Pathology self-adaptive prediction control method for lower limb rehabilitation exoskeleton robot

Abstract

The invention discloses a pathology adaptive prediction control method for a lower limb rehabilitation exoskeleton robot, and belongs to the technical field of rehabilitation robot control. The method comprises the steps that the joint angle, the angular speed and the man-machine interaction torque of the lower limb rehabilitation exoskeleton robot are collected in real time, and a kinetic equation of the lower limb rehabilitation exoskeleton robot is constructed; on the basis of the kinetic equation of the lower limb rehabilitation exoskeleton robot, parameter vectors representing the current pathological features of the patient are estimated online, so that a prediction model in the lower limb rehabilitation exoskeleton robot is updated, and the optimal auxiliary torque is solved. The intelligent brain of the exoskeleton robot can be deeply fused with the real-time state of the body of the patient, and rehabilitation training which is really personalized and self-adaptive and can effectively promote neural function recovery is provided for the patient.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation robot control technology, and more specifically to a pathological adaptive predictive control method for a lower limb rehabilitation exoskeleton robot. Background Technology

[0002] Lower limb rehabilitation exoskeleton robots are key equipment in the field of modern neurorehabilitation, aiming to provide precise gait assistance training for patients with mobility impairments such as stroke and spinal cord injury, thereby promoting neurological functional remodeling. The core challenge in achieving this goal lies in how to coordinate the robot's assistive behavior with the patient's real-time changing motor abilities and intentions, i.e., achieving true "human-machine collaboration." However, patients' pathological conditions, such as insufficient muscle strength, abnormal muscle tone (spasm), and limited joint range of motion, not only vary greatly from person to person but also dynamically evolve throughout a single training session and even the entire rehabilitation cycle. This places extremely high demands on the adaptive capabilities of the control system.

[0003] Currently, traditional exoskeleton control largely employs rigid strategies based on trajectory tracking, such as position control. This method forces the patient's legs to strictly follow a predefined ideal gait trajectory, ensuring proper movement but completely ignoring the patient's active participation. This easily leads to the "passenger effect," where the patient is passively moved by the robot, which not only fails to effectively promote neural recovery but may also cause discomfort due to the stiff human-robot interaction. To address this issue, compliant control methods based on interactive forces, such as impedance control and admittance control, are gradually becoming mainstream. These methods adjust the stiffness and damping of the robot's joints to enable it to respond to force signals emitted by the patient, providing a degree of flexibility. However, their controller parameters usually need to be pre-set, making it impossible to automatically distinguish during training whether a small interactive force originates from the patient's active effort or passive resistance caused by pathological factors such as spasticity. Therefore, this fixed-parameter control strategy struggles to fundamentally adapt to the complex and time-varying pathological characteristics of patients, limiting the accuracy and personalization of its assistance.

[0004] To further enhance the intelligence of control, advanced algorithms such as Model Predictive Control (MPC) have been introduced into this field. MPC can predict the future state of the system based on a model and provide the optimal control command through optimization calculations, demonstrating great potential. However, current applications face a fundamental flaw: the predictive model it relies on is often based on the dynamics of healthy individuals or ideal working conditions, resulting in a significant "model mismatch" with the dynamic characteristics of the real patient-robot coupled system. Specifically, the patient's pathological parameters, such as the equivalent passive stiffness, damping, and active muscle strength of joints, differ by orders of magnitude from the healthy parameters assumed in the model and are constantly changing. When an "intelligent brain" designed based on a healthy human model commands a "body" with pathological characteristics, its predictions will inevitably be inaccurate. The optimized auxiliary torque will either fail to adequately compensate for functional impairments (insufficient assistance) or overcompensate, thereby inhibiting the patient's remaining spontaneous movements (over-assistance), severely limiting the rehabilitation effect.

