Underwater robot motion control method and system, storage medium and program product
Through the method based on learning model prediction control, underwater robot motion control is learned and optimized, the model mismatch caused by current perturbation and parameter perturbation is solved, and the motion control performance of underwater robots is improved.
Patent Information
- Application Number
- CN202410947006.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-08
AI Technical Summary
The existing underwater robot motion control methods have failed to effectively solve the model mismatch caused by current perturbation and parameter perturbation, resulting in insufficient control performance.
Using learning model predictive control (LBMPC) method, the non-linear dynamic function with high uncertainty caused by external ocean current perturbation and parametric perturbation of underwater robots is learned through non-parametric learning methods, and the optimization control problem is constructed. Calculate the control law of converging path deviation and estimate nonlinear terms in real time to improve model matching.
The improvement of the motion control performance of underwater robots is achieved, and the real-time non-parametric learning is ensured through dynamic update of the training set of few-sampled points, effectively solving the mismatch of the control model.
Smart Images

Figure CN118915796A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of robot control technology, and in particular to an underwater robot motion control method based on learning model predictive control. Background Art
[0002] At present, underwater robots need to autonomously complete various tasks in complex marine environments, such as resource exploration, underwater search and rescue, etc. Therefore, the motion control of underwater robots is the key technology to complete their tasks.
[0003] In the prior art, Model Predictive Control (MPC) is a commonly used motion control algorithm for underwater robots, and its control performance depends on an accurate control model. However, the mismatch between the control model and the real model caused by ocean current disturbances and hydrodynamic parameters will degrade the control performance. In summary, the motion control method of autonomous underwater robots based on MPC has obvious shortcomings: it fails to effectively solve the model mismatch problem caused by ocean current disturbances and parameter perturbations.
[0004] In summary, there is at least one of the following technical problems:
[0005] The model mismatch problem caused by ocean current disturbances and parameter perturbations has not been effectively solved; the motion control performance of underwater robots needs to be improved. Summary of the invention
[0006] The main purpose of the present invention is to provide an underwater robot motion control method based on learning-based model predictive control (LBMPC) to solve the problem that the prior art fails to effectively solve the model mismatch problem caused by ocean current disturbances and parameter perturbations and the problem that the motion control performance of the underwater robot needs to be improved.
[0007] In order to achieve the above object, according to one aspect of the present invention, a method for controlling underwater robot motion based on learning model predictive control is provided, comprising:
[0008] Through non-parametric learning methods, the underwater robot learns the nonlinear dynamic function with high uncertainty caused by external ocean current disturbances and parameter perturbations;
[0009] Perform control optimization and construct an optimal control problem based on the LBMPC controller;
[0010] The control law of the convergence path deviation is calculated, and the learned nonlinear dynamic function with high uncertainty is used in the transfer of the rolling optimization state. The control law of the underwater robot is obtained by solving the optimization control problem.
[0011] Preferably, the model mismatch caused by external ocean current disturbances and parameter perturbations is represented as an unknown time-varying nonlinear term, which is estimated in real time by a nonparametric learning method based on the Kinky Inference (KI) prediction equation, and an LBMPC controller is constructed to calculate the motion control law, wherein the nonlinear term estimated by the nonparametric learning method is used for state transfer in rolling optimization to improve the matching of the model.
[0012] Preferably, including:
[0013] Step 1: Construct the state and control quantities of the underwater robot and construct the control equation;
[0014] Step 2: Construct a training set for the nonparametric learning method and establish the estimated Lipschitz constant;
[0015] Step 3: Update the training set of the non-parametric learning method, obtain the state quantity at the current moment, construct and solve the optimization control problem to calculate the control quantity of the underwater robot at the current moment, and send it to the underwater robot for execution.
[0016] Preferably, the step 1 includes: constructing the underwater robot state quantity x and control quantity u, and constructing the control equation
[0017] in: is the state quantity; n x is the state dimension; is the control input; n u is the dimension of the control quantity; the state matrix A and the control matrix B are time-varying; the nonlinear function represents the unmodeled nonlinear dynamics function with bounded unknown disturbances, where is the output space, is a bounded unknown process noise and satisfies Among them, the normal number is the upper bound, is the process noise space, the set and Assume it is a compact set.
[0018] Preferably, step 2 includes: constructing a non-parametric learning method training set And establish the estimated Lipschitz constant L n ;
[0019] in is the measured input value, is the input space, N n is the number of sampling points in the training set, and the input set of sampling is defined as The measured output value of the nonlinear dynamic function f is defined as Lipschitz constant L estimated by offline training n And determine the number of sampling points N in the training set n .
