An Adaptive Steering Control Method and System for Unmanned Logistics Vehicles in Complex Environments

By constructing a comprehensive state vector and operating condition vector, updating control parameters in real time, and using a dual-layer disturbance observer for coordinated compensation of external disturbances and nonlinear errors of the actuator, the problem of steering control accuracy of unmanned logistics vehicles in complex environments is solved, achieving high-precision and stable steering performance.

CN123104770APending Publication Date: 2026-11-17HENAN SUNSHINE LONGRUI NEW ENERGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611518617.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-29
Publication Date
2026-11-17

Smart Images

  • Figure CN123104770A_ABST
    Figure CN123104770A_ABST
Patent Text Reader

Abstract

This invention provides an adaptive steering control method and system for unmanned logistics vehicles in complex environments. The method acquires vehicle operating status information and constructs a comprehensive state vector. Based on the operating status information, it acquires vehicle speed, load mass, path curvature, road surface adhesion coefficient, and the evaluated state information of the actuators. A comprehensive operating condition vector is then constructed based on the evaluated state information. Using the comprehensive operating condition vector as input, an updated control parameter vector is output through an online control parameter update model. Feedback control quantities are obtained based on the control parameter vector, lateral tracking error, and heading error. External disturbances and internal nonlinear errors of the steering actuators are estimated in real time. A comprehensive disturbance estimate is obtained by fusing these with the vehicle's longitudinal speed. A compensation control quantity is then obtained based on the comprehensive disturbance estimate. Finally, the theoretical front wheel steering angle, feedback control quantity, and compensation control quantity are fused to form the final target front wheel steering angle command, which is then output to the steering actuator.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of unmanned logistics vehicles, and in particular to an adaptive steering control method and system for unmanned logistics vehicles in complex environments. Background Technology

[0002] With the rapid development of e-commerce and logistics, unmanned logistics vehicles, as an effective means of "last-mile" delivery, have received widespread attention from academia and industry. These unmanned vehicles mainly operate in closed or semi-closed environments such as parks, warehouses, and factories, with relatively low speeds. For steering control, steer-by-wire systems eliminate the traditional mechanical steering connection, achieving steering through an electronic control system. This provides unmanned vehicles with a better chassis layout and higher steering control precision, and has become the mainstream technology for steering systems in autonomous vehicles.

[0003] Most existing unmanned vehicle steering control systems are designed based on fixed-parameter vehicle dynamics models. They are based on two-degree-of-freedom dynamic models and use algorithms such as PID, LQR, and MPC to solve for the target front wheel steering angle. Some schemes introduce fuzzy adaptive algorithms to adjust PID parameters and rely on multi-layer PID output of "trajectory-angle-current" to output steering control commands. In order to cope with the control effect of vehicle speed, existing technologies use multiple agents to complete the controller switching in different speed ranges, or use adaptive model predictive control (AMPC) to introduce longitudinal speed to build a trajectory tracking controller to achieve precise steering. In the face of parameter uncertainty and external disturbances, sliding mode control is widely used, and path tracking schemes such as backstepping sliding mode and adaptive sliding mode have been derived to achieve disturbance estimation and compensation.

[0004] However, existing steering control technologies still have many shortcomings when dealing with the special operational needs of unmanned logistics vehicles:

[0005] 1. During the transportation process, the quality of goods in unmanned logistics vehicles changes continuously with the loading and unloading tasks, resulting in significant changes in the vehicle's total mass, center of gravity position, and yaw moment of inertia. A dynamic model with fixed parameters is difficult to accurately describe the actual motion characteristics of the vehicle.

[0006] 2. Existing adaptive steering control systems mostly adjust control parameters based on vehicle speed. However, real-world operating conditions are complex and varied, such as straight road sections, continuous curves, and narrow passages. Different path curvatures correspond to different steering requirements. Additionally, there are various road surface conditions, such as dry asphalt roads and wet and slippery roads, with different adhesion coefficients. Therefore, the steering performance of unmanned logistics vehicles is simultaneously constrained by multiple conditions, including load, path curvature, road adhesion coefficient, and actuator status. Adjusting parameters based on a single factor can easily lead to problems such as increased trajectory tracking error, control overshoot, and vehicle body oscillation under complex operating conditions.