[0005] It is evident that the evolution of existing technologies has consistently failed to resolve the core issue of "disconnect between the control model and the patient's real-time pathological state." Whether it is rigid control, compliant control with fixed parameters, or MPC based on a fixed model, none of them have been able to perceive and integrate the patient's pathological characteristics as intrinsic variables of the model online. Therefore, how to broaden our perspective to address this deficiency is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, in order to at least partially solve the above-mentioned technical problems, the present invention provides a pathological adaptive predictive control method for a lower limb rehabilitation exoskeleton robot. The method aims to identify and quantify the patient's key pathological parameters online in real time, and use them to dynamically update the predictive model. This deeply integrates the "intelligent brain" of the exoskeleton robot with the real-time state of the patient's body, providing the patient with truly personalized, adaptive rehabilitation training that can effectively promote the recovery of neurological function.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A pathological adaptive predictive control method for a lower limb rehabilitation exoskeleton robot, comprising the following steps: S1. Real-time acquisition of joint angles, angular velocities, and human-machine interaction torques of the lower limb rehabilitation exoskeleton robot; S2. Determine the dynamic equations of the lower limb rehabilitation exoskeleton robot based on the collected data; S3. Based on the dynamic equations of the lower limb rehabilitation exoskeleton robot, estimate the parameter vector representing the patient's current pathological characteristics online; S4. Update the prediction model in the lower limb rehabilitation exoskeleton robot according to the parameter vector, solve for the optimal auxiliary torque, and apply the optimal auxiliary torque to the lower limb rehabilitation exoskeleton.

[0008] Preferably, in S2, the dynamic equations of the lower limb rehabilitation exoskeleton robot are determined based on the collected data, including: The kinetic energy function and potential energy function of the robot are determined based on the angles and angular velocities of the robot's hip and knee joints, respectively. Substituting the kinetic energy function and potential energy function into the Lagrange equation, and combining them with the human-machine interaction torque, we obtain the standard dynamic equation. Define a pathological parameter vector P, and transform the standard kinetic equation into a linear form with respect to the pathological parameter vector P.

[0009] Preferably, the pathological parameter vector P is:

[0010] in, For the joint's active torque, This is the joint equivalent stiffness. This is the joint equivalent damping.

[0011] Preferably, in S3, the online estimation of the parameter vector representing the patient's current pathological characteristics includes the following steps: S31. Discretize the dynamic equations of the lower limb rehabilitation exoskeleton robot, in the form of:

[0012] In the formula, For observation terms, Let be the vector of pathological parameters to be estimated. The auxiliary torque applied to the robot It is a noise term that includes discretization error, modeling error and external disturbance, where k represents the sampling time; S32. Based on the discretized dynamic equations, the parameter estimates are updated by minimizing the weighted least squares cost function using the recursive least squares method. .

[0013] Preferably, in S32, the steps include: S321. Determine the gain matrix ;

[0014] in, It is a forgetting factor. It is the inverse of the estimated error covariance matrix; the update method is:

[0015] S322. Update the parameter estimate using the following formula. ; .

[0016] Preferably, in S4, updating the prediction model in the lower limb rehabilitation exoskeleton robot based on the parameter vector includes: Based on the current parameter estimates And the current operation point Establish a linear human-machine coupling dynamics model. State vector , To control input ; Discretize the linear human-machine coupling dynamics model; Constructing robot state prediction equations for future multiple steps based on a discretized model; To track the desired gait reference trajectory and minimize the control increment, a constrained quadratic programming problem is constructed, and the optimal control increment sequence is obtained by online solution. U*; The optimal control increment sequence The first element in U* is used as the actual control increment.

[0017] Preferably, the robot's state prediction equations for future multiple steps are constructed based on a discretized model, including: Introducing control increments And define the augmented state vector. The discretized model is then rewritten in augmented state-space form:

[0018] in, ; , and These are the estimated values ​​depending on the current parameters. The Jacobian matrix and constant terms; Based on the augmented state space form, the matrix form of the state prediction equation is derived:

[0019] In the formula, It is a future state prediction sequence. It is the control increment sequence to be optimized. It is by The constructed prediction matrix, It is by The constant vector formed.

[0020] Preferably, constrained quadratic programming problems are constructed, including: Substituting the state prediction equation into the objective function yields a constrained quadratic programming problem:

[0021] subject to:

[0022] in, and It is and Repeat along the diagonal and The block diagonal matrix formed by this process. This is the weight matrix for the state error, which is a positive definite diagonal matrix. It is a weight matrix that controls the incremental changes; it is a positive definite diagonal matrix. It is a reference trajectory sequence. It consists of matrices and vectors composed of control constraints.

[0023] A preferred form of the objective function is:

[0024] subject to:

[0025]

[0026] in, This is the weight matrix for the state error, which is a positive definite diagonal matrix. It is a weight matrix that controls the incremental changes; it is a positive definite diagonal matrix. It is the torque output limit of the actuator. It represents the limit of torque variation between adjacent cycles, used to ensure smoothness. Represents the prediction time domain. represents the control time domain, and j represents the prediction step index.

[0027] Preferably, steps S1-S4 are executed sequentially in each control cycle to achieve adaptive rolling optimization control.

