High-precision control method and system for assisting robots

Abstract

The application provides a high-precision control method for an auxiliary robot, and relates to the field of medical robot motion control. The method is used for an auxiliary robot system with strong nonlinearity, unobservable mode and complex disturbance. Trajectory sample data is collected first, and an original high-dimensional linearization state space model is constructed through a dynamic state feature extraction algorithm. Then, an unobservable subsystem is extracted and stability constraints are applied to obtain an optimized high-dimensional linearization model. A low-dimensional reduced linear state space model is constructed by using a state energy truncation reduction method, and a nominal controller and a state observer are designed based on the model. At the same time, an adaptive estimation learner is constructed to estimate complex disturbance in real time, and a disturbance compensation signal is combined to form an overall robust control system. The application effectively improves the suppression and processing capacity of the auxiliary robot system on strong nonlinearity, unobservable mode and complex disturbance, and realizes high-precision trajectory tracking control.

Description

Technical Field

[0001] This invention relates to the field of motion control for medical robots, and more particularly to a high-precision control method and system for an auxiliary robot. Background Technology

[0002] In the field of modern medical rehabilitation and motor assistance, assistive robot systems have become core equipment for improving patients' motor function and enhancing rehabilitation outcomes. Their control technology is a research hotspot in the field of medical robot motion control, and high-precision trajectory tracking control is a key technology for ensuring the effectiveness and safety of assistive robots in clinical applications, directly affecting the effectiveness of rehabilitation training, the precision of surgical procedures, and the safety of human-machine interaction. With the continuous upgrading of medical rehabilitation needs, assistive robot systems are rapidly developing towards higher reliability, higher safety, and higher precision, placing more stringent requirements on their control performance.

[0003] Currently, existing technologies exist for controlling medical assistive robots. One of the closest solutions involves collecting trajectory sample data from the robot, constructing a linearized state-space model, designing a state observer and a fixed-gain controller, and employing a simple disturbance suppression algorithm to achieve trajectory tracking control. This solution has been applied to medical assistive devices such as knee and ankle rehabilitation orthotics and upper limb rehabilitation robots. Through state estimation and basic disturbance suppression, it has improved the robot's control accuracy to some extent and mitigated the impact of external disturbances on the system.

[0004] However, this existing technology has significant technical shortcomings when applied in actual medical rehabilitation scenarios, making it difficult to meet the high-precision control requirements of assistive robots. Specifically, these shortcomings are as follows: First, the technical solution is insufficient in handling the strong nonlinear characteristics of the assistive robot system, employing only a simple linear modeling approach. It fails to fully consider the nonlinear coupling relationships generated during deep interaction between the robot and the human body, resulting in a large deviation between the model and the actual system. This makes it impossible to accurately represent the system's dynamic behavior, thus affecting control accuracy. Second, the technology does not specifically address unobservable subsystems in the model, nor does it introduce stability constraints. The system is susceptible to instability risks due to unobservable components, reducing control reliability. Finally, the disturbance suppression algorithm used is relatively simple, only able to handle simple, single-type disturbances. It cannot effectively suppress complex disturbances in medical scenarios, such as muscle spasms, joint adhesions, and environmental disturbances, leading to large trajectory tracking errors and poor control performance.

[0005] In summary, the existing technology cannot simultaneously and effectively handle the strong nonlinearity of the assistive robot system, cope with the instability risk caused by the unobservable part, and suppress and eliminate complex disturbances. It can only alleviate some problems from a single perspective, and the overall control performance is limited, making it difficult to meet the application requirements of modern medical assistive robots for high precision and high reliability. Therefore, there is an urgent need for a control method that can solve the above-mentioned core technical problems. Summary of the Invention

[0006] The purpose of this invention is to provide a high-precision control method for an auxiliary robot system, in order to solve the problems in existing auxiliary robot control that cannot effectively handle strong nonlinearity, cope with unobservable components, and achieve complex disturbance suppression and elimination at the same time.

[0007] The above-mentioned objective of the present invention is achieved through the following technical solution:

[0008] S1: Collect trajectory sample data of the assisted robot system and construct the original high-dimensional linear state space model using a dynamic state feature extraction algorithm;

[0009] S2: Extract the unobservable subsystems from the high-dimensional linearized state-space model and introduce stability constraints to obtain an optimized high-dimensional linearized model;

[0010] S3: The optimized high-dimensional linearized model is reduced in order based on the state energy truncation reduction method to construct a low-dimensional reduced-order linear state-space model.

