Real-time tracking data guide type active-disturbance-rejection learning control method for unmanned autonomous vehicle

Abstract

The invention relates to the technical field of automatic control of unmanned autonomous vehicles, in particular to a real-time tracking data guide type active-disturbance-rejection learning control method for an unmanned autonomous vehicle, and the method comprises the steps: determining an expected target trajectory of the unmanned autonomous vehicle; non-linear time-varying dynamic characteristics of the unmanned autonomous vehicle are determined based on the composite disturbance; the motion control parameters are initialized; based on the non-linear time-varying dynamic characteristics and the current motion control parameters, the unmanned autonomous vehicle is controlled to track the expected target trajectory, and a tracking trajectory is obtained; performing instruction filtering on the expected target trajectory based on an attraction rule; based on the expected target trajectory after instruction filtering, the tracking trajectory and the current motion control parameter, updating an estimator of an iterative change disturbance coefficient; until full-time-domain tracking of the expected target trajectory of the unmanned autonomous vehicle is completed; according to the invention, complex disturbance during real-time tracking of the unmanned autonomous vehicle can be reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic control of unmanned autonomous vehicles, and in particular to a real-time tracking data guided auto-disturbance rejection learning control method for unmanned autonomous vehicles. Background Art

[0002] Real-time tracking control of unmanned autonomous vehicles has attracted widespread attention in practical applications. In particular, with the increasing complexity of vehicle systems and the increasing demand for autonomous control accuracy, the application of intelligent control methods in high-precision tracking has become a research hotspot. However, traditional intelligent control methods mostly focus on steady-state tracking tasks, making it difficult to balance the dynamic requirements of control accuracy and energy consumption. Although methods such as finite-time control or predetermined performance control can optimize transient response and accuracy in repetitive tasks, their core flaw is that they cannot take into account the frequency consistency requirements between system components and cannot guarantee delay-free operation within a finite time. In addition, the introduction of fractional exponential terms in the controller can easily lead to singularity problems, complicating the control design and threatening system stability.

[0003] In repetitive mission scenarios such as drone taxiing or vehicle patrolling, iterative learning control (ILC) shows application potential due to its ability to gradually optimize under repetitive trajectories. However, traditional ILC methods and their improved solutions have shortcomings when dealing with non-repetitive disturbances, strongly nonlinear systems, and complex disturbances. Specifically, the existing methods have limited ability to suppress dynamic disturbances, the parameter tuning process is complex and easily affected by the time-varying nature of the system, the predetermined performance control is not robust enough when dealing with disturbances beyond the preset boundaries, and the design of its piecewise performance function may destroy the continuity of the Lyapunov function, making it difficult to ensure theoretical rigor. These problems limit the practical application of existing technologies in the precise control of unmanned vehicles in complex and dynamic environments. Summary of the Invention

[0004] In view of the above problems, the present invention provides a real-time tracking data-guided automatic disturbance rejection learning control method for unmanned autonomous vehicles, which solves the technical problem in the prior art that autonomous vehicles are easily affected by complex disturbances during real-time tracking.

[0005] The present invention provides a real-time tracking data-guided auto-disturbance rejection learning control method for an unmanned autonomous vehicle, comprising the following steps:

[0006] Step S1: Determine the desired target trajectory of the unmanned autonomous vehicle; obtain a composite disturbance based on the extended state learning observer, wherein the composite disturbance includes an iterative invariant component d inv (t) and the iterative change component d var,k (t) determining nonlinear time-varying dynamic characteristics of the unmanned autonomous vehicle based on the composite disturbance; determining motion control parameters of the unmanned autonomous vehicle, and initializing the motion control parameters;

[0007] The expression of the non - linear time - varying dynamic characteristics is as follows:

[0008]

[0009] y k = x 1,k

[0010] where x i,k is the i - th state corresponding to the unmanned autonomous vehicle at the k - th iteration, is the derivative of the state x i,k with respect to time, i is the ordinal number of the state variable, n is the total number of state variables, k ∈ Z + is the number of iterations, Z + is the set of positive integers, u k is the control input of the system at the k - th iteration; represents the system state variables from the 1st to the i - th order; f i 、 is the time - varying system function of the i - th state of the unmanned autonomous vehicle, ∈ is an additional time - varying system function, y k is the position state of the unmanned autonomous vehicle at the k - th iteration, d k is the composite disturbance term; the x i,k 、 u k 、 f i 、 ∈、d k 、y k are all time - variables related to the running time t;

[0011] Step S2: Based on the non - linear time - varying dynamic characteristics and the current motion control parameters, control the unmanned autonomous vehicle to track the desired target trajectory, and obtain the tracking trajectory of the unmanned autonomous vehicle;

[0012] Step S3: Perform command filtering on the desired target trajectory based on the attraction rule; based on the command - filtered desired target trajectory, the tracking trajectory, and the current motion control parameters, use data - guided learning to update the current motion control parameters, including estimating the iterative change disturbance coefficient for updating;

[0013] Step S4: Return to Step S2 for iterative update until the full - time - domain tracking of the desired target trajectory of the unmanned autonomous vehicle is completed.

