Motion control method and system for distributed ultra-flat motion control platform

Abstract

The invention provides a motion control method and system for a distributed ultra-flat motion control platform, and belongs to the technical field of automatic driving. According to the invention, by analyzing the wheel-ground action relationship of the distributed ultra-flat motion control platform, an integrated model of wheels, the whole control platform, an electric drive module and the like is constructed, and a method for integrating road adhesion coefficient estimation and uncertainty thereof into a path tracking control law is designed. A motion platform control high-speed stability influence mechanism is analyzed, and a robust stability control strategy is constructed, so that the robustness and tracking precision of the distributed ultra-flat platform motion controller can be effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of automatic driving, and particularly relates to a motion control method and system for a distributed super-flat motion control platform. BACKGROUND

[0002] As an important direction for the development of future intelligent electric vehicles, the distributed super-flat motion control platform has the technical advantages of low gravity center, strong controllability, and controllable electric drive height. Compared with the traditional centralized power platform, its wheel edge drive architecture provides more space for maximizing the adhesion utilization rate of the vehicle and high dynamic response control.

[0003] However, the platform still faces great challenges in achieving precise control under extreme conditions. Traditional filtering estimation methods rely on sufficient time domain signal excitation, which is difficult to apply to weak motion excitation conditions. In addition, many current studies do not fully evaluate the error range and distribution characteristics of the estimation results, resulting in a lack of sufficient robustness of the path tracking controller, especially in the case of sudden or rapid changes in adhesion coefficients. In addition, measurement noise inevitably leads to adhesion coefficient estimation errors, thereby increasing the uncertainty of system model and controller constraints, and model misalignment and constraint conservation will affect vehicle control performance. Therefore, it is particularly important to carry out research on road adhesion adaptive motion control for the distributed super-flat platform, analyze the wheel-ground interaction of the small-diameter wheels of the motion platform, construct an integrated model of the wheels, vehicle body, and electric drive modules, design a method that can integrate adhesion coefficient estimation and its uncertainty into the path tracking control law, analyze the influence mechanism of high-speed stability of the motion platform and construct its robust stability control strategy, and improve the robustness and tracking accuracy of the motion controller of the distributed super-flat platform.

[0004] Therefore, it is necessary to provide a motion control method and system for a distributed super-flat motion control platform to solve the above problems. SUMMARY

[0005] The application provides a motion control method and system for a distributed super-flat motion control platform, which designs a learning-type stochastic model predictive control path tracking method to solve the influence of road adhesion coefficient estimation uncertainty in the design of the prediction model and the chance constraint, thereby solving at least one technical problem involved in the background art.

[0006] To solve the above technical problems, the application is implemented as follows: A motion control method for a distributed super-flat motion control platform, comprising the following steps: Step S1, based on the dynamics model of the distributed ultra-flat motion control platform and the dynamics model of the wheels, an expression function of the road adhesion coefficient is derived, and a state parameter related to the road adhesion coefficient is determined under the condition that there is only one real solution to the expression function; Step S2, a space-time convolution network is constructed, a plurality of different combinations of state parameters form a group of data, the road adhesion coefficient under the group of data is taken as a label, a plurality of groups of data and corresponding labels are combined to form a data set, the data set is input into the space-time convolution network for learning, high-dimensional features are extracted from time domain signals and time-frequency domain signals respectively, and the output high-dimensional feature vector is taken as an input vector of a sparse variational Gaussian process, a mapping relationship between the state parameters and the road adhesion coefficient is learned by using a sparse variational strategy, after the training is completed, the state parameter combination is input into the space-time convolution network, and an estimated value and an estimated variance of the road adhesion coefficient are output; Step S3, based on the estimated value and the estimated variance of the road adhesion coefficient, a controller for adaptive path tracking is designed to regulate the motion state of the distributed ultra-flat motion control platform.

[0007] As a preferred improvement, the dynamics model of the seven-degree-of-freedom distributed ultra-flat motion control platform is represented as: In the formula, denotes the mass; denotes the longitudinal velocity; denotes the longitudinal acceleration; , denote the longitudinal forces of the left front wheel and the right front wheel respectively; denotes the front wheel steering angle; , denote the lateral forces of the left front wheel and the right front wheel respectively; , denote the longitudinal forces of the left rear wheel and the right rear wheel respectively; denotes the lateral velocity, denotes the yaw angular velocity; , , denote the air resistance, the wheel rolling resistance and the slope resistance respectively; denotes the lateral acceleration; , denote the lateral forces of the left rear wheel and the right rear wheel respectively; denotes the yaw angular acceleration; , denote the distances from the front and rear axles to the center of mass respectively; denotes the wheel track; represents the yaw moment.

