Adaptive parameter tuning method and system based on dynamic window and storage medium

Abstract

The application provides a dynamic window-based adaptive parameter tuning method and system and a storage medium. A raster map containing a farmland operation direction is obtained, a direction-dependent movement cost is set for each raster, a jump point search algorithm is used to plan a global path point sequence from a starting point to an ending point, and a global guide corridor with a variable width is generated based on the center line. In the dynamic window speed space, the angular velocity is non-uniformly sampled, the Gaussian distribution expectation of the angular velocity is the angular velocity required for aligning the corridor center line, and the standard deviation is proportional to the current corridor width. The sampled speed pairs are simulated in a forward trajectory, the speed pairs are scored by an evaluation function, the evaluation function score contains a heading guide, an obstacle distance and a terrain risk term, the speed pair corresponding to the optimal trajectory is selected as a control instruction, and the maximum linear speed in the next period is adjusted according to the terrain risk.

Description

An adaptive parameter tuning method, system, and storage medium based on dynamic windows. Technical Field

[0001] This application belongs to the field of parameter tuning, and in particular relates to an adaptive parameter tuning method, system and storage medium based on dynamic windows. Background Technology

[0002] Agricultural machinery path planning can employ a hybrid framework combining global and local path planning. For global path planning, A* and Dijkstra algorithms are commonly used grid map search algorithms; however, in large, open farmland scenarios, these algorithms suffer from low search efficiency due to the need to expand redundant nodes. The Jump Point Search (JPS) algorithm reduces the number of search nodes by utilizing jump points and pruning strategies; however, when applied to farmland environments, the JPS algorithm, by not considering the specific directional requirements of farmland operations, may result in globally planned paths that do not meet agronomical requirements. Furthermore, the generated single path point sequence provides weak guidance for local planning, making it difficult to achieve a good balance between obstacle avoidance and path tracking.

[0003] In local path planning, the Dynamic Window Method (DWA) performs uniform sampling within the velocity space. This unbiased sampling method generates low-quality candidate trajectories. When a global path needs to be followed, computational resources are wasted on evaluating trajectories that deviate from the desired direction, reducing planning efficiency. The evaluation function typically only includes general evaluation terms such as heading angle, obstacle distance, and speed, failing to incorporate terrain risks such as ditches, slopes, and soil moisture into the evaluation system. This may lead agricultural machinery to make suboptimal or even dangerous decisions under terrain conditions. Furthermore, the upper and lower limits of the dynamic window's speed are fixed and cannot be adjusted based on real-time assessments of terrain passability, limiting the potential for agricultural machinery to improve operational efficiency while ensuring safety. Summary of the Invention

[0004] This invention proposes an adaptive parameter tuning method based on a dynamic window to address the problem that existing technologies cannot adjust parameters according to real-time assessed terrain accessibility, including:

[0005] Obtain a grid map containing the predetermined operation direction of the farmland, and set a direction-dependent movement cost for each grid cell in the grid map, wherein the cost of moving along the predetermined operation direction is less than the cost of moving outside the predetermined operation direction; run a jump point search algorithm based on the direction-dependent movement cost to plan a global path point sequence connecting the start point and the end point, and use this as the center line to generate a global guidance corridor with varying width according to the curvature of each path point in the sequence and the distance to the nearest obstacle;

[0006] Within the velocity space of the dynamic window, velocity pairs consisting of linear velocity and angular velocity are sampled non-uniformly, wherein the sampling of the angular velocity follows a Gaussian distribution;

[0007] Forward trajectory simulations were performed on each velocity pair obtained from the sampling, and all simulated trajectories were scored using an evaluation function that includes heading guidance, obstacle distance, and terrain risk.

[0008] The simulation trajectory with the highest score is selected as the optimal trajectory, and the corresponding velocity pair is output as the current control command. Based on the terrain risk evaluation value of the optimal trajectory, the maximum linear velocity of the dynamic window in the next planning cycle is adjusted.

[0009] In another aspect, this invention proposes an adaptive parameter tuning system based on dynamic windows, comprising the following modules:

[0010] A generation module is used to acquire a grid map containing the predetermined operation direction of farmland, and to set a direction-dependent movement cost for each grid cell in the grid map, wherein the cost of moving along the predetermined operation direction is less than the cost of moving outside the predetermined operation direction; a jump point search algorithm is run based on the direction-dependent movement cost to plan a global path point sequence connecting the start point and the end point, and a global guide corridor with varying width is generated based on the curvature of each path point in the sequence and the distance to the nearest obstacle, using this as the center line;

[0011] The sampling module is used to perform non-uniform sampling of velocity pairs consisting of linear velocity and angular velocity within the velocity space of a dynamic window, wherein the sampling of the angular velocity follows a Gaussian distribution;

[0012] The evaluation module is used to simulate the forward trajectory of each sampled velocity pair and score all simulated trajectories using an evaluation function that includes heading guidance, obstacle distance, and terrain risk.

