A single-rudder AGV control parameter adaptive adjustment method and system

Abstract

The application provides a single-rudder AGV control parameter adaptive adjustment method and system, relates to the AGV control technical field, and comprises the following steps: constructing a two-dimensional gain scheduling model; obtaining real-time running parameters of the single-rudder AGV, including rear wheel midpoint pose information, heading angle information and real-time running speed parameters; determining the position of the nearest projection point corresponding to the current pose, calculating the nearest path curvature parameter; outputting a Stanley control gain coefficient; calculating an expected rudder wheel rotation angle instruction value according to a Stanley path tracking control law; and outputting to a rudder wheel actuator after rotation angle limiting processing, so as to perform path tracking control on the single-rudder AGV. The application solves the technical problem that the prior art usually adopts fixed gain for path tracking control, and the control performance is difficult to adaptively change under different running conditions, resulting in the control instability phenomenon that steering oscillation easily occurs at low speed, path convergence is slow at high speed, and there is outer deviation on a curve.

Description

Technical Field

[0001] This invention relates to the field of AGV control technology, and specifically to an adaptive adjustment method and system for control parameters of a single-rudder AGV. Background Technology

[0002] Automated Guided Vehicles (AGVs) are key material handling equipment in intelligent manufacturing and intelligent warehousing systems, and their path tracking and control performance directly determines the system's operating efficiency and operational accuracy. Among them, single-steering wheel AGVs are widely used in industrial scenarios due to their simple structure and high load-bearing capacity. They typically use the front steering wheel as the drive and steering actuator, and the midpoint of the rear wheel as the kinematic reference point to achieve tracking and control of a preset path.

[0003] In existing technologies, the Stanley path tracking algorithm is widely used in AGV control systems due to its simple structure, ease of implementation, and high path tracking accuracy. However, in practical engineering applications, the control gain in the Stanley algorithm is usually set to a fixed constant, making it difficult for the control performance to adapt to different operating conditions. This results in the system being overly sensitive to lateral error responses at low speeds, leading to steering oscillations, while at high speeds, the control response is insufficient, causing a decrease in path convergence speed or even out-of-bounds deviation on curves. Summary of the Invention

[0004] This application provides an adaptive adjustment method and system for the control parameters of a single-rudder AGV, aiming to solve the technical problem that the existing technology usually uses fixed gain for path tracking control, which makes it difficult for the control performance to adapt to different operating conditions, resulting in steering oscillations at low speeds and slow path convergence at high speeds, accompanied by outward deflection on curves.

[0005] The first aspect disclosed in this application provides an adaptive adjustment method for control parameters of a single-rudder AGV. The method includes: constructing a two-dimensional gain scheduling model on a preset path model; acquiring real-time operating parameters of the single-rudder AGV through an onboard state perception unit, wherein the real-time operating parameters include rear wheel midpoint pose information, heading angle information, and real-time operating speed parameters; determining the position of the nearest projection point on the target path corresponding to the current pose based on the preset path model, and calculating the nearest path curvature parameter at the nearest projection point; inputting the real-time operating parameters and the nearest path curvature parameter into the two-dimensional gain scheduling model, and outputting the Stanley control gain coefficient corresponding to the current operating moment; calculating the desired steering wheel angle command value according to the Stanley path tracking control law based on the Stanley control gain coefficient; and outputting the desired steering wheel angle command value to the steering wheel actuator after angle limiting processing to perform path tracking control of the single-rudder AGV.

[0006] The second aspect of this application discloses an adaptive adjustment system for control parameters of a single-rudder AGV. This system is used in the aforementioned adaptive adjustment method for control parameters of a single-rudder AGV. The system includes: a model building unit for building a two-dimensional gain scheduling model on a preset path model; an operation parameter acquisition unit for acquiring real-time operation parameters of the single-rudder AGV through an onboard state perception unit, wherein the real-time operation parameters include rear wheel midpoint pose information, heading angle information, and real-time operating speed parameters; and a curvature parameter calculation unit for determining the nearest point on the target path corresponding to the current pose based on the preset path model. The system includes a projection point location unit, a nearest path curvature parameter calculation unit, a gain coefficient output unit, and a path tracking control unit. The former inputs the real-time operating parameters and the nearest path curvature parameter into the two-dimensional gain scheduling model and outputs the Stanley control gain coefficient corresponding to the current operating moment. The latter calculates the desired steering wheel angle command value based on the Stanley control gain coefficient and the Stanley path tracking control law. The former performs angle limiting processing on the desired steering wheel angle command value and outputs it to the steering wheel actuator for path tracking control of the single-steering AGV.

