Multi-modal nerve-driven rehabilitation robot control method and system

Abstract

The embodiment of the invention provides a multi-mode nerve-driven rehabilitation robot control method and system, and relates to the technical field of robot control technologies. The method comprises the following steps: acquiring man-machine interaction task data and a real-time user signal of a user; generating a digital signal based on the user signal and a preset identification system; according to a difference value between the digital signal and the simulation signal, or according to a difference value between man-machine motion / force interaction data and simulation motion / force data, carrying out on-line correction on a corresponding preset model; based on the corrected parameters of the model and the obtained historical training data, generating rehabilitation evaluation data; and generating task data and training parameters according to the rehabilitation evaluation data. According to the method and the device, the problem of low training strategy accuracy caused by weak personalized adaptive adjustment capability is solved, and the effect of improving the personalized adaptive adjustment capability is further achieved.

Description

Technical Field

[0001] The present invention relates to the field of rehabilitation robot control technology, and more specifically, to a multimodal neural-driven rehabilitation robot control method and system. Background Technology

[0002] Neurological disorders, such as hemiplegia after stroke, often lead to motor dysfunction, severely impacting patients' quality of life. Rehabilitation training is a crucial step in functional recovery. Traditional rehabilitation training often relies on manual assistance from physical therapists or passive traction from robotic arms. These methods suffer from insufficient personalization of training programs, making it difficult to dynamically adjust training strategies based on the patient's real-time neuromuscular state during training. This can lead to mismatched training intensity or compensatory movement patterns. Summary of the Invention

[0003] This invention provides a multimodal neural-driven control method and system for rehabilitation robots, which at least solves the problem of low accuracy of training strategies in related technologies.

[0004] According to an embodiment of the present invention, a multimodal neural-driven rehabilitation robot control method is provided, comprising: The system acquires human-computer interaction task data and real-time user signals, wherein the user signals include electroencephalogram (EEG) signals, electromyogram (EMG) signals, and human-computer motion / force interaction data, and the human-computer interaction task data includes task type, task objective, and user operation feedback. Based on the user signal and the preset recognition system, a digital signal is generated, wherein the digital signal includes a digital motion signal and a digital activation signal; and a simulation signal is generated based on the preset model and the human-computer interaction task data, wherein the simulation signal includes a simulation motion signal, a simulation activation signal, and simulation motion / force data. Based on the difference between the digital signal and the simulation signal, or based on the difference between the human-computer motion / force interaction data and the simulation motion / force data, the corresponding preset models are corrected online. Rehabilitation assessment data is generated based on the parameters of the corrected model and the acquired historical training data; Task data and training parameters are generated based on the rehabilitation assessment data. The task data is used for task information interaction between the rehabilitation robot and the user, and the training parameters are used for motion / force interaction between the rehabilitation robot and the user.

[0005] In one exemplary embodiment, the step of determining the difference between the digital signal and the simulated signal includes: The digital motion signal is compared with the simulated motion signal output by the cognitive-motor model based on the human-computer interaction task data to obtain a first difference, and the cognitive-motor model is corrected online based on the first difference using a first machine learning algorithm, wherein the preset model includes the cognitive-motor model. The digital activation signal is compared with the simulated activation signal output by the skeletal muscle control model based on the digital motion signal to obtain a second difference. The skeletal muscle control model is then corrected online based on the second difference using a second machine learning algorithm. The preset model includes the skeletal muscle control model. Online correction of the preset model based on the difference between the human-computer motion / force interaction data and the simulated motion / force data includes: The real-time human-computer motion / force interaction data is compared with the simulated motion / force data output by the musculoskeletal model based on the simulation activation signal to obtain a third difference. A third machine learning algorithm is then used to correct the musculoskeletal model online based on the third difference. The preset model includes the musculoskeletal model.

[0006] In an exemplary embodiment, generating a digital signal based on the user signal and a preset identification system includes: Based on the electroencephalogram (EEG) signals and a preset motion signal recognition system, a digital motion signal representing the user's motion intention is generated. Based on the electromyographic signals and a preset activation signal recognition system, a digital activation signal representing the muscle activation state is generated.

[0007] In one exemplary embodiment, the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm each include a recurrent neural network, a long short-term memory network, a Gaussian process regression, a reinforcement learning policy gradient method, or an online adaptive filter.

[0008] In one exemplary embodiment, after performing online correction on a preset model, the method further includes: Periodically obtain model snapshots of the corrected preset model; Predicted physiological response data are generated based on the model snapshot; The predicted physiological response data is compared with the preset baseline response data to calculate the evaluation index of the model snapshot; The meta-parameters are adjusted according to the evaluation indicators, wherein the meta-parameters are used to adjust the preset model online.

[0009] According to another embodiment of the present invention, a multimodal neural-driven rehabilitation robot control system is provided, comprising: The data acquisition module is used to acquire human-computer interaction task data and real-time user signals, wherein the user signals include electroencephalogram (EEG) signals, electromyogram (EMG) signals, and human-computer motion / force interaction data, and the human-computer interaction task data includes task type, task objective, and user operation feedback. The signal generation module is used to generate digital signals based on the user signals and a preset recognition system, wherein the digital signals include digital motion signals and digital activation signals; and to generate simulation signals based on a preset model and the human-computer interaction task data, wherein the simulation signals include simulated motion signals, simulated activation signals, and simulated motion / force data. The correction module is used to correct the corresponding preset models online based on the difference between the digital signal and the simulation signal, or based on the difference between the human-computer motion / force interaction data and the simulation motion / force data. The rehabilitation assessment module is used to generate rehabilitation assessment data based on the parameters of the corrected model and the acquired historical training data. The task generation module is used to generate task data and training parameters based on the rehabilitation assessment data. The task data is used for task information interaction between the rehabilitation robot and the user, and the training parameters are used for motion / force interaction between the rehabilitation robot and the user.

