Rehabilitation robot multi-mode control method and system based on mixed modal signal

Abstract

The application provides a rehabilitation robot multi-mode control method and system based on a mixed mode signal, and the application adopts a discrete determination learning theory to establish an experience-based control library and an identifier library under different rehabilitation modes and rehabilitation actions; then, based on a mode control theory, a plurality of physiological signals and kinematic parameters of a current patient, such as electroencephalogram, electromyogram, human posture and contact force, are identified to identify a rehabilitation mode suitable for a rehabilitation level of the current patient and a rehabilitation action intended by the patient, and then a corresponding controller is selected to drive the patient to train. Moreover, the mixed mode signal of the patient can be analyzed in real time during the training process, and once the intention of the patient changes, the experience-based control library and the identifier library can work quickly to quickly identify and switch the controller, thereby improving the performance of the rehabilitation training system.

Description

Technical Field

[0001] This invention belongs to the field of rehabilitation robot control technology, and relates to a multi-mode control method and system for rehabilitation robots based on mixed-modal signals. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Two-thirds of stroke patients experience varying degrees of motor impairment, severely impacting their daily lives. Rehabilitation training can help patients fully repeat movements and restore motor function. Addressing the shortcomings of traditional rehabilitation therapies, such as high manpower requirements and poor repetitiveness, rehabilitation robots can provide patients with focused and repetitive training. Furthermore, clinical results have demonstrated that robot-assisted rehabilitation training has a positive effect on restoring motor function.

[0004] Rehabilitation for stroke patients is a lengthy process, broadly divided into the flaccid paralysis stage, spasticity stage, and recovery stage, each requiring different rehabilitation training methods. Rehabilitation robots are envisioned to possess human-like intelligent control capabilities, enabling them to learn and recognize different patient conditions and, like humans, adopt precise control strategies based on experience. Inspired by human learning and control concepts, a combined approach of pattern learning, pattern recognition, and pattern control has been proposed—pattern-based control. However, learning in dynamic environments is considered one of the most challenging problems in adaptive and learning control; traditional control methods are either incapable of achieving learning or have limited learning effectiveness. Furthermore, existing pattern recognition based on statistical decision-making relies on known pattern characteristics rather than control performance, and therefore cannot be directly applied to pattern recognition and control.

[0005] Current upper limb rehabilitation robots have developed various rehabilitation modes, such as passive mode, resistance mode, and assisted mode, and involve a wide variety of rehabilitation training movements. Accurate identification of the patient's motor intention is crucial for implementing mode-based rehabilitation training and ensuring safe and accurate human-computer interaction. However, single-modal physiological information (such as EEG, EMG, etc.) or behavioral information (such as body posture, contact force, etc.) cannot fully reflect motor intention, especially in patients with central nervous system or peripheral limb injuries. Due to functional impairment, single-source physiological or behavioral information cannot accurately provide a basis for judging motor intention. Therefore, there is an urgent need to develop rehabilitation robot systems driven by multiple physiological or behavioral information to improve the system's comprehensive and accurate judgment of the patient's motor intention. Summary of the Invention

[0006] To address the aforementioned problems, this invention proposes a multi-modal control method and system for rehabilitation robots based on mixed-modal signals. This invention employs discrete deterministic learning theory to establish an experience-based control library and identifier library for different rehabilitation modes and movements. Then, based on mode-based control theory, it identifies various physiological signals and kinematic parameters of the current patient, such as EEG, EMG, body posture, and contact force, to determine the appropriate rehabilitation mode and desired rehabilitation movement for the current patient's rehabilitation level. The corresponding controller is then selected to guide the patient's training. Furthermore, the patient's mixed-modal signals can be analyzed in real time during training. If the patient's intention changes, the experience-based control library and identifier library can quickly identify and switch controllers, improving the performance of the rehabilitation training system.

[0007] According to some embodiments, the present invention adopts the following technical solution:

[0008] A multi-mode control method for a rehabilitation robot based on mixed-modal signals includes the following steps:

[0009] Using discrete deterministic learning theory, an experience-based controller library is established for different rehabilitation modes and rehabilitation actions;

[0010] Using discrete deterministic learning theory, an experience-based identifier library is established for all rehabilitation modes and rehabilitation actions;

[0011] Based on the current mixed modal signals, the kinematic parameters are compared with an experience-based identifier library to identify the rehabilitation mode suitable for the current patient's rehabilitation level and the rehabilitation movements the patient wants.