[0007] One or more technical solutions provided in this application have at least the following beneficial effects: By constructing a two-dimensional gain scheduling model on a preset path model, the control system is able to uniformly map speed parameters and path curvature parameters to control gain coefficients, thus providing a unified parameter adjustment framework for subsequent control. By acquiring rear wheel midpoint pose information, heading angle information, and real-time speed parameters through an onboard state perception unit, the control system can perceive the motion state of the single-rudder AGV in real time. By determining the nearest projection point of the target path and calculating the path curvature parameters at that point, the control system can accurately obtain the local path geometric features corresponding to the vehicle's current position, thus providing crucial geometric constraint information for gain scheduling and improving the targeting and adaptability of path tracking control. By inputting real-time operating parameters and path curvature parameters into a two-dimensional path... The system employs a gain-based scheduling model and outputs Stanley control gain coefficients, enabling the control system to dynamically adjust the control gain according to the current operating conditions, thereby improving the stability and accuracy of path tracking across all operating conditions. By calculating the desired steering wheel angle using the Stanley path tracking control law based on adaptive control gain, the vehicle's lateral and heading errors can work together to influence the steering control output, thus improving path convergence speed and reducing trajectory deviation error, achieving a smoother path tracking control effect. By limiting the desired steering wheel angle and outputting it to the actuator, the control command is always kept within the mechanically permissible range, thus avoiding actuator saturation or instability caused by excessive control input, and enhancing the reliability of the entire control system.

[0008] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0009] Figure 1 This is a schematic flowchart illustrating an adaptive adjustment method for control parameters of a single-rudder AGV provided in an embodiment of this application.

[0010] Figure 2 This is a schematic diagram of an adaptive adjustment system for control parameters of a single-rudder AGV provided in an embodiment of this application.

[0011] Figure labeling: Model building unit 10, running parameter acquisition unit 20, curvature parameter calculation unit 30, gain coefficient output unit 40, instruction value calculation unit 50, path tracking control unit 60. Detailed Implementation

[0012] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0013] Example 1, as Figure 1 As shown in the figure, this application provides an adaptive adjustment method for the control parameters of a single-rudder AGV, the method comprising: A two-dimensional gain scheduling model is constructed based on the preset path model.

[0014] A two-dimensional gain scheduling model is constructed based on a pre-defined path model. This pre-defined path model is represented by a discrete path point sequence, spline curve, or multi-segment broken line to describe the desired travel trajectory of the AGV. The two-dimensional gain scheduling model takes real-time speed parameters and path curvature parameters as inputs and Stanley control gain coefficients as outputs. Essentially, it establishes a mapping relationship from the input state space to the control gain space, which is used to describe the different requirements of the AGV for control sensitivity and stability under different operating conditions.

[0015] In practical engineering implementation, this two-dimensional gain scheduling model can be constructed in one of three ways: a gain scheduling table based on velocity-curvature discrete partitioning, a rule system based on fuzzy logic reasoning, or a continuous mapping model based on polynomial function fitting, thereby balancing the complexity of engineering implementation with the requirements of control accuracy.

[0016] The real-time operating parameters of the single-rudder AGV are obtained through the on-board state perception unit. The real-time operating parameters include the rear wheel midpoint pose information, heading angle information, and real-time operating speed parameters.

[0017] During the control cycle, the operating status parameters of the single-rudder AGV are collected in real time through an onboard state perception unit, which includes an encoder, an inertial measurement unit (IMU), and an onboard controller. Specifically, the pose information of the midpoint of the AGV's rear wheels is acquired, which serves as the kinematic reference point for the single-rudder AGV; the vehicle's current heading angle is acquired to describe the vehicle's attitude relative to the global coordinate system; and the vehicle's real-time speed is acquired, calculated by a wheel speed encoder or a fusion filtering algorithm. These parameters together constitute the AGV's real-time motion state input vector, used for subsequent path error calculation and gain scheduling calculation.

[0018] Based on the preset path model, determine the position of the nearest projection point on the target path corresponding to the current pose, and calculate the nearest path curvature parameter at the nearest projection point.

[0019] Based on a pre-defined path model, a nearest point search is performed on the path near the current AGV position to determine the nearest projected point of the path corresponding to the current rear wheel midpoint position. After determining this nearest projected point, the local curvature at that point is calculated according to the path geometry model. If the path is a discrete point sequence, the curvature can be estimated using the three-point method or the fitted circular arc method; if the path is a spline curve, the curvature can be directly calculated using analytical derivatives. This curvature parameter is used to characterize the turning intensity of the current path, where a larger curvature corresponds to emergency curves, and a smaller curvature corresponds to straight sections or gentle curves, thus providing path complexity information for gain scheduling.

[0020] The real-time operating parameters and the nearest path curvature parameters are input into the two-dimensional gain scheduling model, and the Stanley control gain coefficient corresponding to the current operating moment is output.

[0021] The acquired real-time velocity and path curvature are input into the two-dimensional gain scheduling model to calculate the Stanley control gain corresponding to the current moment. Under different implementations, this calculation process is as follows: when using a gain scheduling table, the corresponding Stanley control gain coefficient is obtained by directly looking up the table based on the index of the discrete interval containing the real-time velocity and path curvature; when using a fuzzy logic model, the real-time velocity and path curvature are fuzzified, and the continuously output Stanley control gain coefficient is obtained through rule-based reasoning and defuzzification calculation; when using a polynomial fitting model, the Stanley control gain coefficient is directly calculated by substituting pre-calibrated coefficients into the function expression. This adaptive gain can improve control response capability under low-speed, high-curvature conditions and reduce system oscillation risk under high-speed, low-curvature conditions, thereby achieving full-condition stability optimization.