[0014] Preferably, in Step S1, determining the desired target trajectory of the unmanned autonomous vehicle specifically means: setting y d(t) is the expected target trajectory of the unmanned autonomous vehicle, where t is a positive integer representing the running time;

[0015] The motion control parameters described in step S1 include: the actual motion state x i,k , the target motion state θ i,k , the control adjustment parameter c i , the control input u k , the estimator of the time-varying system function the estimator of the estimated parameter the system nonlinear parameter ζ n,k , the estimator of the iterative change perturbation coefficient Set the initial values of the parameters in the motion control parameters to 0 to complete the initialization.

[0016] Preferably, step S3 specifically includes:

[0017] Step S3-1, perform command filtering on the expected target trajectory based on the attraction rule, including: obtaining the target motion state from the expected target trajectory, filtering the target motion state, and obtaining the command filtering value;

[0018] Step S3-2, update the learning estimation law parameters, including: obtaining the actual motion state from the tracking trajectory, updating the estimator of the time-varying system function and the estimator of the estimated parameter based on the actual motion state, and using the projection operator of the perturbation coefficient estimator to update the perturbation coefficient estimator

[0019] Step S3-3, calculate according to the command filtering value and motion control parameters of the unmanned autonomous vehicle, and update the data-guided learning control amount;

[0020] Step S3-4, update the control input of the unmanned autonomous vehicle based on the data-guided learning control amount, the estimator of the iterative invariant component d inv (t) and the perturbation coefficient estimator.

[0021] Preferably, step S3-2 specifically includes:

[0022] Based on the (k - 1)-th iteration of perform iterative update to obtain the k-th iteration of

[0023] Based on the learning law of the estimator of the iterative change perturbation coefficient, perform iterative update on , and the expression is:

[0024]

[0025] Among them, Respectively represent the estimators of the iterative change perturbation coefficients at the (k + 1)-th and k-th iterations, denote the learning law projection operator of the estimator of the iterative change perturbation coefficient, γ m represents the learning rate of the m-th basis function, T represents the single-iteration period of the system, x 1,k (τ), represents the first state variable and its estimator at the k-th iteration, and τ represents the integration time.

[0026] Preferably, step S3-3 specifically includes:

[0027] From the actual motion state x i,k and the target motion state θ i,k calculate the tracking error z i,k ; then, according to the command filtering value of the unmanned autonomous vehicle the tracking error z i,k the control adjustment parameter c i the actual motion state x i,k the updated time-varying system function estimator obtained in step S3-2 the updated estimator of the estimated parameter δ n,k of the estimator and the system nonlinear parameter ζ i,k perform calculations to update the data-guided learning control quantity u DiLC k .

[0028] Preferably, the calculation expression of the data-guided learning control quantity u DiLC k is:

[0029]

[0030] where z n,k is the error of the system variable at the k-th iteration of the n-th state. When n = 1, z 1,k = y k - y d is the position error variable of the system at the k-th iteration, c n is the n-th control adjustment parameter, is the estimator of the ∈ value of the time-varying system function at the k-th iteration, is the estimator of the estimated parameter δ n,k at the k-th iteration of the n-th state, and ζ i,k represents the system nonlinear parameter.

[0031] Preferably, step S3-4 specifically includes:

[0032] Step S3-4-1: The estimator of the iterative invariant component d can be obtained based on the extended state learning observer model inv of The expression of the extended state learning observer model is as follows:

[0033]

[0034] where represents the estimator of the derivative of state x i,k with respect to time, represents the estimator of state x i+1,k , x 1,k represents the first state corresponding to the k-th iteration of the unmanned autonomous vehicle, l i represents the observer parameter corresponding to the i-th state, respectively represent the estimators of the iterative invariant component d inv and the iterative varying component d var,k , and respectively represent the derivatives of n+1 , b n+2 b respectively represent the extended state gain parameters with indices n + 1 and n + 2;

[0035] Step S3-4-2: Update the control input of the unmanned autonomous vehicle based on the data-guided learning control quantity, the estimator of the iterative invariant component d inv (t), and the disturbance coefficient estimator. The expression is as follows:

[0036]

[0037] where M represents the total number of basis functions for decomposing the iterative varying disturbance, and φ m represents the m-th basis function.

[0038] Compared with the prior art, the present invention has at least the following beneficial effects:

[0039] (1) The present invention provides a non-linear time-varying dynamic characteristic with a composite disturbance term, which can more comprehensively represent the uncertain factors in the unmanned autonomous vehicle system. The composite disturbance includes an iterative invariant component and an iterative varying component, and can respectively model and estimate these two different types of disturbances. It improves the ability of the system to adapt to various disturbances in a complex environment, enabling the unmanned autonomous vehicle to effectively track the desired target trajectory even in a changing environment.