[0012] Based on the identified rehabilitation pattern and rehabilitation action, the corresponding controller is selected from the controller library to control the rehabilitation robot to perform the corresponding rehabilitation training.

[0013] As an alternative implementation, the following steps are also included:

[0014] During training, based on the mixed modal signals and kinematic parameters, it is determined whether the current training intention has changed. If so, the appropriate rehabilitation mode and rehabilitation action for the current patient's rehabilitation level are identified, and the system switches to the matching controller from the controller library.

[0015] As an alternative implementation method, the specific process of establishing an experience-based controller library for different rehabilitation modes and rehabilitation actions using discrete deterministic learning theory includes:

[0016] The discrete dynamic model of the rehabilitation robot is transformed into a dynamic model in the form of a canonical system with output feedback;

[0017] Designing a discrete adaptive neural network controller using discrete deterministic geometry theory;

[0018] Based on Lyapunov stability theory and sampling deterministic learning theory, the weight update law of the discrete adaptive neural network controller is determined.

[0019] Based on the discrete adaptive neural network controller and weight update law, the unknown dynamics of the system are learned along the periodic trajectory, and the learned knowledge is used to construct an experience-based controller library under different training modes.

[0020] Furthermore, based on the discrete adaptive neural network controller and the weight update law, the specific process of learning the unknown dynamics of the system along the periodic trajectory and using the learned knowledge to construct an experience-based controller library under different training modes includes: when the neurons of the discrete adaptive neural network controller along the periodic trajectory satisfy the continuous excitation condition, the state error and the neural network weights are bounded and converge exponentially; the stable converged neural network weights are stored in the form of a constant neural network; and then the learned knowledge is used to construct an experience-based controller for different training modes.

[0021] Based on the characteristics of different modes, the experienced controllers are optimized to form a controller library.

[0022] Furthermore, the specific process of optimizing the experienced controller based on the characteristics of different modes includes:

[0023] Passive mode, employing the experienced controller described above;

[0024] In active mode, admittance control is added to the experienced controller, and the parameters are changed to adjust the compliance during the human-computer interaction process.

[0025] In the stop mode, a threshold control is added to the experienced controller. When the detected muscle strength exceeds the set threshold, the controller's controlled movement is forcibly stopped.

[0026] As an alternative implementation method, the specific process of establishing an experience-based identifier library for all rehabilitation modes and rehabilitation actions using discrete deterministic learning theory includes:

[0027] Based on the discrete dynamics model of the rehabilitation robot, a dynamic neural network identifier is built to model the intrinsic dynamics information of the subsystem under different training modes.

[0028] Lyapunov theory is used to update the weights of the identifier and demonstrate the system stability.

[0029] By sampling, the exponential stability principle in the learning theory is determined to ensure that the neural network can converge exponentially to the true value, thereby guaranteeing accurate modeling of the intrinsic dynamic information of the subsystem;

[0030] We construct an experience-based identifier library for all rehabilitation patterns and rehabilitation actions using a converged constant neural network.

[0031] As an alternative implementation method, when establishing an experience-based controller library for different rehabilitation modes and rehabilitation actions, the mixed modal signals during each rehabilitation action under each rehabilitation mode are acquired, identified, and a correspondence between different mixed modal signals and different rehabilitation modes and rehabilitation actions is established.

[0032] As an alternative implementation, the mixed-modal signal includes electroencephalography (EEG), electromyography (EMG), human posture, and contact force signals.

[0033] As an alternative implementation method, based on the current mixed modal signal, the corresponding rehabilitation mode is determined by comparing kinematic parameters with an experience-based identifier library, and then the appropriate controller is selected to drive the patient to follow the rehabilitation robot for movement.

[0034] A multi-mode control system for a rehabilitation robot based on mixed-modal signals, comprising:

[0035] The building module is configured to use discrete deterministic learning theory to establish an experience-based controller library for different rehabilitation modes and rehabilitation actions, as well as an experience-based identifier library for all rehabilitation modes and rehabilitation actions.

[0036] The identification module is configured to identify a rehabilitation mode suitable for the current patient's rehabilitation level based on the current mixed modal signal and kinematic parameters.

