The distance between the front axle and the hinge point. The distance between the rear axle and the hinge point. The forward-looking distance of the forward mode;

[0034] The articulated angular velocity in forward mode is: ;

[0035] in, This is the proportional gain.

[0036] Optionally, when the vehicle's driving direction is reverse, the step of calling the reverse mode of the dual-mode controller to output the corresponding control command includes:

[0037] Define the reference point as ;

[0038] The rear of the vehicle is in the following condition: ;

[0039] ;

[0040] ;

[0041] ;

[0042] in, The rear axle heading angle, For the front axle heading angle, For articulated steering angle, Let x be the x-coordinate of the midpoint of the rear axle. The ordinate of the midpoint of the rear axle is y. Let x be the x-coordinate of the midpoint of the front axle. Let be the ordinate of the midpoint of the front axle. The distance between the front axle and the hinge point. This is the distance between the rear axle and the hinge point.

[0043] Reference point The error angle between the rear vehicle body and the front vehicle body is: ;

[0044] The expected hinge angle in back mode is: ;

[0045] in, The forward distance for the backward mode;

[0046] The hinge angular velocity in backward mode is: ;

[0047] in, This is the optimal feedback gain for LQR.

[0048] Optionally, the motion process of the vehicle in the segment of the infeasible path is simulated using the constructed articulated vehicle kinematic model to obtain the corresponding vehicle pose information, which satisfies the following formula:

[0049]

[0050] in, The lateral component of the velocity at the midpoint of the front axle. The longitudinal component of the velocity at the midpoint of the front axle. The front axle angular velocity, For hinge angular velocity, For the front axle heading angle, For articulated steering angle, For the front frame speed, The distance between the front axle and the hinge point. The distance between the rear axle and the hinge point. This refers to the speed of the entire vehicle.

[0051] Optionally, the loss function used by the neural guided planner during training satisfies the following formula:

[0052]

[0053] in, This is the total loss function value. The weighting coefficients for location loss are... The position vector predicted by the neural network. The true position vector. The weighting coefficient for the heading angle loss. The value of the angle loss function. The weighting coefficient for the hinge angle loss. For the front axle heading angle, This is the articulated steering angle.

[0054] This application provides an articulated vehicle motion planning method based on neural-control hierarchical hybrid planning. The method acquires the vehicle's initial pose, target pose, and environmental obstacle information; inputs these information into a trained neural guided planner for analysis and processing to obtain a first path planning result. The neural guided planner includes an environmental encoder and a planning network. The first path planning result is then segmented for verification according to preset feasibility verification conditions, yielding segmented verification results. Based on the segmented verification results, a constructed kinematic planning controller is used to optimize and adjust the first path planning result, resulting in a final path planning result. The kinematic planning controller includes a dual-mode controller with forward and backward modes. The final path planning result includes vehicle pose information and corresponding control commands. This method improves planning efficiency and path feasibility, and enhances robustness and practicality under complex conditions. Attached Figure Description

[0055] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0056] Figure 1 This is a flowchart illustrating the articulated vehicle motion planning method based on neural-control hierarchical hybrid planning provided in this application embodiment. Figure 1 ;

[0057] Figure 2 This is a flowchart illustrating the articulated vehicle motion planning method based on neural-control hierarchical hybrid planning provided in this application embodiment. Figure 2 ;

[0058] Figure 3 This is a schematic diagram of the bidirectional recursive search process provided in the embodiments of this application;

[0059] Figure 4 This is a schematic diagram of the steady-state steering model of the articulated vehicle provided in the embodiments of this application;

[0060] Figure 5 This is a schematic diagram of an articulated vehicle turning in place model provided in an embodiment of this application;

[0061] Figure 6 This is a preview image of the result provided in the embodiment of this application;

[0062] Figure 7 This is the open-loop response diagram of forward motion provided in the embodiments of this application;

[0063] Figure 8 This is the open-loop response diagram of backward motion provided in the embodiments of this application;

[0064] Figure 9 This is a diagram of axial velocity and desired angle provided in an embodiment of this application;

[0065] Figure 10 This is an LQR gain and desired angle diagram provided in an embodiment of this application;

[0066] Figure 11 This is a schematic diagram of the start and end point configuration provided in the embodiments of this application;

[0067] Figure 12 This is a visualization of the planned path provided in the embodiments of this application.

[0068] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0070] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein.