[0010] In one exemplary embodiment, the step of determining the difference between the digital signal and the simulated signal includes: The digital motion signal is compared with the simulated motion signal output by the cognitive-motor model based on the human-computer interaction task data to obtain a first difference, and the cognitive-motor model is corrected online based on the first difference using a first machine learning algorithm, wherein the preset model includes the cognitive-motor model. The digital activation signal is compared with the simulated activation signal output by the skeletal muscle control model based on the digital motion signal to obtain a second difference. The skeletal muscle control model is then corrected online based on the second difference using a second machine learning algorithm. The preset model includes the skeletal muscle control model. Online correction of the preset model based on the difference between the human-computer motion / force interaction data and the simulated motion / force data includes: The real-time human-computer motion / force interaction data is compared with the simulated motion / force data output by the musculoskeletal model based on the simulation activation signal to obtain a third difference. A third machine learning algorithm is then used to correct the musculoskeletal model online based on the third difference. The preset model includes the musculoskeletal model.

[0011] In an exemplary embodiment, generating a digital signal based on the user signal and a preset identification system includes: Based on the electroencephalogram (EEG) signals and a preset motion signal recognition system, a digital motion signal representing the user's motion intention is generated. Based on the electromyographic signals and a preset activation signal recognition system, a digital activation signal representing the muscle activation state is generated.

[0012] According to yet another embodiment of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is configured to perform the steps in any of the above method embodiments when executed.

[0013] According to yet another embodiment of the present invention, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0014] This invention constructs a digital twin model that is synchronized with and interacts bidirectionally with the real user in real time. It can capture the user's initiative at the level of neural intent and make the digital twin approximate the user's real physiological state through continuous machine learning correction. This generates a highly personalized and adaptive training strategy, which solves the technical problem of low accuracy of training strategies caused by insufficient personalization in traditional rehabilitation training. This achieves the beneficial effect of improving rehabilitation effectiveness and efficiency. Attached Figure Description

[0015] Figure 1 This is a structural block diagram of a multimodal neurally driven rehabilitation robot control system according to an embodiment of the present invention; Figure 2 This is a control flowchart according to a specific embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0017] In the following description, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0018] Furthermore, in this application, directional terms such as "upper," "lower," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and may change accordingly depending on the orientation of the components in the accompanying drawings.

[0019] In this application, unless otherwise expressly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral part; it can be a direct connection or an indirect connection through an intermediate medium. Furthermore, the term "coupled" can refer to an electrical connection that enables signal transmission.

[0020] As used herein, “about,” “approximately,” or “approximately” includes the stated value and the average value within an acceptable range of deviation from the given value, wherein the acceptable range of deviation is determined by a person skilled in the art taking into account the measurement under discussion and the error associated with the measurement of the given quantity (i.e., the limitations of the measurement system).

[0021] like Figure 1 As shown, this application provides a functional block diagram of a multimodal neural-driven rehabilitation robot control system. The system may include a data acquisition module 100, a signal generation module 200, a correction module 300, a rehabilitation assessment module 400, and a task generation module 500. These modules can be connected via a bus or other means to achieve communication between them. The data acquisition module 100 is used to acquire human-computer interaction task data and real-time user signals, wherein the user signals include electroencephalogram (EEG) signals, electromyogram (EMG) signals, and human-computer motion / force interaction data, and the human-computer interaction task data includes task type, task objective, and user operation feedback.

[0022] The signal generation module 200 is used to generate digital signals based on the user signals and a preset recognition system, wherein the digital signals include digital motion signals and digital activation signals; and to generate simulation signals based on a preset model and the human-computer interaction task data, wherein the simulation signals include simulated motion signals, simulated activation signals, and simulated motion / force data.

[0023] The correction module 300 corrects the corresponding preset models online based on the difference between the digital signal and the simulation signal, or based on the difference between the human-computer motion / force interaction data and the simulation motion / force data.

[0024] The rehabilitation assessment module 400 is used to generate rehabilitation assessment data based on the parameters of the corrected model and the acquired historical training data. The task generation module 500 is used to generate task data and training parameters based on the rehabilitation assessment data. The task data is used for task information interaction between the rehabilitation robot and the user, and the training parameters are used for motion / force interaction between the rehabilitation robot and the user.

[0025] Reference Figure 2 This illustrates the overall architecture of the multimodal information fusion-based digital twin and rehabilitation robot closed-loop control system for human motor function described in this application embodiment. The system mainly consists of three interacting core components: the real user, the rehabilitation robot, and the digital user (i.e., the digital twin) acting as a bridge between the two.

[0026] The real user is the starting point and end point of the entire closed loop. Their biological structure includes the cognitive center, motor center, spinal cord, skeletal muscle and skeleton. The system monitors the user's physiological activities in real time through multimodal sensors.

[0027] A rehabilitation robot is a physical entity that performs and interacts with other devices. Its internal functional modules include a vision system, an assessment system, a protocol system, a training system, and a motor system.

[0028] A digital user (digital twin) is a high-fidelity virtual model of a real user's neuromuscular-skeletal system built in a computing environment. It comprises three core hierarchical models: a motor neural model (i.e., a cognitive-motor model), a skeletal muscle control model, and a musculoskeletal simulation system (i.e., a musculoskeletal model).

[0029] The information and control flows among these three elements constitute the core working logic of the system. The real user generates movement intentions, which are captured by electroencephalography (EEG) signals. At the same time, the actual activation state of the muscles is collected by surface electromyography (sEMG) signals. The interaction force / displacement between the user's actual limb movements and the rehabilitation robot is captured by motion sensors. These multimodal data streams from the real user are input into the rehabilitation robot's evaluation system to generate macroscopic training objectives. On the other hand, they serve as "realistic labels" for comparison and correction with the simulation output of the digital twin model.

[0030] The digital twin model receives motion / force targets from the rehabilitation robot as input instructions and runs a simulation. The motor neuron model simulates the activity of the cerebral cortex to generate "digital motion intentions". The skeletal muscle control model converts this intention into control signals for the virtual muscles, i.e., "digital motion signals". Based on these control signals, the musculoskeletal simulation system calculates the motion trajectory of the virtual limb and the forces interacting with the outside world, i.e., "digital motion / force simulation data".

