Smooth speed control method and system based on predictive feedback for laser weeding robot

Abstract

This invention discloses a smooth speed control method and system for a laser weeding robot based on predictive feedback, belonging to the field of automation control. The method includes: a prediction stage, in which weeds are identified and an ideal speed is calculated based on a front-facing camera; a time alignment stage, in which the delay is dynamically calculated based on the distance between the camera and the actuator and the current actual speed, and an alignment speed is selected from historical ideal speeds to eliminate spatiotemporal misalignment; a speed smoothing stage, in which the predicted speed after time alignment is frequency-domain stabilized, and high-frequency disturbances caused by visual recognition fluctuations and changes in the number of weeds are suppressed by low-pass filtering to obtain a continuous and consistent predicted speed command; an acceleration constraint planning stage, in which a physically achievable planned speed is generated under the maximum acceleration limit; and a smooth PID dynamic tracking stage, in which the actuator is driven to track the planned speed by a smooth PID controller, and the actual speed is fed back to the time alignment stage to form a closed loop.

Description

Technical Field

[0001] This invention relates to the field of automated control and intelligent equipment, and in particular to a smooth speed control method and system for a laser weeding robot based on predictive feedback, which can be used in multi-camera collaborative control, target prediction and smooth speed control systems. Background Technology

[0002] In multi-sensor cooperative control systems, a common structure is a combination of a front-mounted predictive camera and a rear-mounted execution camera. The predictive camera identifies information such as the distribution, quantity, and size of the landmass or target ahead, and calculates the ideal speed accordingly; while the execution camera or actuator performs the actual operation at the corresponding time point after traveling a certain spatial distance. However, this structure has the following problems:

[0003] 1. Delay misalignment problem: The ideal speed obtained by the predictive camera is based on an estimate of the future position, while the execution camera has a time delay. Directly applying the ideal speed will cause a mismatch in control response.

[0004] 2. Frequent perturbation problem: The recognition result of the prediction camera changes dynamically in real time, and the ideal speed value v fluctuates constantly, causing frequent jumps in the system input;

[0005] 3. Speed ​​mutation problem: Due to the physical inertia of the controlled object, sudden changes in the ideal speed value can cause the actuator to jitter, overshoot, or even decrease in stability.

[0006] 4. Target disappearance / switching problem: Identifying the appearance and disappearance of a target in the camera's field of view will cause instantaneous discontinuity in speed commands, which traditional PID cannot smoothly transition.

[0007] To address the aforementioned issues, a control strategy is needed that can achieve a smooth transition at an ideal speed while taking into account the prediction-execution time delay. Summary of the Invention

[0008] This invention overcomes the shortcomings of existing technologies and provides a smooth speed control method and system for laser robots based on real-time feedback from predictive cameras. Under the condition that there is a fixed distance between the predictive camera and the execution camera, by introducing time alignment, low-pass filtering, acceleration limiting and smooth PID control, the speed of the vehicle or execution system can be adjusted stably, continuously and smoothly.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a smooth speed control method for laser weeding robots based on predictive feedback, comprising the following steps:

[0010] S1. Prediction phase;

[0011] Based on the target images acquired by the predictive camera, weed distribution is identified using a target detection algorithm, and the physical size S of each target is calculated. i :

[0012] ;

[0013] In the formula, S i Let A be the physical size of the i-th target; i Let K be the projected area of ​​the target in the image; K is the set of camera calibration parameters.

[0014] According to target category C i and size S i The single-target processing time T is calculated using a piecewise function. i :

[0015] ;

[0016] In the formula, T i C is the processing time for the i-th target; i For target category identifiers, such as C=1 for broadleaf grass and C=2 for grass; a1, b1, c1, a2, b2, c2 are experimental calibration parameters;

[0017] Based on the single-target processing time T i The total processing time T for all targets is obtained by summing the results. total :

[0018] ;

[0019] The total processing time T for all targets within the comprehensive prediction window total Calculate the ideal driving speed:

[0020] ;

[0021] In the formula, L is the unit working stroke length; T min This is the system's safe time threshold.