[0007] 3. Existing disturbance compensation methods mostly target specific disturbances and lack unified modeling and coordinated compensation for external disturbances and internal nonlinearities of the actuators. Over long-term operation, unmanned logistics vehicles will experience wear and deterioration of the actuators, and nonlinear factors such as backlash, friction and hysteresis of the steering actuators will gradually intensify. Compensation based solely on the vehicle body model cannot maintain control accuracy.

[0008] 4. Existing solutions are mostly limited to local control optimization, lacking an overall framework that integrates state perception, operating condition identification, parameter adaptation, disturbance compensation, and fusion control. Data interaction and collaboration mechanisms are also lacking, making it impossible to complete full-link closed-loop optimization. Summary of the Invention

[0009] To address the problems existing in the background technology, this invention proposes an adaptive steering control method and system for unmanned logistics vehicles in complex environments.

[0010] An adaptive steering control method for unmanned logistics vehicles in complex environments includes the following steps:

[0011] S100. Obtain the vehicle's operating status information and construct a comprehensive state vector;

[0012] S200. Obtain vehicle speed, load mass, path curvature, road surface adhesion coefficient, and actuator evaluation status information based on the operating status information; construct a comprehensive working condition vector and its mapping relationship based on the evaluation status information.

[0013] S300. Using the comprehensive operating condition vector as input, the updated control parameter vector is output through the online update model of the control parameters, and the feedback control quantity is obtained based on the control parameter vector, the lateral tracking error and the heading error.

[0014] S400: Real-time estimation of external disturbances and internal nonlinear errors of the steering actuator, and fusion of the vehicle's longitudinal speed to obtain a comprehensive disturbance estimate, and obtaining a compensation control quantity based on the comprehensive disturbance estimate;

[0015] S500: After calculating the theoretical front wheel steering angle information from the path curvature, the theoretical front wheel steering angle, the feedback control quantity, and the compensation control quantity are fused into the final target front wheel steering angle command and output to the steering actuator.

[0016] Based on the above, in step S100, the operating status information includes at least the vehicle's position coordinates and longitudinal speed. lateral velocity yaw rate r, lateral acceleration Heading angle Front wheel steering angle Drive motor current i, motor output torque T, current cargo weight and reference trajectory curvature ;

[0017] The overall state vector S is:

[0018]

[0019]

[0020] Where m is the total mass of the vehicle. For the moment of inertia of yaw rotation, For empty vehicle weight, For the current cargo load, The moment of inertia is the yaw inertia under no-load conditions. The increment of inertia caused by the load; It is the centroid sideslip angle.

[0021] Based on the above, in step S200, the comprehensive working condition vector M is:

[0022]

[0023]

[0024]

[0025]

[0026]

[0027]

[0028] in, For speed levels, For load rating, For curvature level, For adhesion level, Health level; This is an estimated value for vehicle weight. The mass threshold that separates unloaded and half-loaded conditions. This is the mass threshold separating half-load and full-load conditions. This is the curvature threshold separating straight lines from ordinary curves. This represents the curvature threshold that separates ordinary curves from sharp turns. Here, H represents the estimated road surface adhesion coefficient, and H is the health index.

[0029] Based on the above, in step S300, the updated control parameter vector for:

[0030]

[0031] Feedback control quantity for:

[0032]

[0033] in, For proportional gain, For integral gain, For differential gain, For heading feedback gain; For lateral tracking error, This represents the heading error.

[0034] Based on the above, the parameter update algorithm in step S300 is as follows:

[0035]

[0036]

[0037]

[0038]

[0039]

[0040] in, These are the adaptive learning rates, both of which are positive definite constants. This is the conversion coefficient from heading error to lateral displacement.