[0031] The system continuously compares the "digital motion intent" with the intent decoded by the real EEG, the "digital motion signal" with the signal acquired by the real sEMG, and the "digital motion / force simulation data" with the data captured by the real sensors. The differences or errors generated by these comparisons are input into their respective machine learning algorithms to adjust and optimize the internal parameters of the three models in the digital twin online and continuously. This process enables the digital twin to continuously "learn" and "adapt", thereby reflecting the individual characteristics and state changes of the real user more and more accurately.

[0032] The modified, highly personalized digital twin provides the decision-making basis for the rehabilitation robot. The rehabilitation robot's assessment system comprehensively analyzes signals from the real user and the state of the digital twin to generate a quantitative rehabilitation assessment report. Based on this assessment, the treatment system formulates a personalized training plan for the next stage (e.g., adjusting task difficulty, assistance level, etc.). The training system translates the plan into specific instructions, driving the motor system to physically interact with the real user. The user's response is then captured by sensors, initiating a new closed loop of "perception-modeling-simulation-comparison-correction-decision-execution," achieving adaptive rehabilitation training that integrates brain, muscle, and machine.

[0033] The following will combine Figure 2 This application provides a detailed description of a multimodal information fusion-based digital twin and rehabilitation robot closed-loop control method for human motor function, based on embodiments of the present application.

[0034] S100: Acquires user signals during rehabilitation training in real time, including real-time EEG signals, real-time EMG signals, and real-time human-computer motion and force interaction data; and acquires human-computer interaction task data, including task type, task objective, and user operation feedback.

[0035] This step is performed by an integrated multimodal data acquisition module. This module includes, but is not limited to, the following devices: Electroencephalography (EEG) acquisition device: typically an EEG cap equipped with multiple (e.g., 32 or 64) electrodes. The electrodes are positioned on the user's scalp according to the international 10-20 system standard for non-invasive recording of electrical activity in the cerebral cortex; the acquired user signals are voltage time series at the microvolt level, reflecting the synchronous activity of a large population of neurons. Optionally, to enhance the capture of blood oxygenation changes in brain regions related to motor intent, the device can integrate a near-infrared spectroscopy (fNIRS) probe for synchronous acquisition with the EEG signals, providing richer information on neural activity.

[0036] The electromyography (sEMG) acquisition device typically consists of a set or array of surface electromyography electrodes, attached to the skin of the muscle belly of the major muscle groups (agonists, antagonists, and synergists) involved in the training task. For example, during upper limb elbow flexion and extension training, electrodes are placed on the biceps and triceps brachii. The acquired signals are millivolt-level voltage time series, reflecting the sum of action potentials of motor units during muscle fiber contraction.

[0037] Human-machine motion and force interaction data acquisition equipment: This part of the data is provided by sensors integrated with the rehabilitation robot. For example, a six-dimensional force / torque sensor installed on the end effector of the rehabilitation robot is used to measure the interactive forces such as pushing, pulling, and twisting applied by the user to the robot in real time; at the same time, the robot's joint encoder or an external optical / inertial motion capture system (such as VICON or Xsens) is used to accurately record the kinematic data of the user's limbs, including joint angles, angular velocities, and the position and posture of the limb end.

[0038] The system provides a unified time reference for all acquisition devices through a master clock signal, ensuring that every frame of EEG data, EMG data, and motion / force data acquired has a precise timestamp, which enables accurate time alignment in subsequent analysis.

[0039] Human-computer interaction task data provides contextual information for rehabilitation training tasks. This information gives the model prior knowledge of "what the task is" and "where the goal is," enabling the model's simulation behavior to have a clear purpose. Human-computer interaction task data mainly includes three aspects: Task type: Used to define the nature of the current rehabilitation training; for example, the task can be "trajectory tracking", "goal arrival", "rhythmic movement" or "functional task simulation" (such as drinking water, combing hair); each task type corresponds to different movement patterns and control strategies.

[0040] Task objective: For a "target arrival" task, the task objective may be a coordinate point in three-dimensional space. For a "trajectory tracking" task, the objective might be a predefined spatial trajectory function. For "rhythmic movement," the task objective might be the desired movement frequency. and amplitude .

[0041] User operation feedback: This part of the data records the user's active control over the task process through interactive devices (such as buttons, touch screens, voice commands). For example, the user can issue commands such as "start task", "pause task" or "confirm completion" through buttons.

[0042] For example, suppose a user is performing a "desktop retrieval" task guided by a rehabilitation robot, at a certain point in time... The multimodal data acquisition module simultaneously recorded the following data: One dimension is EEG data matrix Where 64 is the number of electrode channels. It is the number of sampling points within a short time window, and the data unit is microvolts (µV). ).

[0043] One dimension is electromyography data matrix Where 8 represents the number of channels in the target muscle, and the data unit is millivolts (mV). ).

[0044] A 6-dimensional force / torque vector The force is measured by a force sensor. The unit of force is Newton (N), and the unit of torque is Newton-meter (N·m).

[0045] A kinematic vector containing the angles of the shoulder, elbow, and wrist joints. The angle is measured by an encoder or motion capture system, and the unit of angle is radians (rad).

[0046] All of this data This constitutes the time A complete snapshot of the system's state. Throughout the rehabilitation training process, the system continuously performs this step at a high sampling rate (e.g., 1000Hz for EEG / sEMG and 100Hz for motion / force data), thereby generating a continuous data stream that provides input for all subsequent processing steps.

[0047] Before the task begins, the rehabilitation robot's vision system (or one set by the therapist) presents a target water glass in a virtual reality or augmented reality interface. At this point, the human-computer interaction task data is defined as follows: Task type: Task_Type='Target_Reaching'.

[0048] Task objective: The coordinates of the water cup in the robot's workspace are determined as follows: (Unit: meters), and this vector is then passed to the cognitive-motor model.

[0049] User feedback: When the user is ready to start the task, they confirm by pressing a handheld button. The system then records an event User_Feedback='Start_Signal' with a timestamp of [time stamp missing]. This signal can serve as the initiation signal to trigger the cognitive-motor model to begin generating motion intention simulations.