[0020] Preferably, the motion control operation of the underwater robot is performed by repeating steps 1-3.
[0021] The present invention also provides a computer system, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program and implements learning through a non-parametric learning method, learning the nonlinear dynamic function with high uncertainty of the underwater robot caused by external ocean current disturbances and parameter perturbations; performing control optimization, constructing an optimization control problem based on an LBMPC controller; using the learned nonlinear dynamic function with high uncertainty in the transfer of the rolling optimization state, and obtaining the control law of the underwater robot by solving the optimization control problem.
[0022] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements learning through a non-parametric learning method to learn the nonlinear dynamic function with high uncertainty of the underwater robot caused by external ocean current disturbances and parameter perturbations; performs control optimization and constructs an optimization control problem based on an LBMPC controller; uses the learned nonlinear dynamic function with high uncertainty in the transfer of the rolling optimization state, and obtains the control law of the underwater robot by solving the optimization control problem.
[0023] The present invention also provides a computer program product, including a computer program, characterized in that when the computer program is executed by a processor, it implements learning through a non-parametric learning method to learn the nonlinear dynamic function with high uncertainty caused by external ocean current disturbances and parameter perturbations of the underwater robot; performs control optimization and constructs an optimization control problem based on an LBMPC controller; uses the learned nonlinear dynamic function with high uncertainty in the transfer of the rolling optimization state, and obtains the control law of the underwater robot by solving the optimization control problem.
[0024] The application of the technical solution of the present invention has the following technical effects:
[0025] The present invention provides a novel nonparametric learning method, which realizes accurate estimation of nonlinear dynamic functions with high uncertainty based on the KI prediction equation, wherein a dynamically updated training set with few sampling points is adopted to ensure the real-time performance of nonparametric learning.
[0026] The present invention provides a design method for an underwater robot motion controller based on a learning model predictive control. The estimation result of the non-parametric learning method is used to optimize the state transfer in the control problem, which can effectively solve the mismatch of the control model and play a role in improving the motion control performance of the underwater robot. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:
[0028] Figure 1 A flow chart of an underwater robot motion control method based on learning model predictive control according to the present invention is shown. DETAILED DESCRIPTION
[0029] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0030] like Figure 1 As shown, an embodiment of the present invention provides an underwater robot motion control method based on learning model predictive control, including: learning through a non-parametric learning method, learning the nonlinear dynamic function of the underwater robot with high uncertainty caused by external ocean current disturbances and parameter perturbations; performing control optimization, constructing an optimization control problem based on an LBMPC controller; using the learned nonlinear dynamic function with high uncertainty in the transfer of the rolling optimization state, and obtaining the control law of the underwater robot by solving the optimization control problem.
[0031] In this embodiment, learning is performed by a non-parametric learning method to learn the nonlinear dynamic function with high uncertainty caused by external ocean current disturbances and parameter perturbations of the underwater robot; control optimization is performed to construct an optimization control problem based on the LBMPC controller; the control law of the convergence path deviation is calculated, and the learned nonlinear dynamic function with high uncertainty is used in the transfer of the rolling optimization state. Specifically, step 1: construct the state quantity and control quantity of the underwater robot, and construct the control equation; construct the state quantity χ and control quantity of the underwater robot And construct the control equation in: is the state quantity; n x is the state dimension; is the control input; n u is the dimension of the control quantity; the state matrix A and the control matrix B are time-varying; the nonlinear dynamic function represents the unmodeled nonlinear dynamics function with bounded unknown disturbances, where is the output space, is a bounded unknown process noise and satisfies Among them, the normal number is the upper bound, is the process noise space, the set and Assume that it is a compact set. Step 2: Construct a training set for nonparametric learning methods and establish the estimated Lipschitz constant; Construct a training set for nonparametric learning methods and the estimated Lipschitz constant L n ;in is the measured input value, is the input space, N n is the number of sampling points in the training set, and the input set of sampling is defined as The measured output value of the nonlinear dynamic function f is defined as Lipschitz constant L estimated by offline training n And determine the number of sampling points N in the training set n Step 3: Update the training set D of the non-parametric learning method n Obtain the state quantity χ(k) at the current moment, construct and solve the optimization control problem to calculate the control quantity of the underwater robot at the current moment Send it to the underwater robot for execution. Step 4: Repeat steps 1-3 until the motion control operation of the underwater robot is completed.