[0013] The adjustment module is used to select the simulation trajectory with the highest score as the optimal trajectory, output the corresponding velocity pair as the current control command, and adjust the maximum linear velocity of the dynamic window for the next planning cycle based on the terrain risk evaluation value of the optimal trajectory.

[0014] Finally, the present invention also proposes a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements the method described in the first aspect.

[0015] This invention, by setting direction-dependent movement costs during the global path search phase, can plan a global path that perfectly conforms to the preset farmland operation direction. A global guidance corridor built upon this path sets a safe driving range for the local planner, especially in areas with dense obstacles or requiring turns, avoiding collisions and ensuring a smooth path. In local path planning, computational resources are concentrated on high-value candidate trajectories, improving the real-time performance and solution quality of the planning algorithm. A multi-objective evaluation function integrating heading, obstacle avoidance, and terrain risk allows for a balance between path tracking accuracy, safety, and terrain traversability, resulting in a more optimized agricultural machinery driving trajectory. Adjusting the upper limit of driving speed based on the terrain risk evaluation value ensures the stability and safety of agricultural machinery operating in unknown terrain. Attached Figure Description

[0016] Figure 1 is a flowchart of the first embodiment;

[0017] Figure 2 is a schematic diagram of the global guidance corridor generation;

[0018] Figure 3 is a schematic diagram of the Gaussian sampling distribution of angular velocity based on the corridor width. Detailed Implementation

[0019] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0020] The first embodiment, as shown in Figure 1, is an adaptive parameter tuning method based on a dynamic window, comprising:

[0021] S1. Obtain a grid map containing the predetermined operation direction of the farmland, and set a direction-dependent movement cost for each grid cell in the grid map, wherein the cost of moving along the predetermined operation direction is less than the cost of moving outside the predetermined operation direction; run a jump point search algorithm based on the direction-dependent movement cost to plan a global path point sequence connecting the starting point and the ending point, and use this as the center line to generate a global guide corridor with varying width according to the curvature of each path point in the sequence and the distance to the nearest obstacle;

[0022] Point cloud data of farmland is acquired using vehicle-mounted lidar or BeiDou differential positioning system. This data is projected onto a two-dimensional plane and divided into grid cells, for example, with a side length of 0.2m, to obtain a grid map. The east-west direction is set as the predetermined working direction. The cost of moving from one grid cell to an adjacent east-west grid cell is set to 1, the cost of moving to a north-south grid cell is set to 1.5, and the cost of moving to a diagonal grid cell is set to a value between 1.4 and 2.1 based on the angle between the grid cell and the east-west direction, thus achieving direction-dependent movement costs.

[0023] The direction-dependent movement cost is used as the basis for calculating path cost in the jump point search algorithm. Starting from the starting point, when searching for the next node, priority is given to jumping along the direction with the lowest cost until the sequence of path points with the lowest cost to the destination is found. Specifically, when initializing the grid map, an anisotropic movement cost matrix is ​​constructed, and the weight of moving a unit distance along a predetermined operation direction, such as crop row direction, is set to be lower than the weight of vertical or diagonal movement. For example, the weight is set to 1.0 for vertical and 3.0 for horizontal. During the node expansion and pruning process of the jump point search JPS algorithm, when calculating the cumulative cost from the parent node to the child node, the above weights are dynamically loaded according to the angle between the jump vector and the operation direction. This makes the priority queue tend to pop nodes that extend along the operation direction, and only tolerate high-cost lateral jumps when an obstacle is encountered and detour is necessary, by backtracking to the node with the lowest comprehensive cost.

[0024] Optionally, after obtaining the path point sequence, path smoothing is performed using methods such as B-splines. For each path point in the sequence, the angle between the vectors formed by the two points before and after the path point is calculated as an approximate value of the curvature of that point; simultaneously, the distance to the nearest obstacle grid cell is searched on the grid map with that path point as the center. The corridor width is calculated based on the curvature and obstacle distance; for example, the width is directly proportional to the obstacle distance and inversely proportional to the curvature, ensuring that the corridor narrows at turns or near obstacles and widens on straight sections or in open areas.

