Computer-implemented method for designing state regulator with stochastic optimization

The randomly optimized state regulator design implemented by computer solves the problem of uncertainty and interference in the regulation system, shortens development time, reduces testing costs, and improves robustness and performance, suitable for the horizontal and vertical directions of the vehicle adjust.

CN119024825APending Publication Date: 2024-11-26ROBERT BOSCH GMBH
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410393110.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-04-03
Filing Date
2024-04-02
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Existing regulation systems are difficult to effectively consider external interference and parameter uncertainty in highly automated and autonomous driving, resulting in unknown uncertainty in the development stage and high costs for subsequent testing.

Method used

A computer-implemented method is used to design a state regulator with random optimization, and solve the feedback matrix to reduce the impact of uncertainty and interference by receiving the state space model, describing the feedback matrix, and initial state calculation cost function.

Benefits of technology

In the design stage, uncertainty is considered, adjustment function time to market, verification workload is reduced, robustness and performance trade-offs can be achieved, and confidence intervals of closed-loop adjustment loops can be derived to check the robustness and optimality of existing systems. Adjust deviation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119024825A_ABST
    Figure CN119024825A_ABST
Patent Text Reader

Abstract

A general aspect of the present disclosure relates to a computer-implemented method for designing a state regulator with stochastic optimization. The method comprises receiving a state space model for describing a system to be adjusted, the state space model comprising a system matrix, a state vector comprising one or more state variables, an input matrix and an input parameter vector, wherein the input parameter vector is based on the state vector and a feedback matrix describing the state regulator, and the one or more state variables are described based on one or more probability distributions. The method further includes describing an optimization problem including a cost function calculated using at least the system matrix, the feedback matrix, an initial state, and the input matrix, and solving the optimization problem to determine an entry of the feedback matrix.
Need to check novelty before this filing date? Find Prior Art

Description

Background Art

[0001] Highly automated or autonomous systems are gaining more and more attention, for example in the robotics and automotive industries. Control systems in particular are becoming increasingly important in the operation of autonomous or highly automated systems. The lateral guidance of the vehicle plays a central role in highly automated and autonomous driving. The task of the lateral guidance is to keep the lateral distance between the vehicle and the predefined route or the edge of the lane and / or the edge of the road constant. Although many methods for control have been proposed, most of them are based on nominal models, i.e., they do not take into account uncertainties such as external disturbances, parameter uncertainties or model errors.

[0002] The uncertainty in the initial state of the control system and the uncertainty in the variance in the control loop feedback resulting from said initial state are consequences of these uncertainties. The uncertainty in the variance in the control loop feedback is unknown or difficult to quantify during the development phase and parameterization of the controller. Therefore, the controller must be intensively tested and adapted to the uncertainties later in the application phase. This is both time-consuming and cost-intensive. In particular, due to the ever-shorter development cycles and the increasing software proportion, long product delivery times lead to market-specific disadvantages. For these reasons, new controller design technologies are needed to take the above uncertainties into account and reduce the workload in the subsequent testing phase. Summary of the invention

[0003] A first general aspect of the present disclosure relates to a computer-implemented method for designing a state regulator with stochastic optimization. The method includes receiving a state-space model for describing a system to be regulated, wherein the state-space model includes a system matrix, a state vector including one or more state variables, an input matrix, and an input parameter vector, wherein the input parameter vector is based on the state vector and a feedback matrix describing the state regulator, and describes the one or more state variables based on one or more probability distributions. The method also includes describing an optimization problem including a cost function calculated using at least the system matrix, the feedback matrix, an initial state, and the input matrix, and solving the optimization problem to determine entries of the feedback matrix.

[0004] A second general aspect of the present disclosure relates to a computer system designed to perform a computer-implemented method for designing a state regulator with stochastic optimization according to the first general aspect (or its embodiments).

[0005] A third general aspect of the present disclosure relates to a computer program comprising instructions which, when executed by a computer system, cause the computer system to perform a computer-implemented method for designing a state regulator with stochastic optimization according to the first general aspect (or its embodiments).

[0006] A fourth general aspect of the present disclosure relates to a computer-readable medium or signal storing and / or containing a computer program according to the third general aspect (or an embodiment thereof).