[0011] S4: Design a nominal controller and a state observer for the reduced-order linear state-space model;

[0012] S5: To address the complex disturbances present in the system, an adaptive estimation learner is constructed, and the estimated value of the equivalent input disturbance is obtained in real time through the adaptive estimation learner;

[0013] S6: After filtering the estimated value of the equivalent input disturbance, a disturbance compensation signal is generated. The control output of the nominal controller and the disturbance compensation signal are combined to construct an overall robust control system, thereby realizing high-precision trajectory tracking control of the assisted robot.

[0014] Optionally, step S1 includes:

[0015] The considered assistive robot system has nonlinear characteristics, and its discrete state-space model expression is as follows:

[0016] (1)

[0017] in Indicates that the assistive robot system is in The system state at any given moment; These represent the assistive robot system in The system state, system output, system control input, and complex disturbances experienced by the system at any given time; An unknown function that describes the highly coupled relationship between the dynamic, strongly nonlinear characteristics of a system and complex disturbances; This is the system output matrix.

[0018] collection Group trajectory samples The original state is mapped to a higher dimension using a high-dimensional feature mapping function. Construct a predictor model in this high-dimensional linear space:

[0019] (2)

[0020] in, and They represent the mapped The state of a high-dimensional system at any given moment; Represents a high-dimensional feature mapping function; These represent the state matrix, input matrix, and source mapping matrix of the high-dimensional linearized model, respectively. This represents the predicted estimate of the original system state.

[0021] The following data matrix was constructed using a dynamic state feature extraction algorithm:

[0022] (3)

[0023] in, and These represent the high-dimensional feature data matrices for the current time step and the next time step, respectively. This represents the system control input data matrix; This represents the original system state data matrix.

[0024] The initial system matrix set is identified by solving the following least-squares optimization problem. :

[0025] (4)

[0026] Solving for the results Then, the initial high-dimensional model matrix is ​​extracted according to the block matrix structure. , and :

[0027] (5)

[0028] Optionally, step S2 includes:

[0029] Based on the identified initial high-dimensional feature mapping model, an observation depth is constructed. Extended observability matrix :

[0030] (6)

[0031] right Perform singular value decomposition (SVD) to extract the basis of the unobservable subspace. :

[0032] (7)

[0033] use right By performing orthogonal decoupling, the dynamic matrix of the unobservable subsystem of the assisted robot is obtained:

[0034] (8)

[0035] Establish prediction error loss function Stability loss function With regularization loss function :

[0036] (9)

[0037] Construct the overall objective function with stability constraints :

[0038] (10)

[0039] in, As a stability penalty factor, Let be the regularization coefficient. Minimize using gradient descent. The optimized stable model is obtained:

[0040] (11)

[0041] in, , and These represent the high-dimensional system matrix, high-dimensional input matrix, and high-dimensional output mapping matrix after stability constraint optimization, respectively.

[0042] Optionally, step S3 includes:

[0043] Solve the discrete-time Lyapunov equations of the optimized model to obtain the controllable Gauram matrix. With a considerable personality matrix :

[0044] (12)

[0045] Calculate Hankel singular values Introducing the balance transformation matrix , keep before The dominant state corresponding to the largest Hankel singular value is used to obtain a low-dimensional reduced-order model of the assist robot:

[0046] (13)

[0047] in, and Let represent the low-dimensional dominant state variables at the current and next time steps after the order reduction; These are the reduced-order system state matrix, control input matrix, and output mapping matrix, respectively.

[0048] Optionally, step S4 includes:

[0049] Reduce the dominant state With internal mold state By combining the equations, we can construct the augmented state expression:

[0050] (14)

[0051] definition:

[0052] (15)

[0053] (16)

[0054] in, and These represent the system dynamic matrix and the input matrix, respectively, representing the internal model state.

[0055] By employing the optimal control method, the state feedback gain is obtained: Thus, the nominal control law is derived:

[0056] (17)

[0057] Design a full-dimensional state observer to estimate the system state:

[0058] (18)

[0059] in, For observer gain; To assist the actual output of the robot system; These are the observed estimates of the dominant state after the order reduction.