[0008] As a preferred improvement, the wheel dynamics model is represented as: wherein, represents the wheel moment of inertia, represents the wheel angular velocity; represents the wheel angular acceleration; represents the wheel drive torque output by the in-wheel motor drive system; represents the wheel radius; wheel drive torque is composed of a high frequency signal and a low frequency signal, represented as: wherein, , represent the low frequency signal and the high frequency signal, respectively; represents the amplitude of the signal; represents the frequency of the signal, represents time; Then, the wheel dynamics model in drive mode is represented as: wherein, represents the amplitude of the torque; represents ; represents the imaginary unit; represents the relaxation coefficient; represents the tire longitudinal stiffness; The wheel dynamics model in brake mode can be represented as: As a preferred improvement, in non-steady state maneuvering conditions, parameters closely related to the road adhesion coefficient include: mass , wheel slip ratio , longitudinal wheel stiffness , longitudinal acceleration , wheel vertical force , wherein represents the wheel longitudinal force; In steady state maneuvering conditions, parameters closely related to the road adhesion coefficient include: longitudinal speed , wheel angular velocity , wheel vertical force and .

[0009] As a preferred improvement, the space-time convolution network comprises a non-steady-state maneuvering space-time convolution sub-network and a steady-state maneuvering space-time convolution sub-network, the non-steady-state maneuvering space-time convolution sub-network taking state parameters and labels in a non-steady-state maneuvering condition as input, and the steady-state maneuvering space-time convolution sub-network taking state parameters and labels in a steady-state maneuvering condition as input.

[0010] As a preferred improvement, the non-steady-state maneuvering space-time convolution sub-network comprises three convolution units connected in sequence, each convolution unit comprising a convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence, wherein the convolution kernel sizes of the three convolution layers are 1x4x8, 1x4x16 and 1x4x32 in sequence; the steady-state maneuvering space-time convolution sub-network comprises a time-domain feature extraction branch and a time-frequency domain feature extraction branch arranged in parallel, wherein the time-domain feature extraction branch is used to directly process a training data segment, extract time-domain features through convolution operation, the time-domain feature extraction branch comprises three convolution units connected in sequence, each convolution unit comprising a convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence, wherein the convolution kernel sizes of the three convolution layers are 1x4x8, 1x4x16 and 1x4x32 in sequence; the time-frequency domain feature extraction branch obtains two-dimensional time-frequency domain features through continuous wavelet transform, the time-frequency domain feature extraction branch comprises four convolution units connected in sequence, each convolution unit comprising a convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence, wherein the convolution kernel sizes of the three convolution layers are 4x2x4, 3x4x8, 2x2x16 and 2x2x32 in sequence; the time-domain features and the time-frequency domain features are fused together through a Flatten layer to form an input of a sparse variational Gaussian process.

[0011] As a preferred improvement, the sparse variational Gaussian process specifically comprises the following steps: Step S21, selecting representative sample points from the input vector as inducing points, and regarding the function values at the inducing points as a set of inducing variables to represent the key information of the entire input vector; Step S22, introducing a variational distribution in variational inference to approximately replace the real posterior distribution; Step S23, taking the evidence lower bound maximization of marginal log-likelihood as the optimization target, and using the standard stochastic gradient descent method for optimization to make the variational distribution as close as possible to the real posterior distribution.

