A path planning method for weeding robots based on Beidou positioning

Abstract

The present application relates to the field of machine vision and intelligent agricultural technology, in particular to a weeding robot path planning method based on Beidou positioning. The present application comprises collecting farmland image to construct two-dimensional operation coordinate system, and the robot relies on Beidou differential positioning to collect multi-dimensional operation data and complete time series standardization; based on data, five types of potential field are constructed, and positive gain and reverse constraint potential field are distinguished and fused to generate global total coupling potential field; the gradient of the total potential field is calculated, and the optimized gradient vector is obtained after nonlinear gain correction; the operation path is generated along the gradient descent direction, and the driving speed is adaptively adjusted according to the gradient amplitude; the residual potential of the calculation area is used to identify the operation blind area, the compensation path is generated for the blind area and the low-speed compensation operation speed is matched, and finally the complete and residual-free robot operation path is output. The present application can autonomously avoid obstacles, protect seedlings, smooth trajectories and adaptively adjust speed, can identify operation blind area to realize compensation weeding, is suitable for various farmlands, has high degree of automation, complete weeding and lower energy consumption.

Description

Technical Field

[0001] This invention relates to the fields of machine vision and smart agriculture technology, specifically a path planning method for a weeding robot based on BeiDou positioning. Background Technology

[0002] Laser weeding, as a core technology of green precision agriculture, relies on machine vision recognition, high-energy laser thermal effect disinfection, and BeiDou autonomous navigation technology. It can effectively replace manual and chemical weeding, solving industry pain points such as pesticide residues, soil compaction, labor shortages, and low weeding efficiency. It is a key research and industrialization direction in the field of intelligent agricultural machinery. It uses a visual camera to identify weeds, traditional heuristic algorithms for path planning, preset fixed trajectories for field traversal, and constant laser power and exposure time for targeted weed control.

[0003] While existing technologies can achieve basic weed control, they still have the following technical limitations:

[0004] In areas with sparse weeds and flat terrain, environmental constraints are weak, potential field gradients are small, and robot driving force is insufficient, often resulting in problems such as robot shaking, tail wagging, and multiple trajectory inflection points. Existing methods can only correct the trajectory through post-processing algorithms such as filtering and curve smoothing, which is a post-event remedy and cannot eliminate path oscillations in principle. The existing crop repulsion potential model is a uniform circular isotropic structure, which is effective for protecting individual plants, but in orchards and vegetable fields with dense crops, the superposition of repulsion fields from multiple plants will form a potential energy balance point, causing the robot to stagnate, get stuck, and detour erratically, disrupting continuous operation and reducing weed coverage. Traditional solutions only update the potential field based on visible weeds in the current image, and weeds that are difficult to identify, such as those with shadows, seedlings, and recurring roots, are easily missed in a single scan. The system lacks a re-checking and rework mechanism, resulting in a lot of weed residue in the field and poor weed uniformity. The convergence threshold and shutdown judgment conditions of traditional algorithms are fixed values, which cannot adapt to different working conditions such as fields, orchards, and hills. Special personnel are required to debug the parameters on-site, resulting in high operation and maintenance costs and making it difficult to achieve large-scale fully automated unmanned weeding. Summary of the Invention

[0005] This invention provides a path planning method for a weeding robot based on BeiDou positioning.

[0006] The technical solution of this invention is as follows:

[0007] A path planning method for a weeding robot based on BeiDou positioning includes the following steps:

[0008] S1. Collect farmland images to generate farmland orthophoto maps and construct a two-dimensional operation coordinate system for farmland; the robot uses Beidou differential positioning to synchronously collect multi-dimensional raw operation data in the entire operation area and perform unified time-series standardized processing.

[0009] S2. Calculate the BeiDou positioning advantage potential field, weed driving potential field, crop safety repulsion potential field, terrain damping potential field, and laser energy efficiency potential field based on the processed multi-dimensional raw operation data. Use the BeiDou positioning advantage potential field, weed driving potential field, and laser energy efficiency potential field as the positive gain potential field, and the crop safety repulsion potential field and terrain damping potential field as the reverse constraint potential field. The total coupling potential field of the whole domain is obtained through fusion processing.

[0010] S3. The gradient vector of the original total potential field is obtained by calculating the partial derivative based on the total coupled potential field of the whole domain. The gradient is corrected by nonlinear gain processing to obtain the final path gradient vector after correction.

[0011] S4. Generate the weeding robot's working path along the gradient descent direction of the corrected final path, and adjust the weeding robot's speed according to the gradient magnitude of the corrected final path.