[0060] Optionally, step S5 includes:

[0061] Construct an adaptive estimation learner, with the estimated value of the equivalent input disturbance being:

[0062] (19)

[0063] in, It is the pseudo-inverse of the input matrix.

[0064] Optionally, step S6 includes:

[0065] Filtered by a first-order discrete low-pass filter Noise:

[0066] (20)

[0067] in, These are the filter coefficients of a first-order discrete low-pass filter, with values ​​ranging from... .

[0068] The filtered equivalent input disturbance estimate is directly used to compensate for the control input of the robot system.

[0069] (twenty one)

[0070] The control signal to be solved By incorporating the overall robust control system, the influence of complex disturbance terms on the assisted robot system is suppressed. The strong nonlinearity, unobservable parts and complex disturbances existing in the entire assisted robot system are processed to achieve high-precision control of the entire assisted robot system.

[0071] Another object of the present invention is to provide a high-precision control system for an auxiliary robot, for executing the above-described high-precision control method for an auxiliary robot, the system comprising:

[0072] The data acquisition module is used to collect multiple sets of trajectory sample data from the assisted robot system and output them to the high-dimensional modeling module.

[0073] The high-dimensional modeling module receives trajectory sample data, constructs the original high-dimensional linearized state space model through a dynamic state feature extraction algorithm, and outputs it to the stability optimization module.

[0074] The stability optimization module receives the original high-dimensional linearized state-space model, extracts the unobservable subsystems in the model and introduces stability constraints, obtains the optimized high-dimensional linearized model, and outputs it to the order reduction processing module.

[0075] The order reduction processing module receives the optimized high-dimensional linearized model, performs order reduction processing using the state energy truncation method, constructs a low-dimensional reduced-order linear state-space model, and outputs it to the controller design module.

[0076] The controller design module is used to receive a low-dimensional reduced-order linear state-space model, design a nominal controller and a full-dimensional state observer, and send the control output of the nominal controller to the robust control module.

[0077] The adaptive learning module is used to receive the state estimate value from the full-dimensional state observer and the system operation data, build an adaptive estimation learner, obtain the estimate value of the equivalent input disturbance in real time, and generate a disturbance compensation signal after filtering and output it to the robust control module.

[0078] The robust control module receives the control output and disturbance compensation signal from the nominal controller, synthesizes the final control signal, and inputs it into the auxiliary robot system to build an overall robust closed-loop control system, thereby achieving high-precision trajectory tracking control of the auxiliary robot.

[0079] The beneficial effects of the technical solution provided by this invention are:

[0080] The technical solution of this invention takes into account that the auxiliary robot system inevitably has strong nonlinear characteristics, contains unobservable parts, and is susceptible to complex external disturbances. First, it extracts unobservable subsystems and optimizes stability constraints on the constructed high-dimensional linearized state space model, thereby eliminating spurious unstable modes in the system and fundamentally ensuring the asymptotic stability of the high-dimensional feature mapping model.

[0081] Secondly, the state energy truncation reduction method is used to reduce the dimensionality of the optimized high-dimensional linearized model and obtain a low-dimensional dominant state space model. While preserving the core dynamic characteristics of the system, this greatly reduces the real-time computational complexity of the underlying controller.

[0082] Finally, to address the linearization approximation error, order reduction error, and complex external disturbances in the high-dimensional model, an adaptive estimation learner is designed for real-time observation. This learner, combined with a nominal controller, constructs an overall robust control system, enabling real-time active elimination and compensation of the system's overall disturbances. This invention achieves high-performance trajectory tracking control for assisted robot systems with strong nonlinearity and complex disturbances, effectively handling the instability risks caused by unobservable components and the impact of complex disturbances on the system. Simultaneously, it significantly improves the robustness, real-time computational efficiency, and overall control accuracy of the assisted robot system. Attached Figure Description

[0083] Figure 1 A flowchart illustrating the high-precision control method for an auxiliary robot provided in an embodiment of the present invention;

[0084] Figure 2 This is a block diagram of the control structure of a high-precision control system for an auxiliary robot provided in an embodiment of the present invention.

