Agricultural machinery complex scene dynamic obstacle avoidance path planning method and system for coping with unstructured obstacles

Abstract

The invention relates to the technical field of intelligent agriculture, in particular to an agricultural machinery complex scene dynamic obstacle avoidance path planning method and system for dealing with unstructured obstacles, and the method comprises the steps: constructing a multi-target traveling salesman problem model, solving an optimal operation row sequence, and outputting a global operation path; processing the environment sensing data through a convolutional neural network, generating a rasterized map of an unstructured obstacle, and outputting an initial obstacle avoidance reference path; establishing an optimal control problem including an agricultural machine kinematics constraint and an obstacle distance constraint; and discretizing the optimal control problem into a nonlinear programming problem, and performing numerical optimization solution by taking the initial path of the convolutional neural network as a solution starting point to obtain an optimal control sequence and a state trajectory. According to the invention, the problem that a traditional path planning method is complex in calculation, low in precision and incapable of avoiding obstacles in real time when dealing with unstructured obstacles is effectively solved, and the precision and efficiency of path planning are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent agriculture technology, and in particular to a method and system for dynamic obstacle avoidance path planning in complex scenarios involving agricultural machinery and dealing with unstructured obstacles. Background Technology

[0002] In modern agricultural operations, automated agricultural machinery systems face complex working environments, especially with the presence of unstructured obstacles. Traditional path planning methods struggle to effectively address these dynamic and complex obstacles. Existing path planning techniques, such as graph search-based A* algorithms or Dijkstra's algorithms, are primarily suitable for simple, structured environments. However, when faced with dynamic, unstructured obstacles, the computational complexity of these methods increases dramatically, and they may even fail to find a feasible path solution.

[0003] Especially in farmland environments, the types, shapes, and locations of obstacles are constantly changing, making it difficult for traditional path planning methods to update the work path in real time. This can lead to collisions or inefficiencies during agricultural machinery operations. Furthermore, environmental perception technologies are often affected by signal interference or obstruction, resulting in information loss or inaccuracies, further complicating path planning. Summary of the Invention

[0004] This invention provides a dynamic obstacle avoidance path planning method for agricultural machinery in complex scenarios with unstructured obstacles, which can effectively solve the problems in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A dynamic obstacle avoidance path planning method for agricultural machinery in complex scenarios involving unstructured obstacles, the method comprising: Based on farmland boundaries and the number of operation rows, a multi-objective traveling salesman problem model is constructed, and the optimal operation row sequence is solved to output the global operation path; The system processes environmental perception data through a convolutional neural network, generates a raster map of unstructured obstacles, and outputs an initial obstacle avoidance reference path. Based on the global operation path and the initial obstacle avoidance reference path, an optimal control problem is established that includes agricultural machinery kinematic constraints and obstacle distance constraints. The optimal control problem is discretized into a nonlinear programming problem, and numerical optimization is performed using the initial path of the convolutional neural network as the starting point to obtain the optimal control sequence and state trajectory.

[0006] Furthermore, the multi-objective traveling salesman problem model uses work lines as nodes and turning paths as edges, with the optimization objective being to minimize the total turning distance.

[0007] Furthermore, environmental perception data is processed through a convolutional neural network to generate a rasterized map of unstructured obstacles and output an initial obstacle avoidance reference path, including: Radar point cloud data and visual image data are obtained from the environmental perception data and mapped to the same farmland coordinate system through spatiotemporal registration. The registered data is input into the convolutional neural network, which outputs a semantic grid map containing obstacle outlines and category labels. Morphological dilation is performed on the obstacle contours in the semantic grid map to generate an envelope boundary with a preset safety margin. Key points of the collision-free path are extracted outside the safety envelope boundary, and the initial obstacle avoidance reference path is fitted in the farmland coordinate system based on agricultural machinery kinematic constraints.

[0008] Furthermore, based on the global operation path and the initial obstacle avoidance reference path, an optimal control problem is established that includes agricultural machinery kinematic constraints and obstacle distance constraints, including: The global operation path and the initial obstacle avoidance reference path are aligned in coordinates and fused into a cubic spline curve with arc length as the parameter as the optimization reference trajectory. Based on the structure of agricultural machinery, a nonlinear mapping relationship between the front wheel steering angle and the path curvature is established, and an upper limit for the rate of change of curvature is set. Based on the safety envelope boundary of the rasterized map, a symbolic distance function is established, and the Euclidean distance from the trajectory point to the nearest obstacle boundary is calculated in real time to obtain the dynamic distance field. By integrating curvature variation constraints with the dynamic distance field, an objective function is constructed that includes minimizing trajectory offset, maximizing driving smoothness, and adaptive obstacle avoidance.