[0012] As a preferred improvement, the lateral dynamic equation of the distributed ultra-flat motion control platform is: In the formula, M represents the total mass of the distributed ultra-flat motion control platform; represents the longitudinal velocity; represents the lateral velocity; represents the first derivative of the side slip angle of the center of mass; represents the yaw rate; , represents the front and rear wheel lateral forces, respectively; represents the moment of inertia of the vehicle around the z-axis; , represents the distance from the front and rear axles to the center of mass, respectively; The wheel model is represented as: wherein, , represents the front and rear wheel lateral affine forces of the last control period, respectively; , represents the front and rear wheel equivalent side slip stiffness, respectively, represents the road adhesion coefficient estimate; , represents the front and rear wheel side slip angles at the current time, respectively; , represents the front and rear wheel side slip angles of the last control period, respectively; Substituting the wheel model into the lateral dynamics equation, the lateral dynamics equation is further represented as: ; wherein, , , , , , , , all represent intermediate calculation parameters; wherein: ; ; ; ; ; ; ; ; The differential equation of the path tracking model is represented as: ; wherein, represents the state of the distributed ultra-flat motion control platform, , represents the lateral error; represents the heading angle error; denotes the transpose matrix; denotes the control input of the distributed hyper-flat motion control platform, , denotes the system disturbance; , , denote the weight matrix, wherein: ; ; ; wherein, path curvature; discretizes the differential equation of the path tracking model as: ; wherein, , denote the state of the distributed hyper-flat motion control platform at time and , respectively; denotes the control input of the distributed hyper-flat motion control platform at time ; denotes the system disturbance at time ; , , denote the weight matrix at time ; wherein: wherein, denotes the identity matrix; denotes the sampling time.

[0013] As a preferred improvement, the control prediction model of the adaptive path tracking controller is represented as: ; s.t ; ; ; ; wherein, , , denote the weight matrix; denotes the state matrix; denotes the reference state matrix; denotes a control increment; denotes a terminal state matrix; k denotes a time node; N denotes a total number of time nodes; denote a lateral error maximum value, a heading angle error maximum value, a yaw rate maximum value and a mass center side slip angle maximum value, respectively; denotes a probability function; denotes a probability specific value; denote a minimum limit value and a maximum limit value of a front wheel steering angle denote a minimum limit value and a maximum limit value of a front wheel steering angle increment The constructed control prediction model is solved to obtain an optimal control sequence, and a first control quantity of the optimal control sequence is taken as a current expected front wheel steering angle and is sent to a lower steering actuator, so that path tracking control of the distributed ultra-flat motion control platform can be realized.

[0014] A system for performing the motion control method for the distributed ultra-flat motion control platform is provided, and the system comprises: A parameter determination module is configured to derive an expression function of a road adhesion coefficient based on a dynamics model of the distributed ultra-flat motion control platform and a dynamics model of a wheel, and determine a state parameter related to the road adhesion coefficient under the condition that there is only one real solution in the expression function. A space-time convolution network is configured to form a group of data by using different combinations of a plurality of state parameters, take a road adhesion coefficient under the group of data as a label, combine a plurality of groups of data and corresponding labels to form a data set, input the data set into the space-time convolution network for learning, extract high-dimensional features from time domain signals and time-frequency domain signals respectively, and take an output high-dimensional feature vector as an input vector of a sparse variational Gaussian process. A controller design module is configured to design a controller for adaptive path tracking based on the estimated value and the estimation variance of the road adhesion coefficient, and regulate and control the motion state of the distributed ultra-flat motion control platform.

[0015] The present application has the following advantages: ​​​​​​​​​(1) Through the random variation deep kernel learning estimation strategy, the road adhesion coefficient estimation under the complex motion excitation working condition including the weak motion excitation working condition is realized, and the uncertainty evaluation of the road adhesion coefficient estimation is realized; (2) The input parameter type of the random variation deep kernel learning network is determined according to the difference of the state in the time domain and the frequency domain of the non-steady-state maneuvering working condition and the steady-state maneuvering working condition; (3) Based on the linear path tracking method of the affine force input model, a learning type random model predictive control path tracking method is designed to solve the influence of the road adhesion coefficient estimation uncertainty in the prediction model and the chance constraint design, aiming at the influence of the uncertain road on the vehicle dynamics model and the probability state constraint; (4) By analyzing the wheel-ground interaction relationship of the distributed ultra-flat motion control platform, an integrated model of the wheel, the whole control platform and the electric drive module is constructed, a method of integrating the road adhesion coefficient estimation and its uncertainty into the path tracking control law is designed, the high-speed stability influence mechanism of the motion platform control is analyzed and a robust stability control strategy is constructed, which can effectively improve the robustness and tracking accuracy of the distributed ultra-flat platform motion controller. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Figure 1 The flowchart of the motion control method for the distributed ultra-flat motion control platform provided by the present application is shown. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0018] Please refer to Figure 1 The motion control method for the distributed ultra-flat motion control platform provided by the present embodiment comprises the following steps: Step S1, based on the dynamics model of the distributed ultra-flat motion control platform and the dynamics model of the wheel, the expression function of the road adhesion coefficient is derived, and the state parameters related to the road adhesion coefficient are determined under the condition that there is only one real solution in the expression function.