[0028] As can be seen from the above technical solutions, in order to address the problem that existing control methods for lower limb rehabilitation exoskeleton robots cannot adapt to the dynamic changes in pathology between and within individual patients, this invention provides a pathological adaptive predictive control method for lower limb rehabilitation exoskeleton robots. This method mainly estimates the time-varying biomechanical parameters that characterize the patient's motor dysfunction in real time and updates the predictive model accordingly, thereby achieving precise and personalized adjustment of the auxiliary torque, improving the compliance of human-computer interaction, the safety of training, and the effectiveness of rehabilitation training.

[0029] Compared with the prior art, the beneficial effects of the present invention include: By identifying and fusing time-varying biomechanical parameters that characterize patients’ motor dysfunction online, the predictive model was able to adaptively adjust to individual pathological characteristics and rehabilitation progress, effectively improving the personalization and accuracy of the auxiliary torque output. This method, while ensuring trajectory tracking performance, can smoothly adjust the level of assistance according to changes in the patient's active participation, avoiding both insufficient assistance and suppressing the "passenger effect" caused by excessive assistance, thereby promoting neural function remodeling. Furthermore, by incorporating parameters such as joint equivalent impedance into the model, the robustness to abnormal movement patterns such as spasticity is enhanced, significantly improving the safety and comfort of human-computer interaction. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0031] Figure 1 A flowchart illustrating the pathological adaptive predictive control method for exoskeleton robots used in lower limb rehabilitation. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0034] To address the problem that existing control methods for lower limb rehabilitation exoskeleton robots cannot adapt to the dynamic changes in pathology between and within individual patients, this invention discloses a pathological adaptive predictive control method for lower limb rehabilitation exoskeleton robots.

[0035] A lower limb exoskeleton robot is a wearable human-machine coupling system that integrates mechanical structure, drive system, sensing unit, and intelligent control algorithm. Its core design concept is to enhance or restore human motor function by providing external torque assistance and engaging in dynamic interaction with the wearer's lower limbs.

[0036] In this embodiment, the pathological adaptive prediction control method includes the following steps: S1. Real-time acquisition of joint angles, angular velocities, and human-machine interaction torques of the lower limb rehabilitation exoskeleton robot; S2. Determine the dynamic equations of the lower limb rehabilitation exoskeleton robot based on the collected data; S3. Based on the dynamic equations of the lower limb rehabilitation exoskeleton robot, estimate the parameter vector representing the patient's current pathological characteristics online; S4. Update the prediction model in the lower limb rehabilitation exoskeleton robot according to the parameter vector, solve for the optimal auxiliary torque, and apply the optimal auxiliary torque to the lower limb rehabilitation exoskeleton.

[0037] In some embodiments, in S1, the angle, angular velocity and human-machine interaction torque at the joints of the lower limb rehabilitation exoskeleton robot are acquired in real time by sensors, wherein the joints may include the hip joint and the knee joint.

[0038] In some embodiments, the dynamic equations of the lower limb rehabilitation exoskeleton robot are constructed through step S2; In this embodiment, a dynamic model of the lower limb human-machine system is first established, that is, considering the hip and knee joints in the sagittal plane, the patient's lower limb and exoskeleton are regarded as a coupled system, the Lagrangian method is used for modeling, and parameters representing the patient's pathological characteristics are explicitly introduced.

[0039] In some implementation plans, the specific steps are as follows: S21. Determine the robot's kinetic energy function and potential energy function; Robot Kinetic Energy and potential energy It is the joint angle Functions for (hip, knee):

[0040]

[0041] in, For generalized coordinates, It is the system's inertia matrix. It is the joint angular velocity vector. It is the total mass of the patient's lower leg and foot (for the knee joint model) or thigh, lower leg and foot (for the hip joint model). It is gravitational acceleration. It is the height of the center of mass of the link.

[0042] S22. Construct the standard dynamic equations; Substituting the kinetic energy function and potential energy function into the Lagrange equation, the standard dynamic equation is obtained as follows:

[0043] in, It is the joint angular acceleration vector. It consists of the Coriolis force and centrifugal force. It is a gravity term. It is the active torque generated by the patient's muscles. It is the auxiliary torque (i.e., control input u) applied by the exoskeleton robot. It is an unknown external disturbance.

[0044] S23. Introduce pathological parameter vectors; To characterize pathological features (such as spasm, muscle weakness, and convulsions), Decompose into active and passive components, and parameterize the passive components; that is...

