A synergistic activation-based upper limb synergist assessment and training control method and system

Abstract

The application discloses a kind of upper limb coordination evaluation and training control method and system based on synergistic activation, belong to biomedical engineering man-machine interaction field;Method includes: obtaining the surface myoelectricity and terminal force data of target object execution isometric force task, estimates habit force vector and extracts current synergistic activation coefficient;Obtain theoretical synergistic force vector and reference synergistic activation coefficient;In first mode (W-based), based on difficulty coefficient, habit force and theoretical synergistic force are combined, and first target force vector is generated;Real-time calculation force direction angle error and synergistic specificity index, when angle error is less than threshold value and specificity index meets plateau condition, switch to second mode (C-based);In second mode, difference vector of activation coefficient is projected to terminal force space to obtain force correction vector, and then generate second target force vector;The present application realizes the smooth transition of high-dimensional neural features to low-dimensional physical guidance, effectively improves the accuracy of man-machine interaction and state evaluation.

Description

Technical Field

[0001] This invention belongs to the field of human-computer interaction technology in biomedical engineering, specifically relating to an upper limb collaborative assessment and training control method and system based on collaborative activation. Background Technology

[0002] Muscle synergy theory posits that complex movements can be generated by combining several synergistic primitives with varying time-varying activation coefficients. Existing research indicates that while some motor function-limited subjects retain some stable synergistic structures, the timing of synergistic invocation, activation ratios, and task correspondences change. This manifests in end-effector force control as insufficient force in the primary direction, enhanced coupling in non-target directions, and increased aberrant co-activation. Multi-channel surface electromyography (SEMG) combined with end-effector force modeling provides a technical foundation for converting high-dimensional neuromuscular activation into observable and interactive low-dimensional control quantities.

[0003] Existing upper limb assessment or training equipment mainly relies on trajectory error, joint angle error, or single muscle amplitude threshold as the primary feedback basis, which has three shortcomings: First, it lacks a unified framework for individualized coupling modeling of synergistic structures, synergistic activation, and end force direction; second, it lacks a mechanism to automatically generate progressive target vectors based on the difference between the subject's current habitual force and the reference synergistic target; and third, it lacks a quantifiable mode switching criterion for transitioning from coarse to fine adjustment, resulting in insufficient stability and repeatability of training control.

[0004] Recent work has attempted to reduce abnormal co-activation through electromyography-guided training or muscle activation-force / motor mapping, but existing publicly available solutions are usually more inclined to single-stage training, single feedback variables, or specific experimental procedures, and have not yet formed an integrated system architecture of habit force initialization, integrated musculoskeletal projection, dual-mode target generation, adaptive switching, and phased evaluation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for upper limb coordination assessment and training control based on co-activation, thereby solving the problems in the prior art.

[0006] The objective of this invention can be achieved through the following technical solutions: A method for upper limb coordination assessment and training control based on co-activation includes the following steps: Acquire surface electromyography signals and distal force data of the target object when performing a preset isometric force application task involving multiple task directions; Based on the end force data under targetless feedback conditions, the habitual force vector of the target object is estimated; and the current co-activation coefficient is extracted based on the surface electromyography signal. Obtain the reference collaboration and reference collaboration activation coefficients corresponding to each task direction; Based on the individualized musculoskeletal model and the aforementioned reference coordination, the theoretical coordination force vector of the target object is estimated; In the first mode, based on a preset difficulty coefficient, the habitual force vector and the theoretical collaborative force vector are weighted and combined to generate a first target force vector; The current end force direction of the target object is acquired in real time, the angle error between the current end force direction and the first target force vector is calculated, and the cooperative specificity index is calculated based on the current cooperative activation coefficient. When the included angle error is less than a preset angle threshold, and the synergistic specificity index meets the plateau condition within a preset number of consecutive evaluation windows, the control mode will be switched from the first mode to the second mode. In the second mode, the difference vector between the reference co-activation coefficient and the current co-activation coefficient is calculated, and the difference vector is projected onto the end force space spanned by the theoretical co-activation force vector to obtain the force correction vector. The second target force vector is generated by superposition and normalization of the current end force vector and the force correction vector.