[0025] In an optional embodiment, setting a direction-dependent movement cost for each grid cell in the grid map includes:

[0026] The cost of moving along the predetermined work direction is set as the baseline movement cost, and the cost of moving perpendicular to the predetermined work direction is set as a preset multiple of the baseline movement cost.

[0027] Specifically, the farmland environment is rasterized, for example, divided into 1m × 1m grid cells. Assume the established direction of farmland operations is north-south, i.e., along the Y-axis of the raster map. When the farm machinery moves from one grid cell to an adjacent north-south grid cell, this movement is considered to conform to the regular operational process. Therefore, a low baseline movement cost is set, for example, a value of 1, representing the energy and time costs of traveling along the crops.

[0028] For movements perpendicular to the predetermined work direction, i.e., east-west movements, these movements typically correspond to agricultural machinery turning at both ends of a field or crossing multiple rows of crops. In practice, this is inefficient and should be avoided as much as possible. Therefore, the cost of such movements is set to a preset multiple of the baseline movement cost, for example, 1.5 times. The cost for the agricultural machinery to move from one grid cell to an adjacent east-west grid cell is 1.5. By setting this cost, global path planning algorithms, such as the A* algorithm, will tend to select paths with lower total costs when searching for paths, i.e., traveling as much as possible along the north-south direction, thus generating a global path with fewer twists and turns that meets the requirements.

[0029] For each path point p on the global path, a specific corridor width is calculated to obtain a strip region with continuously varying width. In an optional embodiment, generating a globally guiding corridor with varying width based on the curvature of each path point in the sequence and its distance to the nearest obstacle includes:

[0030] Based on the curvature K(p) of path point p and the distance to the nearest obstacle The width W(p) of the global guidance corridor at the path point is calculated using the following formula:

[0031] ;

[0032] Where k is a dimensionless proportionality constant. This is the curvature influence coefficient with the dimension of length.

[0033] For a path point p, calculate two key geometric and environmental parameters for that point. The first is the local curvature K(p) of the path. For example, the curvature K(p1) of a point p1 on a straight path segment is 0, while the curvature K(p2) of a point p2 on a turning arc with a radius of 10m is 0.1. The second is the distance from that point to the nearest obstacle. This is obtained by querying the environment map; for example, p1 is far from obstacles. The distance is 15m, and p2 is located at the obstacle avoidance turn. It is 5m.

[0034] Assuming the scaling factor k is 0.6, the curvature influence coefficient... The width is 20m. For point p1, the corridor width is calculated as follows: m provides a wide passageway. For point p2, the corridor width is calculated as... m. In areas where the path is straight and far from obstacles, the corridor is wider, allowing agricultural machinery more lateral adjustment space; while in areas where the path has a large curvature or is close to obstacles, the corridor will narrow to ensure safe passage through the section, as shown in Figure 2.

[0035] S2, In the velocity space of the dynamic window, the velocity pairs consisting of linear velocity and angular velocity are sampled non-uniformly, wherein the sampling of the angular velocity follows a Gaussian distribution, the expected value of the Gaussian distribution is set as the angular velocity required to align the agricultural machinery with the center line of the global guide corridor, and the standard deviation is proportional to the width of the global guide corridor at the current position of the agricultural machinery;

[0036] In the Dynamic Window Algorithm (DWA), the velocity space is a two-dimensional space comprised of all theoretically executable motion commands. Its horizontal and vertical coordinates correspond to linear velocity and angular velocity, respectively. Each coordinate point within the window represents a specific pair of control commands. By discretizing these points, the path planning problem can be transformed into an optimization problem of finding the highest-scoring coordinate point within the velocity space. Linear velocity determines the rate at which the agricultural machinery moves forward along its current direction, while angular velocity determines the speed and direction of the machinery's turning around its center of rotation. Non-uniform sampling can resolve the conflict between limited computational resources and control requirements. By using strategies such as Gaussian distribution to concentrate sampling points densely near high-confidence regions such as the desired correction value or the current velocity, it ensures that within a very short decision cycle, the trajectory most likely to conform to the global path direction or smoothing requirements is prioritized for evaluation.