[0045] in It is the active torque command issued by the patient's central nervous system, and it is a key pathological / capacity parameter that needs to be estimated. It is a diagonal matrix, and its diagonal elements are... This represents the equivalent passive stiffness of the joint over time. and joint posture Changes are used to simulate increased muscle tone caused by spasms, etc. It is a diagonal matrix, and its diagonal elements are... This represents the equivalent passive damping of the joint. It is the equilibrium position of the joint (usually the anatomical zero position).

[0046] Furthermore, substituting the parameterized model into the standard dynamic equations and rearranging, we obtain:

[0047] Define the pathological parameter vector P:

[0048] in, For the joint's active torque, This is the joint equivalent stiffness. This is the joint equivalent damping.

[0049] The standard kinetic equations are transformed into a linear form with respect to the pathological parameter vector P:

[0050] in It is a regression matrix that contains state variables.

[0051] In some embodiments, step S3 is used to estimate the parameter vector representing the patient's current pathological characteristics online. In this embodiment, recursive least squares with a forgetting factor is used for online parameter estimation. In some specific implementations, the estimation steps include: S31. At sampling time k, the dynamic equations of the continuous lower limb rehabilitation exoskeleton robot are discretized. It is difficult to measure directly, so a first-order difference approximation is used. .

[0052] The discretized observation equation is in the form of:

[0053] In the formula, The observation item, i.e., the measured value, includes , Let be the vector of pathological parameters to be estimated. The auxiliary torque applied to the robot It is a noise term that includes discretization error, modeling error and external disturbance, and k represents the sampling time.

[0054] S32. Based on the discretized dynamic equations, the parameter estimates are updated by minimizing the weighted least squares cost function using the recursive least squares method. ;

[0055] in It is a forgetting factor, used to give higher weight to new data, thereby tracking time-varying parameters.

[0056] In one alternative implementation, the iterative steps of the recursive least squares method include: S321. Calculate the gain matrix ;

[0057] in, It is the forgetting factor, preferably between 0.95 and 0.995. It is the inverse of the estimation error covariance matrix, reflecting the uncertainty of the estimation; the update method is:

[0058] S322. Update the parameter estimate using the following formula. ; .

[0059] At any moment Estimates of the pathological parameter vector p.

[0060] In one embodiment, step S4 involves updating the prediction model in the lower limb rehabilitation exoskeleton robot according to the parameter vector, solving for the optimal auxiliary torque, and applying the optimal auxiliary torque to the lower limb rehabilitation exoskeleton.

[0061] This step first involves designing an adaptive model predictive controller, including using the current parameter estimates. And the current operation point Establish a linear human-machine coupling dynamics model and discretize it; specifically, deconstruct the nonlinear dynamics model at the current operation point. Linearization and discretization are performed at the point to obtain a linear time-varying model for MPC prediction.

[0062] In this embodiment, a state vector is defined. Control input The linearized continuous state-space model is as follows:

[0063] in It depends on the current parameter estimation The Jacobian matrix and constant terms.

[0064] Furthermore, the zero-order hold method is used, with the sampling time... Discretize:

[0065] in:

[0066]

[0067] It is a constant term in the discretization.

[0068] Secondly, building the future based on discretization models The robot state prediction equation for each step; To handle constant terms Incremental model is introduced. Control increment is defined. and augment the state vector The augmented state-space equations are as follows:

[0069] in, ; , and These are the estimated values ​​depending on the current parameters. The Jacobian matrix and constant terms; Based on the augmented state space form, the matrix form of the state prediction equation is derived:

[0070] In the formula, It is a future state prediction sequence. It is the control increment sequence to be optimized. It is by The constructed prediction matrix, It is by The constant vector formed.

[0071] In some implementations, a constrained quadratic programming problem is constructed with the objective of tracking the desired gait reference trajectory and minimizing the control increment, and the optimal control increment sequence is obtained by solving it online. U*; The optimization objective is expressed as follows:

[0072] subject to:

[0073]

[0074] in, This is the weight matrix for the state error, which is a positive definite diagonal matrix. It is a weight matrix that controls the incremental changes; it is a positive definite diagonal matrix. It is the torque output limit of the actuator. It represents the limit of torque variation between adjacent cycles, used to ensure smoothness. Represents the prediction time domain. represents the control time domain, and j represents the prediction step index.

[0075] Furthermore, the state prediction equation Substitute into the objective function This leads to a constrained quadratic programming problem:

[0076] subject to:

[0077] in, and It is and Repeat along the diagonal and The block diagonal matrix formed by this process. It is a reference trajectory sequence. It consists of matrices and vectors composed of control constraints.