[0037] The control module is configured to select the appropriate controller from the controller library based on the identified rehabilitation mode, and control the rehabilitation robot to perform training of the corresponding rehabilitation actions.

[0038] As an alternative implementation, the control module is also configured to identify, based on the mixed modal signal and kinematic parameters, whether the current training intention has changed, and if so, to identify the rehabilitation mode and rehabilitation action suitable for the current patient's rehabilitation level, and switch to the matching controller from the controller library.

[0039] A terminal device includes a processor and a computer-readable storage medium, the processor being configured to implement instructions; the computer-readable storage medium being configured to store a plurality of instructions adapted to be loaded by the processor and executed in accordance with the steps of the method.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] This invention, based on sampled-deterministic learning theory, establishes two libraries: an experience-based controller library for controlling robots to move according to task requirements under different actions and modes; and an experience-based identifier library. This library compares newly generated multimodal signal features with information in the database to obtain the action or mode corresponding to the signal feature with the highest similarity, and then switches accordingly. Simulating human intelligent control strategies, the invention establishes experience-based controller and identifier libraries for different modes. Then, based on the patient's rehabilitation level and rehabilitation wishes, it performs rapid pattern recognition and rapid controller switching, greatly improving system performance.

[0042] The control system proposed in this invention utilizes the patient's mixed-modal physiological signals to identify the patient's intentions and assess the rehabilitation level, and uses deterministic learning theory to perform experience-based rapid pattern recognition, which greatly improves the recognition accuracy and speed.

[0043] The control method proposed in this invention fully considers the discrete-time characteristics of actual rehabilitation robots. It constructs an experience-based discrete learning controller for passive mode to achieve precise rehabilitation in passive training of patients. To improve the compliance of the active mode process, admittance control is added to the experience-based discrete learning controller to stimulate the patient's active movement intention.

[0044] In this invention, both the experience-based controller and the experience-based identifier can accurately model unknown dynamics and unpredictable disturbances in nonlinear systems locally. The learned experience knowledge is stored in the form of a constant neural network. For subsequent identical or similar control tasks, the stored knowledge can be directly called for control or identification, eliminating the need for online calculation of controller parameters. This saves energy and control time in the rehabilitation training system and further improves dynamic control performance.

[0045] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0046] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0047] Figure 1 This is a block diagram of the multi-mode switching mechanism of the upper limb rehabilitation robot in this embodiment.

[0048] Figure 2 This is a schematic diagram of the working principle of the controller in the passive and active modes of the upper limb rehabilitation robot in this embodiment.

[0049] Figure 3This is a flowchart illustrating the specific implementation of the multi-mode control method for the upper limb rehabilitation robot based on mixed-modal physiological signals in this embodiment. Detailed Implementation

[0050] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0051] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0052] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0053] This embodiment uses an upper limb rehabilitation robot as an example to illustrate the technical solution of the present invention.

[0054] However, this does not mean that the present invention can only be used for upper limb rehabilitation robots.

[0055] like Figure 3 As shown, it includes the following steps:

[0056] Pre-acquire mixed-mode signals under different modes, construct mode-mixed-mode signal relationships, and store them;

[0057] We use discrete deterministic learning theory to establish an experience-based controller library for different rehabilitation modes and rehabilitation actions.

[0058] We use discrete deterministic learning theory to build an experience-based identifier library for all rehabilitation modes and rehabilitation actions.

[0059] Once the patient is involved, the corresponding modal and kinematic signal detection devices are provided. Then, the detected modal signals are compared with the information in the identifier library. Through identification and pattern recognition, the rehabilitation mode that best matches the patient's current rehabilitation level and the rehabilitation movement that best matches the patient's current rehabilitation wishes are selected based on the principle of minimum residual. The corresponding experience-based controller is then selected to guide the patient in rehabilitation training.

[0060] It should be noted that the acquisition of multimodal signals in this invention can be achieved using existing technologies, such as acquiring EEG signals by wearing an EEG cap, acquiring EMG signals by wearing an EMG device, obtaining human posture using image acquisition equipment, and acquiring contact force signals using a force sensor.