[0022] Based on the Stanley control gain coefficient, the desired steering wheel angle command value is calculated according to the Stanley path tracking control law.

[0023] After obtaining the Stanley control gain coefficients, the desired steering wheel angle command is calculated in conjunction with the Stanley path-following control law. Specifically, heading error and lateral error are introduced into the control model, where heading error represents the angular deviation between the vehicle's current heading and the path tangent direction, and lateral error represents the lateral offset from the vehicle's rear wheel midpoint to the path. Speed ​​normalization is performed to enhance control consistency at different speeds. Based on the Stanley control law, the desired steering angle is jointly determined by the heading error term and the lateral error feedback term. The lateral error term, through nonlinear mapping, allows the Stanley control gain coefficients to directly affect the path convergence speed and the aggressiveness of control, thereby achieving dynamic adjustment of the vehicle's trajectory correction capability.

[0024] After the desired steering wheel angle command value is processed by angle limiting, it is output to the steering wheel actuator to perform path tracking control of the single-steering AGV.

[0025] Because the actual steering wheel mechanism has physical angle limitations, after calculating the desired angle, it is limited to meet the maximum steering wheel angle constraint range, such as ±30°. The limited control command is output to the steering wheel actuator through the controller, and the servo drive system executes the steering action, thereby realizing the AGV's real-time tracking control of the target path. In this process, the above steps are executed cyclically through the control cycle to realize a closed-loop adaptive path tracking control system driven by both speed and path curvature.

[0026] Furthermore, the two-dimensional gain scheduling model includes a gain scheduling table based on velocity-curvature partitioning, a gain inference rule based on fuzzy logic, or a gain calculation formula based on polynomial fitting.

[0027] The two-dimensional gain scheduling model includes the following three implementation methods: A gain scheduling table based on velocity-curvature partitioning, which divides velocity and curvature into several discrete intervals and combines them in two dimensions to form a scheduling table. The corresponding gain k value is pre-calibrated according to different operating conditions, enabling fast table-lookup calculation, suitable for scenarios with high real-time requirements; a gain inference rule based on fuzzy logic, which fuzzifies velocity and curvature, establishes different membership sets, and outputs the gain through rule base inference, suitable for scenarios with smooth operating condition changes and requiring flexible control; and a gain calculation formula based on polynomial fitting, which establishes a mapping relationship using experimental or simulation data, and obtains a quadratic polynomial model through least squares fitting, realizing gain calculation in the form of a continuous function, suitable for control systems requiring high smoothness.

[0028] Furthermore, the construction and invocation process of the two-dimensional gain scheduling model based on the gain scheduling table of velocity-curvature partitioning includes: Based on a preset speed range and a preset path curvature range, the speed parameters and path curvature parameters are discretized and graded to form speed parameter grading results and path curvature parameter grading results. These results are then combined and matched to generate a speed-curvature two-dimensional gain scheduling table. Within this table, for each discrete state partition, based on the path deviation characteristics of a single-rudder AGV under different speed and path curvature conditions, a corresponding Stanley control gain coefficient calibration value is established. According to the real-time operating parameters and the nearest path curvature parameter, the corresponding scheduling table index unit is located within the speed-curvature two-dimensional gain scheduling table. The Stanley control gain coefficient is then output from the scheduling table index unit.

[0029] Based on the motion capability boundaries of a single-rudder AGV and its actual application scenarios, preset speed ranges and path curvature ranges are defined. Specifically, speed parameters are divided into ranges based on the AGV's minimum and maximum permissible operating speeds, such as low-speed, medium-speed, and high-speed levels. Path curvature is divided based on path geometric characteristics and the range of turning radius variations, such as small curvature, medium curvature, and large curvature levels. Through this division, continuously changing speed and path curvature parameters are converted into discretized hierarchical representations, resulting in graded speed and curvature parameters. This maps the originally continuous control input space to a finite set of states, providing the foundation for subsequent lookup-based gain scheduling.

[0030] The velocity parameter classification results are Cartesian-combined with the path curvature parameter classification results to form a two-dimensional state grid structure, namely a velocity-curvature two-dimensional gain scheduling table. In this two-dimensional scheduling table, each entry corresponds to a unique combination of velocity and curvature levels, such as "low speed - high curvature" or "high speed - low curvature," and a corresponding Stanley control gain coefficient is assigned to each state unit. This scheduling table structure essentially constructs a mapping relationship from a two-dimensional discrete state space to a control gain output space, transforming the calculation of control gain from a continuous function operation into an index lookup process, thereby significantly reducing real-time computational complexity.