[0071] In this application, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0072] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0073] Figure 1 This is a flowchart illustrating the articulated vehicle motion planning method based on neural-control hierarchical hybrid planning provided in this application embodiment. Figure 1 . Figure 2 This is a flowchart illustrating the articulated vehicle motion planning method based on neural-control hierarchical hybrid planning provided in this application embodiment. Figure 2 .like Figure 1 , Figure 2 As shown, the articulated vehicle motion planning method based on neural-control hierarchical hybrid planning provided in this embodiment includes:

[0074] S1: Acquire the vehicle's initial pose, target pose, and environmental obstacle information.

[0075] The vehicle's initial pose refers to its initial position and attitude when starting a task or planning a path, including the coordinates of the initial position and the angles of the initial attitude. The target pose is the final position and attitude the vehicle intends to reach, including the coordinates of the target position and the angles of the target attitude, clearly defining the specific state the vehicle should be in after completing the task. Environmental obstacle information encompasses information about all objects in the vehicle's environment that may obstruct or affect its movement, including the obstacle's location, shape, size, and type.

[0076] S2: Input the vehicle's initial pose, target pose, and environmental obstacle information into the trained neural guidance planner for analysis and processing to obtain the first path planning result.

[0077] The neural guidance planner can employ an MPNet (Motion Planning Networks) architecture. This architecture consists of an environmental encoder and a planning network. The neural guidance planner is a model built on an artificial neural network, trained with a large amount of data containing the vehicle's initial pose, target pose, environmental obstacle information, and corresponding reasonable paths. During training, the network continuously adjusts its internal parameters to learn the mapping relationship between input information and reasonable path planning results, thus enabling it to perform path planning analysis and processing on new input information. The first path planning result is a preliminary path planning scheme for the vehicle from the starting point to the target point, including the coordinate sequence of the path and the driving direction, output by the neural guidance planner after receiving the vehicle's initial pose, target pose, and environmental obstacle information as input, and undergoing complex internal calculations and analysis.

[0078] The training process of a neural guided planner includes: inputting preprocessed data into the neural network, using optimization algorithms (such as stochastic gradient descent, Adam, etc.) to continuously adjust the network parameters so that the loss function value gradually decreases until the network converges, that is, the difference between the network's output and the actual result reaches an acceptable range.

[0079] Specifically, the loss function used during training of the neural guided planner satisfies the following formula:

[0080]

[0081] in, This is the total loss function value. The weighting coefficients for location loss are... The position vector predicted by the neural network. The true position vector. The weighting coefficient for the heading angle loss. The value of the angle loss function. The weighting coefficient for the hinge angle loss. For the front axle heading angle, For articulated steering angle; angle loss is calculated using cosine distance: , The angle value predicted by the neural network. This represents the actual angle value.

[0082] Specifically, the vehicle's initial pose, target pose, and environmental obstacle information are input into a trained neural guidance planner for analysis and processing to obtain the first path planning result, including:

[0083] The environmental obstacle information is encoded using an environmental encoder to obtain an environmental feature vector, which is represented as follows:

[0084]

[0085] in, For environmental feature vectors, Indicates an environmental encoder. Indicates information about environmental obstacles. Indicates the encoding dimension.

[0086] Furthermore, the vehicle's initial pose, target pose, and environmental feature vectors are input into the planning network, and a bidirectional recursive search is used to determine the forward and backward paths.

[0087] Figure 3This is a flowchart illustrating the bidirectional recursive search process provided in this application embodiment. Bidirectional recursive search is a path generation method that simultaneously explores from the starting point to the target and from the target to the starting point. Bidirectional recursive search is the core mechanism of the neural-guided planner in generating initial paths. Its logic is as follows: the planner starts from both the starting point and the ending point simultaneously, and uses recursive neural networks to predict the state respectively. The forward network takes the current state, the target state, and the environmental feature vector as input, and recursively predicts the next path point, constructing a forward path extending from the starting point to the ending point. The backward network recursively predicts from the ending point to the starting point in a symmetrical manner, constructing a backward path. The two search directions advance synchronously, and each prediction incorporates environmental obstacle information to ensure the path's obstacle avoidance rationality. When the distance between the two paths in the state space is less than a preset threshold, the system determines that the two paths have successfully met, and splices the forward path and the backward path in reverse order to form a complete initial path from the starting point to the ending point. If they fail to meet within the maximum number of steps, the unconnected segments are marked for repair and handed over to the subsequent kinematic planning controller for processing. This bidirectional synchronous exploration mechanism significantly improves path search efficiency and makes the generated initial path more balanced and reasonable in terms of topology. In the neural network implementation, the bidirectional recursive search is performed by two recurrent neural networks (RNNs) that perform forward and backward predictions respectively, and the two networks share the encoding of environmental features.