[0061] In some embodiments, to ensure that rehabilitation training remains consistently appropriate, the patient's mixed-modal signals are continuously monitored in real time during the training process. If the patient's movement intentions change or their rehabilitation level improves as training progresses, a dynamic pattern recognition mechanism based on deterministic learning can quickly identify the change and switch to the appropriate controller in a very short time. This multi-modal control method based on mixed-modal signals fully considers the patient's rehabilitation level and their rehabilitation intentions, and can greatly improve the efficiency of patient rehabilitation.

[0062] The specific implementation technical solution will be described in detail below.

[0063] like Figure 1 As shown, the controller's creation and training process specifically includes:

[0064] Step (1): Design a discrete adaptive neural network based on the discrete upper limb rehabilitation robot dynamics model and discrete deterministic theory;

[0065] Based on the following discrete-time dynamic model of the upper limb rehabilitation robot:

[0066]

[0067] The dynamic model, transformed into a canonical system with output feedback, takes the following form:

[0068]

[0069] In the formula, k is the sampling time point, T is the sampling time interval; q(k) and v(k) represent the joint position and joint velocity, respectively. M(q(k)) and is the inertia matrix; F(q(k),v(k)) is the Coriolis force-centrifugal force and gravitational torque matrix; τ(k) is the control input torque;

[0070]

[0071] x(k) = [q(k), v(k)] T , τ d (k) represents the external force torque;

[0072] Based on a discrete dynamics model of the upper limb rehabilitation robot and sampling deterministic learning, a discrete-time adaptive neural network controller is appropriately designed to accurately identify (learn) the unknown dynamics of the upper limb rehabilitation robot during tracking control. Specifically:

[0073]

[0074] Where c2 is the controller gain constant of the design, and satisfies 0≤1-Tc2≤1, and e2 is the tracking error of the joint velocity. The output of a radial basis function neural network (RBFNN) is... Let S be the weights of the RBFNN, and S be the regression vector of the RBFNN. The input to the RBFNN is α1(k), which is the virtual controller, specifically:

[0075]

[0076] In the formula, e1 is the tracking error of the joint position, c1 is the designed controller gain constant, and satisfies 0≤1-Tc1≤1, x d The reference trajectory (in active mode, the trajectory is corrected by admittance control).

[0077] Based on Lyapunov stability theory and sampling deterministic learning theory, the neural network weight update law is designed as follows:

[0078]

[0079] Where, Γ=Γ T >0 is a positive definite diagonal matrix, and σ>0 is a constant.

[0080] Step (2): Learn the unknown dynamics of the system along the periodic trajectory, and use the learned knowledge to build an upper limb rehabilitation robot controller library under different training modes;

[0081] When neurons in an RBFNN along a periodic trajectory satisfy the persistent excitation (PE) condition, both the state error and the neural network weights are bounded and converge exponentially. The neural network weights after stable convergence... With constant neural networks The learned knowledge is stored in the form of [data], and then an experienced controller is built for different training modes using the learned knowledge, specifically:

[0082]

[0083] in, for:

[0084]

[0085] In the formula, [k a ,k b [ ] represents the time interval after the system has reached stable convergence.

[0086] Step (3): Accurately identify the subsystem dynamics under different training modes and use the learned knowledge to build an identifier library;

[0087] For the discrete nonlinear dynamic system v(k+1)=v(k)+T[f(x)+g(x)τ(k)] of the upper limb rehabilitation robot shown in formula (2), the intrinsic dynamic information of the subsystem under different training modes is modeled by building a dynamic neural network identifier, specifically:

[0088]

[0089] In the formula, s represents the training mode. Let β be the state of the i-th dynamic neural network discriminator. i For the gain of the identifier, The weights to be estimated in the RBFNN of the identifier. This represents the regression vector of the RBFNN in the identifier. It is used to approximate the unknown dynamics of the subsystem f(x)+g(x)τ(k).

[0090] The neural network weight update law based on Lyapunov design is as follows:

[0091]

[0092] Where γ represents the learning gain of the neural network weight update law, z i This indicates the tracking error.

[0093] By employing the aforementioned dynamic neural network identifier and weight update law, the intrinsic dynamic information of the subsystem under different training modes can be accurately identified locally by the constant RBFNN:

[0094]

[0095] In the formula, Let be the state of the i-th discrete dynamic identifier in the s-th training mode. It is a constant neural network.