[0031] For each discrete state partition in the scheduling table, the Stanley control gain coefficient is calibrated based on the path tracking error characteristics of a single-rudder AGV under different speed and path curvature combinations. Specifically, through simulation analysis or real-vehicle experiments, the lateral deviation, heading error, and oscillation characteristics of the AGV are evaluated under different speed and curvature combinations. The gain coefficient is empirically tuned or optimized based on the principle of minimizing path tracking error and optimizing system stability. Specifically, in low-speed and high-curvature conditions, a larger gain calibration value is set to improve path convergence speed and suppress curve inward deviation; in high-speed and low-curvature conditions, a smaller gain calibration value is set to avoid oscillations caused by overly aggressive system response, thereby achieving a dynamic performance balance under different conditions. An exemplary speed-curvature two-dimensional gain scheduling table is shown in Table 1.

[0032] Table 1: Velocity-Curvature Two-Dimensional Gain Scheduling Table

[0033] During system operation, the controller determines the discrete level interval to which each input state belongs based on the real-time acquired velocity parameters and the calculated path curvature parameters, thus mapping the continuous input state to the corresponding discrete state index. This discrete state index is then used to locate the corresponding discrete state in the velocity-curvature two-dimensional gain scheduling table, identifying a unique index unit that corresponds to the gain storage location for the current operating condition. This indexing process is essentially a mapping from a continuous state space to a discrete lookup table space; its calculation involves only interval judgment and index matching, thereby ensuring the real-time response capability of the control system.

[0034] After completing the scheduling table index location, the pre-calibrated Stanley control gain coefficient is read from the corresponding scheduling table index unit and output as the control parameter at the current moment.

[0035] Furthermore, the calibration rules for the scheduling table index unit include: In the scheduling unit corresponding to the low speed parameter range, a gain adjustment value that is increased relative to the reference Stanley control gain coefficient is set; in the scheduling unit corresponding to the high speed parameter range, a gain adjustment value that is decreased relative to the reference Stanley control gain coefficient is set; in the scheduling unit corresponding to the large path curvature parameter, a gain adjustment value that is increased relative to the reference Stanley control gain coefficient is set; in the scheduling unit corresponding to the small path curvature parameter, a gain adjustment value that is decreased relative to the reference Stanley control gain coefficient is set.

[0036] Within the low-speed parameter range, due to the low operating speed of the AGV and the small system inertia, the vehicle's response to steering commands is relatively sensitive, but path tracking lag and steady-state error accumulation are also prone to occur. To address this condition, a gain adjustment value that is increased relative to the baseline Stanley control gain coefficient is set in the corresponding scheduling unit. This appropriately increases the final gain coefficient, thereby enhancing the control weight of the lateral error feedback term, improving the system's ability to correct path deviations, accelerating convergence, and reducing trajectory deviation errors during the low-speed crawling phase.

[0037] Within the high-speed parameter range, AGVs exhibit significantly increased system inertia due to their higher kinetic energy. Using a large control gain can easily lead to oversteering response, resulting in path oscillations or even instability. Therefore, a gain adjustment value that is reduced relative to the baseline Stanley control gain coefficient is set in the corresponding scheduling unit. By lowering the control gain weight, the system maintains stronger stability and robustness during high-speed operation, suppressing high-frequency oscillations caused by sensor noise or minor path disturbances, and improving overall operational smoothness.

[0038] In situations with significant path curvature, such as sharp turns or small-radius curves, AGVs require stronger lateral correction capabilities to prevent deviation from the outer edge of the path or insufficient curve cutting. Therefore, in the scheduling unit corresponding to large path curvature, a gain adjustment value that is increased relative to the baseline Stanley control gain coefficient is also set to enhance the control effect of the lateral error term, improve the vehicle's response to changes in path curvature, thereby reducing trajectory deviation during turning and improving curve-passing accuracy.

[0039] Within the small path curvature parameter range, the path tends towards a straight line or gentle curve, where the vehicle's primary task is to maintain stable straight-line travel. If the control gain is too large, the vehicle may overcorrect small lateral errors, leading to serpentine swaying or high-frequency oscillations. Therefore, a gain adjustment value that is smaller than the baseline Stanley control gain coefficient is set in the corresponding scheduling unit to reduce the sensitivity of the control system, ensuring smooth vehicle operation in straight or gentle-curved conditions and improving the stability and comfort of path tracking.

[0040] Furthermore, the construction and invocation process of the two-dimensional gain scheduling model based on fuzzy logic gain inference rules includes: A fuzzy universe of discourse for velocity parameters and a fuzzy universe of discourse for path curvature parameters are constructed, and sets of velocity membership functions and path curvature membership functions are defined respectively, establishing a preset fuzzy rule base. The real-time running velocity parameters and the nearest path curvature parameters are fuzzified to obtain fuzzy subsets of velocity and path curvature. Based on the preset fuzzy rule base, rule matching inference is performed on the fuzzy subsets of velocity and path curvature to generate fuzzy output results. Aggregation operations are performed on the fuzzy output results to obtain a fuzzy set of gain output. The fuzzy set of gain output is solved using a defuzzification calculation method to obtain the Stanley control gain coefficient.