[0050] This contextual information enables digital twin simulations to be purposeful; for example, a cognitive-motor model can receive a task objective... Then, its internal motion planning module can start calculating the optimized motion trajectory from the current hand position to the target position, thereby generating a simulated motion intention that is highly relevant to the task, and so on.

[0051] S200: Based on user signals and a preset recognition system, generate digital signals, including digital motion signals and digital activation signals; and generate simulation signals based on a preset model and human-computer interaction task data, including simulated motion signals, simulated activation signals, and simulated motion / force data. In this embodiment, the following sub-steps are specifically included: S210 generates digital motion signals that represent the user's motion intentions based on real-time EEG signals and a preset motion signal recognition system.

[0052] From complex, noisy raw EEG signals, the system decodes the user's "high-level commands" regarding movement, i.e., the user's movement intentions. These intentions can be decisions about "when to move," "which direction to move," or "how to move." The decoding result is quantified into a "digital motion signal" and input into a subsequent digital twin model, serving as a "real-world label" for comparison with the model's simulation results. The motion signal recognition system is a pre-trained signal processing and pattern recognition model. Its processing flow typically includes the following stages: Preprocessing: The raw EEG signal is filtered to remove noise and artifacts. Common processing methods include: Bandpass filtering: Filters out frequency components that are not related to neural activity. For example, a 0.5-40Hz bandpass filter can be used to retain most of the motor-related EEG rhythms (such as μ rhythms and β rhythms) while removing DC drift and high-frequency noise.

[0053] Notch filter: filters out power frequency interference of 50Hz or 60Hz.

[0054] Spatial filtering: Utilizing information from multiple electrode channels to enhance the signal-to-noise ratio, such as Laplace filtering or independent component analysis (ICA) to remove artifacts from electrooculography, electromyography, etc.

[0055] Feature extraction: Extracting features from the preprocessed signal that can effectively distinguish different motion intentions. Commonly used features include: Motor-related cortical potentials: Approximately 1-2 seconds before the onset of voluntary movement, a slow negative potential drift can be observed in the central motor cortex. Its amplitude, slope, and latency can serve as characteristics of motor intent.

[0056] Event-related desynchronization / synchronization (ERD / ERS): During motor imagery or actual movement, the energy of the μ rhythm (8-13Hz) and β rhythm (14-30Hz) in the sensorimotor cortex decreases significantly (ERD) or rebounds and increases after movement (ERS). Movements in different limbs or directions correspond to specific ERD / ERS patterns in different brain regions. These rhythmic energy features can be extracted using short-time Fourier transform, wavelet transform, or common spatial pattern (CSP) algorithms.

[0057] Pattern classification: Extracted features are input into a classifier to determine the user's specific movement intention. Commonly used classifiers include Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), or deep neural networks (such as Convolutional Neural Networks CNN).

[0058] The output "digital motion signal" It is a time series vector, the specific form of which depends on the complexity of the decoding task.

[0059] For example, in a simple left- or right-hand motor imagery task, the goal of the recognition system is to determine whether the user intends to move their left or right hand. After preprocessing, the system calculates the energy of the μ rhythm on the C3 and C4 electrodes in the central region. The optimal spatial filter is found using the CSP algorithm to maximize the variance difference between the left and right hand imagery tasks in the μ band, resulting in the CSP feature vector. Subsequently, a trained LDA classifier receives the CSP feature vector and outputs a probability value. To indicate at a point in time The user's intention is the probability of moving their right hand.

[0060] Digital motion signals can be represented as:

[0061] Suppose at a certain moment Classifier output If this vector is detected, the system determines that the user currently has a strong intention to move their right hand. This refers to the digital motion signal at that moment, which will be sent to the cognitive-motor model for comparison. This step realizes the transformation from abstract neural activity to concrete, quantifiable motion commands, and is the first key interface connecting the real user's brain with the digital twin world.

[0062] S220: Based on real-time electromyography signals and a preset activation signal recognition system, it generates digital activation signals that characterize the muscle activation state.

[0063] Here, the raw, oscillating surface electromyography (EMG) signal is transformed into a smooth signal that quantifies muscle activation level or contraction strength—the "digital activation signal." This signal represents muscle-level information and serves as a "real-world label" for comparison with digital twin model simulation results. The processing flow of an activation signal recognition system typically includes: Preprocessing: Similar to EEG, the raw sEMG signal also needs filtering. A bandpass filter (e.g., 20-450Hz) is typically used to preserve the main sEMG spectrum and remove motion artifacts and high-frequency noise; a notch filter is also needed to remove power frequency interference.

[0064] Signal processing: Converting the processed sEMG signal into a measure of activation level, specifically including: Rectification: Taking the absolute value of all negative values ​​of a signal to obtain the instantaneous energy of the signal.

[0065] Smoothing / Envelope Extraction: The rectified signal is low-pass filtered (e.g., a Butterworth filter with a cutoff frequency of 2-10 Hz) or a moving average window is used to obtain a smooth envelope. This envelope reflects the trend of muscle activation level over time.

[0066] Normalization: To make comparisons between different muscles, different users, or different measurements, the sEMG envelope needs to be normalized. A common method is to divide it by the sEMG amplitude measured at maximal voluntary contraction (MVC) of the muscle; the normalized signal value is typically between 0 and 1 (or higher, indicating supermaximal contraction), and can be directly interpreted as a percentage of muscle activation level. In addition, the firing time and frequency of individual motor units can be identified from the sEMG signal, providing more refined neural drive information.

[0067] The output "digital activation signal" It is a vector time series whose dimension is equal to the number of muscle channels monitored.

[0068] For example, during elbow flexion and extension training, the system monitors sEMG of the biceps and triceps.

[0069] At any moment The acquired raw sEMG signal was preprocessed, rectified, and filtered by a 4th-order low-pass filter with a cutoff frequency of 6Hz to obtain the envelope value of the biceps brachii. and the envelope value of the triceps brachii .

[0070] Assuming the sEMG envelope amplitude corresponding to the user's MVC is measured in advance, it is as follows: and .