[0096] Step (4): Use the estimator to quickly identify changes in the control state;

[0097] The test mode data is compared with the discrete dynamic identifier in the estimation library to generate the synchronization error (or recognition error) corresponding to different training modes.

[0098]

[0099] To improve the effectiveness of the identification process, the average L1 norm is used for decision-making.

[0100]

[0101] In the formula, T e It is a predefined range for calculating the average L1 norm.

[0102] Step (5): Switch the corresponding controller according to the principle of minimum average L1 norm.

[0103] Another important part is the design and selection of controllers for different rehabilitation modes.

[0104] like Figure 2 As shown, rehabilitation modes can be broadly categorized into passive mode, active mode, and stationary mode. Passive mode primarily utilizes the experience-based discrete learning controller shown in formula (6) for precise trajectory tracking training. It is suitable for the early stages of rehabilitation when patients lack muscle strength and rely mainly on the upper limb rehabilitation robot to move the affected limb. Active mode is divided into assistive mode and resistance mode, suitable for the middle and late stages of rehabilitation when the patient's affected limb has a certain degree of active movement ability. Assistive mode provides assistance when the patient's movement ability is insufficient, while resistance mode provides resistance when the patient's movement force is excessive.

[0105] In this embodiment, the controller in active mode adds admittance control based on formula (6), realizing the switching from passive mode to active mode, and adjusting the compliance in the human-computer interaction process by changing the admittance parameters of the system.

[0106] The stop mode is mainly used to ensure patient safety. When abnormal patient behavior is detected, i.e., muscle strength is significantly greater than the range of force required for training, the upper limb rehabilitation robot movement is forcibly stopped.

[0107] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0108] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0109] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0110] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0111] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0112] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A multi-mode control method for a rehabilitation robot based on mixed-modal signals, characterized in that, Includes the following steps: Using discrete deterministic learning theory, an experience-based controller library is established for different rehabilitation modes and actions. The specific process includes: converting a discrete rehabilitation robot dynamic model into a dynamic model in the form of a standardized system with output feedback; designing a discrete adaptive neural network controller using discrete deterministic learning theory; determining the weight update law of the discrete adaptive neural network controller based on Lyapunov stability theory and sampled deterministic learning theory; and learning the unknown dynamics of the system along a periodic trajectory based on the discrete adaptive neural network controller and the weight update law, and using the learned knowledge to construct a controller library for different training modes. When establishing an experience-based controller library for different rehabilitation modes and rehabilitation actions, mixed modal signals are acquired during each rehabilitation action under each rehabilitation mode, identified, and a correspondence is established between the mixed modal signals and different rehabilitation modes and rehabilitation actions; or, the mixed modal signals include electroencephalogram (EEG), electromyogram (EMG), human posture, and contact force signals. Using discrete deterministic learning theory, an experience-based identifier library is established for all rehabilitation modes and actions. The specific process includes: building a dynamic neural network identifier based on a discrete rehabilitation robot dynamics model to model the intrinsic dynamic information of the subsystem under different training modes; using Lyapunov theory to update the identifier's weights and verify system stability; ensuring the neural network converges exponentially to the true value through the exponential stability principle in sampling deterministic learning theory, thereby guaranteeing accurate modeling of the subsystem's intrinsic dynamic information; and using the converged constant neural network to construct an experience-based identifier library for all rehabilitation modes and actions. Based on the current mixed modal signals and kinematic parameters, a rehabilitation mode suitable for the current patient's rehabilitation level is identified. During training, based on the mixed modal signals and kinematic parameters, it is determined whether the current training intention has changed. If so, the appropriate rehabilitation mode for the current patient's rehabilitation level is identified, and the system switches to the matching controller from the controller library. The rehabilitation mode is divided into passive mode, active mode and stop mode; according to the characteristics of different modes, the specific process of optimizing the experienced controller includes: stop mode, adding threshold control to the experienced controller, when the detected muscle strength is greater than the set threshold, the controller’s controlled movement is forcibly stopped; Based on the identified rehabilitation pattern, a corresponding controller is selected from the controller library to control the rehabilitation robot to perform the corresponding rehabilitation training actions.