[0040] (2) The present invention estimates the iterative change disturbance coefficient using a learning law based on a projection operator, and corrects the coefficient estimate in the form of integral feedback. This mechanism can dynamically adjust the estimation of the disturbance coefficient to more accurately reflect the variable disturbance currently received by the system. The learning and updating strategy of the present invention for the iterative change disturbance coefficient can improve the response ability of the unmanned autonomous vehicle to the changing dynamic environment and enhance the stability and reliability of the system.

[0041] (3) For the control input of the unmanned autonomous vehicle, the present invention realizes effective compensation for the disturbance by deducting the estimated value of the composite disturbance from the control input. This makes the control input more accurately reflect the actual demand, reduces the control deviation caused by the disturbance, and improves the overall performance of the control of the unmanned autonomous vehicle. Description of the Drawings

[0042] The drawings are only for the purpose of showing specific embodiments and are not considered as a limitation to the present invention.

[0043] Figure 1 It is a flowchart of the real-time tracking data-guided active disturbance rejection learning control method for the unmanned autonomous vehicle provided by the present invention.

[0044] Figure 2 It is a schematic diagram of the dynamic evolution of the position of the unmanned autonomous vehicle provided by the present invention and the expected target trajectory along the time axis before and after data learning of the tracking trajectory.

[0045] Figure 3 It is a schematic diagram of the tracking trajectory of the unmanned autonomous vehicle after the 10th iteration provided by the present invention.

[0046] Figure 4 It is a schematic diagram of the dynamic evolution of the trajectory tracking of the unmanned autonomous vehicle under disturbance provided by the present invention. Detailed Embodiments

[0047] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. In addition, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0048] In order to illustrate the effectiveness of the method proposed by the present invention, the above technical solutions of the present invention will be described in detail below through a specific embodiment. The present invention discloses a real-time tracking data-guided active disturbance rejection learning control method for an unmanned autonomous vehicle, and the specific implementation steps are as follows:

[0049] Step S1: Determine the desired target trajectory of the autonomous vehicle; obtain the composite disturbance based on the extended state learning observer, where the composite disturbance includes the iterative invariant component d inv (t) and the iterative changing component d var,k (t), and determine the non-linear time-varying dynamic characteristics of the autonomous vehicle based on the composite disturbance; determine the motion control parameters of the autonomous vehicle and initialize the motion control parameters.

[0050] To guide the autonomous vehicle to operate as expected, a desired target trajectory is predefined. This trajectory serves as the reference value for the tracking control algorithm, representing the target motion state of the vehicle. The task of the tracking control system is to minimize the deviation between the actual trajectory of the vehicle and the desired target trajectory as much as possible.

[0051] In some embodiments, set y d (t) as the desired target trajectory of the autonomous vehicle, where t ∈ 1, 2, ….

[0052] The present invention provides a composite disturbance d obtained based on an extended state observer (ESO, extended state observer), k where the composite disturbance includes the iterative invariant component d inv and the iterative changing component d var,k . The expression is:

[0053] d k = d inv + d var,k

[0054] The present invention determines the non-linear time-varying dynamic characteristics of the autonomous vehicle based on the composite disturbance, and the expression is:

[0055]

[0056] y k = x 1,k

[0057] where x i,k is the i-th state corresponding to the autonomous vehicle at the k-th iteration, is the derivative of the state x i,k with respect to time, i is the ordinal number of the state variable, n is the total number of state variables, k ∈ Z + is the number of iterations, Z + is the set of positive integers, u k is the control input of the system at the k-th iteration; represents the system state variables from order 1 to i; f i 、 The time-varying system function for the i-th state of the autonomous vehicle, ∈ is an additional time-varying system function, y k is the position state of the autonomous vehicle at the k-th iteration, d k is the composite disturbance term; the x i,k , u k , f i , ∈, d k , y k are all time-varying variables related to the running time t.

[0058] In some embodiments, let the initial state of the autonomous vehicle be x i,k = x i,k+1 ,

[0059] In the above manner, the present invention determines the time-varying dynamic characteristics of the system corresponding to the autonomous vehicle. For the i-th state variable, its time derivative is obtained by summing the (i + 1)-th state variable, the time-varying system function, and the time-varying disturbance term. For the last state variable x n,k , its time derivative is composed of the control input u k (t), the time-varying system function f n , the time-varying disturbance term and the composite disturbance term d k added together.

[0060] The present invention introduces a composite disturbance term into the time-varying dynamic characteristics of the system corresponding to the autonomous vehicle, which can more comprehensively handle the uncertainty factors in the autonomous vehicle system. By decomposing the composite disturbance into an invariant part d inv and a variable part d var,k , the two different types of disturbances can be modeled and estimated separately. This decomposition processing method improves the adaptability of the system to various disturbances in a complex environment.