[0037] A feasible range of linear and angular velocities, defined as a dynamic window, is determined under the constraints of the agricultural machinery's current speed and acceleration. Within this window, linear velocities are sampled at equal intervals. For angular velocities, the tangent direction of the global guidance corridor centerline 3 meters ahead of the agricultural machinery is first calculated. Using this as the target heading, the required angular velocity for the agricultural machinery to align with this heading in the next time step is calculated, and this angular velocity is used as the expected value of a Gaussian distribution. Next, the width of the global guidance corridor at the current position of the agricultural machinery is obtained, and the width is multiplied by a scaling factor to obtain the standard deviation of the Gaussian distribution. Based on this Gaussian distribution, a series of angular velocity sample values ​​are generated and combined with the linear velocity sample values ​​to form velocity pairs.

[0038] In an optional embodiment, the sampling of the angular velocity follows a Gaussian distribution, wherein:

[0039] Calculate the heading deviation between the current actual heading of the agricultural machinery and the tangent direction of the centerline of the global guidance corridor at the current position. Divide the heading deviation by the preset simulation duration to obtain the theoretical correction angular velocity, and set it as the expected value of the Gaussian distribution.

[0040] The standard deviation of the Gaussian distribution is proportional to the width of the global guidance corridor at the current position of the agricultural machinery.

[0041] Specifically, the heading angle of the agricultural machinery at the current moment is obtained, for example, 10°. Simultaneously, the projection point on the center line of the global guidance corridor is obtained, and the tangent direction at that point is calculated, i.e., the desired heading angle, for example, 15°. The difference between the two is the heading deviation. °. Assuming the preset simulation duration is 0.5s, then theoretically, the required angular velocity to eliminate this deviation within this duration is: ° / s. This value is set as the expected value of a Gaussian distribution, representing ideal steering control.

[0042] The standard deviation is proportional to the width of the global guidance corridor in which the agricultural machinery is currently located. For example, when the agricultural machinery is located in an open area with a width of 8m, the standard deviation is... It can be set to a relatively large value, such as 3° / s. The sampled angular velocity will fluctuate within a wide range, centered at 10° / s, thus obtaining a more diverse turning trajectory. When the agricultural machinery enters a narrow area with a width of only 2m, the standard deviation... The angular velocity will decrease accordingly, for example, to 0.5° / s. At this point, the sampled angular velocity will be concentrated around 10° / s. The Gaussian distribution-based sampling model, by adjusting the expected value and standard deviation, makes trajectory sampling exploratory in wide areas and focused on tracking in narrow areas, thus achieving adjustment of steering control under different environments, as shown in Figure 3.

[0043] S3. For each sampled velocity pair, a forward trajectory simulation is performed. All simulated trajectories are scored using an evaluation function that includes a heading guidance term, an obstacle distance term, and a terrain risk term. The heading guidance term assesses alignment with the global guidance corridor centerline, the obstacle distance term assesses obstacle avoidance safety, and the terrain risk term assesses terrain traversability. For each sampled velocity pair, the trajectory for a short future time period is calculated based on the motion model, such as a circular arc lasting 2 seconds. The motion model is a differential drive model or a simplified bicycle model. Using the sampled linear and angular velocities as control inputs through state transition equations, and under the premise of satisfying vehicle physical constraints, the robot's state in the two-dimensional plane is calculated in real-time using discrete-time integration, i.e., the robot's coordinates and heading angle at each future moment are calculated.

[0044] The circular trajectory is evaluated. The heading guidance score is the angle between the heading at the trajectory endpoint and the tangent to the centerline of the global guidance corridor corresponding to that point; a smaller angle results in a higher score. The obstacle distance score is the shortest distance from all points on the trajectory to the nearest obstacle, calculated using Euclidean distance; a larger distance results in a higher score. The terrain risk score requires a pre-made terrain risk map, which assigns a risk value to each grid cell, such as turbulence or soil moisture. The sum of the risk values ​​of all grid cells traversed by the trajectory is the score for this item; a lower risk results in a higher score. The three scores are then normalized and weighted to obtain the trajectory score.

[0045] In an optional embodiment, the scoring of all simulated trajectories using an evaluation function that includes a heading guidance term, an obstacle distance term, and a terrain risk term includes:

[0046] The evaluation value of the heading guidance item is calculated based on the absolute value of the angle between the orientation of the end point of the simulation trajectory and the tangent direction of the point on the center line of the global guidance corridor, and the evaluation value decreases as the absolute value of the angle increases.