[0022] S2. Time alignment phase;

[0023] Calculate the prediction-execution time latency:

[0024] ;

[0025] In the formula, D is the physical distance between the prediction camera and the actuator; v(t) is the current execution speed; To prevent the value from being removed from zero;

[0026] Based on the delay time τ(t), the predicted velocity sequence is obtained from historical data. Select the closest one Speed ​​of time As an alignment reference;

[0027] S3. Speed ​​smoothing phase;

[0028] Speed ​​after time alignment Perform low-pass filtering:

[0029] ;

[0030] In the formula, α represents the filtered velocity; α is the filtering coefficient.

[0031] S4. Acceleration constraint stage;

[0032] Based on the maximum permissible acceleration a max Given the control period Δt, calculate the upper limit of the velocity increment:

[0033] ;

[0034] In the formula, a max Δt is the maximum permissible acceleration of the actuator; Δt is the control period.

[0035] Generate a physically executable planning speed v plan (t):

[0036] ;

[0037] In the formula, To achieve a smooth speed for the target;

[0038] S5. Smoothing the PID dynamic tracking stage;

[0039] Calculate the error between the planned speed and the actual speed. ;

[0040] Control inputs are generated using a smoothed PID controller:

[0041] ;

[0042] In the formula, K p K i K d Here are the parameters for the PID controller; e(t) is the speed tracking error.

[0043] Apply acceleration limiting to the output:

[0044] ;

[0045] In the formula, Δv is the actual velocity increment;

[0046] Update actual execution speed: .

[0047] Furthermore, in step S2, the target time point is calculated using the following formula:

[0048] ;

[0049] And predict velocity sequences from historical data Selecting from and The velocity value at the nearest moment is used as the alignment reference.

[0050] Furthermore, in the low-pass filtering process, the range of the filtering coefficient α is 0 < α < 1, and α is adaptively adjusted or fixedly set according to the system noise characteristics to achieve a smooth transition of the speed signal.

[0051] Furthermore, in the acceleration constraint phase, the maximum permissible acceleration a max Dynamic settings based on the physical inertia of the actuator ensure that speed changes comply with mechanical limits.

[0052] This invention discloses a system based on a predictive feedback-based smooth speed control method for a laser weeding robot, comprising the following modules:

[0053] The prediction module is used to identify weed distribution and calculate the ideal velocity v based on the target image acquired by the prediction camera. ideal (t);

[0054] The time alignment module is used to calculate the prediction-execution time delay τ(t) and perform time alignment with the ideal speed;

[0055] The speed smoothing module is used to perform low-pass filtering on the aligned speed to suppress high-frequency disturbances caused by visual recognition fluctuations.

[0056] The acceleration constraint module is used to generate the planned velocity v under the maximum acceleration constraint. plan (t);

[0057] The smooth PID dynamic tracking module is used for closed-loop control based on the error between the planned speed and the actual speed.

[0058] The execution and feedback module is used to drive the actuator to output control speed and feed back the actual speed to the time alignment module and the prediction module to form closed-loop control.

[0059] Furthermore, the prediction module also includes a target size calculation unit, calculated using the formula: Estimate the physical size of the target.

[0060] Furthermore, the system also includes a dynamic update module, which refreshes the ideal velocity sequence in each control cycle and ensures that each module executes sequentially: prediction → alignment → smoothing → planning → execution → feedback.

[0061] The advantages of this invention compared to existing technologies are as follows: By introducing multi-level mechanisms such as time alignment, low-pass filtering, acceleration constraints, and smooth PID control, this invention effectively solves key problems in multi-sensor cooperative control systems. The advantages of this invention compared to existing technologies are mainly reflected in the following aspects.

[0062] 1. Significantly improves system stability and smoothness, reducing overshoot and oscillation;

[0063] This invention suppresses high-frequency fluctuations and abrupt changes in ideal speed through a speed smoothing module (low-pass filter) and an acceleration constraint module. This results in a smoother system response and avoids the jitter and overshoot caused by traditional PID controllers that directly track abrupt signals.

[0064] 2. Effectively solves the problem of prediction-execution delay misalignment, improving control accuracy;

[0065] In this invention, the time alignment module calculates the prediction-execution time delay and selects the alignment speed from the historical predicted speed sequence, ensuring spatial matching between the ideal speed and the execution position. This dynamic delay compensation mechanism improves the system's robustness in variable speed scenarios, enabling the weeding robot to accurately align with the target position and reduce missed or incorrect weeding.