[0019] The dynamic model of the seven-degree-of-freedom distributed ultra-flat motion control platform is represented as: wherein, denotes mass; denotes longitudinal velocity; denotes longitudinal acceleration; , denote longitudinal force of the left front wheel and the right front wheel respectively; denotes front wheel steering angle; , denote lateral force of the left front wheel and the right front wheel respectively; , denote longitudinal force of the left rear wheel and the right rear wheel respectively; denotes lateral velocity, denotes yaw angular velocity; , , denote air resistance, wheel rolling resistance and slope resistance respectively; denotes lateral acceleration; , denote lateral force of the left rear wheel and the right rear wheel respectively; denotes yaw angular acceleration; , denote distance from the front axle and the rear axle to the center of mass respectively; denotes wheel track; denotes yaw moment.

[0020] According to the seven-degree-of-freedom distributed ultra-flat motion control platform model, it can be obtained that: wherein, , denote longitudinal wheel stiffness and lateral wheel stiffness respectively; , , , denote slip ratio of the left front wheel, the right front wheel, the left rear wheel and the right rear wheel respectively; denotes front wheel steering angle; , denote wheel side slip angle of the left front wheel and the right front wheel respectively; , , , denote state judgment parameter of the left front wheel, the right front wheel, the left rear wheel and the right rear wheel respectively, denotes that the wheel is in linear region; represents that the wheel is in a nonlinear region, represents an intermediate parameter, and is expressed as: wherein, represents a road adhesion coefficient; represents a vertical force of the wheel; represents a slip ratio of the wheel; represents a side slip angle of the wheel; The above formula can be expressed as a quadratic function related to the road adhesion coefficient , and is expressed as: The derivation purpose of the quadratic function is mainly to establish the relationship between the state parameters and the road adhesion coefficient , and in the subsequent learning process of the neural network model, only the relative change relationship between the state parameters and the road adhesion coefficient is usually concerned, and is not dependent on the absolute value, therefore, the present application adopts a way of removing the constant term in the quadratic function to simplify the quadratic function , and will not affect the feature learning and training effect of the neural network model, and this kind of simplification can reduce the data processing amount of the neural network model, improve the training efficiency, and the simplified quadratic function is expressed as: The solution of the quadratic function has two cases: one is no real root, and the other is only one real root. Since the distributed ultra-flat motion control platform is an information physical system, the road adhesion coefficient and all the state parameters have actual physical meaning, therefore there must be only one real solution. Based on this characteristic, the road adhesion coefficient can be indirectly determined by analyzing the slope of the quadratic function curve.

[0021] The distributed ultra-flat motion control platform includes a steady-state maneuvering working condition and a non-steady-state maneuvering working condition, wherein the steady-state maneuvering working condition is a longitudinal driving working condition of the longitudinal acceleration of the distributed ultra-flat motion control platform; and the non-steady-state driving maneuvering working condition is all longitudinal driving working conditions except the steady-state maneuvering working condition.

[0022] ​The road adhesion coefficient The parameters closely related to the road adhesion coefficient are the wheel slip ratio , the longitudinal wheel stiffness , the longitudinal acceleration , the wheel vertical force , and the wheel friction force . .

[0023] Under steady-state maneuvering conditions, all wheels are in the linear working region, and the saturation value of the wheel friction force is positively correlated with the slope of the normalized wheel force-slip ratio curve, which reflects the dynamic response characteristics of the wheel under different road adhesion conditions. Different road adhesion conditions change the slope of the normalized wheel force-slip ratio curve, and when the wheel slip ratio is in the linear region, the wheel force and the slip ratio are in a linear relationship. Therefore, by analyzing the slope of the linear section of the normalized wheel force-slip ratio curve, the characteristic information related to the road adhesion coefficient can be extracted.

[0024] The wheel dynamics model is represented as: wherein represents the wheel rotational inertia, represents the wheel rotational angular velocity; represents the wheel rotational angular acceleration; represents the wheel driving torque output by the in-wheel motor drive system; represents the wheel radius.