[0007] The method according to the first general aspect (or its embodiment) proposed in the present disclosure can be used to provide a computer-implemented method for designing a state regulator with stochastic optimization. One advantage may be to shorten the time to market of highly automated regulation functions, because the relevant uncertainties can be taken into account at the design stage. The workload that must be spent in the verification stage / confirmation stage can be reduced. In addition, a trade-off can be achieved between the robustness of the regulation system for parameter uncertainty and interference and the performance of the regulation system. Another advantage may be that a confidence interval of a closed-loop regulation loop can be derived so that a random judgment can be made on the performance. The technology of the present disclosure can also be beneficial to check the robustness of an existing regulation system or quantify its deviation from the optimal regulation. In addition, the technology of the present disclosure is not limited to the lateral guidance of the vehicle, but can be beneficial to various adjustments. In addition to the lateral guidance of the vehicle, longitudinal regulation can also benefit from the technology of the present disclosure. The present technology is also beneficial to the scheme of simultaneous longitudinal and lateral regulation.

[0008] Certain terms are used in this disclosure as follows:

[0009] A "state regulator" may include an algorithm, i.e., a calculation rule, that feeds back a state variable (i.e., an internal state of a controlled system) to an input parameter. A state regulator may include parameters that may weight the state parameter. In an example, a state regulator may be executed on a computer system. For example, a state regulator may include a hardware module having an input and an output or may be part of a hardware module having an input and an output.

[0010] A "vehicle" may be any device for transporting passengers and / or cargo. A vehicle may be a motor vehicle (e.g. a passenger car or a truck) or a rail vehicle. A vehicle may also be a motorized two- or three-wheeled vehicle. However, floating and flying devices may also be vehicles. A vehicle may operate at least partially autonomously or with assistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1-A to Figure 1-B A computer-implemented method for designing a state regulator with stochastic optimization is schematically shown.

[0012] Figure 2 An exemplary regulation loop according to one or more embodiments of the present disclosure is schematically illustrated.

[0013] Figure 3-A to Figure 3-B Exemplary system responses of a state regulator designed using the method of the present disclosure and system responses of a state regulator designed in a conventional manner are shown. DETAILED DESCRIPTION

[0014] First, refer to Figure 1-A , Figure 1-B and Figure 2 Discuss the technology of the present disclosure. Figure 3-A to Figure 3-D Possible consequences and advantages resulting from the method disclosed herein for designing a state regulator are discussed.

[0015] 1 is a flow chart showing possible method steps of a computer-implemented method 100 for designing a state regulator 20 with stochastic optimization. The computer-implemented method 100 for designing a state regulator 20 with stochastic optimization includes receiving 110 a state space model for describing the system 10 to be regulated. The state space model includes a system matrix A, a state vector x(t,ξ) containing one or more state variables x1(t,ξ), x2(t,ξ), x3(t,ξ),…, an input matrix B and an input parameter vector u(t,ξ). The input parameter vector u(t,ξ) is based on the state vector x(t,ξ) and a feedback matrix K describing the state regulator. The one or more state variables x1(t,ξ), x2(t,ξ), x3(t,ξ),… are described based on one or more probability distributions. The method 100 also includes describing 120 an optimization problem including a cost function J calculated using at least the system matrix A, the feedback matrix K, the initial state x0 and the input matrix B, and solving 130 the optimization problem to determine the entries of the feedback matrix K. The system matrix A may include one or more coefficients, which may be described based on the one or more probability distributions. In one example, the state space model may be represented by the equation Description, where t>0, x(0,ξ)=x0(ξ). In one example, the state vector x(t,ξ) can be determined by measurement. In one example, the state vector x(t,ξ) can be reconstructed with the aid of an observer or an estimator. In the present disclosure, parameter uncertainty is expressed by a random variable ξ. In one example, the regulation law describes the influence on the state vector x(t,ξ) with the aid of a feedback matrix K(t), for example, with the aid of weighting, so as to input the result of the influence as an input parameter vector u(t,ξ) to the system to be regulated. In one example, the regulation law can be expressed with the aid of the equation u(t,ξ)=K(t)x(t,ξ), thereby forming the input parameter vector u(t,ξ). In one example, the feedback matrix K can be time-varying, time-invariant or constant. In one example, the feedback matrix K can be deterministic (optionally depending on the cost function J). In one example, it can be considered that the feedback gain disappears for all j≥1 (i.e., K ij =0) related unmeasurable state (x i(t,ξ)). In this disclosure, i represents the row index of the matrix and j represents the column index. This is a finite definition set .