[0066] 3. Enhance the system's anti-interference capability and suppress frequent disturbances and target switching issues;

[0067] This invention effectively addresses the real-time fluctuations in predictive camera recognition results and the problem of target disappearance / switching through a multi-level smoothing mechanism (low-pass filtering, acceleration constraints, and smoothed PID). For example, the low-pass filter coefficient α filters out recognition noise, while the smoothed PID controller achieves dynamic and smooth transition of errors, avoiding the discontinuous commands of traditional PID controllers when the target suddenly appears or disappears. This allows the system to maintain continuous adjustment even under conditions of frequent changes in ideal speed, improving the adaptability and reliability of agricultural machinery.

[0068] 4. High physical feasibility, ensuring the safety and lifespan of the actuator;

[0069] In this invention, the acceleration constraint module ensures that the planned speed conforms to the mechanical inertia of the actuator by limiting the upper limit of the speed increment. This solves the problem of sudden speed changes and prevents mechanism vibration, overshoot, or decreased stability caused by abrupt changes in ideal speed. This constraint not only improves system stability but also extends the service life of mechanical components, making it suitable for high-intensity agricultural operating environments.

[0070] 5. Achieve adaptive closed-loop optimization to improve the system's intelligence level;

[0071] This invention achieves dynamic self-optimization capabilities through a feedback closed-loop and error analysis module. It can dynamically adjust the delay time based on real-time feedback and optimize parameters according to the difference between the planned speed and the ideal speed. This adaptive mechanism enables the system to cope with different operating scenarios (such as changes in weed density) without human intervention, improving the autonomy and efficiency of intelligent agricultural machinery.

[0072] 6. Wide range of applications, validated in multiple experimental scenarios;

[0073] This invention is applicable to scenarios such as intelligent combat robots, intelligent agricultural machinery, automatic inspection systems, and multi-camera collaborative control. This versatility is due to its modular design (prediction, alignment, smoothing, planning, execution, and feedback), ensuring scalability to other automation fields.

[0074] Through the aforementioned technical advantages, this invention comprehensively improves the control performance of the laser weeding robot, achieving stable, continuous, and smooth speed adjustment. Attached Figure Description

[0075] The present invention will now be further described with reference to the accompanying drawings.

[0076] Figure 1 This is a schematic diagram of the process of the present invention.

[0077] Figure 2 This is a flowchart illustrating the predicted velocity generation module in this invention. Detailed Implementation

[0078] The present invention will be further described below with reference to specific embodiments.

[0079] like Figure 1 , Figure 2 As shown, this invention discloses a laser weeding robot with a smooth speed control method and system based on predictive feedback. The system executes sequentially within each control cycle Δt, forming a closed-loop control chain of prediction → alignment → smoothing → planning → execution → feedback. Initially, parameters such as maximum acceleration a are set. max The following steps describe how to initialize the speed sequence, including setting the filter coefficient α, PID parameters, etc. The steps are explained below. Figure 1 This is a flowchart illustrating the closed-loop control process of the method of the present invention, showing the complete process from prediction, time alignment, velocity smoothing, acceleration constraints to PID dynamic tracking and feedback. Figure 2 for Figure 1 A detailed flowchart illustrating the workflow of the mid-prediction phase (ideal velocity generation module).

[0080] Step 1: Prediction Phase – Ideal Velocity Generation;

[0081] Based on the target images acquired by the predictive camera, weed distribution is identified, and the processing time for a single target is calculated according to the target category and size. Then, the ideal driving speed is calculated by combining the results.

[0082] This stage is performed by the prediction module, and its purpose is to identify targets in front in real time based on the front-facing camera and calculate the ideal speed. The working process is as follows:

[0083] Predictively capture images from the camera and identify weed category C using object detection algorithms (such as YOLO or CNN). i (such as broadleaf grasses or grasses) and image projection area A i .

[0084] Calculate the target physical size S based on the camera calibration parameter K. i :

[0085] Where f scale This is the scaling function, determined by the camera model.

[0086] The single-target processing time T is calculated using a piecewise function based on the target category and size. i :

[0087] ;

[0088] The parameters a1, b1, c1, a2, b2, and c2 were calibrated experimentally.