[0009] Furthermore, the obstacle distance constraint is transformed into an objective term through a penalty function.

[0010] Furthermore, the optimal control problem is discretized into a nonlinear programming problem using the fully simultaneous orthogonal collocation finite element method.

[0011] Furthermore, the optimal control problem is discretized into a nonlinear programming problem, and numerical optimization is performed using the initial path of the convolutional neural network as the starting point to obtain the optimal control sequence and state trajectory, including: Based on the kinematic constraints, the discretization step size is determined, and the optimal control problem is transformed into a finite-dimensional nonlinear programming problem. The initial obstacle avoidance reference path is mapped to an initial guessed trajectory in a discrete state space; Numerical optimization is performed starting from the initial guessed trajectory, wherein the kinematic constraints are transformed into path curvature change rate constraints; When the optimal solution simultaneously satisfies the physical limits of the agricultural machinery steering mechanism and the real-time obstacle avoidance safety threshold, the optimal control sequence and state trajectory are output.

[0012] Furthermore, the method also includes: real-time monitoring of obstacle positions during trajectory execution; when a new obstacle intrudes into the path safety zone or the distance between the planned trajectory and the obstacle is lower than a safety threshold, updating the raster map and re-triggering trajectory optimization.

[0013] A dynamic obstacle avoidance path planning system for agricultural machinery in complex scenarios involving unstructured obstacles, the system comprising: The global operation path output module constructs a multi-objective traveling salesman problem model based on farmland boundaries and the number of operation rows, solves the optimal operation row sequence, and outputs the global operation path; The initial obstacle avoidance path output module processes environmental perception data through a convolutional neural network, generates a raster map of unstructured obstacles, and outputs an initial obstacle avoidance reference path. The optimal control problem construction module establishes an optimal control problem that includes agricultural machinery kinematic constraints and obstacle distance constraints based on the global operation path and the initial obstacle avoidance reference path. The optimal control trajectory generation module discretizes the optimal control problem into a nonlinear programming problem, and uses the initial path of the convolutional neural network as the starting point to perform numerical optimization to obtain the optimal control sequence and state trajectory.

[0014] Furthermore, the initial obstacle avoidance path output module includes: The data spatiotemporal registration unit acquires radar point cloud data and visual image data from the environmental perception data, and maps them to the same farmland coordinate system through spatiotemporal registration; The grid map output unit inputs the registered data into the convolutional neural network and outputs a semantic grid map containing obstacle outlines and category labels. The envelope boundary generation unit performs morphological dilation on the obstacle contours in the semantic grid map to generate an envelope boundary with a preset safety margin. The obstacle avoidance path fitting unit extracts key points of the collision-free path outside the safety envelope boundary and fits the initial obstacle avoidance reference path in the farmland coordinate system based on agricultural machinery kinematic constraints.

[0015] The technical solution of this invention can achieve the following technical effects: This effectively solves the problems of traditional path planning methods being computationally complex, having low accuracy, and being unable to avoid obstacles in real time when dealing with unstructured obstacles, thus improving the accuracy and efficiency of path planning.

[0016] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a dynamic obstacle avoidance path planning method for complex agricultural machinery scenarios involving unstructured obstacles; Figure 2 A flowchart illustrating the process of generating a rasterized map of unstructured obstacles and outputting an initial obstacle avoidance reference path; Figure 3 A flowchart illustrating the optimal control problem that includes kinematic constraints of agricultural machinery and distance constraints of obstacles; Figure 4 A flowchart illustrating the process of obtaining the optimal control sequence and state trajectory. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] Example 1: like Figure 1 As shown, this application provides a dynamic obstacle avoidance path planning method for agricultural machinery in complex scenarios involving unstructured obstacles. The method includes: S1: Based on farmland boundaries and the number of operation rows, construct a multi-objective traveling salesman problem model, solve for the optimal operation row sequence, and output the global operation path; Specifically, in this embodiment, firstly, farmland boundary data and the number of work rows are acquired through a high-precision farmland boundary recognition system and a work row number calculation module. This information is used to determine the geometry of the work area and the number of work rows. In practice, the system defines the farmland boundary as a polygon and uses this boundary information to divide the farmland into several work rows. The number of work rows is determined by the total area of ​​the farmland and the width of the work rows. Next, using this work row information, a multi-objective traveling salesman problem model is constructed. To solve this multi-objective optimization problem, heuristic algorithms, such as genetic algorithms or particle swarm optimization algorithms, can be used to search for the optimal work row sequence. Finally, a global work path is output, which is the connection path of the optimal work row sequence, ensuring that the agricultural machinery performs the work task in the optimal way during the operation.