[0012] S5. Obtain the residual potential within the working area, determine the blind zone based on the residual potential, and if it is determined to be a blind zone, generate a compensation path and compensate the working speed of the weeding robot in the blind zone to obtain the final weeding robot path.

[0013] Furthermore, the working path and process of the weeding robot in S4 are as follows:

[0014] Based on the obtained corrected final path gradient vector, the weeding robot's working path is generated along the direction of maximum gradient descent, and the weeding robot's driving speed is adaptively adjusted.

[0015] Furthermore, the global coupled potential field in S2 is as follows:

[0016] ;

[0017] in, This represents the total coupling potential across the entire domain. For positive gain dynamic weights; For reverse constraint dynamic weights; The advantage of BeiDou positioning; Visual motive value for weeds; This represents the laser energy efficiency potential. The repulsion potential value for crop safety; This represents the terrain damping potential.

[0018] Furthermore, in S5, the blind zone of the operation is determined based on the residual potential, and the process is as follows:

[0019] If the residual potential at a point within the work area is greater than or equal to the residual threshold for weeding operations, then that point is determined to be a blind spot in the work area.

[0020] Furthermore, if the area is a blind spot in S5, a secondary weeding operation mechanism is triggered, and a dedicated compensation path and blind spot compensation operation speed are generated.

[0021] Furthermore, in S3, the gradient is corrected through nonlinear gain processing, as follows:

[0022] ;

[0023] in, This is the final path gradient vector after correction; The gradient vector of the original total potential field; This is the gradient gain enhancement coefficient; The baseline threshold for potential value normalization; The original global total coupling potential value of the region before correction.

[0024] Furthermore, in S1, the collected farmland image data is reconstructed in two dimensions to generate farmland orthophoto maps and construct a two-dimensional farmland operation coordinate system that includes information on plot boundaries, crop planting row distribution, and fixed obstacle locations.

[0025] Furthermore, S1 contains multi-dimensional raw operational data, including positioning accuracy parameters, visual feature parameters, crop parameters, terrain parameters, and operational parameters.

[0026] A path planning system for a weeding robot based on BeiDou positioning includes the following:

[0027] The robot acquires positioning data and preprocesses preprocessing modules, collects farmland images to generate farmland orthophoto maps, and constructs a two-dimensional operation coordinate system for farmland. The robot uses BeiDou differential positioning to synchronously collect multi-dimensional raw operation data across the entire operation area and performs unified time-series standardized processing.

[0028] The multi-dimensional potential field fusion calculation module calculates the BeiDou positioning advantage potential field, weed driving potential field, crop safety repulsion potential field, terrain damping potential field, and laser energy efficiency potential field based on the processed multi-dimensional raw operation data. The BeiDou positioning advantage potential field, weed driving potential field, and laser energy efficiency potential field are used as positive gain potential fields, and the crop safety repulsion potential field and terrain damping potential field are used as reverse constraint potential fields. The total coupling potential field of the whole domain is obtained through fusion processing.

[0029] The gradient adaptive correction module calculates the original total potential field gradient vector based on the global total coupled potential field through partial derivatives, and corrects the gradient through nonlinear gain processing to obtain the corrected final path gradient vector.

[0030] The path generation and speed control module generates the weeding robot's working path along the gradient descent direction of the corrected final path, and adjusts the weeding robot's speed according to the gradient magnitude of the corrected final path.

[0031] The blind spot residual compensation planning module obtains the residual potential within the working area, determines the blind spot based on the residual potential, and if it is determined to be a blind spot, it generates a compensation path and compensates for the working speed of the blind spot weeding robot to obtain the final weeding robot path.

[0032] The path generation and speed control module includes a preliminary path and a speed adjustment module. Based on the obtained corrected final path gradient vector, it generates the weeding robot's working path along the direction of maximum gradient descent and adaptively adjusts the weeding robot's travel speed.

[0033] The beneficial effects of this invention are as follows:

[0034] 1. This invention constructs five types of continuous, differentiable, independent potential fields and dynamically couples them with a global total potential field. The forward potential field quantifies the quality of BeiDou positioning, the degree of weed damage, and the energy efficiency of laser disinfection, guiding the robot to prioritize areas with accurate positioning, dense weeds, and lower laser operation energy consumption for weeding. The reverse potential field quantifies the crop safety repulsion force and terrain damping constraints, and sets a hard protection radius for crops based on a piecewise exponential repulsion potential field. When approaching plants, the laser can be directly shut down and the chassis can be braked to eliminate the risk of seedling damage. At the same time, terrain damping is set according to the slope to automatically avoid high-risk operation areas on steep slopes, protecting the chassis and laser tracking system for stable operation. All types of potential fields are updated in real time according to the field weeds, terrain, and positioning conditions. The dynamic weight balances the operation driving force and the risk avoidance constraint, eliminating the need for manual switching of operation modes.