[0085] Figure 3 The image shows a comparison of trajectory tracking simulations with existing technologies provided in this embodiment of the invention. Detailed Implementation

[0086] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0087] This invention provides a high-precision control method for assistive robots, based on trajectory sample data of assistive robot systems with strong nonlinearity and complex disturbances. It constructs an original high-dimensional linearized state-space model using a dynamic state feature extraction algorithm; extracts unobservable subsystems from the high-dimensional linearized state-space model and introduces stability constraints to obtain an optimized high-dimensional linearized model; reduces the order of the optimized high-dimensional linearized model using a state energy truncation reduction method to construct a low-dimensional reduced-order linear state-space model; designs a nominal controller and a state observer for the reduced-order linear state-space model; constructs an adaptive estimation learner to address the complex disturbances in the system; and combines the compensation signal generated by the adaptive estimation learner to construct an overall robust control system, suppressing complex disturbance terms in the assistive robot system and achieving high-precision trajectory tracking control for assistive robot systems with strong nonlinearity, unobservable components, and complex disturbances. This improves the assistive robot system's ability to handle and suppress / eliminate strong nonlinearity, unobservable components, and complex disturbances.

[0088] This embodiment fully discloses the overall mechanism and collaborative relationship of the high-precision control method for assisted robots. It forms an inseparable organic whole from modeling, optimization, order reduction, control, disturbance compensation to closed-loop robust operation. The steps are not simply superimposed, but rather the preceding steps provide a stable foundation for the following steps, and the following steps provide performance guarantees for the preceding steps. It is a fully coupled and collaborative process, which overcomes the technical difficulties of existing technologies that cannot simultaneously solve the contradiction between high-dimensional complexity, unobservable modes, strong disturbances and real-time performance by using state-space modeling, model order reduction or robust control alone. It has outstanding substantive features and significant progress, and is not a conventional combination of existing technologies.

[0089] I. Overall Working Mechanism

[0090] This invention employs a six-layer progressive collaborative mechanism: "high-dimensional accurate modeling → stability constraint optimization → energy-dominated order reduction → nominal control + state observation → adaptive disturbance learning → robust compensation closed loop".

[0091] 1. Construct a high-dimensional linearized state space using dynamic state feature extraction to fully preserve the robot's complex dynamics;

[0092] 2. Extract unobservable subsystems and apply stability constraints to eliminate potential model divergence risks;

[0093] 3. The state energy truncation method is used to preserve the dominant dynamics and eliminate redundant states, balancing accuracy and computational efficiency;

[0094] 4. Design a nominal controller + state observer based on a reduced-order model to provide basic high-precision tracking;

[0095] 5. Construct an adaptive estimation learner to estimate equivalent input disturbances online;

[0096] 6. The filtered compensation signal is superimposed on the control law to form a robust closed loop, ultimately achieving high-precision control with a trajectory tracking error of less than 0.01.

[0097] In the above process, each step depends on the output of the previous step and provides the necessary prerequisites for the next step. The absence of any link will prevent the achievement of the final effect, forming an inseparable technical collaboration.

[0098] II. Step-by-step collaborative working principle

[0099] 1. Dynamic State Feature Extraction and High-Dimensional Modeling (S1)

[0100] Multiple sets of trajectory samples of the assisted robot are collected, and the original state is mapped to high-dimensional features through high-dimensional feature mapping to construct a predictor model. The current time and the next time high-dimensional feature data matrix, control input matrix and original state matrix are constructed using dynamic state feature extraction algorithm. The initial system matrix is ​​identified by least squares and the initial high-dimensional model matrix is ​​extracted according to the block structure.

[0101] Synergistic effect: It provides a complete and lossless high-dimensional dynamic foundation for subsequent stability optimization, avoiding subsequent control inaccuracies caused by rough modeling.

[0102] 2. Unobservable subsystem extraction and stability constraint optimization (S2)

[0103] An extended observability matrix with an observation depth of r is constructed based on the initial high-dimensional model. The basis of the unobservable subspace is extracted by singular value decomposition. The dynamic matrix of the unobservable subsystem is obtained by orthogonal decoupling the state matrix. Prediction error, stability and regularization loss functions are established, and a total objective function with stability constraints is constructed. The stable model is obtained by gradient descent optimization.

[0104] Synergistic effect: It solves the instability problem caused by unobservable modes in high-dimensional models, and provides a stable and optimized model that can be used for controller design for subsequent order reduction. Without this step, the reduced model will still have the risk of divergence.

