Self-adaptive online track generation method and system for agricultural machinery

Abstract

The invention relates to the technical field of automatic navigation and intelligent operation of agricultural machinery, in particular to a self-adaptive online track generation method and system for agricultural machinery. Comprising the following steps: S1, multi-modal perception and operation semantic analysis; s2, constructing a multi-constraint feasible corridor C based on semantics; s3, a multi-target trajectory evaluation method is adopted; and S4, generating an agronomic-kinematic joint optimization trajectory. Compared with the prior art, the method has the advantages that the core jump from'sensing geometry 'to'understanding semantics' and the technical breakthrough from'single-target optimization 'to'multi-target collaborative tradeoff' can be realized.

Description

Technical Field

[0001] This invention relates to the field of automatic navigation and intelligent operation technology for agricultural machinery, and to a method and system for generating adaptive online trajectories for agricultural machinery, particularly to a method and system for generating adaptive online trajectories for agricultural machinery based on semantic understanding of farmland operation scenarios and multi-objective optimization decision-making. Background Technology

[0002] With the development of precision agriculture, automatic navigation of agricultural machinery has become one of the core technologies of modern agriculture. Currently, mainstream automatic navigation systems can achieve centimeter-level accuracy in the straight-line operation stage. However, in the initial stage of the operation line: entering the first operation line from the field or road, and in the transition stage of the operation line: switching from one operation line to the next, i.e., the online process, its automation and intelligence levels still have significant shortcomings.

[0003] Existing technologies typically employ pre-defined trajectory generation methods based on geometric models: The "straight line-circular arc" combined model: The agricultural machinery travels a straight line from the starting point for a certain distance, then turns through a circular arc of a fixed radius, eventually tangentially aligning with the target work line. This method is simple, but it cannot adapt to irregular terrain and dynamic obstacles.

[0004] Path tracing based on pure tracking or Stanley control: Given a coarse reference path, the controller tracks in real time, but the generation of the reference path itself does not fully consider environmental semantics and global optimization.

[0005] These traditional methods have the following inherent drawbacks: Poor environmental adaptability: Treating farmland as an ideal two-dimensional plane, unable to understand and avoid unstructured obstacles such as temporary crop piles, irrigation facilities, water pits, and steep slopes, which pose safety risks.

[0006] Efficiency and energy consumption are not optimal: Fixed turning radius and path pattern may lead to wasted turning space, excessive driving time, or excessive energy consumption.

[0007] Lack of agronomic intelligence: The specific requirements of different operation types for the overhead line trajectory are not considered. Seeders require the end of the overhead line to be precisely aligned with the work line and in the same direction to reduce double-seeding or missed-seeding; while sprayers require that the overhead line process avoid crushing crops.

[0008] Therefore, developing an intelligent online trajectory generation technology that can deeply understand the operation scenario, comprehensively optimize multiple performance indicators, and make real-time adaptive adjustments is of urgent need and great significance for improving the autonomy, safety, and economy of agricultural machinery operations throughout the entire process. Summary of the Invention

[0009] This invention aims to provide a method and system for generating adaptive online trajectories for agricultural machinery based on operational semantic understanding and multi-objective optimization. This method enables agricultural machinery to "understand" the field environment, "comprehend" the operational task, and "combine" its own characteristics, just like an experienced driver, autonomously generating and executing an online trajectory that achieves an optimal balance in safety, efficiency, energy consumption, and agronomic compliance.

[0010] To address the technical problems mentioned in the background section, the present invention provides the following technical solution: S1: Multimodal perception and task semantic parsing The position and velocity status information of agricultural machinery in the global coordinate system and vehicle coordinate system are obtained through the Global Navigation Satellite System (GNSS) and Inertial Measurement Unit (IMU).

[0011] The system acquires raw point cloud and image data of the environment surrounding the agricultural machinery using one or more of the following: visual sensors, lidar, and millimeter-wave radar.

[0012] Semantic understanding: The above perceptual data is fused and parsed to construct a two-dimensional or three-dimensional environmental map containing semantic labels, specifically parsing out: Static semantic layer: drivable farmland areas, insurmountable fixed obstacles (trees, utility poles), field ridge boundaries, and preset work line geometry information.

[0013] Dynamic semantic layer: moving obstacles (people, other agricultural machinery, animals).

[0014] Task semantic layer: Parses the upper-level instructions to clarify the starting point, target launch point, and core operation type (such as sowing, fertilizing, and harvesting) of this online task.

[0015] Ontology semantic layer: Based on the agricultural machinery model and attached implement information, load its kinematic model (such as Ackermann steering model or articulated model), geometric profile, minimum turning radius, and maximum steering angular velocity.