[0088] Recurrent neural networks (RNNs) are a type of neural network with a recurrent connection structure. Their core characteristic is the ability to memorize historical input information through internal states. This allows them to influence the current output with information from previous moments when processing sequential data (such as time series or path point sequences), achieving sequence modeling with memory. The planning network, on the other hand, is the core functional module of a neural guided planner. It is typically implemented using a recurrent neural network. This network takes the current state, the target state, and environmental features as input and recursively predicts the next state in the path, progressively forming a complete path. Specifically, the planning network contains two branches: a forward branch and a backward branch. These branches simultaneously and recursively predict path points from the starting point to the target and from the target to the starting point, respectively. When the predicted paths from both directions meet in the state space, they are joined to form a preliminary path, thus achieving efficient and global guidance from the initial pose to the target pose.

[0089] Forward path , In order to plan the network, This is the initial state. For the target state, the backward path , , Let x be the x-coordinate of the midpoint of the front axle. Let be the ordinate of the midpoint of the front axle. For the front axle heading angle, For articulated steering angle, Used to indicate the direction of travel (+1 forward, -1 reverse). This represents the cumulative driving distance.

[0090] Furthermore, when the distance between the forward path and the backward path in the state space is less than a preset threshold, connecting the forward path and the backward path yields the first path planning result, expressed as:

[0091]

[0092] in, This is the result of the first path planning. Indicates a connection. Indicates back mode.

[0093] In an optional embodiment, the neural-guided planner is designed for the SE(2) state space [x|y|θ], expanded to accommodate the articulated vehicle state [x]. f |y f |θ f |γ];

[0094] State transition function:

[0095]

[0096] Inverse transform:

[0097]

[0098] in, The true x-coordinate of the midpoint of the front axle, position The true ordinate of the midpoint of the front axle. The predicted x-coordinate of the midpoint of the front axle. The predicted ordinate of the midpoint of the front axle. Let cosine be the angle of the vehicle's front body. Let be the sine of the front vehicle's heading angle. The cosine value of the articulated steering angle. This is the sine value of the articulated steering angle.

[0099] S3: Perform segmented verification of the first path planning result based on the preset feasibility verification conditions to obtain the segmented verification result.

[0100] Specifically, the first path planning result is divided into multiple continuous path segments; it is determined whether the change in steering angle corresponding to each path segment is less than a preset angle threshold. If so, the path segment is deemed to satisfy the steering angle constraint; it is determined whether the curvature corresponding to each path segment is less than a preset curvature threshold. If so, the path segment is deemed to satisfy the curvature constraint; collision detection is performed on each path segment, and the detection result is output, which indicates whether there is a collision risk; path segments that simultaneously satisfy the steering angle constraint, curvature constraint, and have no collision risk are designated as feasible path segments; path segments that do not satisfy the steering angle constraint and / or curvature constraint and have a collision risk are designated as infeasible path segments.

[0101] For example, outputting the path of the neural guidance planner It is divided into (M-1) continuous segments.

[0102] The feasibility function of a segment is expressed as:

[0103] Feasibility assessment criteria include: steering angle constraints: , Preset angle threshold; curvature constraint: , The preset curvature threshold is used; check whether the adjacent straight line connection collides with the obstacle. If so, search for a feasible obstacle avoidance path for the segment.

[0104] S4: Based on the segmented verification results, the constructed kinematic planning controller is used to optimize and adjust the first path planning result to obtain the final path planning result.

[0105] The final path planning result includes: vehicle pose information, corresponding control commands, and timestamps.

[0106] Specifically, for any infeasible path segment, the vehicle's direction of travel is determined based on the initial and target poses corresponding to the infeasible path segment. When the vehicle's direction of travel is forward, the forward mode of the dual-mode controller is invoked to output the corresponding control commands, which include speed and articulation angular velocity. When the vehicle's direction of travel is backward, the backward mode of the dual-mode controller is invoked to output the corresponding control commands. The motion process of the vehicle in the infeasible path segment is simulated using the constructed articulated vehicle kinematic model to obtain the corresponding vehicle pose information. The infeasible path segment is adjusted based on the control commands and vehicle pose information to obtain the adjusted path segment planning result. The adjusted path segment planning result is then fused with the feasible path segment to obtain the final path planning result.