[0105] 3. State energy truncation order reduction (S3)

[0106] Solve the discrete Lyapunov equations of the optimization model to obtain the controllable and observable glympian matrices; calculate the Hankel singular values, introduce the equilibrium transformation matrix, retain the dominant states corresponding to the first r largest Hankel singular values, and obtain a low-dimensional reduced-order model.

[0107] Synergistic effect: Significantly reduces the state dimension without losing the dominant dynamics, solving the problem of high computational cost and inability to run in real time for high-dimensional models; at the same time, it provides a low-dimensional, feasible, and accuracy-preserving model foundation for the design of S4 controllers.

[0108] 4. Design of Nominal Controller and State Observer (S4)

[0109] An augmented state is constructed by combining the reduced dominant state with the internal model state. The state feedback gain is obtained by using optimal control, and the nominal control law is obtained. A full-dimensional state observer is designed to estimate the state of the system that cannot be directly measured.

[0110] Synergistic effect: Based on a stable low-dimensional model, high-precision trajectory tracking is achieved, providing a stable control basis for disturbance compensation; the state observer compensates for the unmeasurable state defect, forming a two-layer state guarantee with the processing of the S2 unobservable subsystem.

[0111] 5. Construction of Adaptive Estimation Learner (S5)

[0112] An adaptive estimation learner is constructed, which calculates the equivalent input disturbance estimate by using the pseudo-inverse of the input matrix and the state error. The model error, external disturbance, and unmodeled dynamics are uniformly equated as input disturbances for online estimation.

[0113] Synergistic effect: It transforms complex uncertainties into compensable equivalent disturbances, providing accurate and real-time disturbance information for robust compensation, and forming a "nominal + compensation" two-layer structure with S4 nominal control.

[0114] 6. Robust compensation and overall closed-loop control (S6)

[0115] The noise in the interference estimate is filtered out by a first-order discrete low-pass filter, and the filtered estimate is used for control quantity compensation. The final control signal is then input into the robot system to suppress strong nonlinearity, unobservable modes and complex disturbances, thereby achieving high-precision trajectory tracking.

[0116] Synergistic effect: Completes the closed-loop process from modeling to control, transforming all previous optimization and estimation results into actual control effects, and ensuring that the trajectory tracking error is consistently less than the preset threshold of 0.01.

[0117] III. Explanation of Overall Synergy and Creativity

[0118] This invention is not a simple patchwork of existing modeling, order reduction, and control methods, but rather an organic whole in which the functions of each step support each other, are causally coupled, and achieve overall efficiency:

[0119] 1. Stability optimization provides a prerequisite for order reduction: Order reduction without stability constraints will retain divergent modes, which cannot be used for control;

[0120] 2. Energy cutoff and order reduction provide the foundation for real-time control: High-dimensional models cannot be calculated online, and without order reduction, controllers cannot be deployed;

[0121] 3. Synergy between nominal control and adaptive compensation: Nominal control alone cannot resist disturbances, and disturbance compensation alone will diverge without a stable base;

[0122] 4. Dual-layer protection of unobservable subsystem processing and state observer: A single observer cannot eliminate model bias caused by unobservable modes.

[0123] The aforementioned multi-layered coupling and indispensable mechanism synergy brings about technical effects that cannot be achieved by existing technologies: under complex disturbances, strong nonlinearity, and unobservable modes, it still maintains stable operation, real-time calculation, and high-precision control with trajectory tracking error ≤0.01. The overall scheme is non-obvious and does not belong to the conventional combination of existing technologies.

[0124] According to the corresponding method flow, the high-precision control system for the assisted robot consists of a data acquisition module, a high-dimensional modeling module, a stability optimization module, a reduction-order processing module, a controller design module, an adaptive learning module, and a robust control module working together in sequence to execute the above method flow.

[0125] The present invention also provides a computer-readable storage medium storing multiple instructions adapted for a processor to load and execute the above-described method, enabling the high-precision control to be achieved by either a general-purpose processor or an embedded controller, thereby expanding its industrial applicability.

[0126] The embodiments of the present invention provide a high-precision control method for an auxiliary robot.