[0061] The original signal obtained by the system corresponding to the autonomous vehicle is the information that the system can directly measure or observe. In some embodiments, these signals may include physical state information such as the position, speed, and acceleration of the vehicle, and these signals can be obtained through sensors (such as GPS receivers, radars, cameras, etc.) and used as input data for the control system.

[0062] The original signal can be processed and transformed to obtain the system state, and the system state includes the actual motion state x i,k and the target motion state θ i,k . The control adjustment parameter c iFor determining how the system state affects the control input and then deriving the control input u k , the control input directly acts on the motion of the autonomous vehicle and adjusts the position of the autonomous vehicle.

[0063] For the motion control parameters of the autonomous vehicle, the motion control parameters determined by the present invention include parameters reflecting the system motion state: the actual motion state x i,k and the target motion state θ i,k , parameters reflecting the real-time feedback of the control system on the motion: the control adjustment parameter c i and the control input u k , parameters reflecting the iteration parameters in the data-guided learning iteration process: the time-varying system function estimator the estimator of the estimated parameter the system nonlinear parameter ζ after smooth approximation n,k , the estimator of the iteration change perturbation coefficient In this step, these parameters are initialized, and in subsequent steps, these parameters will be iteratively updated until the real-time tracking of the target trajectory is completed.

[0064] In some embodiments, the initial values of the parameters in the above motion control parameters are set to 0 to complete the initialization of these parameters.

[0065] Step S2, based on the non-linear time-varying dynamic characteristics and the current motion control parameters, control the autonomous vehicle to track the desired target trajectory and obtain the tracking trajectory of the autonomous vehicle;

[0066] In some embodiments, in order to achieve precise path tracking, the autonomous vehicle needs to consider its own non-linear and time-varying motion characteristics. Continuously correct the position of the vehicle in combination with the current motion control parameters to ensure that it always travels along the target path within the current iteration cycle, and finally obtain the tracking trajectory of the autonomous vehicle.

[0067] In some embodiments, in order to determine the real-time position of the autonomous vehicle, the position of the autonomous vehicle at the current moment can be obtained based on a variety of equipped sensors. For example, a GPS receiver, an inertial measurement unit (IMU), and an optical flow sensor can work together to provide accurate information on the current position of the vehicle.

[0068] In some embodiments, after obtaining the real-time position information of the autonomous vehicle, the current tracking error can be calculated. Compare the real-time position data with the predetermined path, which is achieved by comparing the actual position of the vehicle with its ideal position on the predetermined path, and determine the deviation between the actual position of the autonomous vehicle and the position it should reach as the position tracking error.

[0069] In some embodiments, the calculation method of the tracking error can be flexibly selected. For example, the Euclidean distance between the current position of the vehicle and the nearest point on the target path can be measured. Considering that factors such as wind force and terrain changes may affect the driving of the vehicle, a more complex algorithm can be introduced to optimize the error estimation, thereby improving the navigation performance of the unmanned vehicle.

[0070] In some embodiments, the position tracking error z at the k-th iteration 1,k is calculated as: z 1,k = y k - y d , and the tracking error is used for the calculation of subsequent steps.

[0071] Step S3: Perform command filtering on the desired target trajectory based on the attraction rule; based on the command-filtered desired target trajectory, the tracking trajectory, and the current motion control parameters, use data-guided learning to update the current motion control parameters, including estimating the iterative change perturbation coefficient.

[0072] In specific implementation, in the above control method provided by the embodiments of the present invention, step S3 specifically includes the following steps:

[0073] Step S3-1: Perform command filtering on the desired target trajectory based on the attraction rule, including: obtaining the target motion state from the desired target trajectory, filtering the target motion state, and obtaining the command filtering value;

[0074] The present invention introduces a command filtering operation to avoid the problem of explosion of the calculation complexity of high-order partial derivatives. Among them The calculation represents a command filtering operation introduced to avoid the problem of explosion of the calculation complexity of high-order partial derivatives, and there will be dynamic errors in the command filtering operation.

[0075] To compensate for the error of this filter, an attraction rule is designed and adjusted through the following equations:

[0076]

[0077] Among them, θ i,k is the target value of the state variable, is the derivative of the target value of the state variable and serves as the command filtering value, ω i is the filter bandwidth, α i,k is the corresponding function of the virtual controller, c i is the gain of the i-th order controller, η i,k is the state variable of the filter error compensation signal of the i-th order, is the derivative of the state variable of the filter error compensation signal of the i-th order.

[0078] The attraction rule designs the error of the filter so that the direction of the change in the filtering error during the dynamic change process trends towards decrease, because the design form is similar to traction. In the filter design without the attraction rule, the filtering error cannot be effectively restricted either, and the convergence analysis of the filtering error can only be obtained after the design. In this calculation, a virtual controller is introduced, and the parameter value is α i,k , and there are n - 1 virtual controllers in the nonlinear backstepping design. Through the cascaded design of coordinate transformation, the control value of the real controller is finally obtained.