[0016] In one example, description 120 includes determining 121 one or more expected values ​​and one or more variances of one or more state variables x1(t,ξ), x2(t,ξ), x3(t,ξ) of a state vector x(t,ξ), and wherein the cost function J is also formed by a first contribution and a second contribution, wherein the first contribution corresponds to one or more expected values ​​of one or more state variables x1(t,ξ), x2(t,ξ), x3(t,ξ) of the state vector x(t,ξ), and the second contribution corresponds to one or more variances of one or more state variables x1(t,ξ), x2(t,ξ), x3(t,ξ),… of the state vector x(t,ξ). For example, the first contribution may include a first part of the cost function J E And the second contribution may include the second partial cost function J var In addition, the cost function J can be obtained by the first part of the cost function J E and the second cost function J var In one example, an optimization problem is described 120 for achieving optimal regulation.

[0017] In one example, the description 120 may also include using the first weight matrix W E The first contribution is weighted 122 and the second weight matrix W is used to weight the first contribution var The second contribution is weighted 123. In one example, the cost function J may be given by the equation J=J E (E(x),E(u);W E )+J var (var(x),var(u);W var ). In one example, the optimization problem may include minimizing the cost function J with respect to the entries of the feedback matrix K. The associated optimization problem may then be expressed as follows:

[0018]

[0019] in

[0020] u=Kx (iii)

[0021]

[0022] In one example, the weighting 122 of the first contribution and the weighting 123 of the second contribution may be performed by means of a third weight matrix Q and a fourth weight matrix R. In one example, the feedback matrix K may describe a linear quadratic regulator such that J E Can be and similarly Jvar can be expressed by Here we can calculate Q E =W E Q,R E =W E R,Q var =W var Q, R var =W var R. In one example, the third weight matrix Q and / or the fourth weight matrix R may be positive semidefinite matrices, respectively. In one example, the first weight matrix W E and / or a second weight matrix W var It can be a scalar. For example, the first weight matrix W E and the second weight matrix W var can depend on each other. For example, the first weight matrix W E can be equal to the scalar w, and for the second weight matrix W can be applied var =1-w. For example, by selecting the first weight matrix W E and the second weight matrix W var and / or the ratio of the first weight matrix to the second weight matrix to set the influence of the mean or variance on the cost function J.

[0023] As described above, describing 120 the optimization problem may include determining 121 one or more expected values ​​and one or more variances of one or more state variables x1(t,ξ), x2(t,ξ), x3(t,ξ),… of the state vector x(t,ξ). In one example, the calculation of the one or more expected values ​​and one or more variances may be performed with the aid of sampling, such as based on Monte Carlo simulation or Latin hypercube sampling methods. An exemplary calculation of one or more expected values ​​and one or more variances with the aid of a polynomial chaos expansion (PCE) shall be discussed below. In one example, describing 120 may also include performing one or more series expansions based on one or more probability distributions to describe one or more state variables x1(t,ξ), x2(t,ξ), x3(t,ξ). In one example, the one or more probability distributions may be represented by one or more polynomial bases. For example, the state x i The series expansion of (t,ξ) can be obtained by Here, the term φ(ξ) may include one or more polynomial bases. In one example, It can be generally applied, where the coefficient X i,j It can be calculated, for example, by means of an intrusive solution method or a non-intrusive solution method, φ j (ξ) represents one or more polynomial bases. The objective function x i (t,ξ) can be simulated by one or more polynomial bases using the above calculation rules. In one example, each polynomial base can be orthogonal to the probability distribution it represents. In one example, the one or more polynomial bases can be at least one of Hermite polynomials, Legendre polynomials, Jacobi polynomials, and / or generalized Laguerre polynomials. In one example, the state x i The expected value of can be obtained by E(x i )(t)=m T X i (t) is calculated. Here m is an auxiliary vector derived from the polynomial basis φ(ξ). In one example, the state x i The variance of to calculate, where f is also an auxiliary matrix directly derived from the polynomial basis φ(ξ). The calculation rules based on polynomial chaos expansion described in this article can be used to calculate one or more expected values ​​and one or more variances of one or more state variables x1(t,ξ),x2(t,ξ),x3(t,ξ),… of the state vector x(t,ξ).