[0025] The wheel driving torque is composed of a high-frequency signal and a low-frequency signal, and is represented as: wherein , represent the low-frequency signal and the high-frequency signal, respectively; represents the amplitude of the signal; represents the frequency of the signal, represents time.

[0026] Therefore, the wheel dynamics model under driving mode is represented as: wherein represents the amplitude of the torque; represents ; represents the imaginary unit; represents the relaxation coefficient; represents the tire longitudinal stiffness; The wheel dynamics model in the braking mode can be represented as: It is found by analyzing the wheel dynamics model in the driving mode and the braking mode that there is a different mapping relationship between each state parameter in the amplitude-frequency function and the slope of the linear part of the normalized wheel force-slip ratio curve, and therefore the time-varying characteristic parameters involved in the amplitude-frequency function are the state parameters related to the road adhesion coefficient , including the longitudinal velocity , the wheel rotational angular velocity , the wheel vertical force , and .

[0027] Based on the above analysis, the road adhesion coefficient can be represented as: In the formula, denotes the mapping function between the parameter and the adhesion coefficient .

[0028] Step S2, a space-time convolution network is constructed, a plurality of state parameters are combined to form a group of data, the road adhesion coefficient under the group of data is taken as a label, a plurality of groups of data and corresponding labels are combined to form a data set, the data set is input into the space-time convolution network for learning, high-dimensional features are extracted from time domain signals and time-frequency domain signals respectively, and the output high-dimensional feature vector is taken as an input vector of a sparse variational Gaussian process. The mapping relationship between the state parameters and the road adhesion coefficient is learned by using a sparse variational strategy, after the training is completed, the state parameter combination is input into the space-time convolution network, and an estimated value of the road adhesion coefficient and an estimated variance are output.

[0029] The space-time convolution network includes a non-steady-state maneuvering space-time convolution sub-network and a steady-state maneuvering space-time convolution sub-network, the non-steady-state maneuvering space-time convolution sub-network takes the state parameters and the label under the non-steady-state maneuvering condition as input, and the steady-state maneuvering space-time convolution sub-network takes the state parameters and the label under the steady-state maneuvering condition as input.

[0030] The non-steady-state maneuvering space-time convolution sub-network comprises three convolution units connected in sequence, each convolution unit comprises a convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence, wherein the convolution kernel sizes of the three convolution layers are 1x4x8, 1x4x16 and 1x4x32 in sequence; the steady-state maneuvering space-time convolution sub-network comprises a time domain feature extraction branch and a time-frequency domain feature extraction branch arranged in parallel, wherein the time domain feature extraction branch is used for directly processing the training data segment, extracting the time domain feature through convolution operation, the time domain feature extraction branch comprises three convolution units connected in sequence, each convolution unit comprises a convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence, wherein the convolution kernel sizes of the three convolution layers are 1x4x8, 1x4x16 and 1x4x32 in sequence; the time-frequency domain feature extraction branch obtains two-dimensional time-frequency domain features through continuous wavelet transform, and the time-frequency domain feature extraction branch comprises four convolution units connected in sequence, each convolution unit comprises a convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence, wherein the convolution kernel sizes of the three convolution layers are 4x2x4, 3x4x8, 2x2x16 and 2x2x32 in sequence.

[0031] The time domain feature and the time-frequency domain feature are fused together in the Flatten layer to form the input of the sparse variational Gaussian process, and the sparse variational Gaussian process specifically comprises the following steps: Step S21, representative sample points are selected from the input vector as inducing points, and the function values at the inducing points are regarded as a group of inducing variables to represent the key information of the entire input vector; Step S22, a variational distribution is introduced in the variational inference to approximately replace the real posterior distribution; Step S23, the evidence lower bound of the marginal log-likelihood is maximized as the optimization target, and the standard stochastic gradient descent method is used for optimization to make the variational distribution as close as possible to the real posterior distribution.

[0032] After the training is completed, the road adhesion coefficient estimation value and the estimation uncertainty, i.e., the estimation variance , the estimation value and the estimation variance are output. The design of an adaptive path tracking controller for a distributed ultra-flat motion control platform.

[0033] Step S3, considering the uncertainty in the adhesion coefficient estimation process, a controller for adaptive path tracking is designed to regulate the motion state of the distributed ultra-flat motion control platform.