[0007] Furthermore, the steps for obtaining the theoretical synergistic force vector include: A basic limb musculoskeletal model is selected, which includes skeletal segments, joint degrees of freedom, muscle attachment points, muscle path points, muscle biomechanical parameters, and the mapping relationship between joint posture and end force. Based on the target object's upper arm length, forearm length, hand length, and preset task posture, the scale factor of each skeletal segment is calculated, and the skeletal segments, muscle attachment points, and muscle path points in the basic limb musculoskeletal model are geometrically scaled based on the scale factor to obtain the scaled basic limb musculoskeletal model. Based on the surface electromyography (EMG) signals and distal force data collected from the target object in a preset isometric force application task, the surface EMG envelope is used as the muscle activation input and the measured distal force is used as the output constraint. An error function between the predicted distal force and the measured distal force is constructed. One or more of the muscle force scale factor, lever arm correction coefficient, or equivalent mapping parameters from muscle activation to distal force in the scaled basal musculoskeletal model are constrained and optimized to obtain an individualized musculoskeletal mapping model. The individualized musculoskeletal mapping model is used to calculate the end force vector corresponding to the activation of each muscle unit. The theoretical collaborative force vector for the corresponding task direction is obtained by weighting and summing the collaborative weights of each muscle in the reference collaborative structure matrix with the end force vector when the corresponding muscle unit is activated, and then normalizing the sum.

[0008] Furthermore, when performing decomposition using a nonnegative matrix factorization algorithm, the number of collaborations in the collaborative structure matrix is ​​determined based on the following dual criteria: Calculate the global variance explained rate when reconstructing the surface electromyography signal from the cooperative structure matrix and the current cooperative activation coefficients, wherein the global variance explained rate is greater than a first preset threshold. Furthermore, under the premise of satisfying the aforementioned conditions, when a collaboration is added, the increment of the global variance explanation rate is less than the second preset threshold.

[0009] Furthermore, the process of obtaining the habitual force vector includes: For each task direction, the steady-state portion of the end force data of each test trial is extracted and normalized. The normalized force vector of each trial is averaged by a spherical method to obtain the initial habitual force vector. In each training cycle, the habituation force vector is re-estimated. If the angle between the updated habituation force vector and the habituation force vector of the previous cycle is less than the angle threshold, then the habituation force vector is considered to be the first to be considered. If the previous cycle's habitual force vector is used, then the updated habitual force vector will be used as the starting point for the generation of the target force in the next cycle; otherwise, the updated habitual force vector will be used as the starting point for the generation of the target force in the next cycle.

[0010] Furthermore, the formula for calculating the first target force vector is: in, Let be the first target force vector corresponding to the j-th task direction. Represents the vector normalization operator. This represents the difficulty coefficient corresponding to the j-th task direction. Let represent the habitual force vector in the j-th task direction. This represents the theoretical synergistic force vector corresponding to the j-th task direction.

[0011] Furthermore, the calculation process of the synergistic specificity index includes: The extracted collaborative structure vector within the current evaluation window is matched with collaborative templates in the reference collaborative template library based on similarity. The corresponding collaborative activation coefficients are then categorized into task-specific and object-specific collaborative activation coefficients based on the matching results. The similarity matching includes calculating the cosine similarity or Pearson correlation coefficient between the current collaborative structure vector and the reference collaborative template, and determining the category label of the current collaborative structure vector based on a preset similarity threshold. The collaborative specificity index is calculated using the following formula. : in, This represents the integral value, root mean square value, or weighted area of ​​task-specific coactivation within the evaluation window. This represents the integral, root mean square, or weighted area of ​​object-specific coactivation within the same window. This indicates the correlation between activation coefficients. To prevent tiny constants with a denominator of zero.

[0012] Furthermore, the calculation process for the plateau period condition includes: When satisfied And after L evaluation windows, the synergistic specificity index is determined to have entered a plateau period, wherein This represents the synergistic specificity index corresponding to the nth evaluation window. This is the threshold for the plateau period.

[0013] An upper limb coordination assessment and training control system based on co-activation, executing the above method, includes: The data acquisition and collaborative extraction module is used to acquire surface electromyography (EMG) signals and end force data of the target object when performing a preset isometric force application task involving multiple task directions, and to extract the current collaborative activation coefficient based on the surface EMG signals. The habituation force estimation module is used to calculate and output the habituation force vector corresponding to each task direction based on the end force data under the condition of no target feedback. The target force generation module is used to obtain reference coordination and reference coordination activation coefficients corresponding to each task direction; estimate the theoretical coordination force vector based on the individualized musculoskeletal model; and in the first mode, weight the habitual force vector and the theoretical coordination force vector based on a preset difficulty coefficient to generate a first target force vector; and in the second mode, generate a second target force vector based on the projection operation of the difference vector between the reference coordination activation coefficient and the current coordination activation coefficient to the end force space. The dual-mode control module is used to calculate the angle error between the current end force direction and the first target force vector in real time, and to calculate the cooperative specificity index based on the current cooperative activation coefficient; and to switch the control mode from the first mode to the second mode when the angle error is less than a preset angle threshold and the cooperative specificity index meets the continuous window plateau condition.