[0089] Based on the single-target processing time T i The total processing time T for all targets is obtained by summing the results. total :

[0090] ;

[0091] Calculate the total processing time T by combining all targets within the prediction window. total And derive the ideal velocity v ideal (t):

[0092] ;

[0093] Where L is the unit working stroke length, T min This represents the system's safety threshold. Ideal speed ensures coverage of all targets in the shortest possible time.

[0094] Step 2: Time Alignment Phase – Delay Compensation;

[0095] Based on the predicted physical distance between the camera and the actuator and the current actual speed, the prediction-execution time delay is calculated, and the speed value at the corresponding moment is selected from the historical ideal speed sequence for alignment to eliminate control misalignment caused by spatial distance.

[0096] This stage is executed by the time alignment module, and its purpose is to solve the spatiotemporal delay problem between prediction and execution. The workflow is as follows:

[0097] Calculate the time delay τ(t) based on the current execution speed v(t) and the predicted fixed distance D between the camera and the actuator:

[0098] ;

[0099] in To prevent the value from being zero.

[0100] Calculate the target time point t target =t-τ(t), and from the historical ideal velocity sequence {v ideal (t k Select the closest t from )} target The velocity value at time t is used as the alignment velocity v align (t).

[0101] This step ensures that the ideal speed matches the execution location in time and space, avoiding control misalignment caused by directly using the predicted speed.

[0102] Step 3: Speed ​​Smoothing Stage – Low-Pass Filtering;

[0103] The time-aligned velocity is low-pass filtered to smooth out high-frequency disturbances in the command caused by target recognition noise and minute velocity fluctuations.

[0104] This stage is performed by the speed smoothing module, and its purpose is to suppress high-frequency noise and abrupt changes. The working process is as follows:

[0105] For alignment speed v align (t) Perform first-order low-pass filtering:

[0106] ;

[0107] The filter coefficient α (0 < α < 1) controls the smoothing effect; the smaller α is, the stronger the smoothing effect.

[0108] Smooth output speed after filtering This eliminates the jitter caused by predictive fluctuations.

[0109] Step 4: Acceleration Constraint Stage – Planning Velocity Generation;

[0110] Based on the maximum permissible acceleration and control cycle of the actuator, incremental constraints are applied to the smoothed target velocity to generate a physically achievable planned velocity trajectory.

[0111] This stage is executed by the acceleration constraint module, and its purpose is to ensure that the velocity change is physically feasible. The process is as follows:

[0112] Based on the maximum acceleration a max And the control period Δt, calculate the upper limit of the speed increment. .

[0113] Comparison of smooth speeds Compared with the planned velocity v at the previous moment plan (t-1), generate the current planned speed v plan (t):

[0114] ;

[0115] This step prevents sudden speed changes and ensures that the actuator accelerates / decelerates smoothly.

[0116] Step 5: Smoothing PID Dynamic Tracking Stage – Smoothing PID Control;

[0117] A smooth PID controller generates a control quantity based on the error between the planned speed and the actual speed, and limits the acceleration of the control quantity output to drive the actuator to smoothly track the planned speed. At the same time, the actual speed is fed back to the time alignment stage to form a dynamic closed loop.

[0118] This stage is executed by the smoothed PID control module, with the aim of ensuring that the actual speed accurately tracks the planned speed. The working process is as follows:

[0119] Calculate the velocity error e(t) = v plan (t)-v(t).

[0120] The control quantity u(t) is generated by the PID controller:

[0121] ;

[0122] Parameter K p K i K d Configure settings based on system response.

[0123] Apply acceleration limiting to the control output:

[0124] ;

[0125] Update the actual execution speed: v(t+Δt)=v(t)+Δv.

[0126] This step achieves closed-loop tracking, combining integral and differential terms to eliminate steady-state errors.

[0127] Step 6: Feedback and Update Phase – Closed-Loop Cycle;

[0128] This phase is completed by the execution and feedback module, forming a dynamic closed loop:

[0129] The current actual speed v(t) is fed back to the time alignment module for delay calculation in the next cycle.

[0130] The system updates the historical predicted velocity sequence and repeats steps 1 to 5, cycling once per cycle Δt.