[0071] After normalization, the digital activation signal vector at that moment is obtained: .

[0072] This vector represents the time at time... The user's biceps activation level is approximately 30% of their maximum capacity, while the triceps activation level is approximately 12%, and so on.

[0073] S300: Based on the difference between the digital signal and the simulation signal, perform online correction on the corresponding preset model.

[0074] This step specifically includes the following sub-steps: S310: Compare the digital motion signal with the simulated motion signal output by the cognitive motion model based on the human-computer interaction task data to obtain a first difference, and use a first machine learning algorithm to correct the cognitive-motion model online based on the first difference.

[0075] This step uses motor intentions (digital motion signals) decoded from the real brain to calibrate a cognitive-motor model that simulates brain behavior.

[0076] Cognitive-motor models are used to simulate the neural computation process from receiving a task objective to generating an internal motor intention. These models can be based on optimal control theory or complex deep learning models (such as recurrent neural networks (RNNs) or their variants, LSTMs). Their input is human-computer interaction task data acquired from [the system / platform] (such as target points). The output is a "simulated motion signal". The format of this simulated signal is the same as the previously generated digital motion signal. Completely identical, so as to facilitate direct comparison.

[0077] The correction process is as follows: Simulation Execution: After receiving the task objective, the cognitive-motor model begins internal motion planning and decision-making simulation, and outputs a simulated motion signal sequence over time. .

[0078] Comparison and difference calculation: at each time step The system will output the simulation results of the model. Compared with the results obtained from real EEG decoding By comparing, the first difference is obtained. The calculation method can be a simple vector subtraction, or a more complex distance metric, such as cross-entropy loss (if the signal is a probability distribution):

[0079] Online correction: The first machine learning algorithm receives this difference signal. Then adjust the internal parameters of the cognitive-motor model (e.g., the weights of the neural network). ), so that the output of the next simulation Closer to This reduces the difference. This process typically uses gradient descent algorithms.

[0080] in, L is the learning rate, and L is the loss function defined based on the difference.

[0081] For example, for the left-hand and right-hand recognition task, the cognitive-motor model uses a small LSTM network, which takes a task instruction (e.g., "prepare to turn right") as input and outputs a two-dimensional probability vector:

[0082] At any moment The actual EEG decoding result is .

[0083] Meanwhile, the cognitive-motor model is based on task instructions, and its simulation output under the current parameters is: The model “thinks” that the probability of the user wanting to move their right hand is 70%.

[0084] Therefore, the first difference vector is: This difference indicates that the model underestimated the strength of the user's intention to move their right hand.

[0085] The first machine learning algorithm (e.g., an online backpropagation algorithm) then uses this difference to calculate the loss function (e.g., mean squared error). The gradient of the LSTM network is calculated, and the weights of the LSTM network are updated accordingly. After multiple iterations and corrections, the model's output will gradually shift towards... And so on.

[0086] S320: Compare the digital activation signal with the simulated activation signal output by the skeletal muscle control model based on the digital motion signal to obtain a second difference, and use a second machine learning algorithm to correct the skeletal muscle control model online based on the second difference.

[0087] This step utilizes activation levels (digital activation signals) extracted from real muscle sEMG to "calibrate" the skeletal muscle control model that simulates the conversion process from neural signals to muscle activation. The skeletal muscle control model receives signals from the upper-level cognitive-motor model (or directly uses digitally decoded signals from real muscle). Using the output (which reduces error accumulation) as input, this high-level motion command is transformed into fine-grained activation commands for each relevant muscle. This model can be a mapping network based on muscle synergy theory or a complex neural network, whose output is a "simulated activation signal." Its format is the same as the aforementioned generated digital activation signal. Consistent.

[0088] The correction process includes: Simulation execution: The skeletal muscle control model receives the input motion intention, calculates the theoretical activation level of each muscle, and outputs a simulated activation signal sequence. .

[0089] Comparison and difference calculation: at each time step The system will output the simulation results of the model. Compared with those extracted from real sEMG Compare and calculate the second difference. :

[0090] Online correction: The second machine learning algorithm utilizes the difference To adjust the internal parameters of the skeletal muscle control model

[0091] To minimize the gap between simulated activation and real activation:

[0092] For example, using the elbow flexion and extension task, the skeletal muscle control model employs a feedforward neural network. Its input is a scalar representing the desired elbow flexion torque (generated by the upper-layer model according to task requirements), and its output is a two-dimensional activation vector. .

[0093] At any moment The upper-level model instruction requires an elbow flexion movement. The skeletal muscle control model calculates the simulation activation signal based on its current parameters. The model suggests that in order to complete the task, the biceps brachii needs 45% activation, while the triceps brachii needs 5% activation to maintain joint stability.

[0094] However, the digital activation signal measured from real sEMG is At this point, the second difference vector is This indicates that the model overestimates the amount of biceps activation required to complete the task (possibly because the patient has high biceps efficiency or compensation from other synergistic muscles), while underestimating the co-activation level of the antagonistic muscle (triceps) (which is common in stroke patients and manifests as poor muscle control coordination).

[0095] The second machine learning algorithm then uses this difference. The weights of the neural network are updated, and through continuous correction, the model will learn the patient's unique muscle recruitment patterns and coordination strategies. For example, it will reduce the reliance on the biceps brachii and appropriately increase the activation of the triceps brachii when generating the same torque, so that the muscle control strategy of the digital twin is more in line with the user's real physiological condition.

[0096] S330: Compare the real-time human-computer motion and force interaction data with the simulated motion and force data output based on a musculoskeletal model according to the simulation activation signal to obtain a third difference, and use a third machine learning algorithm to correct the musculoskeletal model online based on the third difference.

[0097] This step utilizes limb movements and interactive forces measured from real sensors (real-time human-computer motion and force interaction data) to calibrate a musculoskeletal model simulating limb biomechanical responses. The musculoskeletal model (also known as a musculoskeletal simulation system) is a complex model based on multi-rigid-body dynamics and biomechanical principles (such as models built on platforms like OpenSim), which receives simulation activation signals output from the upper-level skeletal muscle control model. As input, it includes the geometry of the skeleton, mass and inertia parameters, joint motion constraints, and the origin and insertion points of muscles, force-length-velocity relationships (such as the Hill-type muscle model); based on the input muscle activation signals, the model simulates the kinematics of the limb (joint angles) through forward dynamics calculations. ) and dynamics (interaction forces with the environment) The result is the "simulation motion and force data".