[0041] Based on the above, in step S400, the comprehensive disturbance estimate D is:

[0042]

[0043] in, This is the estimated value of the comprehensive external disturbances to the vehicle. This is an estimate of the nonlinear disturbance within the actuator; These are the weighting coefficients.

[0044] Based on the above, the compensation control quantity : .

[0045] Based on the above, in step S500, the theoretical front wheel steering angle :

[0046]

[0047] Where L is the vehicle wheelbase. The reference path curvature.

[0048] Based on the above, in step S500, the target front wheel steering angle command for:

[0049]

[0050] in, The theoretical front wheel steering angle, For feedback control, The compensation control quantity is denoted as .

[0051] An adaptive steering control system for unmanned logistics vehicles in complex environments includes a processing unit, a storage device, and a computer program stored on the storage device and executable. When the processing unit executes the computer program, it implements any of the aforementioned adaptive steering control methods for unmanned logistics vehicles in complex environments.

[0052] This invention has outstanding substantive features and significant progress compared to the prior art, specifically:

[0053] (1) Working condition identification mechanism for logistics vehicles: In view of the characteristics of frequent load changes, low speed operation and frequent turning of logistics vehicles, the vehicle mass and yaw moment of inertia are introduced into the two-degree-of-freedom dynamic model as time-varying parameters, and combined with the path curvature to form a comprehensive state vector, which is used to reflect the motion characteristics of unmanned logistics vehicles under load changes and different road conditions.

[0054] (2) Multi-factor joint adaptive mechanism: Establish a comprehensive working condition vector including five types of factors such as speed, load, curvature, adhesion coefficient and actuator health status, and realize the online continuous adaptive adjustment of control parameters according to the operating conditions through normalized influence factors and weighted comprehensive evaluation;

[0055] (3) Dual-layer compensation architecture: The vehicle layer extended state observer (ESO) and actuator layer disturbance observer (DOB) are used to form a dual-layer disturbance observation architecture. The external disturbance of the vehicle and the internal nonlinearity of the actuator are observed separately. The two types of disturbances are compensated in a coordinated manner through a speed-related dynamic fusion strategy, forming a dual-layer robust control structure of "actuator compensation + vehicle disturbance compensation".

[0056] (4) Hierarchical adaptive control framework: Construct a three-in-one integrated steering system of "path curvature feedforward + adaptive feedback + disturbance compensation" to ultimately form a unified control process of "state acquisition - working condition identification - parameter adaptation - disturbance compensation - steering control", integrating working condition identification, parameter adaptation and disturbance estimation into a complete steering control closed loop. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the process of this invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] like Figure 1 As shown, an adaptive steering control method for unmanned logistics vehicles in complex environments includes the following steps: S100, acquiring vehicle operating status information and constructing a comprehensive state vector; S200, acquiring vehicle speed, load mass, path curvature, road surface adhesion coefficient, and evaluation status information of the actuator based on the operating status information, and constructing a comprehensive operating condition vector and its mapping relationship based on the evaluation status information; S300, using the comprehensive operating condition vector as input, outputting an updated control parameter vector through an online control parameter update model, and obtaining feedback control quantity based on the control parameter vector, lateral tracking error, and heading error; S400, estimating external disturbances and internal nonlinear errors of the steering actuator in real time, and obtaining a comprehensive disturbance estimate by fusing the vehicle's longitudinal speed, and obtaining a compensation control quantity based on the comprehensive disturbance estimate; S500, calculating the theoretical front wheel steering angle information from the path curvature, fusing the theoretical front wheel steering angle, the feedback control quantity, and the compensation control quantity into a final target front wheel steering angle command and outputting it to the steering actuator.

[0060] Specifically:

[0061] 1. Vehicle parameter acquisition and time-varying dynamics modeling

[0062] To obtain vehicle operating status information, multi-source sensors are used to collect vehicle status information in real time, establishing a system that includes real-time load mass and path curvature. A time-varying two-degree-of-freedom dynamic model is used to construct a comprehensive state vector.