[0127] Please refer to Figure 1, which is a flowchart illustrating the steps of a high-precision control method for an auxiliary robot according to an embodiment of the present invention, including:

[0128] S1: Collect trajectory sample data of the assisted robot system and use dynamic state feature extraction algorithm to construct the original high-dimensional linear state space model;

[0129] Step S1 includes:

[0130] The considered assistive robot system has nonlinear characteristics, and its discrete state-space model expression is as follows:

[0131] (1)

[0132] in Indicates that the assistive robot system is in The system state at any given moment; These represent the assistive robot system in The system state, system output, system control input, and complex disturbances experienced by the system at any given time; An unknown function that describes the highly coupled relationship between the dynamic, strongly nonlinear characteristics of a system and complex disturbances; This is the system output matrix.

[0133] collection Group trajectory samples The original state is mapped to a higher dimension using a high-dimensional feature mapping function. Construct a predictor model in this high-dimensional linear space:

[0134] (2)

[0135] in, and They represent the mapped The state of a high-dimensional system at any given moment; Represents a high-dimensional feature mapping function; These represent the state matrix, input matrix, and source mapping matrix of the high-dimensional linearized model, respectively. This represents the predicted estimate of the original system state.

[0136] The following data matrix was constructed using a dynamic state feature extraction algorithm:

[0137] (3)

[0138] in, and These represent the high-dimensional feature data matrices for the current time step and the next time step, respectively. This represents the system control input data matrix; This represents the original system state data matrix.

[0139] The initial system matrix set is identified by solving the following least-squares optimization problem. :

[0140] (4)

[0141] Solving for the results Then, the initial high-dimensional model matrix is ​​extracted according to the block matrix structure. , and :

[0142] (5)

[0143] S2: Extract the unobservable subsystems from the high-dimensional linearized state-space model and introduce stability constraints to obtain an optimized high-dimensional linearized model;

[0144] Step S2 includes:

[0145] Based on the identified initial high-dimensional feature mapping model, an observation depth is constructed. Extended observability matrix :

[0146] (6)

[0147] right Perform singular value decomposition (SVD) to extract the basis of the unobservable subspace. :

[0148] (7)

[0149] use right By performing orthogonal decoupling, the dynamic matrix of the unobservable subsystem of the assisted robot is obtained:

[0150] (8)

[0151] Establish prediction error loss function Stability loss function With regularization loss function :

[0152] (9)

[0153] Construct the overall objective function with stability constraints :

[0154] (10)

[0155] in, As a stability penalty factor, Let be the regularization coefficient. Minimize using gradient descent. The optimized stable model is obtained:

[0156] (11)

[0157] in, , and These represent the high-dimensional system matrix, high-dimensional input matrix, and high-dimensional output mapping matrix after stability constraint optimization, respectively.

[0158] S3: The optimized high-dimensional linearized model is reduced in order based on the state energy truncation reduction method to construct a low-dimensional reduced-order linear state-space model.

[0159] Step S3 includes:

[0160] Solve the discrete-time Lyapunov equations of the optimized model to obtain the controllable Gauram matrix. With a considerable personality matrix :

[0161] (12)

[0162] Calculate Hankel singular values Introducing the balance transformation matrix , keep before The dominant state corresponding to the largest Hankel singular value is used to obtain a low-dimensional reduced-order model of the assist robot:

[0163] (13)

[0164] in, and Let represent the low-dimensional dominant state variables at the current and next time steps after the order reduction; These are the reduced-order system state matrix, control input matrix, and output mapping matrix, respectively.

[0165] S4: Design a nominal controller and a state observer for the reduced-order linear state-space model;

[0166] Step S4 includes:

[0167] Reduce the dominant state With internal mold state By combining the equations, we can construct the augmented state expression:

[0168] (14)

[0169] definition:

[0170] (15)

[0171] (16)

[0172] in, and These represent the system dynamic matrix and the input matrix, respectively, representing the internal model state.

[0173] By employing the optimal control method, the state feedback gain is obtained: Thus, the nominal control law is derived:

[0174] (17)

[0175] Design a full-dimensional state observer to estimate the system state:

[0176] (18)

[0177] in, For observer gain; To assist the actual output of the robot system; These are the observed estimates of the dominant state after the order reduction.

[0178] S5: Construct an adaptive estimation learner to address the complex perturbations present in the system;

[0179] Step S5 includes:

[0180] Construct an adaptive estimation learner, with the estimated value of the equivalent input disturbance being:

[0181] (19)

[0182] in, It is the pseudo-inverse of the input matrix.

[0183] S6: Combine the compensation signal generated by the adaptive estimation learner to construct an overall robust control system and realize high-precision trajectory tracking control of the assisted robot.