[0035] 2. This invention corrects the potential field gradient vector through a nonlinear gain function, automatically amplifying the driving force in low-gradient regions to solve problems such as robot low-speed jitter, trajectory inflection points, and potential energy deadlock. In high-gradient regions, the original gradient value is maintained to avoid sudden path changes and violent shaking of the laser spot. The path is directly generated along the corrected gradient, eliminating the need for preset trajectories and post-smoothing steps, thus reducing computational overhead. The weeding robot adaptively and continuously adjusts its speed based on the gradient amplitude using a hyperbolic tangent function. Low-speed movement in densely weeded areas ensures weed eradication accuracy, while high-speed movement in open areas balances weeding quality and work efficiency, reducing equipment impact damage.

[0036] 3. This invention introduces the operation residual potential quantification of weed residue blind spots, accurately identifies areas missed in weed control due to complex terrain and sparse weeds, and adaptively adjusts the re-operation speed according to the severity of residue, achieving differentiated and refined secondary weeding and eliminating the defects of single fixed-path operations. At the same time, it abandons fixed shutdown thresholds and integrates the dynamic calculation of operation convergence standards based on weed control, laser, positioning, and slope parameters. After the entire area is controlled and the standard is met, the laser is automatically shut down and the machine is reset, achieving fully unmanned closed-loop operation. It takes into account stable positioning, crop protection, energy consumption optimization, weed control without dead angles, and all-scenario adaptability, effectively reducing the cost of manual intervention, equipment failure rate, and operation energy consumption, and improving the overall benefits of autonomous weeding operations in farmland.

[0037] 4. This invention relies on BeiDou RTK high-precision positioning and multi-sensor fusion navigation, combined with a complete technical system including multi-dimensional coupled artificial potential field, gradient adaptive correction, gradient linkage speed regulation, residual blind zone compensation, and dynamic convergence threshold. This system can comprehensively improve the positioning reliability, operational safety, weeding integrity, and operational economy of farmland weeding robots. At the same time, it uses UAVs and ground mapping equipment to construct a standardized grid map containing plots, crops, and obstacles, and uses BeiDou differential positioning to achieve centimeter-level positioning. It also uses filtering algorithms to correct pose deviations and continuously and stably outputs the robot's position, heading, and attitude information. This fundamentally solves the problems of easy drift and navigation failure due to signal interruption in single satellite positioning. The synchronous and standardized processing of multi-dimensional operational data also provides complete and reliable data support for intelligent path planning. Attached Figure Description

[0038] Figure 1 This is a cloud map of the total coupled potential field across the entire domain. Detailed Implementation

[0039] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. It should also be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the scope of the invention.

[0040] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0041] This embodiment provides a path planning method for a weeding robot based on BeiDou positioning, including the following steps:

[0042] S1. Collect farmland images and reconstruct farmland orthophoto maps to construct a two-dimensional coordinate system for farmland operations; the robot uses BeiDou differential positioning to synchronously collect multi-dimensional raw operation data across the entire operation area and perform unified time-series standardized processing.

[0043] In this embodiment of the invention, remote sensing images or ground image data of the target farmland are acquired by an image sensor mounted on a drone or ground mapping equipment. The acquired farmland image data is reconstructed in two dimensions to generate an orthophoto map of the farmland. A two-dimensional coordinate system for farmland operations is constructed, which includes information such as plot boundaries, crop planting row distribution, and the location of fixed obstacles, ensuring that the potential field is continuous and uninterrupted throughout the entire domain without any computational blind spots.