[0022] S2: Process environmental perception data through a convolutional neural network to generate a raster map of unstructured obstacles and output an initial obstacle avoidance reference path; Specifically, environmental sensing devices are used to acquire obstacle data in farmland. To ensure the accuracy and real-time performance of path planning, all sensing data is mapped to a unified farmland coordinate system through spatiotemporal registration technology. Through spatiotemporal registration, the system can fuse data from different sensors, thereby ensuring comprehensive acquisition of obstacle information. Next, a convolutional neural network is used to train and recognize the processed sensing data. The CNN can identify different types of obstacles in the farmland, such as trees, rocks, and bushes, and generate a raster map containing the outlines of the obstacles. These raster maps not only represent the location of the obstacles but also label the category and morphological characteristics of each obstacle. Based on this, the system further outputs an initial obstacle avoidance reference path. This path is based on the generated raster map and takes into account the distribution and morphological characteristics of the obstacles to initially plan an obstacle avoidance path. Although this path is not the final optimized path, as an initial reference path, it can effectively avoid obvious obstacles and provide a starting point for subsequent path optimization.

[0023] S3: Based on the global operation path and the initial obstacle avoidance reference path, establish an optimal control problem that includes agricultural machinery kinematic constraints and obstacle distance constraints; Specifically, the global operation path is aligned with the initial obstacle avoidance reference path to ensure a smooth path that meets operational requirements. Next, considering the kinematic characteristics of the agricultural machinery, the system plans the path based on its physical characteristics such as steering angle and speed. Simultaneously, to ensure path safety, the distance from each path point to the nearest obstacle needs to be calculated in real time. This ensures the path maintains an appropriate safe distance near obstacles; this process is based on a gridded map of obstacles and is calculated in real time to ensure that each path avoids collisions with obstacles. By combining these constraints, an optimal control problem can be constructed, providing clear objectives and constraints for subsequent path optimization and trajectory adjustment.

[0024] S4: Discretize the optimal control problem into a nonlinear programming problem, and use the initial path of the convolutional neural network as the starting point for numerical optimization to obtain the optimal control sequence and state trajectory.

[0025] Specifically, the optimal control problem is discretized into a nonlinear programming problem. By determining the discretization step size, the continuous path is transformed into a finite-dimensional numerical problem. The initial obstacle avoidance reference path is used as the starting point for optimization, and the solution is obtained through a numerical optimization algorithm. During the optimization process, the path is adjusted through an iterative optimization algorithm to ensure that the path conforms to the kinematic constraints of the agricultural machinery and avoids all obstacles. The numerical optimization process is adjusted according to the dynamic changes in the actual environment to ensure that the optimization results are operable in different scenarios. Finally, the optimization process will output the optimal control sequence and state trajectory. These trajectories will guide the agricultural machinery to perform the operation task along the optimal path. The optimization results not only ensure that the agricultural machinery can avoid obstacles, but also improve the operation efficiency and accuracy.

[0026] This invention effectively solves the problems of traditional path planning methods being computationally complex, having low accuracy, and being unable to avoid obstacles in real time when dealing with unstructured obstacles, thereby improving the accuracy and efficiency of path planning.

[0027] As a preferred embodiment of the above, the multi-objective traveling salesman problem model uses work lines as nodes and turning paths as edges, with the optimization objective being to minimize the total turning distance.

[0028] Specifically, the selection of work rows and the arrangement of turning paths directly affect the efficiency of agricultural machinery operations. By treating each work row as a node and the turning path as an edge, the model can clearly show the spatial relationship when the agricultural machinery moves from one work row to another, and quantify the cost of turning. The optimization objective is to minimize the total turning distance, reduce unnecessary energy consumption and time loss caused by turning, and ensure that the agricultural machinery can complete its work tasks efficiently. Reducing the turning distance not only reduces the energy consumption of the agricultural machinery, but also improves the speed and accuracy of operation. By minimizing the turning distance, the agricultural machinery can maintain a relatively stable operating state and avoid excessively frequent turning, which is particularly important for agricultural operations, because frequent turning will lead to a decrease in operating efficiency and an increase in mechanical wear. When optimizing the path, considering the differences in turning shapes, such as right-angle turns and circular turns, the selection of turning paths can be further refined to achieve more precise control.