[0098] The correction process is as follows: Simulation execution: The musculoskeletal model receives the simulation activation signal, runs a forward dynamics simulation, and outputs simulation motion and force data. .

[0099] Comparison and difference calculation: at each time step The system combines the simulation output of the model with the motion and force data measured by the real sensors in the S100. Compare and calculate the third difference. :

[0100] Online correction: A third machine learning algorithm utilizes this difference. To adjust the internal parameters of the musculoskeletal model These parameters may include the muscle's maximum isometric contractile force, tendon relaxation length, bone segment mass, or center of mass location, etc.

[0101] For example, in an elbow flexion-extension task, the rehabilitation robot applies a constant resistance; at time... The simulated activation signal output by the skeletal muscle control model is The musculoskeletal model then receives this signal and, based on its current biomechanical parameters (e.g., maximum biceps force),... The angular acceleration of the elbow joint was calculated through simulation, and the simulated joint angle was obtained by integration. rad, and the simulation force of the end effector interacting with the robot. .

[0102] However, the data measured by the actual sensors is: the actual joint angle. rad, actual interaction force At this point, the third difference is This indicates that, at the same level of muscle activation, the model's simulated movements (with greater angles and forces) are more impactful than those of a real user, which may imply the maximum biceps strength set in the model. This is too high for the user (e.g., due to muscle atrophy).

[0103] Then a third machine learning algorithm (e.g., a Kalman filter or gradient descent optimizer) will adjust the algorithm based on this error. Make a negative adjustment, for example, update it to 580N; in this way, the musculoskeletal model can continuously adjust its biomechanical parameters to match the user's actual physical condition, such as muscle strength, joint stiffness, etc., thereby building a truly personalized, high-fidelity biomechanical digital twin.

[0104] S400: Acquires corrected cognitive-motor digital model parameters, skeletal muscle control digital model parameters, and musculoskeletal digital model parameters, as well as historical training data; and based on this data, generates user rehabilitation assessment data through the rehabilitation robot assessment system.

[0105] Through continuous online adjustments, the three-tiered model parameters of the digital twin can accurately reflect the user's current neuromuscular-skeleton state. These model parameters, combined with long-term training data, are then used for rehabilitation assessments. The acquired data includes: Corrected model parameters: Cognitive-Motor Model Parameters : Used to indicate cognitive features such as a user's reaction time, decision-making strategies, and motor planning abilities.

[0106] Parameters of skeletal muscle control model : Used to indicate characteristics at the neural control level, such as the user's muscle coordination patterns, activation efficiency, and co-activation level.

[0107] Musculoskeletal model parameters : Used to indicate biomechanical characteristics such as muscle strength, joint range of motion, and tissue viscoelasticity of users.

[0108] Historical training data includes all training records of the user over a period of time, such as task completion time, success rate, trajectory tracking error, force generated, sEMG integral value, etc.

[0109] Based on this data, the assessment system calculates rehabilitation assessment indicators: Motor function score: A functional score obtained by running standard clinical assessment tasks (such as a virtual version of the Fugl-Meyer assessment) in a digital twin model.

[0110] Muscle activation symmetry index: For hemiplegic patients, the activation patterns of muscles on the affected and unaffected sides can be compared (or compared with a standard database); for example, the activation ratio of the biceps and triceps during training can be calculated and compared with the normal pattern to quantify their coordination impairment.

[0111] Task completion efficiency: By analyzing energy consumption in the digital twin model, the economic efficiency of user movement is evaluated.

[0112] Indicators of neuroplasticity: These quantify changes in cortical representations by tracking changes in parameters of a cognitive-motor model over time; for example, if the model shows that the activation areas required to complete the same task become closer to the normal pattern, this may be direct evidence of neural remodeling.

[0113] For example, the evaluation system found that during the past week of training, the maximum strength parameter of the affected biceps brachii in the user's musculoskeletal model... The resistance level steadily increased from 450N to 480N. Simultaneously, in the skeletal muscle control model, the triceps co-activation parameter required to complete the standard elbow flexion task decreased from 30% to 18%. Historical data showed a 20% reduction in task completion time. Based on this data, the system generated a rehabilitation assessment report: "User muscle strength improved by approximately 6.7%, muscle control coordination improved, and movement efficiency increased; it is recommended to proceed to the next stage and appropriately increase resistance training," and so on.

[0114] S500: Based on the rehabilitation assessment data, a personalized rehabilitation training plan is generated through the rehabilitation robot solution system; and the rehabilitation robot is driven to execute the plan, forming a closed-loop control.

[0115] This step transforms all the results of the aforementioned perception, modeling, simulation, correction, and evaluation into actual rehabilitation robot actions. This step is completed collaboratively by the rehabilitation robot's solution system, training system, and motor system.

[0116] Solution generation: The rehabilitation robot solution system receives the aforementioned rehabilitation assessment data. The system contains an expert knowledge base or a reinforcement learning decision model. Then, based on the assessment report, it automatically generates or adjusts the rehabilitation training plan for the next stage.

[0117] Training intensity: If the assessment shows increased muscle strength, the program system may increase the resistance applied by the rehabilitation robot during movement.

[0118] Training frequency: If the assessment shows that the user is prone to fatigue (e.g., sEMG spectrum shifts to lower frequencies), the system may suggest reducing the training frequency or increasing rest time.

[0119] Task difficulty level: If the assessment shows improved motor coordination, the system may introduce more complex tasks, such as tracking curved trajectories in three-dimensional space, or requiring faster movement speeds.

[0120] Assisted / Resistance Mode: The system can dynamically switch modes based on the user's real-time performance. For example, it can provide an "assisted start" mode when the user has difficulty initiating the movement, and switch to "resistance training" mode to strengthen muscles once the user is able to exert force actively.