[0032] In this embodiment, a path tracking control method for an underwater robot is combined with model predictive control and non-parametric learning. In the control model, the model mismatch caused by ocean current disturbances and parameter perturbations is estimated in real time by a non-parametric learning method based on the Kinky Inference (KI) prediction equation. An LBMPC controller is constructed to calculate the motion control law, in which the nonlinear dynamic function estimated by the non-parametric learning method is used for state transfer in rolling optimization to improve the matching of the model.
[0033] Detailed implementation of this example:
[0034] Definition 1 (Kinky Inference (KI) prediction function): When the sampling data set D n After obtaining, the estimated function It is calculated as follows:
[0035]
[0036] Where Ln is the estimated Lipschitz constant. is the upper noise bound. Represents the set {1,2,3,…N n},N n represents the number of sampling points in the training set, s(r) represents the input value in the sampling data set, and s(k) represents the input value at the current moment. Denotes the measured output value in the training set. n It is the training set for non-parametric learning methods. and Represents an intermediate step in computing an estimate function.
[0037] Definition 2 (Lazily Adapted Constant Kinky Inference (LACKI) rule): The estimated Lipschitz constant is updated based on the LACKI rule:
[0038]
[0039] Where λ>0 is a hyperparameter. L0 can be set to a small constant. and represents the measured output value in the training set. s(x) and s(y) represent the measured input values in the training set. Represents the set {1,2,3,…N n}.
[0040] Definition 3 (Minimum Input Space): Define the space is the minimum input space, which contains all possible sampling points. And the minimum number of sampling points is defined as
[0041] Comment 3: Real-time performance of KI prediction function and LACKI rule and the number of sampling points N in the sampled data set n If the number of sampling points in the sampling data set is N n The computational complexity of the KI prediction function and the LACKI rule is reduced. In order to achieve better real-time performance, the number of sampling points in the sampling data set is the nominal minimum number of sampling points obtained by offline calculation. Now define the space is the nominal minimum input space.
[0042] Learning rule: Nominal minimum number of sampling points The estimated Lipschitz constant is obtained through an offline training algorithm described below. At this time, the sampled data set is dynamically updated:
[0043]
[0044] Where s(r) represents the measured input value of the rth sample, Represents the measured output value of the rth sample. Representing a collection Indicates the nominal minimum number of sampling points.
[0045] The sampling set at this time The sampled data in consists of the sampled data of a continuous system arranged in time order and having the same sampling period. With such a learning rule, the distance between two adjacent sampling points can be considered bounded.
[0046] When the nonlinear dynamic function f(χ(k)) is learned by the nonparametric learning method, the state transfer equation of the LBMPC controller can be obtained:
[0047]
[0048] in represents the estimated value of the nonlinear dynamics function at time k. χ(k) is the state quantity, is the control quantity at time k. A is the state matrix, B is the control matrix. s(k) represents the measured input value at time k. L n is the estimated Lipschitz constant. A non-sampled dataset that is updated dynamically.
[0049] In order to minimize the state deviation and meet the smoothness of the control input, the increment of the control quantity is defined The following cost function is designed:
[0050]
[0051] Where N n is the prediction time domain. n and R n is the weight matrix, P n is the terminal matrix. is the control quantity increment, and χ is the state quantity. is the control quantity at time k-1. is the control quantity at time k. χ(i|k) represents the state quantity of the oth prediction step in the prediction time domain at time k. represents the i-th control quantity increment of the i-th prediction step in the prediction time domain at time k. n |k) represents the Nth prediction in the time domain at time k n A predicted state. Considering the state quantity and control quantity constraints:
[0052]
[0053] Where χ is the state quantity, χ min and χ max Indicates the minimum and maximum values of the state quantity, Represents a set of state quantity constraints. To control the amount, is the control quantity increment, and Indicates the minimum and maximum values of the control quantity. and Indicates the minimum and maximum values of the control amount increment, Represents a set of control quantity constraints.
[0054] Therefore, an optimization control problem is constructed to calculate the control input:
[0055]
[0056] The solution of the optimization control problem (9) is defined as represents the set {0,1,…,N n -1},N n is the prediction time domain. At this time, the control input is: in represents the control quantity of the i-th prediction step in the prediction time domain at time k, where X(i|k) represents the state quantity of the i-th prediction step in the prediction time domain at time k. Represents the control increment of the i-th prediction step in the prediction time domain at time k. represents the set of control quantity constraints, Represents a set of state quantity constraints. A is the state matrix and B is the control matrix. is a dynamically updated non-sampled data set. n is the estimated Lipschitz constant. s(i|k) represents the input value of the sampled data set of the i-th prediction step in the prediction time domain at time k. Represents the estimated value of the nonlinear dynamic function of the i-th prediction step in the prediction domain at time k.