[0047] The evaluation value of the obstacle distance item is determined by calculating the minimum distance between a point on the simulation trajectory and the nearest obstacle;

[0048] When the minimum distance is less than the preset safety distance, the evaluation value is proportional to the minimum distance; when the minimum distance is not less than the preset safety distance, the evaluation value is the maximum value.

[0049] The evaluation value of the terrain risk item is calculated based on the maximum terrain risk value in the grid cells traversed by the simulation trajectory, and the evaluation value decreases as the maximum terrain risk value increases.

[0050] For a velocity sampling pair, namely linear velocity v and angular velocity The generated simulation trajectory determines the position of the endpoint and the orientation angle of the agricultural machinery at that point. For example, at the end of a trajectory simulation, the orientation angle of the agricultural machinery is 85°.

[0051] Find the point closest to the endpoint on the centerline of the global guidance corridor, and use the tangent direction of this closest point as the desired heading angle for this location. For example, 90°. Calculate the absolute value of the angle between the heading angle and the desired heading angle, i.e., the heading deviation. °. The absolute value of the included angle reflects the degree of deviation between the orientation of the simulated trajectory's end and the ideal heading.

[0052] The evaluation value of the heading guidance item is calculated based on the absolute value of this included angle. The evaluation value is a value related to... A monotonically decreasing function. For example, a normalized evaluation function can be used. ,in It is a positive constant. The trajectory that follows a certain angle will receive a higher evaluation value than the other trajectory. The trajectory is 360°. The smaller the angle, that is, the closer the trajectory end is to the guiding direction, the higher the evaluation value, so that the selected trajectory can guide the agricultural machinery back and continue the direction of the global path.

[0053] Calculate the minimum distance to all obstacles in the environment for each simulated trajectory. This is achieved by sampling a series of discrete points on the trajectory, calculating the distance to the nearest obstacle for each point, and taking the minimum of these distances. For example, for a candidate trajectory, if a point on the trajectory is calculated to be 1.2m from an obstacle, then the trajectory is selected. It is 1.2m.

[0054] This minimum distance With a preset safe distance For comparison, this safety distance represents the minimum clearance that must be maintained between agricultural machinery and obstacles, for example, set at 2.0m. The evaluation value is calculated in two cases. When... Less than At that time, the trajectory intruded into the safety buffer, and the evaluation value was different from... Proportional. For example, a linear function can be used. Perform normalization. For The trajectory of m has a rating of 0.6. If the other trajectory... With a mere 0.5m, the evaluation value is 0.25, resulting in a more severe penalty.

[0055] when When the distance is ≥2.0m, it indicates that the trajectory maintains a sufficient safe distance from obstacles throughout the entire movement and is considered completely safe. In this case, a fixed maximum value is assigned to the evaluation score, such as 1.0. This segmented evaluation can penalize and differentiate trajectories of different degrees of danger, while giving the same highest evaluation to all trajectories that meet the safety requirements, thereby guiding the planner to prioritize safe paths.

[0056] Each grid cell in the pre-built terrain risk raster map is assigned a numerical value to represent the terrain risk of that area, such as slope, soil moisture, and degree of potholes. The risk value is normalized to between 0 and 1, where 0 represents the safest and flattest terrain and 1 represents the most dangerous or impassable area.

[0057] For each simulated trajectory, the series of grid cells traversed by that trajectory is determined. For example, if a trajectory passes through 5 grid cells from the start to the end, after consulting the terrain risk map, the risk values ​​corresponding to these cells are found to be 0.1, 0.1, 0.2, 0.4, and 0.2. The maximum value is then identified from this series of risk values. In this example It is 0.4.

[0058] The evaluation value is random A function that increases and decreases, penalizing trajectories that pass through dangerous areas. A simple implementation is... .for The trajectory has a terrain risk assessment value of 0.6. If another trajectory passes through an area with a risk value of 0.8, If the score is 0.8, the evaluation value drops to 0.2. Through this method, the evaluation system guides the local path planner to choose trajectories that avoid high-risk terrain, thereby improving the safety and stability of agricultural machinery operations. In another embodiment, the evaluation value is obtained by weighted summation after normalization of each evaluation item, using the following formula: Where N is the number of evaluation items, Let i be the weight of the i-th evaluation item. The values ​​are the normalized evaluation items. Optionally, to avoid collisions with personnel or field facilities, and to prevent agricultural machinery from overturning or getting stuck in potholes or steep slopes, the obstacle distance item and terrain risk item have dominant weights of 0.4 and 0.4 respectively. The sum of the weights of the heading guidance item, obstacle distance item, and terrain risk item is 1, so the weight of the heading guidance item is 0.2. However, those skilled in the art should understand that the above data is only a preferred embodiment, and the present invention is not limited to the specific weight values ​​mentioned above.