[0041] A fuzzy universe of discourse for the velocity parameter and a fuzzy universe of discourse for the path curvature parameter are constructed separately to map continuous variables to a fuzzy set space. The velocity universe of discourse is divided according to the actual operating range of the AGV and multiple linguistic variables, such as "low speed," "medium speed," and "high speed," are set to describe different operating states. The path curvature universe of discourse is divided into linguistic variables such as "small curvature," "medium curvature," and "large curvature" based on the geometric characteristics of the path to characterize the degree of path turning.

[0042] Based on this, a set of membership functions is defined for each linguistic variable to describe the degree of membership of the input variable to each fuzzy subset, and a complete fuzzy rule base is constructed. The fuzzy rule base is established based on the influence of speed and curvature on the AGV path tracking performance. For example, a higher gain rule is used under low-speed and high-curvature conditions, and a lower gain rule is used under high-speed and low-curvature conditions, thereby forming an empirical control mapping relationship between input and output.

[0043] During control operation, real-time acquired speed and path curvature parameters are input to the fuzzy processing module. This process calculates membership functions to obtain the membership degrees of the speed parameters to each speed fuzzy subset and the path curvature parameters to each curvature fuzzy subset, thus converting deterministic input variables into fuzzy set representations. For example, a speed value may simultaneously belong to both the "medium speed" and "high speed" fuzzy subsets, thereby achieving a flexible expression of continuous states and giving the control system a stronger ability to smoothly transition between states.

[0044] After fuzzification, rule matching and inference calculations are performed on the velocity fuzzy subset and the path curvature fuzzy subset based on a pre-built fuzzy rule library. During this process, each fuzzy rule calculates its activation strength based on its input membership degree and generates a corresponding control output sub-conclusion, such as "high gain," "medium gain," or "low gain." Multiple rules may be activated simultaneously, forming multiple parallel fuzzy output results based on their activation strength, thus reflecting the superimposed effects of control requirements under complex operating conditions.

[0045] After obtaining multiple fuzzy rule outputs, all rule outputs are aggregated to form a unified fuzzy set of gain outputs. This process typically employs max-min synthesis or weighted superposition to fuse the output membership functions under different rules, resulting in a comprehensive fuzzy output distribution. This fuzzy output set reflects the overall tendency distribution of control gain under the current operating state, providing a foundation for subsequent defuzzification calculations.

[0046] After fuzzy output aggregation, the fuzzy output set is converted into deterministic Stanley control gain coefficients k using a defuzzification calculation method. In a preferred embodiment, the centroid method is used for defuzzification, i.e., the centroid position of the output membership function distribution is calculated to obtain a unique control gain value. This process realizes the mapping from the fuzzy set space to the deterministic control quantity space, so that the output gain can reflect the comprehensive influence of multiple rules and has continuity and smoothness. Finally, the Stanley control gain coefficients are obtained.

[0047] Furthermore, the construction and invocation process of the two-dimensional gain scheduling model based on the gain calculation formula of polynomial fitting includes: Historical operating data of a single-rudder AGV under different combinations of speed and path curvature parameters are collected to construct a training sample set. The training sample set is normalized, and the mapping relationship between sample speed parameters, sample path curvature parameters, and sample control gain coefficients is extracted. The pre-constructed quadratic polynomial gain function is solved using the least squares parameter estimation method to obtain the gain analytical expression. The real-time operating parameters and the nearest path curvature parameters are input into the gain analytical expression, and the Stanley control gain coefficient is output.

[0048] By collecting historical operating data of single-rudder AGVs under different operating conditions, a sample dataset for model training is constructed. Specifically, under different combinations of speed and path curvature parameters, the actual operating state of the AGV is recorded, and the Stanley control gain coefficient k after control optimization under the corresponding operating condition is simultaneously acquired as the label value, thus forming a triplet sample data set. The operating conditions cover typical path scenarios such as low-speed straight lines, high-speed straight lines, low-speed sharp curves, and high-speed gentle curves to ensure that the sample data can comprehensively reflect the gain variation patterns under different control requirements.

[0049] After constructing the training sample set, the sample data is normalized to eliminate the influence of differences in dimensions and numerical ranges between the velocity and path curvature parameters, thereby improving the stability and fitting accuracy of parameter identification. Based on this, a functional mapping relationship is established between the velocity, path curvature, and control gain coefficients, and a quadratic polynomial is selected as the basic structural model of the gain function to describe the nonlinear coupling relationship among the three. The least squares parameter estimation method is used to solve for the coefficients of the quadratic polynomial model, minimizing the sum of squared errors between the model output and historical sample data, thus obtaining the optimal fitting parameters. Through this process, a Stanley control gain function with an analytical expression is obtained, realizing the transformation from discrete empirical data to a continuous analytical model.