[0078] Finally, use an online QP solver (such as OSQP) to solve efficiently. U*.

[0079] In a preferred embodiment, the optimal control increment sequence is... The first element in U* is used as the actual control increment. Specifically, the first element of the optimization sequence is taken as the actual control quantity:

[0080] The actual torque applied to the exoskeleton at the current moment is:

[0081] Will The output is sent to the robot's underlying driver, and the control input from the previous moment is updated. Used for calculation in the next cycle.

[0082] In one exemplary embodiment, such as Figure 1 The complete predictive control method steps are as follows: 1. After the system (lower limb rehabilitation exoskeleton robot) is initialized, rehabilitation training begins; 2. During each control cycle, sensor data, including joint angles, is acquired in real time. angular velocity Human-computer interaction torque ; 3. Online pathological feature recognition and parameter estimation (RLS algorithm) to estimate parameters representing the patient's current pathological state in real time (such as joint equivalent stiffness and damping). 4. Update the internal model of the Model Predictive Controller (MPC), including utilizing... Linearize and discretize the dynamic model to update the MPC prediction model. ; 5. MPC rolling optimization to solve for the optimal auxiliary torque; including constructing and solving the QP problem to obtain the optimal control increment sequence. ; 6. Output the first optimal torque to the joint actuator (exoskeleton actuator) to assist the patient; 7. Determine whether training should continue. If it should continue, return to step 2 and begin the next control cycle to achieve continuous online adaptation. If not, training ends and the process terminates.

[0083] A specific application example is: A stroke survivor with a spastic gait on the affected side wore a lower limb rehabilitation exoskeleton with torque output and sensing functions in both the hip and knee joints for flat walking training. After system initialization, joint angles, angular velocities, and human-machine interaction torques were collected in real time during each control cycle (e.g., 10 milliseconds). A recursive least squares method with a forgetting factor was used to identify the time-varying high-damping parameter characterizing the patient's difficulty in extending the affected knee joint at the end of the swing phase. The adaptive model predictive controller then used this updated parameter to refresh its internal predictive model and calculated the optimal auxiliary torque sequence that could compensate for the high-damping effect by solving a constrained quadratic programming problem. Finally, the first torque value in the sequence was applied to the exoskeleton knee joint actuator to generate a moderate auxiliary extension torque, helping the patient to complete a gait closer to physiological state. At the same time, the level of assistance was smoothly reduced as subsequent parameters decreased, achieving the goal of "on-demand assistance" in rehabilitation training.

[0084] This application quantifies abstract and complex patient pathological states (such as muscle weakness and spasticity) into identifiable dynamic parameters (such as joint equivalent stiffness, damping, and active muscle torque), and embeds them as explicit variables into the human-machine system dynamic model to provide a basis for adaptive control. During rehabilitation training, online estimation algorithms such as recursive least squares (RLS) are used to estimate the above-mentioned pathological parameters from sensor data (joint angles, interaction torques, etc.) in real time, so that the controller can "sense" the patient's current actual ability. Furthermore, the pathological parameters identified online are immediately used to update the prediction model within the MPC. This makes the controller's "brain" no longer fixed, but a "digital twin" that dynamically evolves with the patient's condition, ensuring the accuracy of predictions. The MPC performs rolling optimization based on an updated model that highly matches the patient's current condition, resulting in more personalized, forward-looking, and safer auxiliary torques.

[0085] The core of this invention lies in estimating parameters representing the pathological characteristics of patients online and using these parameters to dynamically adjust the prediction model of the model prediction controller, thereby achieving adaptive control. In specific applications, it can be implanted into a lower limb rehabilitation exoskeleton robot system, where data is collected by sensor units and the above-mentioned adaptive control process is achieved using a computing processing unit.

[0086] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0087] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A pathological adaptive predictive control method for a lower limb rehabilitation exoskeleton robot, characterized by the following steps: include: S1. Real-time acquisition of joint angles, angular velocities, and human-machine interaction torques of the lower limb rehabilitation exoskeleton robot; S2. Determine the dynamic equations of the lower limb rehabilitation exoskeleton robot based on the collected data; S3. Based on the dynamic equations of the lower limb rehabilitation exoskeleton robot, estimate the parameter vector representing the patient's current pathological characteristics online; S4. Update the prediction model in the lower limb rehabilitation exoskeleton robot according to the parameter vector, solve for the optimal auxiliary torque, and apply the optimal auxiliary torque to the lower limb rehabilitation exoskeleton.