[0029] As a preferred embodiment of the above, such as Figure 2 As shown, step S2 involves processing the environmental perception data using a convolutional neural network to generate a rasterized map of unstructured obstacles and outputting an initial obstacle avoidance reference path, including: S21: Obtain radar point cloud data and visual image data from environmental perception data, and map them to the same farmland coordinate system through spatiotemporal registration; S22: Input the registered data into the convolutional neural network and output a semantic grid map containing obstacle outlines and category labels; S23: Perform morphological dilation on the obstacle contours in the semantic grid map to generate an envelope boundary with a preset safety margin; S24: Extract collision-free path key points outside the safety envelope boundary, and fit an initial obstacle avoidance reference path in the farmland coordinate system based on agricultural machinery kinematic constraints.

[0030] Specifically, the process begins by acquiring environmental perception data of obstacles in the farmland, including radar point cloud data and visual image data. Radar point cloud data provides the spatial location of obstacles, while visual image data supplements information such as shape and color. To fuse this data within the same coordinate system, spatiotemporal registration of the radar point cloud data and visual image data is required. Through this registration technique, the data acquired by radar and visual sensors are precisely mapped to the same farmland coordinate system, ensuring accurate fusion of data from different sensors and providing consistent foundational data for subsequent obstacle recognition and path planning. Next, after data registration, this data is input into a convolutional neural network (CNN). Through training, the CNN can accurately identify different obstacles in the farmland environment and assign a category label to each obstacle. Through learning, the CNN can extract the contour information of obstacles from the radar point cloud and visual images, generating a semantic grid map. This semantic grid map not only marks the location of obstacles but also assigns a category label to each obstacle, helping to identify different types of obstacles such as trees, shrubs, and rocks. This semantic grid map provides a comprehensive view of the farmland environment. Detailed information about obstacles in the environment supports subsequent path planning and obstacle avoidance tasks. Then, after obtaining the semantic grid map, morphological dilation is performed on the obstacle contours. This operation generates an envelope boundary with a preset safety margin by expanding the edges of the obstacles. This envelope boundary provides an additional safe area for the agricultural machinery's obstacle avoidance path planning, ensuring that the agricultural machinery maintains a sufficient distance from the obstacles to avoid collisions. The morphological dilation process dynamically adjusts the degree of dilation according to the physical characteristics of the agricultural machinery and the required safety margin, thereby generating an accurate safety envelope boundary. Finally, after generating the safety envelope boundary, path key points that will not collide with obstacles are extracted from it. These key points represent the areas that the agricultural machinery can safely pass through and provide a basis for path planning. During the path planning process, the kinematic constraints of the agricultural machinery, such as the maximum steering angle and driving speed, are considered. Based on these key points, a preliminary obstacle avoidance reference path is fitted. This path not only avoids all obstacles but also fully considers the kinematic characteristics of the agricultural machinery, ensuring the feasibility of the path and the operational stability of the agricultural machinery. Finally, the generated preliminary obstacle avoidance reference path provides a safe and efficient path planning basis for the agricultural machinery to perform its tasks.

[0031] As a preferred embodiment of the above, such as Figure 3 As shown, step S3 involves establishing an optimal control problem based on the global operation path and the initial obstacle avoidance reference path, which includes constraints on the kinematics of the agricultural machinery and distance constraints to obstacles. S31: Align the global operation path with the initial obstacle avoidance reference path in coordinates and merge them into a cubic spline curve with arc length as the parameter as the optimization reference trajectory; S32: Based on the structure of agricultural machinery, establish a nonlinear mapping relationship between the front wheel steering angle and the path curvature, and set an upper limit for the rate of change of curvature; S33: Based on the safety envelope boundary of the raster map, establish a symbolic distance function, calculate the Euclidean distance from the trajectory point to the nearest obstacle boundary in real time, and obtain the dynamic distance field; S34: Integrating curvature variation constraints and dynamic distance fields, constructing an objective function that includes minimizing trajectory offset, maximizing driving smoothness, and adaptive obstacle avoidance.