[0016] S2: Semantic-based construction of multi-constraint feasible corridors Based on the constraints of the static and dynamic semantic layers, adaptive expansion is performed to both sides to form a continuous channel region with a safe boundary, namely "feasible corridor C". This corridor defines the boundary of subsequent refined trajectory search in geometric space.

[0017] S3: Construction of Multi-Objective Trajectory Evaluation Method Within the feasible corridor C constructed in S2, the trajectory generation problem is modeled as a constrained multi-objective optimization problem.

[0018] The sub-target method includes: Safety cost J_safety: The inverse or negative correlation function of the distances of all points on the trajectory to the nearest obstacle, encouraging the trajectory to stay away from obstacles.

[0019] Efficiency cost J_efficiency: The weighted sum of the total length of the trajectory and the estimated travel time.

[0020] Smoothness cost J_smoothness: the square integral of the curvature of the trajectory and its rate of change, to reduce mechanical wear and energy consumption.

[0021] Agronomic adaptation cost J_agronomy: dynamically adjusted based on the job type in the task semantic layer.

[0022] For sowing / transplanting: significantly increase the weights of the positional error and heading angle error between the trajectory endpoint and the target work line.

[0023] J_agronomy(τ) = γ·Δx²(τ) + (1-γ)·Δθ²(τ) (where γ is the position and heading error weight, Δx(τ) is the position error between the trajectory endpoint and the target work line, and Δθ(τ) is the heading angle error).

[0024] For spraying / fertilizing: Increase the weight of trajectory to avoid crushing crop areas.

[0025] S4: Agronomic-Kinematic Joint Optimization Trajectory Generation Define the parameterized trajectory as τ(s) = [x(s), y(s), θ(s), κ(s)]^T, where s is the arc length parameter, which is the natural parameter of the trajectory τ(s) (also called the path length parameter, simply put, it is the actual distance traveled along the trajectory curve from the starting point of the trajectory to the current point τ(s), x(s) and y(s) are the position coordinates, θ(s) is the heading angle, and κ(s) is the curvature.

[0026] The objective function is designed as a linear weighted sum of multiple weighted sub-objectives, and its general form is: min J(τ) = w_s * J_safety(τ) + w_e * J_efficiency(τ) + w_m * J_smoothness(τ) + w_a * J_agronomy(τ); stτ(s) ∈ C, s ∈ [0,S] (the entire trajectory lies within a feasible corridor); θ_min ≤ θ(s) ≤ θ_max, s ∈ [0,S] (heading angle constraint, adapting to mechanical steering capability); κ_min ≤ κ(s) ≤ κ_max, s ∈ [0,S] (curvature constraint, corresponding to minimum turning radius); Where θ_min and θ_max are the minimum and maximum heading angles of the agricultural machinery, respectively; κ_min and κ_max are the minimum and maximum curvatures, respectively (determined by the minimum turning radius of the machinery, κ_max=1 / R_min, where R_min is the minimum turning radius); w_s, w_e, w_m, and w_a are the weight coefficients of the sub-objectives of safety, efficiency, smoothness, and agronomic adaptability, respectively, which can be dynamically configured according to the operation mode, satisfying w_s, w_e, w_m, w_a ≥ 0 and w_s+w_e+w_m+w_a=1.

[0027] Output a smooth upper trajectory {τ0,τ1,…,τ} containing M trajectory points. M}

[0028] The beneficial effects of this application are as follows: 1. Achieving a core leap from "perceiving geometry" to "understanding semantics": Unlike existing technologies that only acquire environmental geometric information, this invention constructs a multi-layered semantic system of static, dynamic, task, and ontology through multimodal perception data fusion and semantic parsing. This enables the system to accurately understand "where it is safe to drive," "where there are dangers," and "what the task of this operation is" just like a human driver. This provides a comprehensive semantic information foundation for intelligent trajectory decision-making and significantly improves the adaptability and operational reliability of agricultural machinery in complex and unstructured field environments.

[0029] 2. Achieving a technological breakthrough from "single-objective optimization" to "multi-objective synergistic trade-off": For the first time, key requirements such as safety, operational efficiency, energy consumption control, and agronomic adaptability are systematically incorporated into the trajectory optimization framework. Through weighted multi-objective function modeling, the optimal balance of multi-dimensional objectives is achieved. The generated trajectory not only meets geometric passage requirements but also closely matches the comprehensive needs of actual field operations. This solves the problems of poor agronomic compatibility and high energy consumption caused by existing technologies that only pursue a single objective (such as the shortest path). Attached Figure Description

[0030] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the outer boundary and internal obstacles. Figure 2 This is a schematic diagram after adaptive expansion; Figure 3 This represents the point between starting point A and target point B on the work line; Figure 4 This represents the generated trajectory graph; Figure 5 This is a flowchart of an embodiment. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] This invention aims to provide a method and system for generating adaptive online trajectories for agricultural machinery based on operational semantic understanding and multi-objective optimization. This method enables agricultural machinery to "understand" the field environment, "comprehend" the operational task, and "combine" its own characteristics, just like an experienced driver, autonomously generating and executing an online trajectory that achieves an optimal balance in safety, efficiency, energy consumption, and agronomic compliance.