[0034] The lateral dynamics equation of the distributed ultra-flat motion control platform is: wherein, denotes the total mass of the distributed ultra-flat motion control platform; denotes the longitudinal velocity; denotes the lateral velocity; denotes the first derivative of the side slip angle of the center of mass; denotes the yaw rate; , denote the wheel lateral forces of the front and rear wheels, respectively; denotes the moment of inertia of the whole vehicle around the z-axis; , denote the distances from the front and rear axles to the center of mass, respectively.

[0035] The wheel model is represented as: wherein, , denote the front and rear wheel lateral affine forces of the last control cycle, respectively; , denote the front and rear wheel equivalent cornering stiffness, respectively, denotes the road adhesion coefficient estimate; , denote the front and rear wheel side slip angles at the current time instant, respectively; , denote the front and rear wheel side slip angles of the last control cycle, respectively.

[0036] Substituting the wheel model into the lateral dynamics equation, the lateral dynamics equation is further represented as: ; wherein, , , , , , , , all denote intermediate calculation parameters; wherein: ; ; ; ; ; ; ; .

[0037] The differential equation of the path tracking model is represented as: ; wherein, denotes the state of the distributed ultra-flat motion control platform, , denotes the lateral error; denotes the heading angle error; denotes the transpose matrix; denotes the control input of the distributed ultra-flat motion control platform, , denotes the system disturbance; , , all denote weight matrices, wherein: ; ; ; wherein, path curvature.

[0038] The differential equations of the path following model are discretized and represented as: ; wherein, , denote the state of the distributed ultra-flat motion control platform at time and , respectively; denotes the control input of the distributed ultra-flat motion control platform at time ; denotes the system disturbance at time ; , , denote the weight matrices at time ; wherein: wherein, denotes the identity matrix; denotes the sampling time.

[0039] The motion of the distributed ultra-flat motion control platform needs to satisfy the following constraints: (1) The lateral stability constraint is represented as: ; wherein, ; , , , , respectively represent the maximum value of lateral error, the maximum value of heading angle error, the maximum value of yaw rate, the maximum value of side slip angle of the center of mass.

[0040] (2) The chance constraint is expressed as: ; In the formula, represents a probability function; represents a probability specific value; (3) The physical constraint of the actuator is expressed as: ; In the formula, , respectively represent the minimum limit value and the maximum limit value of the front wheel steering angle ; , respectively represent the minimum limit value and the maximum limit value of the front wheel steering angle increment .

[0041] Based on the above path tracking model and the constraint, a control prediction model is constructed and expressed as: ; s.t ; ; ; ; In the formula, , , represent a weight matrix; represents a state matrix; represents a reference state matrix; represents a control amount increment; represents a terminal state matrix; k represents a time node; and N represents the total number of time nodes.

[0042] The constructed control prediction model is solved to obtain an optimal control sequence, and the first control amount of the optimal control sequence is taken as a current expected front wheel steering angle and sent to a lower steering actuator, so that the path tracking control of the distributed ultra-flat motion control platform can be realized.

[0043] The embodiments of the application are described above in combination with the drawings, but the application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative but not restrictive. Those skilled in the art can make many forms under the inspiration of the application without departing from the purpose of the application and the scope protected by the claims, and all of them belong to the protection of the application.

Claims

1. A motion control method for a distributed ultra-flattened motion control platform, characterized in that, Comprising the following steps: Step S1, based on the dynamics model of the distributed ultra-flat motion control platform and the dynamics model of the wheel, the expression function of the road adhesion coefficient is derived, and the state parameters related to the road adhesion coefficient are determined under the condition that there is only one real solution to the expression function; Step S2, constructing a space-time convolution network, forming a group of data with different combinations of a plurality of state parameters, taking the road adhesion coefficient under the group of data as a label, combining a plurality of groups of data and corresponding labels to form a data set, inputting the data set into the space-time convolution network for learning, extracting high-dimensional features from time domain signals and time-frequency domain signals respectively, and taking the output high-dimensional feature vector as the input vector of the sparse variational Gaussian process, learning the mapping relationship between the state parameters and the road adhesion coefficient by using the sparse variational strategy, and after training, inputting the state parameter combination into the space-time convolution network to output the road adhesion coefficient estimate value and the estimation variance; Step S3, based on the road adhesion coefficient estimate value and the estimation variance, designing a controller for adaptive path tracking to regulate the motion state of the distributed ultra-flat motion control platform.