[0032] Specifically, firstly, the global operation path and the initial obstacle avoidance reference path are aligned in coordinates to ensure accurate correspondence within the same coordinate system. Since the global operation path and the initial obstacle avoidance reference path may differ in accuracy and coordinate system, alignment ensures the continuity and consistency of path planning. Then, a cubic spline curve method is used to smoothly fit the path. The cubic spline curve serves as the baseline trajectory for optimization, ensuring good smoothness of the path globally and providing a stable starting point for subsequent optimization. This fitting method uses arc length as a parameter to reasonably express the curvature changes of the path, avoiding excessive curvature. To ensure the stability and smoothness of agricultural machinery during operation, considering the complexities of the terrain, a nonlinear mapping relationship between the front wheel steering angle and the path curvature is established, taking into account the physical structure of the machinery, especially the impact of the front wheel steering angle on the path. In actual operation, the front wheel steering angle directly affects the path curvature; an excessively large steering angle can lead to severe path bending, affecting the stability of the machinery. Therefore, through this nonlinear mapping relationship, the relationship between the steering angle and the path curvature can be accurately described. Furthermore, an upper limit for the rate of change of curvature is set based on the structure and performance limitations of the machinery. This aims to prevent the path from becoming excessively curved, ensuring the smooth operation of the machinery during operation. The machine can travel smoothly without becoming unstable or reducing work efficiency due to excessive steering angles. Then, using the safety envelope boundary generated in the raster map as a benchmark, a signed distance function is established. This function helps quickly calculate the distance from a path point to the nearest obstacle boundary. In this way, the distance between each trajectory point and obstacle can be obtained in real time, generating a dynamic distance field. This dynamic distance field provides real-time obstacle information for path planning, ensuring that the agricultural machinery can avoid obstacles during operation. As the working environment changes, obstacle distances are constantly updated, and the dynamic distance field reflects these changes in real time, ensuring the accuracy of path planning. The process emphasizes flexibility and adaptability. Finally, curvature variation constraints are integrated with the dynamic distance field to construct an optimization objective function. This objective function is designed to balance multiple factors, including minimizing trajectory offset, maximizing path smoothness, and adaptive obstacle avoidance. Minimizing trajectory offset ensures that the path does not deviate from the predetermined route, maximizing driving smoothness reduces the turning and vibration of the agricultural machinery, and improves the stability and efficiency of operation. Adaptive obstacle avoidance ensures that the agricultural machinery can adjust the path in a timely manner in a dynamic environment to avoid collisions. By comprehensively considering these constraints, the objective function ultimately achieves efficient and stable operation path planning.

[0033] As a preferred embodiment of the above, the obstacle distance constraint is transformed into an objective term through a penalty function.

[0034] Specifically, in the path planning optimization process, obstacle distance constraints are a key factor in ensuring that agricultural machinery avoids obstacles and maintains a safe distance. In traditional path planning algorithms, obstacle distance constraints are usually addressed by directly calculating the distance between the path and obstacles and then modifying the path to avoid collisions. However, this method can lead to complex calculations and cannot effectively handle changes in obstacles in dynamic environments. To simplify constraint processing and effectively integrate it into path optimization, this embodiment employs a penalty function approach, transforming obstacle distance constraints into an optimization objective. Specifically, a signed distance function is first defined, which evaluates safety based on the distance between a path point and the nearest obstacle boundary. When a path point approaches an obstacle, the value of the signed distance function decreases, indicating that the path is in a danger zone. In the optimization objective function, obstacle distance constraints are represented by a penalty function. The penalty function is a punishment mechanism; when the distance between a path point and an obstacle is less than a preset safety margin, the penalty function adds an additional penalty term to the optimization objective. The strength of this penalty term is inversely proportional to the distance between the path point and the obstacle; the closer the distance, the greater the penalty. In this way, the optimization process automatically adjusts the path to avoid getting too close to obstacles, and forces the optimization algorithm to choose a path that is farther away from obstacles by adding a penalty term.

[0035] As a preferred embodiment of the above, the optimal control problem is discretized into a nonlinear programming problem using the fully simultaneous orthogonal collocation finite element method.

[0036] Specifically, in the optimization process of path planning, to effectively solve the optimal control problem, this embodiment employs the fully simultaneous orthogonal finite element method (FEM). The core of this method is to discretize the optimal control problem into a nonlinear programming problem, which can then be solved using numerical optimization techniques. First, the optimal control problem typically involves continuous decision variables. To solve it numerically, these continuous variables need to be transformed into discrete variables. The fully simultaneous orthogonal finite element method divides the path into multiple discrete intervals, each represented by a basis function, thus parameterizing the continuous path into discrete points. These discrete points constitute the control variables in path optimization and become the basis for subsequent solutions. The key to the fully simultaneous orthogonal finite element method is the use of orthogonal basis functions to interpolate the path. This allows for the construction of approximate solutions to the path within each discrete interval using orthogonal basis functions. This method can accurately describe the curvature, steering angle, and other characteristics of the path without requiring excessive calculations at each path point. By introducing multiple interpolation nodes during the finite element discretization process, an accurate representation of the path can be obtained while maintaining the efficiency of the optimization calculation. In this process, the discretization of the path is transformed into a finite-dimensional nonlinear programming problem, which allows traditional continuous optimization algorithms to be effectively applied to solve the problem. The fully parallel orthogonal finite element method can also adaptively adjust the accuracy of path discretization, so that computational efficiency and path accuracy can be balanced during the optimization process.