[0059] To ensure that local planning strictly follows the global guidance corridor with varying width generated in S1, and to prevent agricultural machinery from going out of the safe operating boundary during obstacle avoidance or correction, such as crushing crops, if the lateral deviation of any point on the trajectory exceeds half the width of the current corridor, the trajectory is determined to be an illegal trajectory, and its total score is directly set to negative infinity or it is removed.

[0060] S4. Select the simulation trajectory with the highest score as the optimal trajectory, output the corresponding velocity pair as the current control command, and adjust the maximum linear velocity of the dynamic window for the next planning cycle according to the terrain risk evaluation value of the optimal trajectory.

[0061] The algorithm iterates through the scores of all simulated trajectories and finds the trajectory with the highest score. The linear velocity and angular velocity of this trajectory are then sent as control commands to the underlying controller of the agricultural machinery. Simultaneously, the terrain risk evaluation value of this optimal trajectory is extracted. If the evaluation value is higher than the risk threshold, indicating poor road capacity ahead, the maximum permissible linear velocity is reduced by a fixed value in the dynamic window calculation of the next planning cycle, for example, from 2 meters per second to 1 meter per second. Conversely, if the evaluation value is lower than the threshold, the maximum permissible linear velocity is appropriately increased or restored to the default maximum value. The threshold is determined based on historical data. The algorithm is run on typical terrains of known safety and complexity to collect a large number of terrain risk score samples. By analyzing the probability distribution of the scores, high-scoring segments, such as the top 15% of high-scoring points, are used as the threshold for entering the low-speed zone.

[0062] In an optional embodiment, adjusting the maximum linear velocity of the dynamic window for the next planning period based on the terrain risk evaluation value of the optimal trajectory includes:

[0063] The maximum linear velocity for the next planning cycle is reduced based on the terrain risk evaluation value of the optimal trajectory, and the reduction in the maximum linear velocity is inversely proportional to the terrain risk evaluation value.

[0064] Specifically, at the end of the current planning cycle, an optimal trajectory has been selected from all simulation trajectories, and corresponding evaluation values ​​have been obtained, including the terrain risk evaluation value. Assuming the maximum linear velocity is set to 2.5 m / s, the terrain risk assessment value of the optimal trajectory in the current cycle is 0.7.

[0065] The dynamic window parameters for the next planning cycle, especially the maximum linear velocity, are updated based on this evaluation value. A lower rating indicates a higher risk level as the terrain is approaching or being entered, thus requiring a speed reduction. The reduction is inversely proportional to the rating; that is, the lower the rating, the greater the reduction. A concrete implementation model could be... ,in This is the baseline maximum linear velocity. In this example, the maximum linear velocity for the next cycle will be adjusted to... m / s.

[0066] At the start of the next planning cycle, the dynamic window method will limit the sampling range of linear velocity during velocity sampling to within [a certain range]. Within this range, the maximum linear velocity does not exceed 1.75 m / s. If the terrain risk assessment value of the optimal trajectory rises back to 0.95, the maximum linear velocity will be adjusted accordingly to 2.375 m / s, allowing the agricultural machinery to recover a relatively high speed on safe terrain. This adjustment model enables the agricultural machinery to autonomously decelerate according to terrain risk, becoming more cautious when encountering steep slopes or soft ground, thereby improving the safety of autonomous operation.

[0067] In the second embodiment, the present invention also proposes an adaptive parameter tuning system based on a dynamic window, comprising the following modules:

[0068] A generation module is used to acquire a grid map containing the predetermined operation direction of farmland, and to set a direction-dependent movement cost for each grid cell in the grid map, wherein the cost of moving along the predetermined operation direction is less than the cost of moving outside the predetermined operation direction; a jump point search algorithm is run based on the direction-dependent movement cost to plan a global path point sequence connecting the start point and the end point, and a global guide corridor with varying width is generated based on the curvature of each path point in the sequence and the distance to the nearest obstacle, using this as the center line;

[0069] The sampling module is used to perform non-uniform sampling of velocity pairs consisting of linear velocity and angular velocity within the velocity space of a dynamic window, wherein the sampling of the angular velocity follows a Gaussian distribution;

[0070] The evaluation module is used to simulate the forward trajectory of each sampled velocity pair and score all simulated trajectories using an evaluation function that includes heading guidance, obstacle distance, and terrain risk.