[0107] Specifically, the kinematic model of the articulated vehicle can be decomposed into a steady-state steering motion model and a stationary steering motion model.

[0108] Figure 4 This is a schematic diagram of the steady-state steering model of the articulated vehicle provided in an embodiment of this application. Where O is the instantaneous center of gravity. and These are the midpoints of the front and rear axles, respectively. and These are the distances between the front and rear axles and the hinge point, respectively. and The heading angle is the angle between the front and rear of the vehicle body. This is the articulated steering angle. It is based on the midpoint of the front axle of the articulated vehicle. This serves as a reference point for the overall vehicle status. The speed at this point is consistent with the forward direction of the articulated vehicle, which is beneficial for analysis and calculation.

[0109] Define the vehicle speed for: , Refers to the speed of the front frame;

[0110] Front axle midpoint Speed ​​is expressed as: , The lateral component of the velocity at the midpoint of the front axle. The longitudinal component of the velocity at the midpoint of the front axle. For the front axle heading angle;

[0111] The rate of change of the front axle heading angle, i.e., the front axle angular velocity, is: , The distance between the front axle and the hinge point. The distance between the rear axle and the hinge point. For articulated steering angle;

[0112] The position and orientation of the front axle during the steady-state steering process of an articulated vehicle can be represented as:

[0113]

[0114] in, This refers to the hinge angular velocity.

[0115] Figure 5 This is a schematic diagram of an articulated vehicle model for stationary steering provided in an embodiment of this application. Due to the midpoint of the front axle... As the reference point for the overall vehicle status, it is assumed that this point is stationary relative to the ground during the stationary turning process.

[0116] Then the midpoint of the front bridge The velocity components in the XY plane are: ;

[0117] The rate of change of heading angle is: ;

[0118] The pose state of the front axle during the steady-state steering process of an articulated vehicle is represented as follows:

[0119]

[0120] Combining the steady-state steering model and the stationary steering model, the articulated vehicle's position and posture can be represented as follows:

[0121]

[0122] Figure 7 This is an open-loop response diagram of forward motion provided in an embodiment of this application, showing the curve of the change of the internal angle between the trailer and the tractor over time during forward motion. Figure 9 This is a diagram of axial velocity and desired angle provided in an embodiment of this application, illustrating the relationship between the vehicle's axial velocity and internal angle.

[0123] Understandably, the forward mode of the dual-mode controller employs a "pure tracking" control algorithm. This algorithm simulates the anti-aiming behavior of human drivers, directly calculating smooth steering commands based on the geometric relationship between the vehicle's current position and the target point ahead, allowing the vehicle to travel along a smooth curve towards the target.

[0124] Specifically, the reference point is defined as ;

[0125] The front of the vehicle is in the following condition: ;

[0126] Reference point The error angle between the front vehicle body and the front vehicle body state is: ;

[0127] in, Let x be the x-coordinate of the midpoint of the front axle. Let be the ordinate of the midpoint of the front axle. The x-coordinate of the reference point, The ordinate of the reference point, For the front axle heading angle;

[0128] The desired hinge angle for forward mode is: ;

[0129] in, For articulated steering angle, The distance between the front axle and the hinge point. The distance between the rear axle and the hinge point. The forward-looking distance of the forward mode;

[0130] The articulated angular velocity in forward mode is: ;

[0131] in, For proportional gain: ;

[0132] Figure 8 This is an open-loop response diagram of backward motion provided in an embodiment of this application, showing the curve of the internal angle changing with time during backward motion.

[0133] Understandably, in reverse mode, in addition to aiming at the target position, the controller stabilizes the articulation angle dynamics using a linear quadratic regulator (LQR). Controller parameters are automatically adjusted and gain scheduled according to different steering amplitudes. Figure 10 This is an LQR gain and desired angle diagram provided in the embodiments of this application, which shows the curve of LQ feedback gain changing with the desired steering angle. It is used for gain scheduling to ensure the stability of the internal angle during reverse motion.

[0134] Specifically, the reference point is defined as ;

[0135] The rear of the vehicle is in the following condition: ;

[0136]

[0137]

[0138]

[0139] in, The rear axle heading angle, For the front axle heading angle, For articulated steering angle, Let x be the x-coordinate of the midpoint of the rear axle. The ordinate of the midpoint of the rear axle is y. Let x be the x-coordinate of the midpoint of the front axle. Let be the ordinate of the midpoint of the front axle. The distance between the front axle and the hinge point. This is the distance between the rear axle and the hinge point.