[0037] As a preferred embodiment of the above, such as Figure 4 As shown, in step S4, the optimal control problem is discretized into a nonlinear programming problem, and numerical optimization is performed using the initial path of the convolutional neural network as the starting point to obtain the optimal control sequence and state trajectory, including: S41: Determine the discretization step size based on kinematic constraints, transforming the optimal control problem into a finite-dimensional nonlinear programming problem; S42: Map the initial obstacle avoidance reference path to an initial guessed trajectory in the discrete state space; S43: Numerical optimization is performed starting from the initial guessed trajectory, where the kinematic constraints are transformed into path curvature rate of change constraints; S44: When the optimal solution simultaneously satisfies the physical limits of the agricultural machinery steering mechanism and the real-time obstacle avoidance safety threshold, the optimal control sequence and state trajectory are output.

[0038] Specifically, the first step is to determine the discretization step size based on the kinematic characteristics of the agricultural machinery. The kinematic constraints of the agricultural machinery include maximum steering angle, maximum acceleration, and maximum turning radius, all of which limit its movement. By analyzing these constraints, a suitable discretization step size can be set. The selection of the discretization step size directly affects the precision of the path and the computational complexity. An excessively large step size may lead to low path fitting accuracy, resulting in abrupt turns or paths that do not conform to the actual driving capabilities of the agricultural machinery; an excessively small step size will increase computational load and reduce efficiency. Therefore, a suitable step size can balance computational efficiency and path planning accuracy. Through this discretization process, the continuous optimal control problem is transformed into a finite-dimensional nonlinear programming problem. In this way, each point on the path can be used as a decision variable, and the path parameters are gradually adjusted during the optimization process, making the path gradually approach the optimal solution. Then, in this step, the initial obstacle avoidance reference path is discretized and mapped to a discrete state space. The initial obstacle avoidance reference path is usually obtained by processing perception data through a convolutional neural network; it represents a preliminary safe path. To adapt to the optimization algorithm, this path needs to be transformed into a discrete path. Each path point represents the position of the agricultural machinery at a certain moment, and the details of the path can be adjusted according to kinematic constraints. This process ensures that the path can be effectively represented in the discrete state space and provides an initial starting point for subsequent optimization. Through this mapping, the initial path becomes the starting point of the optimization process and can be further adjusted through numerical optimization to ultimately achieve the best obstacle avoidance effect. Next, in the numerical optimization process, the path is adjusted by the optimization algorithm, starting from the initial guessed trajectory. The main goal of the optimization is to ensure that the path conforms to the kinematic constraints of the agricultural machinery, such as the maximum steering angle and the minimum turning radius. The limitation of the rate of change of the path curvature is one of the key kinematic constraints. An excessive rate of change of the steering angle will lead to instability in the operation of the agricultural machinery and may even lead to slippage or other unsafe situations. Therefore, by introducing a constraint on the rate of change of curvature during the optimization process, the curvature of the path can be effectively controlled, preventing the agricultural machinery from becoming unstable due to excessively sharp turns. In each iteration, the numerical optimization algorithm adjusts the path parameters, such as the curvature of the path and the specific location of the path points, until the optimization result meets the physical constraints of the agricultural machinery and can achieve smooth driving in actual operation. Finally, during the numerical optimization process, when the optimized solution satisfies the physical limits of the agricultural machinery's steering mechanism, such as the maximum steering angle, minimum turning radius, and the real-time obstacle avoidance safety threshold, the optimization process ends and the optimal control sequence and state trajectory are output. The optimal control sequence refers to the specific control commands that the agricultural machinery needs to execute during operation, such as steering angle and driving speed.The state trajectory describes the state of the agricultural machinery as it travels along the path, including parameters such as position, speed, and acceleration at each moment. The optimization results ensure that the agricultural machinery can travel smoothly along the planned path while avoiding obstacles in the path, meeting the physical performance limitations of the agricultural machinery, and ensuring that no collisions or other dangerous situations occur during operation. Through this method, the final optimal control sequence and state trajectory not only meet the dynamic constraints of the agricultural machinery, but also enable efficient and accurate dynamic obstacle avoidance path planning in actual operating environments.

[0039] As a preferred embodiment of the above, the method further includes: real-time monitoring of obstacle positions during trajectory execution; when a new obstacle intrudes into the path safety zone or the distance between the planned trajectory and the obstacle is lower than a safety threshold, updating the raster map and re-triggering trajectory optimization.