[0121] Execution and Interaction: 1. The training system decomposes the newly generated scheme into specific task data (such as new target points and new trajectories) and training parameters (such as resistance magnitude). (Assist force threshold).

[0122] 2. Task data is presented to the user through an interactive system (such as screen display) to realize task information interaction.

[0123] 3. The training parameters are sent to the motion system (i.e., the robot's underlying controller). Based on these parameters, the motion system precisely controls the motor output to drive the robot links and enable real-time, controlled physical motion and force interaction with the user.

[0124] For example, based on the evaluation results, the scheme system generates a new training scheme: Task type: Upgraded from two-dimensional "straight line arrival" to three-dimensional "spiral trajectory tracking".

[0125] Training intensity: Increase the baseline resistance from 5N to 8N.

[0126] Assist / Resistance Mode: When the user's biceps brachii receives digital activation signals When the value is below 0.2, the robot provides The robot applies an auxiliary force of N; when the signal is higher than 0.2, it applies a constant resistance of 8N.

[0127] The rehabilitation robot then begins implementing the new program, with the user training under new tasks and biomechanical conditions. During this process, all the aforementioned steps continue to run at high frequency in the background, constantly collecting new data, refining the digital twin model, and performing microscopic evaluations, thus forming a complete closed loop. If, under the new program, the system detects an undesirable compensatory pattern in the user (e.g., overactivation of shoulder muscles detected by sEMG), the closed-loop system reacts immediately, potentially fine-tuning the assist / resistance parameters within seconds to guide the user back to the correct movement pattern, and so on.

[0128] Example 2 After obtaining the corrected model, the correction process may result in overfitting, leading to reduced prediction accuracy. Unlike Example 1, this example performs a test on the corrected model. Specifically: S341: Constructing a virtual exploration task set and benchmark response database The system pre-constructs a standardized Virtual Exploration Task Set (VPTS). This task set is independent of the user's daily rehabilitation training tasks and aims to stimulate various response characteristics of the user's neuromuscular-skeleton system in a standardized manner. At the same time, for each exploration task, the system maintains a benchmark database extracted from large-scale clinical data. This database describes the expected physiological signal patterns (such as ERD / ERS patterns in EEG, co-activation patterns in sEMG, etc.) of healthy individuals or similar patient groups under a specific exploration task.

[0129] S342: Periodically perform online model validation The system initiates an online model validation process at a low frequency (e.g., after each complete rehabilitation training session, or after every 30 minutes of training). This process does not involve physical interaction with real users and is conducted entirely in the computing environment.

[0130] Model Snapshot: The system first obtains a complete "snapshot" of the digital twin model (including the cognitive-motor model, skeletal muscle control model, and musculoskeletal model) at the current moment after it has been fully corrected.

[0131] Virtual exploration tasks: The system drives this model snapshot and executes all virtual exploration tasks in the VPTS one by one in the simulation environment.

[0132] Generate predictive responses: For each exploration task, the model snapshot generates a complete set of predictive, multi-level physiological response data, including predicted EEG patterns, muscle activation patterns, and limb movement trajectories.

[0133] S343: Calculate model evaluation metrics and generate validation reports. The system compares the predicted responses generated above with the baseline response database corresponding to VPTS in multiple dimensions, and calculates a series of evaluation indicators for quantifying model quality: Generalization performance score ( ): Used to measure the model’s performance on untrained exploration tasks, specifically by calculating the difference between the predicted response and the baseline response (e.g., Dynamic Time Warped (DTW) distance); the smaller the difference, the higher the score.

[0134] Predicted stability score ( ): Used to evaluate the consistency of a model's output under similar inputs. It is calculated by applying a small perturbation to the input of the exploration task and observing the degree of jitter in the model's output. The smaller the jitter, the higher the score.

[0135] Physiological explainability score ( This section examines whether the model's internal working mechanism conforms to known physiological principles. For example, when simulating a rapid elbow flexion task, does the output of the skeletal muscle control model clearly reflect the "triphasic activation pattern" where the agonist muscle (biceps brachii) is activated first, followed by braking of the antagonist muscle (triceps brachii)? The higher the conformity, the higher the score.

[0136] The system combines these scores to generate an online verification report on the current quality of the digital twin model.

[0137] S344: Adaptive adjustment strategy for model correction based on validation report Based on the validation report generated by S343, the system dynamically adjusts the "meta-parameters" (especially the strength of the regularization term) used in subsequent model corrections. Regularization is a method used in machine learning (such as L1 / L2 regularization) to prevent model overfitting. Specifically, it involves adding a penalty term to the loss function to constrain the complexity of the model.

[0138] The adaptive adjustment logic is as follows: If the validation report shows a generalization performance score Lower values ​​indicate that the model may have begun to overfit the current training task. In this case, the system will automatically increase the strength of the regularization term in the aforementioned loss function (e.g., increase the L2 regularization coefficient). This will allow subsequent model correction processes to focus more on maintaining the smoothness and simplicity of model parameters while reducing training errors, thereby improving its adaptability to new tasks.

[0139] If the validation report shows a generalization performance score High: This indicates that the model is in good condition and has good generalization ability. At this time, the system can appropriately reduce the regularization strength, allowing the model to fit the user's individual characteristics more precisely in order to pursue higher personalized accuracy.

[0140] For example, a user performs repetitive "horizontal trajectory tracking" training for a whole week.

[0141] Initially: Through continuous refinement, the model's prediction error for this task gradually decreases.

[0142] Model validation triggered: After one week of training, the system initiates online model validation.

[0143] Problem identified: When performing the "vertical anti-gravity lift" task in the virtual exploration task set, the muscle activation patterns predicted by the model snapshots were severely inconsistent with physiological benchmarks, demonstrating extremely poor generalization ability. In this case, the validation report gave a low score. Fraction.

[0144] Adaptive adjustment: The system identifies the risk of overfitting and automatically adjusts the L2 regularization coefficient in the skeletal muscle control model algorithm. Increased from 0.001 to 0.01.