[0063] The autonomous vehicle uses various sensor modules to collect its position coordinates and longitudinal velocity in real time. (In this embodiment, longitudinal refers to the vehicle's direction of travel), lateral speed (In this embodiment, "lateral" refers to the direction perpendicular to the vehicle's direction of travel within the vehicle's driving plane), yaw rate r, lateral acceleration Heading angle Front wheel steering angle Drive motor current i, motor output torque T, current cargo weight The acquired raw signal is then passed through a first-order low-pass filter. The preprocessed data is then input into the vehicle controller. The original input signal, This is the filtered output signal. These are the filter coefficients, which are calibrated based on sensor noise in practice.

[0064] Under conditions with a small steering angle, based on the linear tire model, the vehicle's lateral dynamics and yaw motion satisfy the following:

[0065]

[0066] Where m is the vehicle mass. Let the vehicle's yaw moment of inertia be... , These are the lateral forces of the front and rear wheels, respectively. , These are the distances from the center of mass to the front and rear axles, respectively.

[0067] According to the linear tire theory, the lateral forces of the front and rear wheels are expressed as follows:

[0068]

[0069] in , For the lateral stiffness of the front and rear wheels, It is the centroid sideslip angle.

[0070] A state-space model is established based on the above equations:

[0071]

[0072] Where the state vector , This is the first derivative with respect to time X (in this embodiment, the dot above the parameter represents the first derivative of that parameter, and so on), controlling the input. Matrices A and B are determined in real time based on the current vehicle parameters. The aforementioned two-degree-of-freedom dynamic model characterizes the basic physical constraints of the vehicle's lateral motion and provides a basis for subsequent calculations of the vehicle's total mass m and yaw moment of inertia. The time-varying parameters introduced into the control system provide a theoretical basis; the core control quantity of this invention is directly calculated based on the error feedback and disturbance compensation in subsequent steps, without the need for online solution of the state-space model.

[0073] Considering the real-time changes in load during unmanned logistics vehicle transportation, the total vehicle mass m and yaw moment of inertia are... As a real-time variable:

[0074]

[0075] in For empty vehicle weight, For the current cargo load, The moment of inertia is the yaw inertia under no-load conditions. This represents the increase in inertia caused by the load.

[0076] For subsequent working condition identification, a comprehensive state vector is defined. . The reference trajectory curvature is provided in real time by the vehicle's own path planning module.

[0077] 2. Operating condition identification and status assessment

[0078] To improve the steering control performance of unmanned logistics vehicles under different operating environments, a multi-condition recognition module was established. This module identifies the vehicle's operating conditions by comprehensively considering five factors: vehicle speed, load mass, path curvature, road surface adhesion coefficient, and actuator status. A comprehensive condition vector is then established to enable online identification of the vehicle's operating conditions.

[0079] First, based on the commonly used operating speeds of logistics vehicles, speed levels are determined. Divided into:

[0080]

[0081] Secondly, the vehicle mass is estimated based on the output current of the drive motor. The longitudinal dynamic equation is: The combined estimated vehicle mass value is ,in The efficiency coefficient of the transmission system. The torque constant of the motor. Where R is the driving current and R is the wheel rolling radius. Let the acceleration be the vehicle's longitudinal acceleration. When the acceleration... When the value approaches zero, retain the valid estimate from the previous time step.

[0082] Load level Divided into three levels:

[0083]

[0084] in, The mass threshold that separates unloaded and half-loaded conditions. The mass thresholds for half-load and full-load are calibrated by the vehicle's factory parameters.

[0085] And referencing the curvature provided by the path planning module. (path arc length), curvature level Divided into:

[0086]

[0087] in, This is the curvature threshold separating straight lines from ordinary curves. The curvature thresholds for ordinary curves and sharp turns are preset based on road design specifications and the minimum turning radius of vehicles.