[0184] Step S6 includes:

[0185] Filtered by a first-order discrete low-pass filter Noise:

[0186] (20)

[0187] in, These are the filter coefficients of a first-order discrete low-pass filter, with values ​​ranging from... .

[0188] The filtered equivalent input disturbance estimate is directly used to compensate for the control input of the robot system.

[0189] (twenty one)

[0190] In one embodiment of the present invention, the control signal obtained by solving equation (21) is... By incorporating the overall robust control system, the influence of complex disturbance terms on the assisted robot system is suppressed. The strong nonlinearity, unobservable parts and complex disturbances existing in the entire assisted robot system are processed to achieve high-precision control of the entire assisted robot system.

[0191] In one embodiment of the present invention, the control structure diagram of the assistive robot system is shown in Figure 2. The present invention addresses the problems of strong nonlinearity, unobservable components, and complex disturbances present in existing assistive robot control systems. Based on high-dimensional feature mapping and stability constraint optimization of the unobservable subsystems in the system model, state energy truncation is used to reduce the order, resulting in a reduced-order linear state-space model. Subsequently, a state observer, nominal controller, and adaptive estimation learner are designed to achieve high-precision control of the assistive robot system. First, the method analyzes the system model, constructing a high-dimensional linearized state-space model by collecting sample data. Based on this, the unobservable subsystems are extracted and stability constraints are introduced, thus handling the strong nonlinearity and unobservable components of the system.

[0192] In one embodiment of the present invention, an adaptive estimation learner is designed to further handle interference in the system. By solving and feeding forward the disturbance compensation signal estimated by the adaptive estimation learner, the influence of complex disturbance terms on the system is suppressed, thereby achieving the suppression and elimination of complex disturbances in the system. This makes the system have strong anti-interference performance against comprehensive disturbances, meeting the application requirements of high-precision trajectory tracking control of assisted robot systems in complex human-machine interaction environments.

[0193] In one embodiment of the present invention, in the field of motion control requirements for assistive robots in medical rehabilitation, elderly care, and disability assistance, this solution can provide significant improvements in control accuracy, enhanced disturbance rejection, and applicability in complex dynamic environments. These technological advancements lay the foundation for achieving highly safe and stable assistive robot systems, and bring significant advantages to high-precision human-computer interaction and rehabilitation applications.

[0194] In one embodiment of the present invention, a comparison of simulation results is shown in Figure 3. All three control strategies in the figure perform high-dimensional feature mapping on the system model to address strong nonlinearity. Comparing the different response curves: the dashed line represents the control method without introducing stability constraints for the unobservable subsystem. Due to the lack of constraints and handling of the unobservable part in the high-dimensional space, its tracking curve exhibits significant phase lag and large errors; the dotted line represents the control method that introduces unobservable constraints but does not design an adaptive estimation learner. Due to the lack of a disturbance compensation mechanism, it is highly susceptible to complex disturbances during operation, resulting in high-frequency chattering and fluctuations; in contrast, the solid line represents the method of the present invention, which, after fully integrating high-dimensional feature mapping, unobservable subsystem extraction optimization, state dimensionality reduction, and an adaptive estimation learner, smoothly matches the target reference trajectory (dotted line). The above simulation comparison demonstrates that the method provided by the present invention simultaneously handles the strong nonlinearity of the system, addresses the unobservable part, and achieves the suppression and elimination of complex disturbances, ultimately realizing high-precision, high-performance, and robust control of the assisted robot system.

[0195] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will readily conceive of those skilled in the art upon consideration of the specification and the disclosure of practical truths.

[0196] This invention is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and examples are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A high-precision control method for an auxiliary robot, characterized in that, Includes the following steps: S1: Collect trajectory sample data of the assisted robot system and construct the original high-dimensional linear state space model using a dynamic state feature extraction algorithm; S2: Extract the unobservable subsystems from the high-dimensional linearized state-space model and introduce stability constraints to obtain an optimized high-dimensional linearized model; S3: The optimized high-dimensional linearized model is reduced in order based on the state energy truncation reduction method to construct a low-dimensional reduced-order linear state-space model. S4: Design a nominal controller and a state observer for the reduced-order linear state-space model; S5: To address the complex disturbances present in the system, an adaptive estimation learner is constructed, and the estimated value of the equivalent input disturbance is obtained in real time through the adaptive estimation learner; S6: After filtering the estimated value of the equivalent input disturbance, a disturbance compensation signal is generated. The control output of the nominal controller and the disturbance compensation signal are combined to construct an overall robust control system, thereby realizing high-precision trajectory tracking control of the assisted robot.