[0050] The real-time acquired velocity and path curvature parameters are input into the established analytical expression for control gain, and the Stanley control gain coefficient corresponding to the current moment is directly obtained through function calculation. This process does not require table lookup or rule-based reasoning; instead, it achieves control gain output through real-time calculation of the analytical function, enabling the control system to possess continuity, smoothness, and good numerical differentiability.

[0051] Furthermore, the analytical expression for the gain is as follows: ; in, This is the Stanley control gain coefficient. These are real-time operating parameters, where k is the nearest path curvature parameter. , , , , , The coefficients, determined by the least squares method, are the gain reference constant, velocity linear term coefficient, curvature linear term coefficient, velocity-curvature coupling term coefficient, velocity quadratic term coefficient, and curvature quadratic term coefficient.

[0052] The analytical expression for the gain is used to describe the nonlinear mapping relationship between the Stanley control gain coefficient and the vehicle's real-time speed and path curvature. Here, k(v,κ) represents the adaptive control gain within the current control cycle, serving as a key adjustment parameter in the path tracking control law.

[0053] This expression consists of a constant term, a linear term, a coupling term, and a quadratic term: where The basic gain bias term is used to determine the basic control strength of the system; Used to describe the linear effect of speed changes on control gain, reflecting the change in the system's dynamic response requirements as speed increases; Used to describe the effect of path curvature on control gain, reflecting the need for correction capability under curve conditions; This is a coupling term of velocity and curvature, used to characterize the combined effects under complex conditions such as high-speed turning; and These are used to characterize the nonlinear enhancement effects of velocity and curvature, respectively, giving the model stronger fitting ability and generalization performance.

[0054] By identifying the coefficients using the least squares method, the expression can continuously estimate the control gain under different operating conditions based on historical operating data, thereby avoiding the problem of insufficient adaptability of the traditional fixed gain method under complex path conditions.

[0055] Furthermore, the mathematical expression of the Stanley path tracking control law is as follows: ; in, The desired steering wheel angle command value. This is the heading error parameter between the vehicle's current heading and the tangent direction of the target path. This is the Stanley control gain coefficient. The lateral error parameter is the vehicle's lateral offset after speed normalization.

[0056] This expression is used to calculate the desired steering wheel angle command for a single-steering AGV, where the desired steering wheel angle command represents the steering angle that the vehicle should perform in the current control cycle, and is the direct output quantity for achieving path tracking control.

[0057] The expression consists of two parts: a heading error term and a lateral error feedback term. It represents the angular error between the vehicle's current heading angle and the tangent direction of the target path, and is used to correct the overall deviation of the vehicle's driving direction. This represents a nonlinear correction term based on lateral bias, where It represents the lateral offset of the vehicle's rear wheel midpoint relative to the target path, and is normalized by speed to eliminate the scaling effect caused by speed changes.

[0058] In this structure, adaptive gain The lateral error feedback strength can be directly adjusted to enable the system to automatically adjust its correction capability under different speed and path curvature conditions. When k is larger, the system responds more strongly to path deviations, making it suitable for low-speed or high-curvature conditions; when k is smaller, the system response is smoother, which helps to suppress oscillations and overshooting phenomena during high-speed operation.

[0059] This control law enables the synergistic effect of heading correction and lateral deviation correction, thereby ensuring that the single-rudder AGV has a stable and continuous path tracking capability in complex path environments.

[0060] Example 2, based on the same inventive concept as the adaptive adjustment method for single-rudder AGV control parameters in the aforementioned examples, such as... Figure 2 As shown in the figure, this application provides an adaptive adjustment system for the control parameters of a single-rudder AGV, the system comprising: The model building unit 10 is used to build a two-dimensional gain scheduling model on a preset path model; the running parameter acquisition unit 20 is used to acquire the real-time running parameters of the single-rudder AGV through the vehicle state perception unit, wherein the real-time running parameters include the rear wheel midpoint pose information, heading angle information, and real-time running speed parameters; the curvature parameter calculation unit 30 is used to determine the position of the nearest projection point on the target path corresponding to the current pose based on the preset path model, and calculate the nearest path curvature parameter at the nearest projection point; the gain coefficient output unit 40 is used to input the real-time running parameters and the nearest path curvature parameter into the two-dimensional gain scheduling model, and output the Stanley control gain coefficient corresponding to the current running time; the command value calculation unit 50 is used to calculate the desired steering wheel angle command value according to the Stanley path tracking control law based on the Stanley control gain coefficient; and the path tracking control unit 60 is used to perform angle limiting processing on the desired steering wheel angle command value and output it to the steering wheel actuator for path tracking control of the single-rudder AGV.

[0061] Furthermore, the two-dimensional gain scheduling model includes a gain scheduling table based on velocity-curvature partitioning, a gain inference rule based on fuzzy logic, or a gain calculation formula based on polynomial fitting.