[0088] Then, the road adhesion coefficient is estimated using the vehicle's lateral acceleration. , For gravitational acceleration, the adhesion level will be determined. Divided into:

[0089]

[0090] The actuator's condition is assessed, and the steering angle tracking error is defined. ,in Turn the corner towards your target. For actual turning points. Establish health indicators. , This is the normalized mean square value of the steering error. This refers to the maximum operating angle of the steering actuator. [Health level] Divided into:

[0091]

[0092] Finally, a hierarchical simplification strategy is adopted to quantify the operating condition information. The comprehensive operating condition vector is: Each element represents a discrete level (1, 2, 3). Each discrete level is converted into a corresponding continuous physical quantity value according to a preset monotonic mapping relationship, which is pre-stored in the controller's parameter table and determined during the vehicle calibration phase. After conversion, the aforementioned continuous physical quantity values ​​are input into the parameter update module of subsequent steps.

[0093] 3. Multi-factor adaptive updating of control parameters

[0094] After identifying the vehicle's operating conditions, the control system needs to adjust the steering control parameters in real time based on the vehicle's current operating status to ensure good lateral stability under different operating conditions. Therefore, the continuous physical quantity values ​​obtained after the transformation in the above steps are taken as the operating condition inputs to construct an online update model for the control parameters, enabling continuous and adaptive adjustment of the control parameters.

[0095] First, to unify the dimensions of different factors, a normalized influencing factor is established:

[0096]

[0097] All factors satisfy , forming a normalized vector .

[0098] Secondly, a weight matrix is ​​introduced. Conduct a comprehensive working condition evaluation To avoid setting the weights solely based on experience, vehicle speed, vehicle load, road curvature, road surface adhesion coefficient, and vehicle health status were normalized to a range of 0-1. Secondly, using 0.2 as the initial weighting benchmark for each factor, and ensuring the sum of the five weights always equals 1, individual weights were perturbed by a factor of 0.5-1.5, with corresponding adjustments to the weights of other factors. Finally, under the same vehicle parameters, road curvature, road surface adhesion conditions, and target path conditions, simulation tests were conducted on different weight combinations. Lateral tracking error, heading error, steering angle tracking error, and disturbance recovery time were used as evaluation indicators. The weight combination with the optimal comprehensive control performance and that satisfies vehicle stability constraints was selected as the comprehensive operating condition evaluation. (in fact) middle It is a normalized influence factor vector, and the input must be a continuous physical quantity. Therefore, the level mapping in the aforementioned steps is to map the corresponding continuous physical quantity values, which is to give... (Input provided)

[0099]

[0100] After the above calibration of the logistics vehicle, the final weights were determined as follows: speed 0.25, load 0.3, curvature 0.2, adhesion 0.15, and health 0.1. These weights were determined based on the degree of influence of each factor on the vehicle's lateral dynamic response, path tracking performance, and steering performance. Among these, vehicle load has the most direct impact on the vehicle's center of gravity position, yaw inertia, and tire side-slip characteristics, therefore it has the largest weight. Vehicle speed directly affects the vehicle's lateral inertia and steering response speed, hence its weight is 0.25. Road curvature determines the vehicle's required lateral acceleration and target steering input, hence its weight is 0.2. The road adhesion coefficient affects the tire lateral force and the vehicle's stability boundary, hence its weight is 0.15. Vehicle health reflects the degree of performance degradation of the steering actuator, therefore its impact is minimal.

[0101] Furthermore, to avoid the problem of parameter abrupt changes caused by traditional segmented switching, a continuous mapping mechanism is adopted:

[0102]

[0103] in, , , , These are the base values ​​for proportional, integral, derivative, and heading feedback gains (calibrated values ​​under no-load, straight-line, high-adhesion, and good health conditions). , , , These correspond to the adjustment coefficients for each parameter. Each gain must meet preset upper and lower bound constraints to ensure system stability.

[0104] To further improve adaptability, an online learning mechanism based on tracking error is introduced. Lateral tracking error and heading error are defined as follows: , This refers to the actual horizontal position. For the desired horizontal coordinate, This is the vehicle's actual heading angle. Let the desired heading angle be denoted. Define the overall error. ,in The conversion factor from heading error to lateral displacement. It has the dimension of length and converts heading error into lateral displacement deviation. (Combined error rate of change) First-order difference approximation , To control the sampling period, and to perform low-pass filtering on the differential results to suppress noise amplification.