[0040] Specifically, when a new obstacle encroaches on the safe zone of the work path, or when the distance between the planned trajectory and the obstacle falls below a preset safety threshold, it indicates that the current path may collide with the obstacle. In this case, the raster map must be updated immediately. By updating the raster map, the location and changes of new obstacles can be identified in a timely manner, thereby adjusting the path planning. After the raster map is updated, path optimization must be retried. At this time, the safety and feasibility of the path are recalculated based on the new environmental data. During the optimization process, the impact range of the obstacle is reassessed, and the path is corrected in conjunction with the kinematic constraints of the agricultural machinery. This ensures that the new path avoids obstacles and meets the safety and efficiency requirements of agricultural machinery operations.

[0041] Example 2: Based on the same inventive concept as the dynamic obstacle avoidance path planning method for complex agricultural machinery scenarios dealing with unstructured obstacles in the foregoing embodiments, the present invention also provides a dynamic obstacle avoidance path planning system for complex agricultural machinery scenarios dealing with unstructured obstacles, including: The global operation path output module constructs a multi-objective traveling salesman problem model based on farmland boundaries and the number of operation rows, solves the optimal operation row sequence, and outputs the global operation path; The initial obstacle avoidance path output module processes environmental perception data through a convolutional neural network, generates a raster map of unstructured obstacles, and outputs an initial obstacle avoidance reference path. The optimal control problem construction module establishes an optimal control problem that includes agricultural machinery kinematic constraints and obstacle distance constraints based on the global operation path and the initial obstacle avoidance reference path. The optimal control trajectory generation module discretizes the optimal control problem into a nonlinear programming problem, and uses the initial path of the convolutional neural network as the starting point for numerical optimization to obtain the optimal control sequence and state trajectory.

[0042] The adjustment system described above in this invention can effectively realize a dynamic obstacle avoidance path planning method for agricultural machinery in complex scenarios with unstructured obstacles. The technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0043] As a preferred embodiment of the above, the initial obstacle avoidance path output module includes: The data spatiotemporal registration unit acquires radar point cloud data and visual image data from environmental perception data, and maps them to the same farmland coordinate system through spatiotemporal registration; The grid map output unit inputs the registered data into the convolutional neural network and outputs a semantic grid map containing obstacle outlines and category labels. The envelope boundary generation unit performs morphological dilation on the obstacle contours in the semantic grid map to generate an envelope boundary with a preset safety margin. The obstacle avoidance path fitting unit extracts key points of the collision-free path outside the safety envelope boundary and fits the initial obstacle avoidance reference path in the farmland coordinate system based on agricultural machinery kinematic constraints.

[0044] Similarly, the above-mentioned optimization schemes for the system can also achieve the optimization effects corresponding to the methods in Embodiment 1, which will not be repeated here.

[0045] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A dynamic obstacle avoidance path planning method for agricultural machinery in complex scenarios involving unstructured obstacles, characterized in that, The method includes: Based on farmland boundaries and the number of operation rows, a multi-objective traveling salesman problem model is constructed, and the optimal operation row sequence is solved to output the global operation path; The system processes environmental perception data through a convolutional neural network, generates a raster map of unstructured obstacles, and outputs an initial obstacle avoidance reference path. Based on the global operation path and the initial obstacle avoidance reference path, an optimal control problem is established that includes agricultural machinery kinematic constraints and obstacle distance constraints. The optimal control problem is discretized into a nonlinear programming problem, and numerical optimization is performed using the initial path of the convolutional neural network as the starting point to obtain the optimal control sequence and state trajectory.

2. The dynamic obstacle avoidance path planning method for agricultural machinery in complex scenarios with unstructured obstacles as described in claim 1, characterized in that, The multi-objective traveling salesman problem model uses work lines as nodes and turning paths as edges, with the optimization objective being to minimize the total turning distance.

3. The dynamic obstacle avoidance path planning method for agricultural machinery in complex scenarios with unstructured obstacles as described in claim 1, characterized in that, The system processes environmental perception data using a convolutional neural network to generate a rasterized map of unstructured obstacles and outputs an initial obstacle avoidance reference path, including: Radar point cloud data and visual image data are obtained from the environmental perception data and mapped to the same farmland coordinate system through spatiotemporal registration. The registered data is input into the convolutional neural network, which outputs a semantic grid map containing obstacle outlines and category labels. Morphological dilation is performed on the obstacle contours in the semantic grid map to generate an envelope boundary with a preset safety margin. Key points of the collision-free path are extracted outside the safety envelope boundary, and the initial obstacle avoidance reference path is fitted in the farmland coordinate system based on agricultural machinery kinematic constraints.