[0062] Furthermore, the gain coefficient output unit 40 is used to perform the following operation steps: Based on a preset speed range and a preset path curvature range, the speed parameters and path curvature parameters are discretized and graded to form speed parameter grading results and path curvature parameter grading results. These results are then combined and matched to generate a speed-curvature two-dimensional gain scheduling table. Within this table, for each discrete state partition, based on the path deviation characteristics of a single-rudder AGV under different speed and path curvature conditions, a corresponding Stanley control gain coefficient calibration value is established. According to the real-time operating parameters and the nearest path curvature parameter, the corresponding scheduling table index unit is located within the speed-curvature two-dimensional gain scheduling table. The Stanley control gain coefficient is then output from the scheduling table index unit.

[0063] Furthermore, the gain coefficient output unit 40 is used to perform the following operation steps: In the scheduling unit corresponding to the low speed parameter range, a gain adjustment value that is increased relative to the reference Stanley control gain coefficient is set; in the scheduling unit corresponding to the high speed parameter range, a gain adjustment value that is decreased relative to the reference Stanley control gain coefficient is set; in the scheduling unit corresponding to the large path curvature parameter, a gain adjustment value that is increased relative to the reference Stanley control gain coefficient is set; in the scheduling unit corresponding to the small path curvature parameter, a gain adjustment value that is decreased relative to the reference Stanley control gain coefficient is set.

[0064] Furthermore, the gain coefficient output unit 40 is used to perform the following operation steps: A fuzzy universe of discourse for velocity parameters and a fuzzy universe of discourse for path curvature parameters are constructed, and sets of velocity membership functions and path curvature membership functions are defined respectively, establishing a preset fuzzy rule base. The real-time running velocity parameters and the nearest path curvature parameters are fuzzified to obtain fuzzy subsets of velocity and path curvature. Based on the preset fuzzy rule base, rule matching inference is performed on the fuzzy subsets of velocity and path curvature to generate fuzzy output results. Aggregation operations are performed on the fuzzy output results to obtain a fuzzy set of gain output. The fuzzy set of gain output is solved using a defuzzification calculation method to obtain the Stanley control gain coefficient.

[0065] Furthermore, the gain coefficient output unit 40 is used to perform the following operation steps: Historical operating data of a single-rudder AGV under different combinations of speed and path curvature parameters are collected to construct a training sample set. The training sample set is normalized, and the mapping relationship between sample speed parameters, sample path curvature parameters, and sample control gain coefficients is extracted. The pre-constructed quadratic polynomial gain function is solved using the least squares parameter estimation method to obtain the gain analytical expression. The real-time operating parameters and the nearest path curvature parameters are input into the gain analytical expression, and the Stanley control gain coefficient is output.

[0066] Furthermore, the analytical expression for the gain is as follows: ; in, This is the Stanley control gain coefficient. These are real-time operating parameters, where k is the nearest path curvature parameter. , , , , , The coefficients, determined by the least squares method, are the gain reference constant, velocity linear term coefficient, curvature linear term coefficient, velocity-curvature coupling term coefficient, velocity quadratic term coefficient, and curvature quadratic term coefficient.

[0067] Furthermore, the mathematical expression of the Stanley path tracking control law is as follows: ; in, The desired steering wheel angle command value. This is the heading error parameter between the vehicle's current heading and the tangent direction of the target path. This is the Stanley control gain coefficient. The lateral error parameter is the vehicle's lateral offset after speed normalization.

[0068] Through the foregoing detailed description of an adaptive adjustment method for control parameters of a single-rudder AGV, those skilled in the art can clearly understand the adaptive adjustment system for control parameters of a single-rudder AGV in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be found in the method section.

[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An adaptive adjustment method for control parameters of a single-rudder AGV, characterized in that, The method includes: A two-dimensional gain scheduling model is constructed based on the preset path model; The real-time operating parameters of the single-rudder AGV are obtained through the vehicle-mounted state perception unit, wherein the real-time operating parameters include the rear wheel midpoint pose information, heading angle information and real-time operating speed parameters. Based on the preset path model, determine the position of the nearest projection point on the target path corresponding to the current pose, and calculate the nearest path curvature parameter at the nearest projection point; The real-time operating parameters and the nearest path curvature parameters are input into the two-dimensional gain scheduling model, and the Stanley control gain coefficient corresponding to the current operating moment is output. Based on the Stanley control gain coefficient, the desired steering wheel angle command value is calculated according to the Stanley path tracking control law; After the desired steering wheel angle command value is processed by angle limiting, it is output to the steering wheel actuator to perform path tracking control of the single-steering AGV.

2. The adaptive adjustment method for control parameters of a single-rudder AGV as described in claim 1, characterized in that, The two-dimensional gain scheduling model includes a gain scheduling table based on velocity-curvature partitioning, a gain inference rule based on fuzzy logic, or a gain calculation formula based on polynomial fitting.