[0044] A high-precision BeiDou receiver and satellite antenna are installed on the weeding robot to receive BeiDou satellite navigation signals. Real-time dynamic differential positioning technology such as RTK-BDS is employed, combined with differential data from a ground reference station, to achieve centimeter-level positioning accuracy. Simultaneously, the weeding robot, equipped with a binocular vision camera, terrain slope sensor, and laser tracking system, collects multi-dimensional raw operational data across the entire work area. This data includes: positioning accuracy parameters such as BeiDou satellite signal-to-noise ratio, number of effective observation satellites, and positioning mean square error; visual characteristic parameters such as weed density, variety, root and stem thickness, growth height, and spatial distribution; crop parameters such as crop core coordinates, row spacing, canopy morphology, and crop height protection parameters; terrain parameters such as surface slope, terrain undulation, and road surface smoothness; and operational parameters such as standard laser disinfection dosage, real-time tracking angle, and laser output energy efficiency. All multi-dimensional raw operational data are synchronized and time-aligned, and standardized to provide raw data support for multi-dimensional potential field initialization, iterative optimization, and adaptive parameter adjustment.

[0045] In optimized environments, such as hilly areas and dense orchards, where BeiDou signals are severely obstructed, inertial measurement unit (IMU) and wheeled odometer data are fused with BeiDou positioning data using an extended Kalman filter algorithm to continuously output the position, heading, and attitude information of the weeding robot. A combined navigation system using dual-antenna orientation and gyroscopes is employed to obtain the robot's real-time heading angle. Lateral and heading deviations between the robot's actual pose and the planned path are corrected in real-time using a compensated Kalman filter algorithm.

[0046] S2. Calculate the BeiDou positioning advantage potential field, weed driving potential field, crop safety repulsion potential field, terrain damping potential field, and laser energy efficiency potential field based on the processed multi-dimensional raw operation data. Use the BeiDou positioning advantage potential field, weed driving potential field, and laser energy efficiency potential field as positive gain potential fields, and the crop safety repulsion potential field and terrain damping potential field as reverse constraint potential fields. Obtain the total coupling potential field of the entire domain through weighted coupling processing.

[0047] In this embodiment of the invention, a continuously differentiable physical potential field is constructed based on the multi-dimensional original operation data obtained in S1.

[0048] Specifically, this is used to represent the reliability of BeiDou positioning at various points in farmland, accurately avoiding signal blockage and positioning drift blind spots. The higher the positioning accuracy, the higher the operation priority. The calculation process is as follows:

[0049] ;

[0050] in, The BeiDou positioning advantage potential value at point P; The average signal-to-noise ratio of the satellite; To effectively observe the number of satellites; The mean square error of the positioning coordinates; The normalization coefficient is the coefficient for the entire domain. For the set unit reference constant (e.g., take...) Furthermore, the BeiDou positioning advantage potential value is a positioning quality excitation potential field, representing the reliability and accuracy advantage of BeiDou RTK positioning at any spatial point P in the farmland. This potential field does not characterize geographical location, but only indicates the quality of positioning reliability at the current location.

[0051] Specifically, the model is built based on the multi-dimensional features of weeds identified in real time by machine vision. Furthermore, the weights of weed type, root and stem thickness, and growth height are considered. The higher the degree of weed damage and the longer the period without operation, the stronger the driving force, and the robot will prioritize its operation. The robot is also precisely adapted to the differentiated laser pest control requirements. The process is as follows:

[0052] ;

[0053] in, Let P be the visual driving force value of weeds; The surface density of weeds; The comprehensive damage coefficient of weeds is calculated by weighting visual characteristics such as weed species, root and stem thickness, and growth height. This refers to the interval between regional operations. For scene adaptive correction coefficients; The standard weed areal density is the standard weed density threshold set by this system, used for dimensionless processing.

[0054] Specifically, an exponentially decaying repulsive potential field is constructed with the crop core as the center. Furthermore, the closer to the crop, the higher the repulsive potential value, forcing the robot and laser spot to stay away from the crop area, thus avoiding the risk of laser burns and mechanical crushing damage to seedlings from the source. The process is as follows:

[0055] ;

[0056] in, Let P be the crop safety repulsion potential value, which is the dangerous repulsion force between the weeding robot and the crop; The straight-line distance to the nearest crop core; The maximum repulsion amplitude coefficient is obtained by calibrating the field seedling protection safety threshold. The preferred calibration value in this invention is 0.85, which is used to limit the maximum repulsion intensity in the crop area. The repulsive potential space attenuation coefficient, unit: m −1 It is obtained by inversely solving the crop safety protection radius; This indicates the safe hard protection radius of the laser spot, which is the physical distance from the crop core to the hard protection boundary. Once triggered, it directly forces an emergency stop on the chassis and instantly shuts off the laser. Furthermore, It can be obtained by summing the actual radius of the laser spot, the vibration error of the weeding robot's robotic arm or chassis, and the safety redundancy. The maximum effective radius of the repulsive potential field is the farthest distance from which the core of the crop can affect the robot's operation. Beyond this radius, the crop becomes completely invisible to the robot.