2. The motion control method for the distributed ultra-flattened motion control platform according to claim 1, wherein, The dynamics model of the seven-degree-of-freedom distributed ultra-flat motion control platform is represented as: wherein denotes the mass; denotes the longitudinal velocity; denotes the longitudinal acceleration; , denote the longitudinal forces of the left and right front wheels, respectively; denotes the front wheel steering angle; , denote the lateral forces of the left and right front wheels, respectively; , denote the longitudinal forces of the left and right rear wheels, respectively; denotes the lateral velocity, denotes the yaw angular velocity; , , denote the air resistance, the wheel rolling resistance and the slope resistance, respectively; denotes the lateral acceleration; , denote the lateral forces of the left and right rear wheels, respectively; denotes the yaw angular acceleration; , denote the distances of the front and rear axles to the center of mass, respectively; denotes the wheel track; denotes the yaw moment.

3. The motion control method for the distributed ultra-flattened motion control platform according to claim 1, wherein, The wheel dynamics model is represented as: In the formula, denotes the wheel rotational inertia, denotes the wheel rotational angular velocity; denotes the wheel rotational angular acceleration; denotes the wheel drive torque output by the wheel hub motor drive system; denotes the wheel radius; Wheel drive torque consisting of a high frequency signal and a low frequency signal, expressed as: wherein, , respectively represent low frequency signals and high frequency signals; represents the amplitude of the signal; represents the frequency of the signal, represents time; Then, the wheel dynamics model in the driving mode is represented as: In the formula, denotes the amplitude of the torque; denotes ; denotes the imaginary unit; denotes the relaxation coefficient; denotes the tire longitudinal stiffness; The wheel dynamics model in the braking mode can be represented as: 。 4. The motion control method for the distributed ultra-flattened motion control platform according to claim 3, wherein, The parameters closely related to the road adhesion coefficient include: mass , wheel slip , longitudinal wheel stiffness , longitudinal acceleration , wheel vertical force , , wherein Fw denotes the wheel longitudinal force; Steady state maneuvering conditions and road adhesion coefficient Parameters of close relevance include: longitudinal velocity , wheel angular velocity , wheel vertical force and .

5. The motion control method for distributed ultra-flattened motion control platform according to claim 1, wherein, The space-time convolution network includes a non-steady-state maneuvering space-time convolution sub-network and a steady-state maneuvering space-time convolution sub-network, the non-steady-state maneuvering space-time convolution sub-network takes the state parameters and labels under the non-steady-state maneuvering condition as input, and the steady-state maneuvering space-time convolution sub-network takes the state parameters and labels under the steady-state maneuvering condition as input.

6. The motion control method for the distributed ultra-flattened motion control platform according to claim 5, wherein, The non-steady-state maneuvering space-time convolution sub-network includes three convolution units connected in sequence, each convolution unit includes a convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence, wherein the convolution kernel sizes of the three convolution layers are 1x4x8, 1x4x16 and 1x4x32 respectively; the steady-state maneuvering space-time convolution sub-network includes a time domain feature extraction branch and a time-frequency domain feature extraction branch arranged in parallel, wherein the time domain feature extraction branch is used to directly process the training data segment and extract the time domain feature through convolution operation, the time domain feature extraction branch includes three convolution units connected in sequence, each convolution unit includes a convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence, wherein the convolution kernel sizes of the three convolution layers are 1x4x8, 1x4x16 and 1x4x32 respectively; the time-frequency domain feature extraction branch obtains two-dimensional time-frequency domain features through continuous wavelet transform, the time-frequency domain feature extraction branch includes four convolution units connected in sequence, each convolution unit includes a convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence, wherein the convolution kernel sizes of the three convolution layers are 4x2x4, 3x4x8, 2x2x16 and 2x2x32 respectively; the time domain feature and the time-frequency domain feature are fused together through the Flatten layer to form the input of the sparse variational Gaussian process.