[0057] The present invention also provides a computer system, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program and implements learning through a non-parametric learning method to learn the nonlinear dynamic function with high uncertainty caused by external ocean current disturbances and parameter perturbations of an underwater robot; performs control optimization and constructs an optimization control problem based on an LBMPC controller; calculates the control law of the convergence path deviation, and uses the learned nonlinear dynamic function with high uncertainty in the transfer of a rolling optimization state.
[0058] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements learning through a non-parametric learning method to learn the nonlinear dynamic function with high uncertainty of the underwater robot caused by external ocean current disturbances and parameter perturbations; performs control optimization and constructs an optimization control problem based on an LBMPC controller; uses the learned nonlinear dynamic function with high uncertainty in the transfer of the rolling optimization state, and obtains the control law of the underwater robot by solving the optimization control problem.
[0059] The present invention also provides a computer program product, including a computer program, characterized in that when the computer program is executed by a processor, it implements learning through a non-parametric learning method to learn the nonlinear dynamic function with high uncertainty caused by external ocean current disturbances and parameter perturbations of the underwater robot; performs control optimization and constructs an optimization control problem based on an LBMPC controller; calculates the control law of the convergence path deviation, and uses the learned nonlinear dynamic function with high uncertainty in the transfer of the rolling optimization state.
[0060] From the above description, it can be seen that the above embodiments of the present invention achieve the following technical effects:
[0061] The present invention provides a novel nonparametric learning method, which realizes accurate estimation of nonlinear dynamic functions with high uncertainty based on the KI prediction equation, wherein a dynamically updated training set with few sampling points is adopted to ensure the real-time performance of nonparametric learning.
[0062] The present invention provides a design method for an underwater robot motion controller based on a learning model predictive control. The estimation result of the non-parametric learning method is used to optimize the state transfer in the control problem, which can effectively solve the mismatch of the control model and play a role in improving the motion control performance of the underwater robot.
[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A motion control method for an underwater robot based on learning model predictive control, characterized in that: include: Through non-parametric learning methods, the underwater robot learns the nonlinear dynamic function with high uncertainty caused by external ocean current disturbances and parameter perturbations; Perform control optimization and construct an optimal control problem based on the LBMPC controller; The learned nonlinear dynamic function with high uncertainty is used in the transfer of rolling optimization state, and the control law of the underwater robot is obtained by solving the optimization control problem.
2. The underwater robot motion control method based on learning model predictive control as claimed in claim 1, characterized in that ,The model mismatch caused by external ocean current disturbance and parameter perturbation is estimated in real time by a non-parametric learning method based on the KI prediction equation, and the LBMPC controller is constructed to calculate the motion control law, where the non-linear term estimated by the non-parametric learning method is used for the state transfer in the rolling optimization, so as to improve the matching of the model.
3. The underwater robot motion control method based on learning model predictive control as claimed in claim 1, characterized in that: include: Step 1: Construct the state and control quantities of the underwater robot and construct the control equation; Step 2: Construct a training set for the nonparametric learning method and establish the estimated Lipschitz constant; Step 3: Update the training set of the non-parametric learning method, obtain the state quantity at the current moment, construct and solve the optimization control problem to calculate the control quantity of the underwater robot at the current moment, and send it to the underwater robot for execution.
4. The underwater robot motion control method based on learning model predictive control as claimed in claim 1, characterized in that: The step 1 includes: constructing the underwater robot state quantity X and control quantity u, and constructing the control equation in: is the state quantity; n x is the state dimension; is the control input; n u is the dimension of the control quantity; the state matrix A and the control matrix B are time-varying; the nonlinear function represents the unmodeled nonlinear dynamics function with bounded unknown disturbances, where is the output space, is a bounded unknown process noise and satisfies Among them, the normal number is the upper bound, is the process noise space, the set and Assume it is a compact set.
5. The underwater robot motion control method based on learning model predictive control as claimed in claim 4, characterized in that: The step 2 includes: constructing a non-parametric learning method training set And establish the estimated Lipschitz constant L n ; in is the measured input value, is the input space, N n is the number of sampling points in the training set, and the input set of sampling is defined as The measured output value of the nonlinear dynamic function f is defined as Lipschitz constant L estimated by offline training n And determine the number of sampling points N in the training set n .
6. The underwater robot motion control method based on learning model predictive control as claimed in claim 3, characterized in that: The motion control operation of the underwater robot is performed by repeating steps 1-3.
7. The underwater robot motion control method based on learning model predictive control as claimed in claim 1, characterized in that: described.
8. A computer system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program and implements the steps of the method according to claim 1.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to claim 1 are implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to claim 1 are implemented.