[0105] A parameter update algorithm is designed using the gradient descent method:

[0106]

[0107]

[0108]

[0109]

[0110] in The adaptive learning rate is a positive definite constant.

[0111] By analyzing the above rate of change By integrating, the updated control parameter values ​​at the current moment can be obtained. The final output is the updated control parameter vector. .

[0112] 4. Nonlinear compensation and disturbance observation of the actuator

[0113] After completing vehicle operating condition identification and online update of control parameters, this invention establishes a two-layer disturbance observation and compensation system. It estimates external disturbances to the vehicle and internal nonlinear errors of the steering actuator in real time, and achieves comprehensive compensation through speed-related fusion.

[0114] Conventional steering control is mostly based on vehicle dynamics models, neglecting the effects of actuator wear, backlash, lead screw friction, and reducer clearance. Long-term vehicle operation continuously amplifies the nonlinear errors of the actuators, making it difficult to guarantee control accuracy relying solely on vehicle model compensation.

[0115] To improve the trajectory tracking accuracy of logistics vehicles in complex environments:

[0116] First, a vehicle-level disturbance observer (ESO) is established to measure the total disturbance at the vehicle level. (Including road slope, crosswind, modeling errors, etc.) are expanded to the system state. Let... Establish a third-order extended state observer:

[0117]

[0118] in This is the error in the observation of yaw rate. They are respectively The observed values. The observer gain is tuned using the bandwidth method: ,in This refers to the observer bandwidth. ESO output. .

[0119] Secondly, an actuator-level disturbance observer (DOB) is established to unify and equate nonlinearities such as actuator backlash, friction, dead zone, and hysteresis into disturbances. The actual turning angle satisfies Defined from the expected turning angle. To the actual turning angle The nominal model of the transfer function is The disturbance observer is:

[0120]

[0121] in, This is an estimate of the internal disturbances of the executing agency. This is the transfer function for a low-pass filter. This is the Laplace transform of the actual front wheel steering angle. For the Laplace transform of the desired front wheel steering angle, Let be the nominal transfer function from the desired rotation angle to the actual rotation angle, and its physical meaning is the input-output dynamic characteristics of the actuator in a healthy state. Due to nonlinearities such as backlash and friction in actual actuators, the actual rotation angle... With expected corner There is a deviation between them, and this deviation is the actuator disturbance. This is a manifestation of [the principle / principle].

[0122] The nominal model The method of acquisition is as follows: during the vehicle's factory calibration phase or when the actuator is in good health, it is identified through frequency sweep experiments, and its form is a first-order inertial element. ,in For the gain of the actuator, It is a time constant. For the Laplace operator. It is a low-pass filter. Based on the selection of the actuator response bandwidth, the preferred method is... .

[0123] Finally, the proportion of vehicle dynamics disturbances increases at high speeds, while the nonlinear effects of actuators are more significant at low speeds. Therefore, after obtaining the corresponding disturbance estimation results at the vehicle layer ESO and the actuator layer DOB, the disturbances are estimated based on the vehicle's longitudinal speed. Dynamic weighted fusion of the two types of disturbance observation results ensures that external disturbances at the vehicle level and internal nonlinearities at the actuator level can be compensated to a certain extent under different speed conditions.

[0124] Define the fusion weight of the vehicle layer ESO as follows: The fusion weight of the execution layer DOB is Both conditions are met:

[0125]

[0126] in The minimum vehicle disturbance weight is set (0.3 is preferred in this example). This is the maximum reference speed within the vehicle's operating speed range. When the vehicle is operating at low speed... near The vehicle layer ESO maintains a disturbance compensation ratio of no less than 0.3, while the execution layer DOB undertakes a higher proportion of disturbance compensation; when When λ=1, the vehicle layer ESO undertakes all disturbance compensation.