[0079] In some embodiments, each controller gain c1,..., c i together constitute the control adjustment parameters. By adjusting the controller state and gain in the above manner, the error of the unmanned vehicle during the movement process will not gradually increase and can be effectively controlled and reduced.

[0080] Step S3 - 2: Update the learning estimation law parameters, including: obtaining the actual motion state from the tracking trajectory, updating the time - varying system function estimator and the estimator of the estimation parameters based on the actual motion state, and using the projection operator of the disturbance coefficient estimator to update the disturbance coefficient estimator

[0081] The present invention updates the data - guided learning control quantity u based on the guided adaptive adjustment of the time - varying system function estimator and the estimator of the estimation parameters . Before updating u DiLC k . This step first performs iterative update on DiLC k beforehand. for iteration update.

[0082] The data - guided learning control learning estimation law is:

[0083]

[0084] wherein, is the learning law projection operator of the ∈ value of the time - varying system function, is the adaptive parameter, and the calculation method of κ i,k is κ i,k = z i,k - η i,k , where i = 1,…, n, representing the compensated coordinate transformation obtained from the coordinate transformation and the attraction rule, is the learning law projection operator of the estimation parameter δ i,k , and the calculation method of the system nonlinear parameter ζ i,k is: where \(i = 1,\ldots,n\), representing the system's non - linear function after approximation and smoothing by the hyperbolic tangent function, where \(j\) is the summation index, is the summation weight of the \(j\) - th state variable at the \(k\) - th iteration, \(x\) j,k is the \(j\) - th state variable at the \(k\) - th iteration, \(\tanh\) is the hyperbolic tangent function, \(\rho\) k is the scaling factor at the \(k\) - th iteration, used to adjust the speed of change of the inflection point of the smoothing function.

[0085] In the above - mentioned way, update the parameters of the current iteration based on the previous estimated values and Conduct data - guided learning, and the control system of the unmanned autonomous vehicle can achieve adaptive adjustment to better cope with the influence of environmental changes and system internal dynamics.

[0086] This step also iteratively updates the estimator of the iterative change perturbation coefficient For the purpose of tracking the iterative change perturbation of the unmanned autonomous vehicle, the present invention decomposes and models the iterative change perturbation to obtain the expression:

[0087]

[0088] where, \(a\) m (k) represents the iterative change perturbation coefficient of the \(m\) - th basis function at the \(k\) - th iteration, \(M\) represents the total number of basis functions used to decompose the perturbation, \(\varphi\) m represents the \(m\) - th basis function.

[0089] Based on the above - mentioned iterative change perturbation modeling, the present invention determines the learning law of the estimator of the iterative change perturbation coefficient, and the expression is:

[0090]

[0091] where, respectively represent the estimators of the iterative change perturbation coefficients at the \((k + 1)\) - th and \(k\) - th iterations, represents the learning - law projection operator of the estimator of the iterative change perturbation coefficient, \(\gamma\) m represents the learning rate of the \(m\) - th basis function, \(T\) represents the single - iteration period of the system, \(x\) 1,k (\(\tau\)), represents the first state variable and its estimator at the \(k\) - th iteration, and \(\tau\) represents the integration time.

[0092] Step S3 - 3: Calculate according to the command filtering value and motion control parameters of the unmanned autonomous vehicle, and update the data - guided learning control quantity;

[0093] The specific calculation process of this step is as follows: calculate the tracking error from the actual motion state and the target motion state; then calculate based on the command filtering value of the unmanned autonomous vehicle, the tracking error, the control adjustment parameter, the actual motion state, the updated time-varying system function estimator obtained in step S3-2, the estimator of the updated estimated parameter, and the updated system nonlinear parameter to update the data-guided learning control amount.

[0094] According to the system matrix corresponding to the unmanned autonomous vehicle, obtain its corresponding coordinate transformation z 1,k = y k - y d , z i,k = x i,k - θ i,k . Among them, the coordinate transformation z 1,k = y k - y d is the position error variable of the t system in the kth iteration, and the coordinate transformation z i,k = x i,k - θ i,k is the system position error variable in the kth iteration of the i-th state.

[0095] In some embodiments, the system matrix is determined based on the dynamic and control system models of the unmanned autonomous vehicle, and the system matrix includes various parameters and state variables of vehicle motion. The coordinate transformation z 1,k = y k - y d represents the difference between the actual position and the target position of the vehicle, while z i,k = x i,k - θ i,k represents the difference between the actual value and the target value of the vehicle in other state variables (such as speed, acceleration, etc.). This coordinate transformation is for facilitating subsequent error analysis and control strategy design.