[0140] Reference point The error angle between the rear vehicle body and the front vehicle body is:

[0141] The expected hinge angle in back mode is:

[0142] in, The forward distance for the backward mode;

[0143] The hinge angular velocity in backward mode is: ;

[0144] in, The optimal feedback gain for LQR is obtained by solving the algebraic Riccati equation, and is usually based on the desired hinge angle. Perform gain scheduling, using different gain values ​​for different steering amplitudes, and store the results in a lookup table.

[0145] In an optional embodiment, the method further includes employing a three-level replanning mechanism.

[0146] Level 1: Minor adjustments, modifying neural pathway point locations to maintain topological structure.

[0147]

[0148] in, Refers to the adjusted path points. Points to be adjusted Let be the perturbation vector. This refers to the maximum disturbance amplitude.

[0149] Level 2: Local replanning, replanning between the start and end points of infeasible segments:

[0150]

[0151] in, Refers to the path segment after local adjustments. Refers to space with obstacles.

[0152] Level 3: Global backup, fallback to classic planner:

[0153]

[0154] in, Refers to the path after global adjustment.

[0155] The three adjustment levels are determined step by step according to the repair cost: First, try level 1, which applies a small perturbation to the path points; if it is ineffective, upgrade to level 2, which calls the kinematic planning controller to perform local replanning on the infeasible segments; if it still fails, activate level 3, which falls back to the classic planner for global planning.

[0156] Figure 11 This is a schematic diagram of the start and end point configuration provided in the embodiments of this application. Figure 12This is a visualization of the planned path provided in an embodiment of this application. In an optional embodiment, after the user sets the vehicle's initial and target poses, the system activates the constructed neural-control hierarchical hybrid planner. The neural-guided planner quickly generates a global coarse path. Then, the system calls the underlying kinematic planner to verify and precisely execute each connection of this coarse path. If a path proposed by the neural-guided planner is impractical, the kinematic planner immediately takes over locally, using its kinematic model and controller to replan a feasible alternative path. Finally, a complete motion trajectory from the starting point to the ending point is output. This trajectory is not only globally reasonable, but each segment also includes precise control commands such as speed and steering angle, as well as timing information, which can be directly issued to the vehicle's underlying control system for execution. Through direct suggestions from the neural network, the kinematic planner verifies and repairs in a collaborative mode, balancing the speed and reliability of the planning.

[0157] This application provides an articulated vehicle motion planning method based on neural-control hierarchical hybrid planning. The method acquires the vehicle's initial pose, target pose, and environmental obstacle information; inputs these information into a trained neural guided planner for analysis and processing to obtain a first path planning result. The neural guided planner includes an environmental encoder and a planning network. The first path planning result is then segmented and verified according to preset feasibility verification conditions to obtain segmented verification results. Based on the segmented verification results, a constructed kinematic planning controller is used to optimize and adjust the first path planning result to obtain the final path planning result. The kinematic planning controller includes a dual-mode controller with a forward mode and a backward mode. The final path planning result includes vehicle pose information and corresponding control commands. This method improves planning efficiency and path feasibility, and enhances robustness and practicality under complex conditions.

[0158] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0159] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. An articulated vehicle motion planning method based on neural-control hierarchical hybrid planning, characterized in that, The method includes: Acquire the vehicle's initial pose, target pose, and environmental obstacle information; The vehicle's initial pose, target pose, and environmental obstacle information are input into a trained neural guidance planner for analysis and processing to obtain the first path planning result. The neural guidance planner includes an environmental encoder and a planning network. The first path planning result is verified in segments according to the preset feasibility verification conditions, and the segment verification results are obtained. Based on the segmented verification results, the constructed kinematic planning controller is used to optimize and adjust the first path planning result to obtain the final path planning result. The kinematic planning controller includes a dual-mode controller, which includes a forward mode and a backward mode. The final path planning result includes vehicle pose information and corresponding control commands.

2. The method according to claim 1, characterized in that, The process involves inputting the vehicle's initial pose, target pose, and environmental obstacle information into a trained neural guidance planner for analysis and processing to obtain the first path planning result, including: Environmental obstacle information is encoded using an environmental encoder to obtain an environmental feature vector; The vehicle's initial pose, target pose, and environmental feature vectors are input into the planning network, and a bidirectional recursive search is used to determine the forward and backward paths. When the distance between the forward path and the backward path in the state space is less than a preset threshold, the forward path and the backward path are connected to obtain the first path planning result.