[0033] S1: Multimodal perception and task semantic parsing The position and velocity status information of agricultural machinery in the global coordinate system and vehicle coordinate system are obtained through the Global Navigation Satellite System (GNSS) and Inertial Measurement Unit (IMU).

[0034] The system acquires raw point cloud and image data of the environment surrounding the agricultural machinery using one or more of the following: visual sensors, lidar, and millimeter-wave radar.

[0035] Semantic understanding: The above perceptual data is fused and parsed to construct a two-dimensional or three-dimensional environmental map containing semantic labels, specifically parsing out: Static semantic layer: drivable farmland areas, insurmountable fixed obstacles (trees, utility poles), field ridge boundaries, and preset work line geometry information.

[0036] Dynamic semantic layer: moving obstacles (people, other agricultural machinery, animals).

[0037] Task semantic layer: Parses the upper-level instructions to clarify the starting point, target launch point, and core operation type (such as sowing and fertilizing) of this online task.

[0038] Ontology semantic layer: Based on the agricultural machinery model and attached implement information, load its kinematic model (such as Ackermann steering model or articulated model), geometric profile, minimum turning radius, and maximum steering angular velocity.

[0039] S2: Semantic-based construction of multi-constraint feasible corridors Based on the constraints of the static and dynamic semantic layers, adaptive expansion occurs to both sides, forming a continuous channel region with safe boundaries, i.e., a "feasible corridor." This corridor geometrically defines the boundary for subsequent refined trajectory search. Figure 1 This is a schematic diagram of the outer boundary and internal obstacles. Figure 2 This is a schematic diagram after adaptive expansion.

[0040] S3: Multi-object trajectory evaluation method Within the feasible corridor C constructed in S2, the trajectory generation problem is modeled as a constrained multi-objective optimization problem.

[0041] Figure 3 This represents the starting point A and the target point B on the work line.

[0042] The sub-target method includes: Safety cost J_safety: The inverse or negative correlation function of the distances of all points on the trajectory to the nearest obstacle, encouraging the trajectory to stay away from obstacles.

[0043] Efficiency cost J_efficiency: The weighted sum of the total length of the trajectory and the estimated travel time.

[0044] Smoothness cost J_smoothness: the square integral of the curvature of the trajectory and its rate of change, to reduce mechanical wear and energy consumption.

[0045] Agronomic adaptation cost J_agronomy: dynamically adjusted based on the job type in the task semantic layer.

[0046] For sowing / transplanting: significantly increase the weights of the positional error and heading angle error between the trajectory endpoint and the target work line.

[0047] For spraying / fertilizing: Increase the weight of trajectory to avoid crushing crop areas.

[0048] S4: Agronomic-Kinematic Joint Optimization Trajectory Generation Define the parameterized trajectory as τ(s) = [x(s), y(s), θ(s), κ(s)]^T, where s is the arc length parameter, which is the natural parameter of the trajectory τ(s) (also called the path length parameter, simply put, it is the actual distance traveled along the trajectory curve from the starting point of the trajectory to the current point τ(s), x(s) and y(s) are the position coordinates, θ(s) is the heading angle, and κ(s) is the curvature.

[0049] The objective function is designed as a linear weighted sum of multiple weighted sub-objectives, and its general form is: min J(τ) = w_s * J_safety(τ) + w_e * J_efficiency(τ) + w_m * J_smoothness(τ) + w_a * J_agronomy(τ); stτ(s) ∈ C, s ∈ [0,S] (the entire trajectory lies within a feasible corridor); θ_min ≤ θ(s) ≤ θ_max, s ∈ [0,S] (heading angle constraint, adapting to mechanical steering capability); κ_min ≤ κ(s) ≤ κ_max, s ∈ [0,S] (curvature constraint, corresponding to minimum turning radius); Where θ_min and θ_max are the minimum and maximum heading angles of the agricultural machinery, respectively; κ_min and κ_max are the minimum and maximum curvatures, respectively (determined by the minimum turning radius of the machinery, κ_max=1 / R_min, where R_min is the minimum turning radius); w_s, w_e, w_m, and w_a are the weight coefficients of the sub-objectives of safety, efficiency, smoothness, and agronomic adaptability, respectively, which can be dynamically configured according to the operation mode, satisfying w_s, w_e, w_m, w_a ≥ 0 and w_s+w_e+w_m+w_a=1.