[0096] Determine the data-guided learning control amount u at this iteration through the following formula DiLC k :

[0097]

[0098] Among them, z n,k = x n,k - θ n,k is the error of the system variable in the kth iteration of the nth state. When n = 1, z 1,k = y k - y d is the position error variable of the system in the kth iteration, and c n is the gain coefficient of the nth, used to adjust the control intensity. is the estimator of the ∈ value of the time-varying system function at the k-th iteration, is the estimated parameter δ at the k-th iteration of the n-th state n,k The estimator of. It should be noted that for z in the control design i,k , x i,k , θ n,k The function values such as etc. are all time-varying variables related to t, and t is omitted in the representation.

[0099] Step S3-4, update the control input of the unmanned autonomous vehicle based on the data-guided learning control quantity, the estimator of the iterative invariant component d inv (t) and the estimator of the disturbance coefficient;

[0100] In this step, update the control input u of the unmanned autonomous vehicle based on the estimator of the iterative invariant component d inv of the data-guided learning control quantity and the estimator of the disturbance coefficient k .

[0101] The present invention completes the calculation of the relationship between the estimated values of the motion state, the iterative invariant component and the iterative change component by establishing an extended state learning observer equation, and the expression is:

[0102]

[0103] Among them, represents the estimator of the derivative of the state x i,k with respect to time, represents the estimator of the state x i+1,k , x 1,k represents the first state corresponding to the unmanned autonomous vehicle at the k-th iteration, l i represents the observer parameter corresponding to the i-th state, respectively represent the estimators of the iterative invariant component d inv and the iterative change component d var,k , and respectively represent derivatives, b n+1 , b n+2 respectively represent the extended state gain parameters with indexes n + 1 and n + 2.

[0104] The estimator of the iterative invariant component d can be obtained by solving the extended state learning observer equation inv On this basis, finally, the control input u of the unmanned autonomous vehicle at this iteration is determined by the following formula k :

[0105] ​

[0106] Step S4. Return to Step S2 for iterative update until the full-time domain tracking of the desired target trajectory of the unmanned autonomous vehicle is completed.

[0107] It should be noted that in the above control method provided by the embodiments of the present invention, there is no fixed order for the execution of each step in Step S1, and it is not limited herein.

[0108] The above real-time tracking iterative learning control system and method for a time-varying unmanned autonomous vehicle provided by the embodiments of the present invention can overcome the influence brought by the time-varying dynamics of the unmanned autonomous vehicle, enabling the unmanned autonomous vehicle with unknown non-linear time-varying characteristics to be able to track the trackable target trajectory in real time at all times within a limited time period, and having extremely high tracking accuracy.

[0109] There are three uncertainties in the unmanned autonomous vehicle under interference: the unknown non-linear function f i (·), which represents the inaccuracy of our internal dynamics modeling of the system; the time-varying parameter ∈(t), such as parameter drift caused by environmental changes or system aging; and the iteration-related external disturbance d k (t), which may have periodicity or some iterative law in repetitive tasks; the present invention analyzes the above three uncertainties and constructs an R-DiLC control framework based on an instruction filter, an extended state observer (ESO), and direct iterative learning control (DiLC).

[0110] The present invention conducts a strict stability analysis of the R-DiLC control framework theoretically. By constructing a composite Lyapunov function, it can be proven through formula derivation that this Lyapunov function continuously decreases during the iteration process and finally approaches zero, thereby proving the stability of the R-DiLC scheme closed-loop system and the convergence of the compensation error, and proving that in the case of ideal filtering, the tracking error can asymptotically converge to zero; in actual situations, due to the existence of filtering deviation, the tracking error will converge to a small bounded region. Compared with traditional adaptive iterative learning control methods, the error convergence speed of the present invention is faster, and the scheme provided by the present invention has stronger superiority theoretically, providing a solid theoretical basis for practical applications.

[0111] The following provides a simulation case of applying the real-time high-precision tracking iterative learning control system and method for an unmanned autonomous vehicle under interference provided by the present invention.

[0112] Embodiment 1

[0113] Consider a time-varying unmanned autonomous vehicle with the following non-linear dynamics:

[0114]

[0115] Among them, the system parameters are

[0116] ∈=1+0.2sin(t)

[0117] f1(x 1,k )=cos(t)·x 1,k ,

[0118] f2(x 1,k ,x 2,k )=sin(t)[x 1,k ·cos(x 1,k )+x 2,k ·sin(x 2,k )].

[0119] f1 and f2 represent the nonlinear part of the system. These terms introduce the nonlinear dynamic behavior of the system. ∈=1+0.2sin(t) represents the time-varying interference intensity of the system. and is an additional nonlinear function, which further increases the complexity of the system. k (t) include:

[0120] d inv (t)=10cos(2πt)+0.1randn(t) (iteration remains unchanged)

[0121] d var,k (t) = 0.05cos(k)cos(2πt) (iterative variation)

[0122] Where randn(t) represents Gaussian white noise.