2. The pathological adaptive predictive control method for lower limb rehabilitation exoskeleton robot according to claim 1, characterized in that, In S2, the dynamic equations of the lower limb rehabilitation exoskeleton robot are determined based on the collected data, including: The kinetic energy function and potential energy function of the robot are determined based on the angles and angular velocities of the robot's hip and knee joints, respectively. Substituting the kinetic energy function and potential energy function into the Lagrange equation, and combining them with the human-machine interaction torque, we obtain the standard dynamic equation. Define a pathological parameter vector P, and transform the standard kinetic equation into a linear form with respect to the pathological parameter vector P.

3. The pathological adaptive predictive control method for lower limb rehabilitation exoskeleton robot according to claim 2, characterized in that, The pathological parameter vector P is: in, For the joint's active torque, This is the equivalent stiffness of the joint. This is the joint equivalent damping.

4. The pathological adaptive predictive control method for lower limb rehabilitation exoskeleton robot according to claim 1 or 2, characterized in that, In S3, the online estimation of the parameter vector representing the patient's current pathological characteristics includes the following steps: S31. Discretize the dynamic equations of the lower limb rehabilitation exoskeleton robot, in the form of: In the formula, For observation terms, Let be the vector of pathological parameters to be estimated. The auxiliary torque applied to the robot It is a noise term that includes discretization error, modeling error and external disturbance, where k represents the sampling time; S32. Based on the discretized dynamic equations, the parameter estimates are updated by minimizing the weighted least squares cost function using the recursive least squares method. .

5. The pathological adaptive predictive control method for lower limb rehabilitation exoskeleton robot according to claim 4, characterized in that, In S32, the steps include: S321. Determine the gain matrix ; in, It is a forgetting factor. It is the inverse of the estimated error covariance matrix; the update method is: S322. Update the parameter estimate using the following formula. ; 。 6. The pathological adaptive predictive control method for lower limb rehabilitation exoskeleton robot according to claim 1, characterized in that, In S4, the prediction model in the lower limb rehabilitation exoskeleton robot is updated according to the parameter vector, including: Based on the current parameter estimates And the current operation point Establish a linear human-machine coupling dynamics model. For state vectors, For control input; Discretize the linear human-machine coupling dynamics model; Constructing robot state prediction equations for future multiple steps based on a discretized model; To track the desired gait reference trajectory and minimize the control increment, a constrained quadratic programming problem is constructed, and the optimal control increment sequence is obtained by online solution. U*; The optimal control increment sequence The first element in U* is used as the actual control increment.

7. The pathological adaptive predictive control method for a lower limb rehabilitation exoskeleton robot according to claim 6, characterized in that, Based on a discretized model, equations for predicting the robot's state over multiple future steps are constructed, including: Introducing control increments And define the augmented state vector. The discretized model is then rewritten in augmented state-space form: in, ; , and These are the estimated values ​​depending on the current parameters. The Jacobian matrix and constant terms; Based on the augmented state space form, the matrix form of the state prediction equation is derived: In the formula, It is a future state prediction sequence. It is the control increment sequence to be optimized. It is by The constructed prediction matrix, It is by The constant vector formed.

8. The pathological adaptive predictive control method for a lower limb rehabilitation exoskeleton robot according to claim 6, characterized in that, Construct constrained quadratic programming problems, including: Substituting the state prediction equation into the objective function yields a constrained quadratic programming problem: subject to: in, , and It is and Repeat along the diagonal and The block diagonal matrix formed by this process. This is the weight matrix for the state error, which is a positive definite diagonal matrix. It is a weight matrix that controls the incremental changes; it is a positive definite diagonal matrix. It is a reference trajectory sequence. It consists of matrices and vectors composed of control constraints.

9. The pathological adaptive predictive control method for a lower limb rehabilitation exoskeleton robot according to claim 6 or 8, characterized in that, The objective function is expressed in the following form: subject to: in, This is the weight matrix for the state error, which is a positive definite diagonal matrix. It is a weight matrix that controls the incremental changes; it is a positive definite diagonal matrix. It is the torque output limit of the actuator. It is the limit of torque variation between adjacent cycles. Represents the prediction time domain. represents the control time domain, and j represents the prediction step index.

10. The pathological adaptive predictive control method for a lower limb rehabilitation exoskeleton robot according to claim 1, characterized in that, Steps S1-S4 are executed sequentially in each control cycle to achieve adaptive rolling optimization control.

Images