[0127] In the speed-related fusion strategy, the fusion weight λ of the vehicle-level ESO increases monotonically with increasing longitudinal vehicle speed, while the fusion weight 1-λ of the execution-level DOB decreases accordingly. When the longitudinal vehicle speed is below a preset low-speed threshold, the impact of internal nonlinearities of the actuator (such as backlash, friction, etc.) on steering accuracy is greater than the impact of vehicle dynamics disturbances; in this case, the fusion weight of DOB is higher than that of ESO. When the longitudinal vehicle speed is above a preset high-speed threshold, the impact of vehicle dynamics disturbances on steering accuracy is greater than the impact of internal nonlinearities of the actuator; in this case, the fusion weight of ESO is higher than that of DOB. When the longitudinal vehicle speed is between the low-speed threshold and the high-speed threshold, the fusion weights of ESO and DOB change continuously according to the vehicle speed. The specific values ​​of the fusion weights can be determined through calibration based on actual vehicle parameters and operating conditions.

[0128] Under the three typical speed conditions mentioned above, the optimal fusion weights of ESO and DOB are obtained through simulation. The corresponding vehicle lateral tracking error, heading error, error between the actual steering angle and the target steering angle, and disturbance recovery time are recorded to evaluate the effectiveness of the speed-related dynamic fusion strategy. Furthermore, the estimated external disturbance from the vehicle-level ESO and the estimated internal nonlinear disturbance from the actuator-level DOB are weighted and fused to obtain the final estimated comprehensive disturbance value D.

[0129]

[0130] in, This represents the estimated external disturbance of the vehicle output by the vehicle-level ESO. This is the estimated value of the internal nonlinear disturbance of the actuator output by the execution layer DOB. D is then further used to collaboratively compensate for external disturbances to the vehicle and the internal nonlinearity of the actuator.

[0131] 5. Feedforward-feedback-compensation integrated steering control

[0132] After obtaining the desired path curvature, the real-time motion state of the vehicle, and the predicted lateral error, a steering control mechanism that integrates path curvature feedforward, adaptive feedback adjustment, and external disturbance compensation is constructed to achieve high-precision, adaptive steering control of unmanned logistics vehicles in complex road environments.

[0133] First, based on the kinematic model, the theoretical front wheel steering angle is calculated using the path curvature. :

[0134]

[0135] Where L is the vehicle wheelbase. To reference the path curvature, the vehicle steers in advance before entering the curve, reducing lag error. It exhibits high accuracy at low to medium speeds; however, in high-speed, sharp curves, only a basic steering angle is provided, with residual errors corrected by the feedback module.

[0136] Secondly, updated control parameters are adopted. Feedback control quantity for:

[0137]

[0138] This feedback gain integrates five factors: velocity, load, curvature, adhesion, and health, eliminating the need for independently designed gain rules.

[0139] Then, the estimated combined disturbance D is directly used to obtain the compensation control quantity. :

[0140]

[0141] The compensation amount incorporates external disturbances at the vehicle layer and internal nonlinearities at the actuator layer.

[0142] The final turn instruction is the sum of three items:

[0143]

[0144]

[0145] in, The final output is the target front wheel steering angle command. The feedforward term is used to predict road change trends and achieve active steering; the feedback term is used to eliminate the actual trajectory error of the vehicle; and the compensation term is used to resist environmental disturbances.

[0146] Control output target front wheel steering angle command The data is sent to the steering actuator, and the vehicle performs the steering maneuver. Simultaneously, onboard sensors continuously collect feedback values ​​for the actual vehicle position, heading angle, yaw rate, lateral acceleration, and steering angle, and recalculate accordingly. , D, enter the next control cycle, and realize the complete closed loop of "path planning - state perception - operating condition identification - parameter update - disturbance estimation - fusion control - steering execution - feedback update".

[0147] This embodiment also provides an adaptive steering control system for unmanned logistics vehicles in complex environments, including a processing unit, a storage device, and a computer program stored on the storage device and executable. When the processing unit executes the computer program, it implements any of the aforementioned adaptive steering control methods for unmanned logistics vehicles in complex environments.