3. The adaptive adjustment method for control parameters of a single-rudder AGV as described in claim 2, characterized in that, The construction and invocation process of the two-dimensional gain scheduling model based on the velocity-curvature partitioning gain scheduling table includes: Based on the preset speed range and preset path curvature range, the speed parameters and path curvature parameters are discretized and graded respectively to form the speed parameter grading results and the path curvature parameter grading results. The velocity parameter classification results are combined and matched with the path curvature parameter classification results to generate a velocity-curvature two-dimensional gain scheduling table. In the speed-curvature two-dimensional gain scheduling table, for each discrete state partition, based on the path deviation characteristics of the single-rudder AGV under different speed and different path curvature conditions, the corresponding Stanley control gain coefficient calibration value is established. Based on the real-time operating parameters and the nearest path curvature parameters, locate the corresponding scheduling table index unit in the velocity-curvature two-dimensional gain scheduling table; The Stanley control gain coefficient is output from the scheduling table index unit.

4. The adaptive adjustment method for control parameters of a single-rudder AGV as described in claim 3, characterized in that, The calibration rules for the scheduling table index unit include: In the scheduling unit corresponding to the low speed parameter range, a gain adjustment value that is increased relative to the baseline Stanley control gain coefficient is set. In the scheduling unit corresponding to the high-speed parameter range, a gain adjustment value that is reduced relative to the baseline Stanley control gain coefficient is set; In the scheduling unit corresponding to the large path curvature parameter, set a gain adjustment value that is increased relative to the baseline Stanley control gain coefficient; In the scheduling unit corresponding to the small path curvature parameter, a gain adjustment value that is reduced relative to the baseline Stanley control gain coefficient is set.

5. The adaptive adjustment method for control parameters of a single-rudder AGV as described in claim 2, characterized in that, The construction and invocation process of the two-dimensional gain scheduling model based on fuzzy logic gain inference rules includes: Construct fuzzy universes of discourse for velocity parameters and fuzzy universes of discourse for path curvature parameters, and define sets of velocity membership functions and sets of path curvature membership functions respectively, and establish a preset fuzzy rule base; The real-time running speed parameters and the nearest path curvature parameters are fuzzified to obtain a speed fuzzy subset and a path curvature fuzzy subset. Based on the preset fuzzy rule library, rule matching reasoning is performed on the speed fuzzy subset and the path curvature fuzzy subset to generate fuzzy output results. The fuzzy output results are aggregated to obtain a gain output fuzzy set; The Stanley control gain coefficient is obtained by solving the fuzzy set of the gain output using a defuzzification calculation method.

6. The adaptive adjustment method for control parameters of a single-rudder AGV as described in claim 2, characterized in that, The construction and invocation process of the two-dimensional gain scheduling model based on the gain calculation formula of polynomial fitting includes: Historical operating data of single-rudder AGVs under different combinations of speed parameters and path curvature parameters were collected to construct a training sample set. The training sample set is normalized, and the mapping relationship between sample velocity parameters, sample path curvature parameters and sample control gain coefficients is extracted. The least squares parameter estimation method is used to solve the pre-constructed quadratic polynomial gain function to obtain the analytical expression of gain. The real-time operating parameters and the nearest path curvature parameters are input into the gain analytical expression, and the Stanley control gain coefficient is output.

7. The adaptive adjustment method for control parameters of a single-rudder AGV as described in claim 6, characterized in that, The analytical expression for the gain is as follows: ; in, This is the Stanley control gain coefficient. These are real-time operating parameters, where k is the nearest path curvature parameter. , , , , , The coefficients, determined by the least squares method, are the gain reference constant, velocity linear term coefficient, curvature linear term coefficient, velocity-curvature coupling term coefficient, velocity quadratic term coefficient, and curvature quadratic term coefficient.

8. The adaptive adjustment method for control parameters of a single-rudder AGV as described in claim 1, characterized in that, The mathematical expression for the Stanley path tracking control law is as follows: ; in, The desired steering wheel angle command value, This is the heading error parameter between the vehicle's current heading and the tangent direction of the target path. This is the Stanley control gain coefficient. This is the lateral error parameter after the vehicle's lateral offset has been normalized to speed.

9. An adaptive adjustment system for control parameters of a single-rudder AGV, characterized in that, An adaptive adjustment method for control parameters of a single-rudder AGV according to any one of claims 1-8, the system comprising: The model building unit is used to build a two-dimensional gain scheduling model on a preset path model. The operation parameter acquisition unit is used to acquire the real-time operation parameters of the single-rudder AGV through the vehicle status perception unit. The real-time operation parameters include the rear wheel midpoint pose information, heading angle information, and real-time operation speed parameters. The curvature parameter calculation unit is used to determine the position of the nearest projection point on the target path corresponding to the current pose based on the preset path model, and to calculate the curvature parameter of the nearest path at the nearest projection point. The gain coefficient output unit is used to input the real-time running parameters and the nearest path curvature parameters into the two-dimensional gain scheduling model and output the Stanley control gain coefficient corresponding to the current running time. The command value calculation unit is used to calculate the desired steering wheel angle command value based on the Stanley control gain coefficient and in accordance with the Stanley path tracking control law. The path tracking control unit is used to perform angle limiting processing on the desired steering wheel angle command value and then output it to the steering wheel actuator to perform path tracking control of the single-steering AGV.

Images