2. The multi-mode control method for a rehabilitation robot based on mixed-modal signals as described in claim 1, characterized in that, Based on the aforementioned discrete adaptive neural network controller and weight update law, the specific process of learning the unknown dynamics of the system along the periodic trajectory and using the learned knowledge to construct a controller library under different training modes includes: when the neurons of the discrete adaptive neural network controller along the periodic trajectory satisfy the continuous excitation condition, the state error and the neural network weights are bounded and converge exponentially; the stable converged neural network weights are stored in the form of a constant neural network; and then the learned knowledge is used to construct an experienced controller for different training modes. Based on the characteristics of different modes, the experienced controllers are optimized to form a controller library.

3. The multi-mode control method for a rehabilitation robot based on mixed-modal signals as described in claim 2, characterized in that, The specific process of optimizing the experienced controller based on the characteristics of different modes also includes: Passive mode, employing the experienced controller described above; In active mode, admittance control is added to the experienced controller, and the parameters are changed to adjust the compliance during the human-machine interaction process.

4. The multi-mode control method for a rehabilitation robot based on mixed-modal signals as described in claim 1, characterized in that, Based on the current mixed modal signals, the corresponding rehabilitation mode is determined by comparing kinematic parameters with an experience-based identifier library, and then the appropriate controller is selected to drive the patient to follow the rehabilitation robot for movement.

5. A multi-mode control system for a rehabilitation robot based on mixed-modal signals, characterized in that it includes: The building module is configured to use discrete deterministic learning theory to establish an experience-based controller library for different rehabilitation modes and rehabilitation actions, as well as an experience-based identifier library for all rehabilitation modes and rehabilitation actions. The specific process of establishing an experience-based controller library for different rehabilitation modes and actions using discrete deterministic learning theory includes: converting the discrete dynamic model of the rehabilitation robot into a dynamic model in the form of a canonical system with output feedback; designing a discrete adaptive neural network controller using discrete deterministic learning theory; determining the weight update law of the discrete adaptive neural network controller based on Lyapunov stability theory and sampled deterministic learning theory; and learning the unknown dynamics of the system along a periodic trajectory based on the discrete adaptive neural network controller and the weight update law, and using the learned knowledge to construct a controller library under different training modes. The specific process of establishing an experience-based identifier library for all rehabilitation modes and actions using discrete deterministic learning theory includes: building a dynamic neural network identifier based on a discrete rehabilitation robot dynamics model to model the intrinsic dynamic information of the subsystem under different training modes; using Lyapunov theory to update the weights of the identifier and verify the system stability; ensuring that the neural network can converge exponentially to the true value through the exponential stability principle in sampling deterministic learning theory, thereby guaranteeing accurate modeling of the intrinsic dynamic information of the subsystem; and using the converged constant neural network to construct an experience-based identifier library for all rehabilitation modes and actions. When establishing an experience-based controller library for different rehabilitation modes and rehabilitation actions, mixed modal signals are acquired during each rehabilitation action under each rehabilitation mode, identified, and a correspondence is established between the mixed modal signals and different rehabilitation modes and rehabilitation actions; or, the mixed modal signals include electroencephalogram (EEG), electromyogram (EMG), human posture, and contact force signals. The identification module is configured to identify a rehabilitation mode suitable for the current patient's rehabilitation level based on the current mixed modal signal and kinematic parameters. During training, based on the mixed modal signals and kinematic parameters, it is determined whether the current training intention has changed. If so, the appropriate rehabilitation mode for the current patient's rehabilitation level is identified, and the system switches to the matching controller from the controller library. The rehabilitation mode is divided into passive mode, active mode and stop mode; according to the characteristics of different modes, the specific process of optimizing the experienced controller includes: stop mode, adding threshold control to the experienced controller, when the detected muscle strength is greater than the set threshold, the controller’s controlled movement is forcibly stopped; The control module is configured to select the appropriate controller from the controller library based on the identified rehabilitation mode, and control the rehabilitation robot to perform training of the corresponding rehabilitation actions.

6. The multi-mode control system for a rehabilitation robot based on mixed-modal signals as described in claim 5, characterized in that, The control module is also configured to identify, based on mixed modal signals and kinematic parameters, whether the current training intention has changed, and if so, to identify the rehabilitation mode and rehabilitation action suitable for the current patient's rehabilitation level and switch to the matching controller from the controller library.

Images