[0131] Optionally, the error analysis module monitors the difference between the planned speed and the ideal speed and adaptively adjusts parameters (such as α or τ(t)) to improve robustness.

[0132] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A smooth speed control method for laser weeding robots based on predictive feedback, characterized in that, Includes the following steps: S1. Prediction phase; Based on the target images acquired by the predictive camera, weed distribution is identified using a target detection algorithm, and the physical size S of each target is calculated. i : ； In the formula, S i Let A be the physical size of the i-th target; i Let K be the projected area of ​​the target in the image; K is the set of camera calibration parameters. According to target category C i and size S i The single-target processing time T is calculated using a piecewise function. i : ； In the formula, T i C is the processing time for the i-th target; i For target category identifiers, such as C=1 for broadleaf grass and C=2 for grass; a1, b1, c1, a2, b2, c2 are experimental calibration parameters; Based on the single-target processing time T i The total processing time T for all targets is obtained by summing the results. total : ； The total processing time T for all targets within the comprehensive prediction window total Calculate the ideal driving speed: ； In the formula, L is the unit working stroke length; T min This is the system's safe time threshold. S2. Time alignment phase; Calculate the prediction-execution time latency: ； In the formula, D is the physical distance between the prediction camera and the actuator; v(t) is the current execution speed; To prevent the value from being removed from zero; Based on the delay time τ(t), the predicted velocity sequence is obtained from historical data. Select the closest one Speed ​​of time As an alignment reference; S3. Speed ​​smoothing phase; Speed ​​after time alignment Perform low-pass filtering: ； In the formula, α represents the filtered velocity; α is the filtering coefficient. S4. Acceleration constraint stage; Based on the maximum permissible acceleration Given the control period Δt, calculate the upper limit of the velocity increment: ； In the formula, a max Δt is the maximum permissible acceleration of the actuator; Δt is the control period. Generate a physically executable planning speed v plan (t): ； In the formula, To achieve a smooth speed for the target; S5. Smoothing the PID dynamic tracking stage; Calculate the error between the planned speed and the actual speed. ; Control inputs are generated using a smoothed PID controller: ； In the formula, K p K i K d Here are the parameters for the PID controller; e(t) is the speed tracking error. Apply acceleration limiting to the output: ； In the formula, Δv is the actual velocity increment; Update actual execution speed: .

2. The laser weeding robot smooth speed control method based on predictive feedback according to claim 1, characterized in that, In the low-pass filtering process, the value range of the filtering coefficient α is 0 < α < 1, and α is adaptively adjusted or fixedly set according to the system noise characteristics to achieve a smooth transition of the speed signal.

3. The laser weeding robot smooth speed control method based on predictive feedback according to claim 1, characterized in that, During the acceleration constraint phase, the maximum permissible acceleration a max Dynamic settings based on the physical inertia of the actuator ensure that speed changes comply with mechanical limits.

4. A system for a smooth speed control method based on predictive feedback for a laser weeding robot according to any one of claims 1 to 3, characterized in that, Includes the following modules: The prediction module is used to identify weed distribution and calculate the ideal velocity v based on the target image acquired by the prediction camera. ideal (t); The time alignment module is used to calculate the prediction-execution time delay τ(t) and perform time alignment with the ideal speed; The speed smoothing module is used to perform low-pass filtering on the time-aligned predicted speed to suppress high-frequency disturbances caused by visual recognition fluctuations. The acceleration constraint planning module is used to generate physically realizable planned velocities v under maximum acceleration constraints. plan (t); The smooth PID dynamic tracking module is used for closed-loop control based on the error between the planned speed and the actual speed. The execution and feedback module is used to drive the actuator to output control speed and feed back the actual speed to the time alignment module and the prediction module to form closed-loop control.

5. The system of the laser weeding robot based on the predictive feedback-based smooth speed control method according to claim 4, characterized in that, The prediction module also includes a target size calculation unit, calculated using the formula: Estimate the physical size of the target.

6. The system of the laser weeding robot based on the predictive feedback-based smooth speed control method according to claim 4, characterized in that, The system also includes a dynamic update module, which refreshes the ideal velocity sequence in each control cycle and ensures that each module executes sequentially: prediction, alignment, smoothing, planning, execution, and feedback.

Images