[0057] Specifically, for hilly and undulating farmland terrain, the impact of terrain slope on the driving stability and laser tracking accuracy of weeding robots is quantified. The greater the slope and the more severe the terrain undulation, the higher the damping potential value, which inhibits operation in high-risk areas and ensures the accuracy of laser disinfection and equipment safety. The process is as follows:

[0058] ;

[0059] in, Let P be the topographic damping potential value; The slope of the ground; This is the normalized coefficient for terrain damping; The maximum slope angle that the weeding robot can safely climb; This represents the theoretical maximum value of the topographic damping potential field.

[0060] Specifically, by combining the multi-dimensional features of weeds identified in real-time using machine vision with the working conditions of the work area, the optimal laser operation efficiency at each point is quantified, representing the laser disinfection adaptability of the work area. The higher the energy efficiency adaptability, the higher the potential value, driving the weeding robot to prioritize and complete precise disinfection in the optimal working area. At the same time, it provides the underlying basis for dynamic adjustment of laser parameters, completely solving the problem of energy efficiency imbalance with fixed laser parameters. The process is as follows:

[0061] ;

[0062] in, Let P be the laser energy efficiency potential value; The optimal laser dose for the weeds at this location; Standard laser dose; The laser tracking and aiming adaptation coefficient is obtained by weighting the area flatness and positioning accuracy, where the laser tracking and aiming system is existing technology. This is the laser energy efficiency normalization coefficient; To prevent the fine-tuning constant from being zero in the denominator, a very small positive number is taken.

[0063] Furthermore, this potential field corresponds to the optimal dosage required by the weeds. A negative correlation is observed, causing the chassis to prioritize approaching low-energy-consumption, high-accuracy tracking work areas, thus achieving priority scheduling. Simultaneously, the vision system will... The parameters are sent to the laser driver layer in real time, replacing the fixed calibration parameters, and enabling bidirectional dynamic adaptation between scheduling decisions and energy execution.

[0064] Specifically, each independent potential field is merged, and a globally continuous, differentiable, and real-time iterative total potential field space is generated based on the forward operation gain potential field and the reverse constraint potential field. The specific process is as follows:

[0065] ;

[0066] in, This represents the total coupling potential across the entire domain. The positive gain dynamic weight controls the operational tendency of the positive potential field in the total potential field; the larger the value, the more aggressively the robot goes to the task. The reverse constraint dynamic weight controls the defensive or risk-avoidance tendency of the reverse potential field in the total potential field. The larger the value, the more conservatively the robot will detour or park.

[0067] Furthermore, the BeiDou positioning advantage field, the weed visual driving potential field, and the laser energy efficiency adaptation field serve as positive gain potential fields, superimposed to enhance the priority of weeding operations in the region; the crop safety repulsion potential field and the terrain damping potential field serve as negative constraint potential fields, avoiding risks and violations in weeding operations. Real-time synchronous updates ensure the overall potential field dynamically evolves with field conditions, achieving adaptation across the entire operational area.

[0068] S3. The original total potential field gradient vector is obtained by calculating the partial derivative of the global total coupled potential field. The gradient is then corrected by nonlinear gain processing to obtain the corrected final path gradient vector.

[0069] The gradient vector acting on the original total potential field does not change the scalar value of the original potential field, and is specifically designed to solve problems such as weak gradients in smooth regions, robot low-speed jitter, and trajectory inflection points. Through adaptive correction of the gradient vector using a nonlinear gain function, weak gradients are enhanced and strong gradients are stabilized, achieving smoothness of the original trajectory in the pure potential field.

[0070] The gradient correction process is shown below:

[0071] ;

[0072] in, This is the final path gradient vector after correction, and it serves as the sole basis for path generation. Given the original total potential field gradient vector, we obtain the vector gradient by taking the two-dimensional partial derivative using spatial coordinates; This is the gradient gain enhancement coefficient, with a value ranging from 0.2 to 0.6; The baseline threshold for potential value normalization; The original global total coupling potential value of the region before correction.

[0073] In this embodiment of the invention, based on the total potential field scalar value And combined with potential value normalization benchmark threshold Precisely define the types of weeding operation areas and establish standardized judgment rules:

[0074] When the regional potential satisfies When the positive and negative potential fields at the current position cancel each other out, the weeding robot is at a potential energy equilibrium point or a region with a very small gradient. At this time, the adaptive exponential term... The gradient adaptive gain mechanism takes effect, dynamically amplifying the weak original gradient, which can fundamentally solve the problem of low-speed jitter and trajectory inflection distortion caused by insufficient driving force in flat areas, and enhance the robot's ability to escape low potential deadlock areas.