[0123] Consider the operation period as T = 10, and select the initial state of the unmanned autonomous vehicle as x k (0)=[0,0] T , And the initial input is u0(t)=[0,0] T , An initial iteration k=0 is performed.

[0124] Step S10: Determine the expected target trajectory of the unmanned autonomous vehicle as y d (t) = sin 3 (t), t∈{1,2,…,10}. This step sets the expected path that the unmanned autonomous vehicle needs to follow within a certain time, that is, from 1 to 10 seconds.

[0125] Step S20: Obtain the system matrix corresponding to the unmanned autonomous vehicle, and obtain the corresponding system function that satisfies the following relationship:

[0126]

[0127] Where λ1 and λ2 are constants, indicating the tightness of the inequality.

[0128] Step S30: Obtain the position y of the unmanned autonomous vehicle at the current iteration k k (t), And calculate the tracking error z under the current iteration k (t) = y d (t)-y k (t),

[0129] Step S40: Update the control input of the unmanned autonomous vehicle for the next iteration based on the control input, tracking error, and selected control adjustment parameters of the unmanned autonomous vehicle in the current iteration, and control the unmanned autonomous vehicle to execute the next iteration. This specifically includes the following steps:

[0130] Step S401: Determine the control input of the unmanned autonomous vehicle based on the control input and tracking error of the unmanned autonomous vehicle in the current iteration. The control input is calculated in the same manner as in the above method.

[0131] Step S402: The unmanned autonomous vehicle learns and tracks the error according to the control input data of the current iteration to adjust the learning estimation law to obtain the learning estimation law of the next iteration, update the controller, adjust its own position, and finally achieve real-time tracking of the expected target trajectory, i.e., lim k→∞ y k (t) = y d (t), t∈{1,2,…,10}.

[0132] Step S50: Return to step S30 and perform multiple iterations until the real-time tracking of the target trajectory is completed.

[0133] In order to further illustrate the effectiveness of the real-time tracking iterative learning control system and method for the time-varying unmanned autonomous vehicle provided by the present invention, Figure 2 The dynamic evolution of the position of the unmanned autonomous vehicle and the expected target trajectory along the time axis is shown before and after data learning. Figure 3 The figure shows the dynamic evolution of the position of the unmanned autonomous vehicle and the expected target trajectory along the time axis after the 10th iteration. Figure 4 The dynamic evolution of observation noise for tracking trajectories of unmanned autonomous vehicles is shown.

[0134] Therefore, the time-varying unmanned autonomous vehicle real-time high-precision tracking iterative learning control system and method provided by the present invention can effectively solve the problems caused by the time-varying dynamics of unmanned autonomous vehicles, enabling unmanned autonomous vehicles with unknown nonlinear dynamic characteristics to accurately track a predetermined trajectory in real time within a finite time and achieving extremely high tracking accuracy.

[0135] Although the specific embodiments of the present invention depict various actions or steps in a particular order, this should be understood as requiring such actions or steps to be performed in the particular order shown or in a sequential order, or requiring all illustrated actions or steps to be performed to achieve the desired result. In certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although a number of specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the present disclosure. Certain features described in the context of separate embodiments may also be implemented in combination in a single implementation. Conversely, the various features described in the context of a single implementation may also be implemented separately or in any suitable sub-combination in multiple implementations.

[0136] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A real-time tracking data-guided active disturbance rejection learning control method for an autonomous vehicle, characterized in that including the following steps: Step S1: Determine the desired target trajectory of the unmanned autonomous vehicle; obtain the composite disturbance including the iterative invariant component d inv (t) and the iterative varying component d var,k (t), and determine the non-linear time-varying dynamic characteristics of the unmanned autonomous vehicle based on the composite disturbance; determine the motion control parameters of the unmanned autonomous vehicle, and initialize the motion control parameters; The expression of the non-linear time-varying dynamic characteristics is: y k = x 1,k where, x i,k is the i-th state corresponding to the unmanned autonomous vehicle at the k-th iteration, is the derivative of the state x i,k with respect to time, i is the ordinal number of the state variable, n is the total number of state variables, k ∈ Z + is the iteration number, Z + is the set of positive integers, u k is the control input of the system at the k-th iteration; represents the system state variables from the 1st to the i-th order; f i 、 is the time-varying system function of the i-th state of the unmanned autonomous vehicle, ∈ is an additional time-varying system function, y k is the position state of the unmanned autonomous vehicle at the k-th iteration, d k is the composite disturbance term; the x i,k 、 u k 、 f i 、 ∈、d k 、y k are all time-varying variables related to the running time t; Step S2, based on the non-linear time-varying dynamic characteristics and the current motion control parameters, control the unmanned autonomous vehicle to track the desired target trajectory, and obtain the tracking trajectory of the unmanned autonomous vehicle; Step S3, perform command filtering on the desired target trajectory based on the attraction rule; Based on the expected target trajectory after instruction filtering, the tracking trajectory, and the current motion control parameters, the current motion control parameters are updated using data-guided learning, including estimating the iterative change perturbation coefficient for update; Step S4, return to Step S2 for iterative update until the full-time domain tracking of the desired target trajectory of the unmanned autonomous vehicle is completed.