[0050] Output a smooth upper trajectory {τ0,τ1,…,τ} containing M trajectory points. M} Figure 4 This represents the generated trajectory graph.

[0051] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0052] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An agricultural machine adaptive on-line trajectory generation method, characterized in that, Comprise the following steps: S1: multi-modal perception and task semantic parsing; S2: semantic-based multi-constraint drivable corridor C construction; S3: multi-objective trajectory evaluation method; S4: agronomic-kinematic joint optimization trajectory generation.

2. The adaptive on-line trajectory generation method for agricultural machinery of claim 1, wherein, The S1 specifically comprises the following steps: Obtain the pose and speed state information of the agricultural machine in the global coordinate system and the vehicle coordinate system through the global navigation satellite system and the inertial measurement unit; Obtain the original point cloud and image data of the surrounding environment of the agricultural machine through one or more of the visual sensor, the laser radar and the millimeter wave radar; Semantic understanding: fuse and parse the above perception data to construct a two-dimensional or three-dimensional environment map containing semantic labels, and specifically parse out: Static semantic layer: drivable farmland area, impassable fixed obstacles, ridge boundary, and preset work line geometric information; Dynamic semantic layer: moving obstacles; Task semantic layer: parse the upper-layer instruction to determine the starting point, target on-line point and core work type of the current on-line task; Ontology semantic layer: load the kinematic model, geometric profile, minimum turning radius and maximum turning angular velocity of the agricultural machine according to the model and the information of the attached implement.

3. The adaptive on-line trajectory generation method for agricultural machinery of claim 1, wherein, The S2 specifically comprises the following steps: According to the constraints of the static semantic layer and the dynamic semantic layer, the two sides are adaptively inflated to form a continuous channel area with a safety boundary, i.e. a "drivable corridor C", which defines the boundary of the subsequent fine trajectory search in the geometric space.

4. The adaptive on-line trajectory generation method for agricultural machinery of claim 1, wherein, The S3 specifically comprises the following steps: In the drivable corridor C constructed in the S2, the trajectory generation problem is modeled as a multi-objective optimization problem with constraints. The sub-objective method comprises: Safety cost J_safety: the reciprocal or negative correlation function of the distance between all points on the trajectory and the nearest obstacle, which encourages the trajectory to be far away from the obstacle; Efficiency cost J_efficiency: the weighted sum of the total length of the trajectory and the expected travel time; Smoothness cost J_smoothness: the square integral of the curvature and its rate of change of the trajectory to reduce mechanical wear and energy consumption; Agronomic adaptation cost J_agronomy: dynamically adjusted according to the work type in the task semantic layer.

5. The adaptive on-line trajectory generation method for agricultural machinery according to claim 4, wherein, The agronomic adaptation cost J_agronomy; For seeding / transplanting: significantly increase the weight of the alignment error between the end point of the trajectory and the target work line. For pesticide / fertilizer spraying: increase the weight of the trajectory to avoid crushing the crop area.

6. The adaptive on-line trajectory generation method for agricultural machinery of claim 1, wherein, The S3 specifically comprises the following steps: Define the parameterized trajectory as τ(s) = [x(s), y(s), θ(s), κ(s)]^T, where s is the arc length parameter, x(s) and y(s) are the position coordinates, θ(s) is the heading angle, and κ(s) is the curvature; The objective function is designed as a linear weighted sum of multiple weighted sub-objectives, and its general form is: min J(τ) = w_s * J_safety(τ) + w_e * J_efficiency(τ) + w_m * J_smoothness(τ) + w_a * J_agronomy(τ); s.t. τ(s) ∈ C, s ∈ [0, S] (the trajectory falls within the walkable corridor throughout) θ_min ≤ θ(s) ≤ θ_max, s ∈ [0,S] (heading angle constraint, adapting to mechanical steering capabilities); κ_min < κ(s) < κ_max, s e [0, S] (curvature constraint, corresponding to minimum turning radius); Wherein, θ_min, θ_max are minimum and maximum heading angles of the agricultural machine respectively; κ_min, κ_max are minimum and maximum curvatures respectively; w_s, w_e, w_m, w_a are weight coefficients of safety, efficiency, smoothness and agronomic adaptability sub-targets respectively, which can be dynamically configured according to the operation mode, and meet w_s, w_e, w_m, w_a ≥ 0 and w_s+w_e+w_m+w_a=1; The output is a smooth up-line trajectory {τ0, τ1, …, τM} containing M trajectory points. M} containing M trajectory points.

7. An agricultural machine adaptive on-line trajectory generation system, characterized by, Run an agricultural machine adaptive online trajectory generation method according to any one of claims 1-6.

Images