[0075] When the regional potential satisfies When the current location exhibits significant positive operational attraction, such as in dense grass areas, or negative safety constraints, such as near-crop / steep slope, the adaptive exponent term is indicated. The gradient gain coefficient approaches 1, maintaining a stable output of the original gradient solution value, effectively avoiding path abrupt changes and laser tracking system spot jitter caused by excessive amplification under high gradient conditions.

[0076] Furthermore, when Only then will there be no benefit at all.

[0077] S4. Generate the weeding robot's working path along the gradient descent direction of the corrected final path, and adjust the weeding robot's speed according to the gradient magnitude of the corrected final path.

[0078] Based on the obtained corrected final path gradient vector The system generates a working path along the direction of maximum gradient descent, without preset trajectories, iterative optimization, or post-smoothing. Simultaneously, it adaptively adjusts the driving speed based on the gradient magnitude, achieving a dynamic balance between precise disinfection and efficient passage.

[0079] Adaptive speed regulation, the process is as follows:

[0080] ;

[0081] in, The adaptive baseline operating speed for the weeding robot's first operation at point P; For maximum and minimum operating speeds; This is the gradient magnitude scaling normalization factor, used to map the original gradient magnitude to... The effective sensitive interval of the function is [0,3]. It is a hyperbolic tangent normalized function. The input gradient magnitude is dimensionless, and the output result is dimensionless. The velocity difference is multiplied by the dimensionless coefficient and then the dimensions are unified to ensure a smooth velocity transition.

[0082] Furthermore, when the robot is in an area with dense, high-gradient weeds, its speed automatically approaches the lower limit. To ensure laser tracking accuracy; when in a low-gradient open area, the vehicle speed automatically increases to the maximum limit. This enables efficient scene switching between no-load operation and the weeding robot simply walking around without doing any work.

[0083] S5. Obtain the residual potential within the working area, determine the blind zone based on the residual potential, and if it is determined to be a blind zone, generate a compensation path and compensate the working speed of the weeding robot in the blind zone to obtain the final weeding robot path.

[0084] In this embodiment of the invention, only a single fixed-path weeding operation is completed, which cannot identify blind spots in low-speed, complex terrain, or sparsely weeded areas. This step, based on the potential field gradient decay characteristics and speed operation coverage characteristics, performs full-area residual identification and secondary supplementary operation planning. By quantifying the completeness of regional operation coverage, the area of ​​weed residue is accurately determined, achieving full-area coverage of weeding operations without dead corners.

[0085] The residual potential of the regional operation is shown below:

[0086] ;

[0087] in, Let P be the operational residual potential, used to represent the risk of weed residue and the severity of operational blind spots at that point; Let P be the visual driving force value of weeds, representing the original degree of weed damage and the intensity of weeding operation demand; is the global residual decay coefficient, and is a fixed calibration constant used to normalize the decay weights of operation coverage time and speed; The duration of a single complete operation in the current work area; The real-time time variable for the task has a value range of [range missing]. ; The real-time spatial operation point of the weeding robot at time s; The weeding robot is at the work site at time s. Adaptive adjustment of work speed at the location; Let s be the total potential field gradient vector after correction at time s, representing the real-time operation driving strength and path guidance accuracy; The magnitude of the gradient vector of the total potential field after correction represents the effective driving force of weeding operations at this location; the integral is a quantitative value of the sufficiency of regional operation coverage, and the larger the value, the more complete the initial operation coverage of the region. The higher the value, the higher the probability of weed residue and the more significant the blind spot characteristics.

[0088] In this embodiment of the invention, the blind spot compensation determination process is as follows:

[0089] when If the location is automatically identified as a blind spot, a secondary weeding operation mechanism will be triggered, and a dedicated compensation path will be generated. To preset the residual threshold for weeding operations, the parameters are calibrated based on the type of crop in the field, the characteristics of the weeds, and the accuracy of laser pest control; the compensation operation speed is as follows:

[0090] ;

[0091] in, To compensate for the blind spot at point P; The adaptive baseline operation speed for the first operation at point P; The preset residual judgment threshold is set. Let P be the real-time residual potential value of the operation. This formula can achieve a lower compensation operation speed in areas with higher residuals and more severe blind spots, thus achieving refined low-speed weeding operations; in areas with no blind spots and where the residuals meet the standards, the normal operation speed is maintained, realizing differentiated intelligent operations across the entire area, and completely solving the problems of weed residue and blind spots in traditional fixed-path single operations.