[0024] In one example, the first contribution and / or the second contribution each include the addition Q+K T In one example, the cost function J can be simplified by the equation where X0=X(0) can express the initial state of the system. In one example, P can be expressed by Calculate, where Depend on Derive and and In one example, the polynomial chaos expansion (PCE) can be performed in the preprocessing step, which can help reduce the computational workload when solving the optimization problem 130. In the first weight matrix W E =w and the second weight matrix W var =1-w, can be calculated using the equation Results of polynomial chaos expansion (PCE).

[0025] In one example, the optimization problem may also include optimality conditions, stability conditions, and / or a definition set At least one of the restrictions on the possible entries of the feedback matrix K. In one example, for the optimization problem it may apply:

[0026]

[0027] in

[0028]

[0029]

[0030] In one example, equation (ii) is the optimality condition, equation (iii) is the stability condition, and equation (iv) is the definition set Restrictions on the possible entries of the feedback matrix K. The stability conditions here can be based on the Lyapunov equations.

[0031] In one example, the optimization problem can be solved 130 by means of a global search method, a simplex method or a gradient method. If, for example, a gradient method is implemented, the initial value K required for this ini It may be determined, for example, by a linear quadratic regulator that does not take uncertainty into account, expert opinion, or a global search. In one example, the quality of the solution may be arbitrary, ie, it does not have to reach a certain level.

[0032] In one example, the system 10 to be adjusted can be designed to be arranged in a vehicle and / or designed to control vehicle functions (especially driving functions). For example, the vehicle function can be a function for autonomous and / or assisted driving. In some examples, the state regulator 20 can be designed to be executed on a computer system of a vehicle (e.g., autonomous driving, highly automated driving, or assisted driving vehicle). For example, the computer system can be implemented locally in the vehicle or (at least partially) in a back end connected to the vehicle communication ground. For example, the computer system can include a control device on which the state regulator 20 can be executed. In some examples, the vehicle can include a computer system with a communication interface that enables communication with the back end. For example, the state regulator 20 can be executed in the back end. In one example, the system 10 to be adjusted can be a system for lateral guidance and / or longitudinal guidance of the vehicle. The uncertainty described by means of one or more probability distributions can include, for example, the load-dependent yaw inertia and mass of the vehicle, the distance between the center of gravity of the vehicle and the axle, or the road friction (or a combination thereof) that affects stiffness. In one example, one or more random variables x1(t,ξ), x2(t,ξ), x3(t,ξ), ... may be variables for at least one of a steering angle, an azimuth angle, a yaw rate, a sideslip angle, and / or a lateral error. In one example, an entry of the input parameter vector u(t,ξ) may be a steering speed.

[0033] The present disclosure also relates to a method of controlling a vehicle using a state regulator designed by means of the method of the present disclosure. In some examples herein, the state regulator 20 and / or the system to be regulated 10 may be designed as described above.

[0034] In other examples, the system 10 to be regulated may be arranged in the robot and / or designed to control robot functions (in particular control the movement functions of the robot). For example, the system to be regulated may be a system for lateral guidance and / or longitudinal guidance of the robot. In some examples, the state regulator 20 may be executed on a computer system of the robot. For example, the computer system may be implemented locally in the robot or (at least partially) in a backend communicatively connected to the robot.

[0035] The present disclosure also relates to a method of controlling a robot using a state regulator designed by means of the method of the present disclosure.In some examples herein, the state regulator 20 and / or the system to be regulated 10 may be designed as described above.

[0036] The following should be assisted by Figure 3-A to Figure 3-D To explain the advantages of the technology of the present disclosure over the prior art. Figure 3-A to Figure 3-D1 shows the regulator response of a controlled lane change of a vehicle. Here, response 60 is the regulator response of a conventional linear quadratic regulator designed according to the techniques of the prior art, while response 50 is the regulator response of a state regulator designed with the aid of the method of the present disclosure. As the value w decreases, the state regulator becomes more conservative, but has a smaller variance. At w=10 -3 When , a rise time comparable to that of a conventional regulator can be achieved, but with less variance and without the excess.