2. The real-time tracking data-guided active disturbance rejection learning control method for an unmanned autonomous vehicle according to claim 1, wherein, The determination of the desired target trajectory of the driverless autonomous vehicle in step S1 is specifically as follows: Set y d (t) as the expected target trajectory of the driverless autonomous vehicle, where t is a positive integer representing the running time; The motion control parameters in Step S1 including: actual motion state x i,k , target motion state θ i,k , control adjustment parameter c i , control input u k , time-varying system function estimator estimator of estimated parameter system nonlinear parameter ζ n,k , estimator of iterative change perturbation coefficient Let the initial values of the parameters in the motion control parameters be 0 to complete the initialization.

3. The real-time tracking data-guided active disturbance rejection learning control method for the unmanned autonomous vehicle according to claim 2, characterized in that, Step S3 specifically includes: Step S3-1, perform command filtering on the desired target trajectory based on the attraction rule, including: obtaining the target motion state from the desired target trajectory, filtering the target motion state, and obtaining the command filtering value; Step S3-2, updating the learning estimation law parameters, including: obtaining the actual motion state from the tracking trajectory, updating the estimator of the time-varying system function and the estimator of the estimation parameters based on the actual motion state, and using the projection operator of the disturbance coefficient estimator Updating the disturbance coefficient estimator Step S3-3, calculate according to the command filtering value and the motion control parameters of the unmanned autonomous vehicle, and update the data-guided learning control amount; Step S3-4: Update the control input of the unmanned autonomous vehicle based on the data-guided learning control quantity, the estimator of the iterative invariant component d inv (t), and the estimator of the disturbance coefficient.

4. The real-time tracking data-guided active disturbance rejection learning control method for an unmanned autonomous vehicle according to claim 3, characterized in that Step S3-2 specifically includes: Based on the (k - 1)-th iteration perform iterative update to obtain the (k)-th iteration of The learning law of the estimator based on the iterative change perturbation coefficient is used to perform iterative updates, and the expression is: Among them, respectively represent the estimators of the iterative change perturbation coefficients at the (k + 1)-th and k-th iterations, represents the learning law projection operator of the estimator of the iterative change perturbation coefficient, γ m represents the learning rate of the m-th basis function, T represents the single-iteration period of the system, represents the first state variable and the estimator at the k-th iteration, and τ represents the integration time.

5. The real-time tracking data-guided active disturbance rejection learning control method for an unmanned autonomous vehicle according to claim 4, wherein, Step S3-3 specifically includes: From the actual motion state x i,k and the target motion state θ i,k calculate the tracking error z i,k ; then, according to the command filtering value of the unmanned autonomous vehicle the tracking error z i,k , the control adjustment parameter c i , the actual motion state x i,k , the updated time-varying system function estimator obtained in step S3-2 the estimated value of the updated estimated parameter δ n,k and the system nonlinear parameter ζ i,k perform calculations to update the data-guided learning control amount u DiLC k k .

6. The real-time tracking data-guided active disturbance rejection learning control method for an unmanned autonomous vehicle according to claim 5, wherein The calculation expression of the data-guided learning control quantity u DiLC k is as follows: where z n,k is the error of the system variable at the k-th iteration of the n-th state. When n = 1, z 1,k = y k - y d is the position error variable of the system at the k-th iteration, c n is the n-th control adjustment parameter, is the estimator of the ∈ value of the time-varying system function at the k-th iteration, is the estimator of the estimation parameter δ at the k-th iteration of the n-th state n,k , ζ i,k represents the system nonlinear parameter.

7. The real-time tracking data-guided active disturbance rejection learning control method for the unmanned autonomous vehicle according to claim 6, characterized in that, Step S3-4 specifically includes: Step S3-4-1: The iterative invariant component d inv can be obtained based on the extended state learning observer model inv estimator The expression of the extended state learning observer model is: Among them, represents the estimator of the derivative of the state x with respect to time, i,k represents the estimator of the state x, i+1,k x 1,k represents the first state corresponding to the k-th iteration of the unmanned autonomous vehicle, i represents the observer parameter corresponding to the i-th state, respectively represent the invariant component d of the iteration inv and the varying component d of the iteration var,k estimators, and respectively represent the derivative of, n+1 , n+2 respectively represent the extended state gain parameters with indices n + 1 and n + 2;​ Step S3-4-2: Update the control input of the unmanned autonomous vehicle based on the data-guided learning control quantity, the estimator of the iterative invariant component d inv (t), and the estimator of the disturbance coefficient. The expression is as follows: Among them, M represents the total number of basis functions used to decompose the iterative change perturbation, and φ m represents the m-th basis function.

Images