[0092] Furthermore, this invention can also eliminate the fixed termination threshold and dynamically solve the operation convergence conditions based on multi-dimensional working condition parameters across the entire domain, adapting to working conditions in all scenarios such as fields, orchards, and hilly areas.

[0093] Dynamic threshold formula: ;

[0094] in, The threshold for real-time adaptive job termination; , The average value of laser energy efficiency potential for all weeds in the area; Correction coefficients are used to adjust the average positioning error of BeiDou. ; For the average slope of the entire area, a corresponding correction factor is applied. ; This is the weighted correction factor.

[0095] Real-time traversal of the six-dimensional spatiotemporal total potential value of all spatial points in the entire domain When the potential values ​​of all points in the entire region are less than the current scene's adaptive dynamic convergence threshold Once it is determined that all visible weeds, hidden weeds, and weeds remaining at the edges have been eradicated, the system automatically shuts down the laser eradication module, stops moving, and resets to standby mode, completing a fully unmanned, fully automatic closed-loop weeding operation.

[0096] This embodiment provides a path planning system for a weeding robot based on BeiDou positioning, including the following:

[0097] The robot acquires positioning data and preprocesses preprocessing modules, collects farmland images to generate farmland orthophoto maps, and constructs a two-dimensional operation coordinate system for farmland. The robot uses BeiDou differential positioning to synchronously collect multi-dimensional raw operation data across the entire operation area and performs unified time-series standardized processing.

[0098] The multi-dimensional potential field fusion calculation module calculates the BeiDou positioning advantage potential field, weed driving potential field, crop safety repulsion potential field, terrain damping potential field, and laser energy efficiency potential field based on the processed multi-dimensional raw operation data. The BeiDou positioning advantage potential field, weed driving potential field, and laser energy efficiency potential field are used as positive gain potential fields, and the crop safety repulsion potential field and terrain damping potential field are used as reverse constraint potential fields. The total coupling potential field of the whole domain is obtained through fusion processing.

[0099] The gradient adaptive correction module calculates the original total potential field gradient vector based on the global total coupled potential field through partial derivatives, and corrects the gradient through nonlinear gain processing to obtain the corrected final path gradient vector.

[0100] The path generation and speed control module generates the weeding robot's working path along the gradient descent direction of the corrected final path, and adjusts the weeding robot's speed according to the gradient magnitude of the corrected final path.

[0101] The blind spot residual compensation planning module obtains the residual potential within the working area, determines the blind spot based on the residual potential, and if it is determined to be a blind spot, it generates a compensation path and compensates for the working speed of the blind spot weeding robot to obtain the final weeding robot path.

[0102] See attached document Figure 1 This is a cloud map of the total coupled potential field fusion in an embodiment of the present invention. It fuses the positioning, weed, and laser positive potential fields with the crop and terrain constraint potential fields. The color scale represents the overall operation priority, with high potential areas being priority operation areas. Among them, a two-dimensional farmland grid coordinate system is used as the basis, and gradient color scales are used to represent the magnitude of the total coupled potential value. Warm color areas represent high total potential values ​​and high operation priority of weeding robots, while cool color areas represent low total potential values, the existence of avoidance constraints, or no operation needs.

Claims

1. A path planning method for a weeding robot based on BeiDou positioning, characterized in that, Includes the following steps: S1. Collect farmland images to generate farmland orthophoto maps and construct a two-dimensional operation coordinate system for farmland; the robot uses Beidou differential positioning to synchronously collect multi-dimensional raw operation data in the entire operation area and perform unified time-series standardized processing. S2. Calculate the BeiDou positioning advantage potential field, weed driving potential field, crop safety repulsion potential field, terrain damping potential field, and laser energy efficiency potential field based on the processed multi-dimensional raw operation data. Use the BeiDou positioning advantage potential field, weed driving potential field, and laser energy efficiency potential field as the positive gain potential field, and the crop safety repulsion potential field and terrain damping potential field as the reverse constraint potential field. The total coupling potential field of the whole domain is obtained through fusion processing. S3. The gradient vector of the original total potential field is obtained by calculating the partial derivative based on the total coupled potential field of the whole domain. The gradient is corrected by nonlinear gain processing to obtain the final path gradient vector after correction. S4. Generate the weeding robot's working path along the gradient descent direction of the corrected final path, and adjust the weeding robot's speed according to the gradient magnitude of the corrected final path. S5. Obtain the residual potential within the working area, determine the blind zone based on the residual potential, and if it is determined to be a blind zone, generate a compensation path and compensate the working speed of the weeding robot in the blind zone to obtain the final weeding robot path.