2. The method according to claim 1, characterized in that, Step S1 includes: Considering the discrete state-space model of the assisted robot system, P sets of trajectory samples are collected, and the original state is mapped to high-dimensional features through a high-dimensional feature mapping function to construct a predictor model. A high-dimensional feature data matrix for the current time step, a high-dimensional feature data matrix for the next time step, a system control input data matrix, and an original system state data matrix are constructed using a dynamic state feature extraction algorithm. By solving the least squares optimization problem, the initial system matrix set is identified, and the initial high-dimensional model matrix is ​​extracted according to the block matrix structure.

3. The method according to claim 2, characterized in that, Step S2 includes: Based on the initial high-dimensional feature mapping model obtained from the identification, an extended observability matrix with an observation depth of r is constructed. Singular value decomposition is performed on the extended observability matrix to extract the basis of the unobservable subspace; By orthogonally decoupling the state matrix using the unobservable subspace basis, the dynamic matrix of the unobservable subsystem of the assisted robot is obtained; Establish the prediction error loss function, the stability loss function, and the regularization loss function; Construct a total objective function with stability constraints, and minimize the total objective function using gradient descent to obtain the optimized stable model.

4. The method according to claim 3, characterized in that, Step S3 includes: Solve the discrete-time Lyapunov equations of the optimized model to obtain the controllable character Lamb matrix and the observable character Lamb matrix; Calculate the Hankel singular values, introduce a balance transformation matrix, and retain the dominant states corresponding to the first r largest Hankel singular values ​​to obtain a low-dimensional reduced-order model of the assist robot.

5. The method according to claim 4, characterized in that, Step S4 includes: By combining the reduced-order dominant state with the internal model state, an augmented state expression is constructed. By employing the optimal control method, the state feedback gain is obtained, and the nominal control law is derived. Design a full-dimensional state observer to estimate the system state.

6. The method according to claim 5, characterized in that, Step S5 includes: An adaptive estimation learner is constructed to obtain an estimate of the equivalent input disturbance, which is calculated from the pseudo-inverse of the input matrix and the state error.

7. The method according to claim 6, characterized in that, Step S6 includes: The noise in the equivalent input interference estimate is filtered out by a first-order discrete low-pass filter. The filter coefficients of the first-order discrete low-pass filter range from 0 to 1. The filtered equivalent input disturbance estimate is used to compensate for the control input of the robot system. By substituting the solved control signals into the overall robust control system, the influence of complex disturbance terms on the auxiliary robot system is suppressed, thereby achieving high-precision control.

8. A high-precision control system for an auxiliary robot, characterized in that, include: The data acquisition module is used to collect trajectory sample data of the assisted robot system; The high-dimensional modeling module is used to construct the original high-dimensional linear state-space model using a dynamic state feature extraction algorithm. The stability optimization module is used to extract the unobservable subsystems in the high-dimensional linearized state-space model and introduce stability constraints to obtain an optimized high-dimensional linearized model. The order reduction module is used to reduce the order of the optimized high-dimensional linearized model based on the state energy truncation order reduction method, and construct a low-dimensional reduced-order linear state space model. The controller design module is used to design the nominal controller and the state observer for the reduced-order linear state-space model. The adaptive learning module is used to build an adaptive estimation learner to address complex perturbations in the system. The robust control module is used to combine the compensation signal generated by the adaptive estimation learner to build an overall robust control system, thereby realizing high-precision trajectory tracking control of the assisted robot. The system is used to perform the method according to any one of claims 1 to 7.

9. A high-precision control method for an auxiliary robot, characterized in that, include: The overall robust control system is constructed using the method described in any one of claims 1 to 7; By applying a robust overall control system to the trajectory tracking task of an assisted robot, the effects of strong nonlinearity, unobservable components, and complex disturbances on the system are suppressed, thereby achieving high-precision trajectory tracking control.

10. The method according to claim 9, characterized in that, The trajectory tracking error of the high-precision trajectory tracking control is less than a preset threshold, which is 0.01.

Images