[0037] A computer system is also disclosed, which is designed to perform a computer-implemented method 100 for designing a state regulator 20 with stochastic optimization. The computer system may include at least one processor and / or at least one working memory. The computer system may also include a (non-volatile) memory.

[0038] Furthermore, a computer program is disclosed, comprising instructions which, when executed by a computer system, cause the computer system to perform a computer-implemented method 100 for designing a state regulator 20 with stochastic optimization. The computer program may, for example, be in an interpretable or compiled form. The computer program may be loaded (even in parts) into the RAM of a computer for execution, for example as a sequence of bits or bytes.

[0039] A computer readable medium or signal storing and / or containing a computer program or at least a part thereof is also disclosed. The medium may include, for example, one of RAM, ROM, EPROM, HDD, SDD, etc., on which / in which the signal is stored.

Claims

1. A computer-implemented method (100) for designing a state regulator (20) with stochastic optimization, wherein the method (100) comprises the following steps: - receiving (110) a state space model describing the system (10) to be regulated, wherein the state space model comprises a system matrix (A), a state vector (x(t,ξ)) containing one or more state variables (x1(t,ξ), x2(t,ξ), x3(t,ξ), ...), an input matrix (B) and an input parameter vector (u(t,ξ)), wherein the input parameter vector (u(t,ξ)) is based on the state vector (x(t,ξ)) and a feedback matrix (K) describing the state regulator and describes the one or more state variables (x1(t,ξ), x2(t,ξ), x3(t,ξ), ...) based on one or more probability distributions, - describing (120) an optimization problem, said optimization problem comprising a cost function (J) computed using at least said system matrix (A), said feedback matrix (K), an initial state (x0) and said input matrix (B), and - Solving (130) the optimization problem to determine the entries of the feedback matrix (K).

2. The computer-implemented method (100) of claim 1, wherein the description (120) further comprises -Determine (121) one or more expected values ​​and one or more variances of one or more state variables (x1(t,ξ), x2(t,ξ), x3(t,ξ), ...) of the state vector (x(t,ξ)), and wherein the cost function J is also formed by a first contribution and a second contribution, the first contribution corresponding to one or more expected values ​​of one or more state variables (x1(t,ξ), x2(t,ξ), x3(t,ξ), ...) of the state vector (x(t,ξ)), and the second contribution corresponding to one or more variances of one or more state variables (x1(t,ξ), x2(t,ξ), x3(t,ξ), ...) of the state vector (x(t,ξ)).

3. The computer-implemented method (100) of any one of claims 1 or 2, wherein describing (120) the optimization problem further comprises - With the help of the first weight matrix (W E )) weighting the first contribution (122), and - With the help of the second weight matrix (W var ) weights the second contribution (123).

4. The computer-implemented method (100) according to any of the preceding claims, wherein the optimization problem comprises minimizing the cost function J with respect to the entries of the feedback matrix (K).

5. The computer-implemented method (100) of any one of the preceding claims, wherein the description (120) further comprises - performing one or more series expansions based on the one or more probability distributions to describe the one or more state variables (x1(t,ξ), x2(t,ξ), x3(t,ξ), ...).

6. The computer-implemented method (100) of any preceding claim, wherein the one or more probability distributions are represented by one or more polynomial bases.

7. The computer-implemented method (100) according to any of the preceding claims, wherein the optimization problem further comprises an optimality condition, a stability condition and / or a definition set At least one of restrictions on the possible entries of the feedback matrix (K).

8. The computer-implemented method (100) according to any one of the preceding claims, wherein the system (10) to be regulated is designed to be arranged in a vehicle and / or to control a vehicle function.

9. The computer-implemented method (100) according to claim 8, wherein the system to be regulated is a system for lateral guidance and / or longitudinal guidance of the vehicle.

10. A computer system designed to execute the computer-implemented method (100) for designing a state regulator (20) with stochastic optimization according to any one of the preceding claims 1 to 9.

11. A computer program comprising instructions which, when executed by a computer system, cause the computer system to perform a computer-implemented method (100) for designing a state regulator (20) with stochastic optimization according to any one of claims 1 to 9.

12. A computer readable medium or signal storing and / or containing a computer program according to claim 11.

Citation Information

Cited By

  • Random high-order network containment control method based on simple complex

    CN119758732A