2. The path planning method for a weeding robot based on BeiDou positioning according to claim 1, characterized in that, The working path and process of the weeding robot in S4 are as follows: Based on the obtained corrected final path gradient vector, the weeding robot's working path is generated along the direction of maximum gradient descent, and the weeding robot's driving speed is adaptively adjusted.

3. The path planning method for a weeding robot based on BeiDou positioning according to claim 1, characterized in that, The global coupled potential field in S2 is as follows: ； in, This represents the total coupling potential across the entire domain. For positive gain dynamic weights; For reverse constraint dynamic weights; The advantage of BeiDou positioning; Visual motive value for weeds; This represents the laser energy efficiency potential. The repulsion potential value for crop safety; This represents the terrain damping potential.

4. The path planning method for a weeding robot based on BeiDou positioning according to claim 1, characterized in that, In S5, the blind zone of the operation is determined based on the residual potential, and the process is as follows: If the residual potential at a point within the work area is greater than or equal to the residual threshold for weeding operations, then that point is determined to be a blind spot in the work area.

5. The path planning method for a weeding robot based on BeiDou positioning according to claim 4, characterized in that, If the area is a blind spot in S5, a secondary weeding operation mechanism will be triggered, and a dedicated compensation path and blind spot compensation operation speed will be generated.

6. The path planning method for a weeding robot based on BeiDou positioning according to claim 1, characterized in that, In S3, the gradient is corrected through nonlinear gain processing, as follows: ； in, This is the final path gradient vector after correction; The gradient vector of the original total potential field; This is the gradient gain enhancement coefficient; The baseline threshold for potential value normalization; The original global total coupling potential value of the region before correction.

7. The path planning method for a weeding robot based on BeiDou positioning according to claim 1, characterized in that, In S1, the collected farmland image data is reconstructed in two dimensions to generate farmland orthophoto maps and construct a two-dimensional farmland operation coordinate system that includes information on plot boundaries, crop planting row distribution, and fixed obstacle locations.

8. The path planning method for a weeding robot based on BeiDou positioning according to claim 1, characterized in that, S1 contains multi-dimensional raw operational data, including positioning accuracy parameters, visual feature parameters, crop parameters, terrain parameters, and operational parameters.

9. A path planning system for a weeding robot based on BeiDou positioning, used to implement the path planning method for a weeding robot based on BeiDou positioning as described in claim 1, characterized in that, Includes the following: The robot acquires positioning data and preprocesses preprocessing modules, collects farmland images to generate farmland orthophoto maps, and constructs a two-dimensional operation coordinate system for farmland. The robot uses BeiDou differential positioning to simultaneously collect multi-dimensional raw operation data across the entire operation area and performs unified time-series standardized processing. The multi-dimensional potential field fusion calculation module calculates the BeiDou positioning advantage potential field, weed driving potential field, crop safety repulsion potential field, terrain damping potential field, and laser energy efficiency potential field based on the processed multi-dimensional raw operation data. The BeiDou positioning advantage potential field, weed driving potential field, and laser energy efficiency potential field are used as positive gain potential fields, and the crop safety repulsion potential field and terrain damping potential field are used as reverse constraint potential fields. The total coupling potential field of the whole domain is obtained through fusion processing. The gradient adaptive correction module calculates the original total potential field gradient vector based on the global total coupled potential field through partial derivatives, and corrects the gradient through nonlinear gain processing to obtain the corrected final path gradient vector. The path generation and speed control module generates the weeding robot's working path along the gradient descent direction of the corrected final path, and adjusts the weeding robot's speed according to the gradient magnitude of the corrected final path. The blind spot residual compensation planning module obtains the residual potential within the working area, determines the blind spot based on the residual potential, and if it is determined to be a blind spot, it generates a compensation path and compensates for the working speed of the blind spot weeding robot to obtain the final weeding robot path.

10. A path planning system for a weeding robot based on BeiDou positioning according to claim 9, characterized in that, The path generation and speed control module includes a preliminary path and a speed adjustment module. Based on the obtained corrected final path gradient vector, it generates the weeding robot's working path along the direction of maximum gradient descent and adaptively adjusts the weeding robot's travel speed.

Images