3. The method according to claim 1, characterized in that, The step of performing segmented verification on the first path planning result according to preset feasibility verification conditions to obtain segmented verification results includes: The first path planning result is divided into multiple continuous path segments; Determine whether the change in steering angle corresponding to each path segment is less than a preset angle threshold. If so, determine that the path segment satisfies the steering angle constraint condition. Determine whether the curvature of each path segment is less than a preset curvature threshold. If so, determine that the path segment satisfies the curvature constraint condition. Collision detection is performed on each path segment, and the detection results are output, which are used to indicate whether there is a collision risk. The path segment that simultaneously satisfies the steering angle constraint, the curvature constraint, and has no collision risk is considered a feasible path segment. Path segments that do not meet the steering angle constraint and / or curvature constraint, or that pose a collision risk, are designated as infeasible path segments.

4. The method according to claim 3, characterized in that, The step of optimizing and adjusting the first path planning result using the constructed kinematic planning controller based on the segmented verification results to obtain the final path planning result includes: For any infeasible path segment, determine the vehicle's driving direction based on the starting pose and target pose corresponding to the infeasible path segment; When the vehicle is traveling in the forward direction, the forward mode of the dual-mode controller is invoked to output the corresponding control command, which includes the articulated angular velocity. When the vehicle is traveling in the reverse direction, the reverse mode of the dual-mode controller is invoked to output the corresponding control command; The motion process of the vehicle in the segment of the infeasible path is simulated using the constructed kinematic model of the articulated vehicle to obtain the corresponding vehicle pose information; The infeasible path segments are adjusted according to the control commands and vehicle pose information to obtain the adjusted path segmentation planning results. The adjusted path segmentation planning results and feasible path segmentation are integrated to obtain the final path planning result.

5. The method according to claim 4, characterized in that, When the vehicle's direction of travel is forward, the method of calling the forward mode of the dual-mode controller to output the corresponding control command includes: Define the reference point as ; The front of the vehicle is in the following condition: ; Reference point The error angle between the front vehicle body and the front vehicle body state is: ; in, Let x be the x-coordinate of the midpoint of the front axle. Let be the ordinate of the midpoint of the front axle. The x-coordinate of the reference point, The ordinate of the reference point, For the front axle heading angle; The desired hinge angle for forward mode is: ; in, For articulated steering angle, The distance between the front axle and the hinge point. The distance between the rear axle and the hinge point. The forward-looking distance of the forward mode; The articulated angular velocity in forward mode is: ; in, This is the proportional gain.

6. The method according to claim 4, characterized in that, When the vehicle's direction of travel is reverse, the reverse mode of the dual-mode controller is invoked to output the corresponding control command, including: Define the reference point as ; The rear of the vehicle is in the following condition: ; ； ； ； in, The rear axle heading angle, For the front axle heading angle, For articulated steering angle, Let x be the x-coordinate of the midpoint of the rear axle. The ordinate of the midpoint of the rear axle is y. Let x be the x-coordinate of the midpoint of the front axle. Let be the ordinate of the midpoint of the front axle. The distance between the front axle and the hinge point. This is the distance between the rear axle and the hinge point. Reference point The error angle between the rear vehicle body and the front vehicle body is: ; The expected hinge angle in back mode is: ; in, The forward distance for the backward mode; The hinge angular velocity in backward mode is: ; in, This is the optimal feedback gain for LQR.

7. The method according to claim 4, characterized in that, The kinematic model of the articulated vehicle is used to simulate the motion process of the vehicle in the segment of the infeasible path, and the corresponding vehicle pose information is obtained, which satisfies the following formula: in, The lateral component of the velocity at the midpoint of the front axle. The longitudinal component of the velocity at the midpoint of the front axle. The front axle angular velocity, For hinge angular velocity, For the front axle heading angle, For articulated steering angle, For the front frame speed, The distance between the front axle and the hinge point. The distance between the rear axle and the hinge point. This refers to the speed of the entire vehicle.

8. The method according to claim 1, characterized in that, The loss function used during training of the neural guided planner satisfies the following formula: in, This is the total loss function value. These are the weighting coefficients for the location loss. The position vector predicted by the neural network. The true position vector. The weighting coefficient for the heading angle loss. The value of the angle loss function. This is the weighting coefficient for the hinge angle loss. For the front axle heading angle, This is the articulated steering angle.

Images