[0145] Subsequent impact: In subsequent training, even on horizontal tasks, the model correction process will tend to learn muscle coordination strategies that are more in line with the basic principles of biomechanics due to stronger regularization constraints, rather than simply memorizing specific activation patterns for horizontal tasks, and so on.

[0146] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0147] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0148] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to perform the steps in any of the above method embodiments when executed.

[0149] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0150] Embodiments of the present invention also provide an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0151] In one exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0152] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0153] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0154] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0155] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0156] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0157] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multimodal neural-driven control method for a rehabilitation robot, characterized in that, include: The system acquires human-computer interaction task data and real-time user signals, wherein the user signals include electroencephalogram (EEG) signals, electromyogram (EMG) signals, and human-computer motion / force interaction data, and the human-computer interaction task data includes task type, task objective, and user operation feedback. Based on the user signal and the preset recognition system, a digital signal is generated, wherein the digital signal includes a digital motion signal and a digital activation signal; and a simulation signal is generated based on the preset model and the human-computer interaction task data, wherein the simulation signal includes a simulation motion signal, a simulation activation signal, and simulation motion / force data. Based on the difference between the digital signal and the simulation signal, or based on the difference between the human-computer motion / force interaction data and the simulation motion / force data, the corresponding preset models are corrected online. Rehabilitation assessment data is generated based on the parameters of the corrected model and the acquired historical training data; Task data and training parameters are generated based on the rehabilitation assessment data. The task data is used for task information interaction between the rehabilitation robot and the user, and the training parameters are used for motion / force interaction between the rehabilitation robot and the user.

2. The method according to claim 1, characterized in that, The difference between the digital signal and the simulated signal includes: The digital motion signal is compared with the simulated motion signal output by the cognitive-motor model based on the human-computer interaction task data to obtain a first difference, and the cognitive-motor model is corrected online based on the first difference using a first machine learning algorithm, wherein the preset model includes the cognitive-motor model. The digital activation signal is compared with the simulated activation signal output by the skeletal muscle control model based on the digital motion signal to obtain a second difference. The skeletal muscle control model is then corrected online based on the second difference using a second machine learning algorithm. The preset model includes the skeletal muscle control model. Online correction of the preset model based on the difference between the human-computer motion / force interaction data and the simulated motion / force data includes: The real-time human-computer motion / force interaction data is compared with the simulated motion / force data output by the musculoskeletal model based on the simulation activation signal to obtain a third difference. A third machine learning algorithm is then used to correct the musculoskeletal model online based on the third difference. The preset model includes the musculoskeletal model.

3. The method according to claim 1, characterized in that, The generation of digital signals based on the user signals and the preset recognition system includes: Based on the electroencephalogram (EEG) signals and a preset motion signal recognition system, a digital motion signal representing the user's motion intention is generated. Based on the electromyographic signals and a preset activation signal recognition system, a digital activation signal representing the muscle activation state is generated.

4. The method according to claim 2, characterized in that, The first, second, and third machine learning algorithms each include a recurrent neural network, a long short-term memory network, a Gaussian process regression, a reinforcement learning policy gradient method, or an online adaptive filter.

5. The method according to claim 1, characterized in that, After performing online corrections on the preset model, the method further includes: Periodically obtain model snapshots of the corrected preset model; Predicted physiological response data are generated based on the model snapshot; The predicted physiological response data is compared with the preset baseline response data to calculate the evaluation index of the model snapshot; The meta-parameters are adjusted according to the evaluation indicators, wherein the meta-parameters are used to adjust the preset model online.

6. A multimodal neural-driven rehabilitation robot control system, characterized in that, include: The data acquisition module is used to acquire human-computer interaction task data and real-time user signals, wherein the user signals include electroencephalogram (EEG) signals, electromyogram (EMG) signals, and human-computer motion / force interaction data, and the human-computer interaction task data includes task type, task objective, and user operation feedback. The signal generation module is used to generate digital signals based on the user signals and a preset recognition system, wherein the digital signals include digital motion signals and digital activation signals; and to generate simulation signals based on a preset model and the human-computer interaction task data, wherein the simulation signals include simulated motion signals, simulated activation signals, and simulated motion / force data. The correction module is used to correct the corresponding preset models online based on the difference between the digital signal and the simulation signal, or based on the difference between the human-computer motion / force interaction data and the simulation motion / force data. The rehabilitation assessment module is used to generate rehabilitation assessment data based on the parameters of the corrected model and the acquired historical training data. The task generation module is used to generate task data and training parameters based on the rehabilitation assessment data. The task data is used for task information interaction between the rehabilitation robot and the user, and the training parameters are used for motion / force interaction between the rehabilitation robot and the user.

7. The system according to claim 6, characterized in that, The difference between the digital signal and the simulated signal includes: The digital motion signal is compared with the simulated motion signal output by the cognitive-motor model based on the human-computer interaction task data to obtain a first difference, and the cognitive-motor model is corrected online based on the first difference using a first machine learning algorithm, wherein the preset model includes the cognitive-motor model. The digital activation signal is compared with the simulated activation signal output by the skeletal muscle control model based on the digital motion signal to obtain a second difference. The skeletal muscle control model is then corrected online based on the second difference using a second machine learning algorithm. The preset model includes the skeletal muscle control model. Online correction of the preset model based on the difference between the human-computer motion / force interaction data and the simulated motion / force data includes: The real-time human-computer motion / force interaction data is compared with the simulated motion / force data output by the musculoskeletal model based on the simulation activation signal to obtain a third difference. A third machine learning algorithm is then used to correct the musculoskeletal model online based on the third difference. The preset model includes the musculoskeletal model.

8. The system according to claim 6, characterized in that, The generation of digital signals based on the user signals and the preset recognition system includes: Based on the electroencephalogram (EEG) signals and a preset motion signal recognition system, a digital motion signal representing the user's motion intention is generated. Based on the electromyographic signals and a preset activation signal recognition system, a digital activation signal representing the muscle activation state is generated.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program is configured to perform the method described in any one of claims 1 to 5 when executed.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method described in any one of claims 1 to 5.

Images