[0148] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An adaptive steering control method for unmanned logistics vehicles in complex environments, characterized in that, Including the following steps: S100. Obtain the vehicle's operating status information and construct a comprehensive state vector; S200. Based on the operating status information, obtain the vehicle speed, load mass, path curvature, road surface adhesion coefficient, and the evaluation status information of the actuator; and construct a comprehensive working condition vector and its mapping relationship based on the evaluation status information. S300. Using the comprehensive operating condition vector as input, the updated control parameter vector is output through the online update model of the control parameters, and the feedback control quantity is obtained based on the control parameter vector, the lateral tracking error and the heading error. S400: Real-time estimation of external disturbances and internal nonlinear errors of the steering actuator, and fusion of the vehicle's longitudinal speed to obtain a comprehensive disturbance estimate, and obtaining a compensation control quantity based on the comprehensive disturbance estimate; S500: After calculating the theoretical front wheel steering angle information from the path curvature, the theoretical front wheel steering angle, the feedback control quantity, and the compensation control quantity are fused into the final target front wheel steering angle command and output to the steering actuator.

2. The adaptive steering control method for unmanned logistics vehicles in complex environments according to claim 1, characterized in that, In step S100, the operating status information includes at least the vehicle's position coordinates and longitudinal speed. lateral velocity yaw rate r, lateral acceleration Heading angle Front wheel steering angle Drive motor current i, motor output torque T, current cargo weight and reference trajectory curvature ; The overall state vector S is: Where m is the total mass of the vehicle. For the moment of inertia of yaw rotation, For empty vehicle weight, For the current cargo load, The moment of inertia is the yaw inertia under no-load conditions. This refers to the increase in inertia caused by the load. It is the centroid sideslip angle.

3. The adaptive steering control method for unmanned logistics vehicles in complex environments according to claim 1, characterized in that, In step S200, the comprehensive working condition vector M is: in, For speed levels, For load rating, For curvature level, For adhesion level, Health level; This is an estimated value for vehicle weight. The mass threshold that separates unloaded and half-loaded conditions. This is the mass threshold separating half-load and full-load conditions. This is the curvature threshold separating straight lines from ordinary curves. This represents the curvature threshold that separates ordinary curves from sharp turns. Here, H represents the estimated road surface adhesion coefficient, and H is the health index.

4. The adaptive steering control method for unmanned logistics vehicles in complex environments according to claim 1, characterized in that, In step S300, the updated control parameter vector for: Feedback control quantity for: in, For proportional gain, For integral gain, For differential gain, For heading feedback gain; For lateral tracking error, This represents the heading error.

5. The adaptive steering control method for unmanned logistics vehicles in complex environments according to claim 4, characterized in that, In step S300, the parameter update algorithm is as follows: in, These are the adaptive learning rates, both of which are positive definite constants. This is the conversion coefficient from heading error to lateral displacement.

6. The adaptive steering control method for unmanned logistics vehicles in complex environments according to claim 1, characterized in that, In step S400, the comprehensive disturbance estimate D is: in, This is the estimated value of the comprehensive external disturbances to the vehicle. This is an estimate of the nonlinear disturbance within the actuator; These are the weighting coefficients.

7. The adaptive steering control method for unmanned logistics vehicles in complex environments according to claim 6, characterized in that: Compensation control quantity : .

8. The adaptive steering control method for unmanned logistics vehicles in complex environments according to claim 1, characterized in that, In step S500, the theoretical front wheel steering angle : Where L is the vehicle wheelbase. The reference path curvature.

9. The adaptive steering control method for unmanned logistics vehicles in complex environments according to claim 1, characterized in that, In step S500, the target front wheel steering angle command for: in, The theoretical front wheel steering angle, For feedback control, The compensation control quantity is denoted as .

10. An adaptive steering control system for unmanned logistics vehicles in complex environments, comprising a processing unit, a storage device, and a computer program stored on the storage device and executable thereon, characterized in that: When the processing unit executes the computer program, it implements any one of the adaptive steering control methods for unmanned logistics vehicles in complex environments as described in claims 1 to 9.