7. The motion control method and system for distributed ultra-flattened motion control platform according to claim 1, wherein, The sparse variational Gaussian process specifically comprises the following steps: Step S21, selecting a representative sample point from the input vector as an inducing point, and regarding the function value at the inducing point as a set of inducing variables to induce the key information of the entire input vector; Step S22, introducing a variational distribution to approximate the real posterior distribution in the variational inference; Step S23, taking the evidence lower bound maximization of the marginal log-likelihood as the optimization goal, and using the standard stochastic gradient descent method to optimize the variational distribution to be as close as possible to the real posterior distribution.

8. The motion control method and system for distributed ultra-flattened motion control platform according to claim 1, wherein, The lateral dynamic equation of the distributed ultra-flat motion control platform is: wherein denotes the total mass of the distributed ultra-flat motion control platform; denotes the longitudinal velocity; denotes the lateral velocity; denotes the first derivative of the side slip angle of the center of mass; denotes the yaw rate; , denote the wheel lateral forces of the front and rear wheels, respectively; denotes the moment of inertia of the entire vehicle about the z-axis; , denote the distances of the front and rear axles to the center of mass, respectively; The wheel model is represented as: wherein, , respectively represent the front and rear wheel lateral affine forces of the last control period; , respectively represent the front and rear wheel equivalent cornering stiffnesses, represents the road adhesion coefficient estimate; , respectively represent the front and rear wheel side slip angles at the current time instant; , respectively represent the front and rear wheel side slip angles of the last control period; The lateral dynamic equation is further represented by substituting the wheel model into the lateral dynamic equation: ； wherein , , , , , , , all represent intermediate calculation parameters; Wherein: ； ； ； ； ； ； ； ； The differential equation of the path tracking model is represented as: ； wherein denotes the state of the distributed ultra-flat motion control platform, , denotes the lateral error; denotes the heading angle error; denotes the transpose matrix; denotes the control input of the distributed ultra-flat motion control platform, , denotes the system disturbance; , , all denote weight matrices, wherein: ； ； ； In the formulae, Path curvature; The differential equation of the path tracking model is discretized and represented as: ； In the formula, , respectively represent and the state of the distributed ultra-flat motion control platform at the moment; represent the control input of the distributed ultra-flat motion control platform at the moment; represent the system disturbance at the moment; , , represent the weight matrix at the moment; wherein: In the formula, denotes the unit matrix; denotes the sampling time.

9. The motion control method and system for distributed ultra-flattened motion control platform according to claim 1, wherein, The control prediction model of the adaptive path tracking controller is represented as: ； s.t ; ； ； ； In the formula, , , denotes a weight matrix; denotes a state matrix; denotes a reference state matrix; denotes a control amount increment; denotes a terminal state matrix; k denotes a time node; N denotes a total number of time nodes; ; , , , , respectively denote a lateral error maximum value, a heading angle error maximum value, a yaw rate maximum value, and a center of mass side slip angle maximum value; denotes a probability function; denotes a probability specific value; , respectively denote a minimum limit value and a maximum limit value of a front wheel steering angle ; , respectively denote a minimum limit value and a maximum limit value of a front wheel steering angle increment ; Solving the constructed control prediction model obtains the optimal control sequence, and taking the first control quantity of the optimal control sequence as the current expected front wheel steering angle and sending it to the bottom steering actuator can realize the path tracking control of the distributed ultra-flat motion control platform.

10. A system for performing the motion control method of any one of claims 1-9, wherein, Comprise: The parameter determination module is used to derive an expression function of the road adhesion coefficient based on the dynamic model of the distributed ultra-flat motion control platform and the dynamic model of the wheel, and determine the state parameters related to the road adhesion coefficient under the condition that there is only one real solution to the expression function; The spatio-temporal convolution network is used to form a set of data with different combinations of multiple state parameters, take the road adhesion coefficient under the set of data as the label, combine multiple sets of data and corresponding labels to form a data set, input the data set into the spatio-temporal convolution network for learning, extract high-dimensional features from time domain signals and time-frequency domain signals respectively, and take the output high-dimensional feature vector as the input vector of the sparse variational Gaussian process. The mapping relationship between the state parameters and the road adhesion coefficient is learned by using the sparse variational strategy. After the training is completed, the state parameters are combined to input into the spatio-temporal convolution network, and the road adhesion coefficient estimation value and estimation variance are outputted; The controller design module is used to design the adaptive path tracking controller based on the road adhesion coefficient estimation value and the estimation variance, and to regulate and control the motion state of the distributed ultra-flat motion control platform.

Images