4. The dynamic obstacle avoidance path planning method for agricultural machinery in complex scenarios with unstructured obstacles as described in claim 1, characterized in that, Based on the global operation path and the initial obstacle avoidance reference path, an optimal control problem is established that includes agricultural machinery kinematic constraints and obstacle distance constraints, including: The global operation path and the initial obstacle avoidance reference path are aligned in coordinates and fused into a cubic spline curve with arc length as the parameter as the optimization reference trajectory. Based on the structure of agricultural machinery, a nonlinear mapping relationship between the front wheel steering angle and the path curvature is established, and an upper limit for the rate of change of curvature is set. Based on the safety envelope boundary of the rasterized map, a symbolic distance function is established, and the Euclidean distance from the trajectory point to the nearest obstacle boundary is calculated in real time to obtain the dynamic distance field. By integrating curvature variation constraints with the dynamic distance field, an objective function is constructed that includes minimizing trajectory offset, maximizing driving smoothness, and adaptive obstacle avoidance.

5. The dynamic obstacle avoidance path planning method for agricultural machinery in complex scenarios with unstructured obstacles as described in claim 4, characterized in that, The obstacle distance constraint is transformed into an objective term through a penalty function.

6. The method for dynamic obstacle avoidance path planning in complex agricultural machinery scenarios to cope with unstructured obstacles as described in claim 1, characterized in that, The optimal control problem is discretized into a nonlinear programming problem using the fully simultaneous orthogonal collocation finite element method.

7. The method for dynamic obstacle avoidance path planning in complex agricultural machinery scenarios to cope with unstructured obstacles as described in claim 1, characterized in that, The optimal control problem is discretized into a nonlinear programming problem, and numerical optimization is performed using the initial path of the convolutional neural network as the starting point to obtain the optimal control sequence and state trajectory, including: Based on the kinematic constraints, the discretization step size is determined, and the optimal control problem is transformed into a finite-dimensional nonlinear programming problem. The initial obstacle avoidance reference path is mapped to an initial guessed trajectory in a discrete state space; Numerical optimization is performed starting from the initial guessed trajectory, wherein the kinematic constraints are transformed into path curvature change rate constraints; When the optimal solution simultaneously satisfies the physical limits of the agricultural machinery steering mechanism and the real-time obstacle avoidance safety threshold, the optimal control sequence and state trajectory are output.

8. The method for dynamic obstacle avoidance path planning in complex agricultural machinery scenarios to cope with unstructured obstacles as described in claim 1, characterized in that, The method further includes: real-time monitoring of obstacle positions during trajectory execution; when a new obstacle intrudes into the path safety zone or the distance between the planned trajectory and the obstacle is lower than a safety threshold, updating the raster map and re-triggering trajectory optimization.

9. A dynamic obstacle avoidance path planning system for agricultural machinery in complex scenarios involving unstructured obstacles, characterized in that: The system includes: The global operation path output module constructs a multi-objective traveling salesman problem model based on farmland boundaries and the number of operation rows, solves the optimal operation row sequence, and outputs the global operation path; The initial obstacle avoidance path output module processes environmental perception data through a convolutional neural network, generates a raster map of unstructured obstacles, and outputs an initial obstacle avoidance reference path. The optimal control problem construction module establishes an optimal control problem that includes agricultural machinery kinematic constraints and obstacle distance constraints based on the global operation path and the initial obstacle avoidance reference path. The optimal control trajectory generation module discretizes the optimal control problem into a nonlinear programming problem, and uses the initial path of the convolutional neural network as the starting point to perform numerical optimization to obtain the optimal control sequence and state trajectory.

10. The dynamic obstacle avoidance path planning system for agricultural machinery in complex scenarios with unstructured obstacles as described in claim 9, characterized in that, The initial obstacle avoidance path output module includes: The data spatiotemporal registration unit acquires radar point cloud data and visual image data from the environmental perception data, and maps them to the same farmland coordinate system through spatiotemporal registration; The grid map output unit inputs the registered data into the convolutional neural network and outputs a semantic grid map containing obstacle outlines and category labels. The envelope boundary generation unit performs morphological dilation on the obstacle contours in the semantic grid map to generate an envelope boundary with a preset safety margin. The obstacle avoidance path fitting unit extracts key points of the collision-free path outside the safety envelope boundary and fits the initial obstacle avoidance reference path in the farmland coordinate system based on agricultural machinery kinematic constraints.

Images