[0071] The adjustment module is used to select the simulation trajectory with the highest score as the optimal trajectory, output the corresponding velocity pair as the current control command, and adjust the maximum linear velocity of the dynamic window for the next planning cycle based on the terrain risk evaluation value of the optimal trajectory.

[0072] In an optional embodiment, setting a direction-dependent movement cost for each grid cell in the grid map includes:

[0073] The cost of moving along the predetermined work direction is set as the baseline movement cost, and the cost of moving perpendicular to the predetermined work direction is set as a preset multiple of the baseline movement cost.

[0074] In an optional embodiment, generating a globally guiding corridor with varying width based on the curvature of each path point in the sequence and its distance to the nearest obstacle includes:

[0075] Based on the curvature K(p) of path point p and the distance to the nearest obstacle The width W(p) of the global guidance corridor at the path point is calculated using the following formula:

[0076] ;

[0077] Where k is a dimensionless proportionality constant. This is the curvature influence coefficient with the dimension of length.

[0078] In an optional embodiment, the sampling of the angular velocity follows a Gaussian distribution, wherein:

[0079] Calculate the heading deviation between the current actual heading of the agricultural machinery and the tangent direction of the centerline of the global guidance corridor at the current position. Divide the heading deviation by the preset simulation duration to obtain the theoretical correction angular velocity, and set it as the expected value of the Gaussian distribution.

[0080] The standard deviation of the Gaussian distribution is proportional to the width of the global guidance corridor at the current position of the agricultural machinery.

[0081] In an optional embodiment, the scoring of all simulated trajectories using an evaluation function that includes a heading guidance term, an obstacle distance term, and a terrain risk term includes:

[0082] The evaluation value of the heading guidance item is calculated based on the absolute value of the angle between the orientation of the end point of the simulation trajectory and the tangent direction of the point on the center line of the global guidance corridor, and the evaluation value decreases as the absolute value of the angle increases.

[0083] The evaluation value of the obstacle distance item is determined by calculating the minimum distance between a point on the simulation trajectory and the nearest obstacle;

[0084] When the minimum distance is less than the preset safety distance, the evaluation value is proportional to the minimum distance; when the minimum distance is not less than the preset safety distance, the evaluation value is the maximum value.

[0085] The evaluation value of the terrain risk item is calculated based on the maximum terrain risk value in the grid cells traversed by the simulation trajectory, and the evaluation value decreases as the maximum terrain risk value increases.

[0086] In an optional embodiment, adjusting the maximum linear velocity of the dynamic window for the next planning period based on the terrain risk evaluation value of the optimal trajectory includes:

[0087] The maximum linear velocity for the next planning cycle is reduced based on the terrain risk evaluation value of the optimal trajectory, and the reduction in the maximum linear velocity is inversely proportional to the terrain risk evaluation value.

[0088] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0089] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0090] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. An adaptive parameter tuning method based on dynamic windows, characterized in that, Includes the following steps: A grid map containing the predetermined operation direction of the farmland is obtained, and a direction-dependent movement cost is set for each grid cell in the grid map, wherein the cost of moving along the predetermined operation direction is less than the cost of moving outside the predetermined operation direction; a jump point search algorithm is run based on the direction-dependent movement cost to plan a global path point sequence connecting the start and end points, and a global guidance corridor with varying width is generated based on the curvature of each path point in the sequence and the distance to the nearest obstacle; in the velocity space of the dynamic window, velocity pairs consisting of linear velocity and angular velocity are non-uniformly sampled, wherein the sampling of the angular velocity follows a Gaussian distribution; the sampling of the angular velocity following a Gaussian distribution includes: calculating the agricultural machinery's... The heading deviation between the actual heading and the tangent direction of the global guidance corridor centerline at the current position is calculated. This heading deviation is divided by a preset simulation duration to obtain the theoretical correction angular velocity, which is then set as the expected value of the Gaussian distribution. The standard deviation of the Gaussian distribution is proportional to the width of the global guidance corridor at the current position of the agricultural machinery. Forward trajectory simulation is performed on each sampled velocity pair, and all simulated trajectories are scored using an evaluation function that includes heading guidance, obstacle distance, and terrain risk. The simulation trajectory with the highest score is selected as the optimal trajectory, and the corresponding velocity pair is output as the current control command. Based on the terrain risk evaluation value of the optimal trajectory, the maximum linear velocity of the dynamic window in the next planning cycle is adjusted.

2. The method according to claim 1, characterized in that, The step of setting a direction-dependent movement cost for each grid cell in the grid map includes: setting the cost of moving along the predetermined operation direction as a baseline movement cost, and setting the cost of moving perpendicular to the predetermined operation direction as a preset multiple of the baseline movement cost.

3. The method according to claim 1, characterized in that, The step of generating a globally guiding corridor with varying width based on the curvature of each path point in the sequence and its distance to the nearest obstacle includes: based on the curvature K(p) of path point p and its distance to the nearest obstacle... The width W(p) of the global guidance corridor at the path point is calculated using the following formula: Where k is a dimensionless proportionality constant. This is the curvature influence coefficient with the dimension of length.

4. The method according to claim 1, characterized in that, The evaluation function, which includes a heading guidance item, an obstacle distance item, and a terrain risk item, scores all simulated trajectories. Specifically: the heading guidance item's evaluation value is calculated based on the absolute value of the angle between the orientation of the endpoint of the simulated trajectory and the tangent direction of that point on the center line of the global guidance corridor; the evaluation value decreases as the absolute value of the angle increases. The obstacle distance item's evaluation value is determined by calculating the minimum distance between a point on the simulated trajectory and the nearest obstacle; when the minimum distance is less than a preset safety distance, the evaluation value is proportional to the minimum distance; when the minimum distance is not less than the preset safety distance, the evaluation value is the maximum value. The terrain risk item's evaluation value is calculated based on the maximum terrain risk value among the grid cells traversed by the simulated trajectory; the evaluation value decreases as the maximum terrain risk value increases.

5. The method according to claim 1, characterized in that, The step of adjusting the maximum linear velocity of the dynamic window for the next planning period based on the terrain risk evaluation value of the optimal trajectory includes: reducing the maximum linear velocity for the next planning period based on the terrain risk evaluation value of the optimal trajectory, wherein the reduction in the maximum linear velocity is inversely proportional to the terrain risk evaluation value.

6. An adaptive parameter tuning system based on dynamic windows, characterized in that, It includes the following modules: a generation module, used to obtain a grid map containing the predetermined operation direction of farmland, and set a direction-dependent movement cost for each grid cell in the grid map, wherein the cost of moving along the predetermined operation direction is less than the cost of moving outside the predetermined operation direction; a jump point search algorithm is run based on the direction-dependent movement cost to plan a global path point sequence connecting the start point and the end point, and using this as the center line, a global guide corridor with varying width is generated according to the curvature of each path point in the sequence and the distance to the nearest obstacle; The sampling module performs non-uniform sampling of velocity pairs consisting of linear velocity and angular velocity within the velocity space of the dynamic window. The angular velocity sampling follows a Gaussian distribution. This Gaussian distribution includes: calculating the heading deviation between the current actual heading of the agricultural machinery and the tangent direction of the global guidance corridor centerline at the current position; dividing the heading deviation by a preset simulation duration to obtain the theoretical correction angular velocity; and setting this as the expected value of the Gaussian distribution. The standard deviation of the Gaussian distribution is proportional to the width of the global guidance corridor at the current position of the agricultural machinery. The evaluation module performs forward trajectory simulation on each sampled velocity pair and scores all simulated trajectories using an evaluation function that includes a heading guidance term, an obstacle distance term, and a terrain risk term. The adjustment module selects the highest-scoring simulated trajectory as the optimal trajectory, outputs the corresponding velocity pair as the current control command, and adjusts the maximum linear velocity of the dynamic window for the next planning cycle based on the terrain risk term evaluation value of the optimal trajectory.

7. The system according to claim 6, characterized in that, The step of setting a direction-dependent movement cost for each grid cell in the grid map includes: setting the cost of moving along the predetermined operation direction as a baseline movement cost, and setting the cost of moving perpendicular to the predetermined operation direction as a preset multiple of the baseline movement cost.

8. The system according to claim 6, characterized in that, The step of generating a globally guiding corridor with varying width based on the curvature of each path point in the sequence and its distance to the nearest obstacle includes: based on the curvature K(p) of path point p and its distance to the nearest obstacle... The width W(p) of the global guidance corridor at the path point is calculated using the following formula: Where k is a dimensionless proportionality constant. This is the curvature influence coefficient with the dimension of length.

9. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when executed by a processor, implements the method as described in any one of claims 1-5.

Images