[0014] Furthermore, the system also includes: The individualized musculoskeletal mapping module is used to perform cooperative decomposition of the surface electromyography signal to obtain a cooperative structure matrix, select a basic limb musculoskeletal model including skeletal segments, joint degrees of freedom, muscle paths, muscle mechanical parameters and the mapping relationship between joint posture and end force, and geometrically scale the basic limb musculoskeletal model according to the target object's upper arm length, forearm length, hand length and preset task posture. The individualized musculoskeletal mapping module is also used to constrain and calibrate one or more of the muscle force scale factor, lever arm correction coefficient, or equivalent mapping parameters from muscle activation to end force in the scaled basal musculoskeletal model based on the surface electromyography signals and end force data collected by the target object in a preset isometric force application task, so as to obtain an individualized musculoskeletal mapping model. The individualized musculoskeletal mapping module is also used to calculate the end force vector corresponding to the activation of each muscle unit based on the individualized musculoskeletal mapping model, and to perform weighted summation and normalization of the muscle weights in the reference synergistic structure matrix with the corresponding end force vectors to obtain the theoretical synergistic force vector.

[0015] Furthermore, the system also includes: The phase evaluation module is used to calculate and output coupling indices in the corresponding orthogonal directions based on the end force data. : in, The force component applied in the direction of the target. These are the force components in the orthogonal directions. This indicates the correlation between force curves, and d represents the orthogonal direction.

[0016] The beneficial effects of this invention are: 1. The system of the present invention uses an individualized musculoskeletal mapping module to geometrically scale a general musculoskeletal model according to the limb size and joint posture parameters of the target object, and performs parameter calibration by combining multi-channel surface electromyography and end-effector force data. Subsequently, the system maps the extracted high-dimensional reference synergistic structure onto the individualized model and calculates the theoretical synergistic force vector. By converting the abstract and difficult-to-understand high-dimensional muscle synergistic state into an intuitive and perceptible low-dimensional end-effector target force vector, the target object in different motion-restricted states can more easily understand the guidance intent of the device, thus improving the efficiency of human-computer interaction.

[0017] 2. In the W-based (cooperative vector driven) mode of this invention, the system first evaluates the habitual force vector of the target object under the condition of no target feedback; subsequently, the system evaluates the habitual force vector of the target object based on the dynamically adjusted difficulty coefficient. The system generates a first target force vector by weighting the habitual force vector representing the current natural state and the theoretical synergistic force vector representing the ideal state. By introducing habitual force as the starting point and gradually adjusting the difficulty coefficient, the system can generate a target force that matches the current ability of the target object. This overcomes the defect of the initial training target being too difficult due to directly using the ideal trajectory as the fixed target, and improves the accessibility, smoothness and individual consistency of the target force vector generation. It also helps to reduce abnormal compensatory movements induced by unreasonable target setting.

[0018] 3. During operation, the system of this invention calculates external physical characteristics (the angle error between the current end force direction and the target vector) and internal physiological characteristics (the plateau period of the coordination-specific index) in real time. Only when the direction error is less than the preset angle threshold and the coordination-specific index (including the comprehensive ratio of target coordination and abnormal coordination characteristics) remains stable within a continuous window will the system automatically trigger the switch from W-based mode to C-based mode. This joint judgment mechanism based on internal and external dual quantitative indicators establishes an objective adaptive switching criterion from coarse to fine adjustment. It reduces the dependence of mode switching on human experience, improves the problem of insufficient stability and repeatability of training control, and makes different evaluation sessions more comparable.

[0019] 4. In this invention, when the system switches to C-based (co-activation driven) mode, the system calculates the difference vector between the reference co-activation coefficient and the current co-activation coefficient, and projects this difference vector inversely into the end force space spanned by the theoretical co-activation force vector to generate a second target force vector for fine-tuning the current end force. In the later stages of training, even if the error in the macroscopic force direction of the target object is small, the coordination ratio of its internal neuromuscular system may still deviate from the normal reference state. Through spatial projection calculation, the subtle muscle activation ratio imbalance is amplified and mapped into a visualized force correction vector, effectively compensating for the problem of insufficient target resolution in the later stages.

[0020] 5. The system of this invention not only outputs target-guided vector data, but also defines and calculates coupling force indices (i.e., the ratio or correlation between the residual force component in the orthogonal plane and the main force component in the target direction) to characterize the degree of compensation in non-target directions, while recording the phased changes of synergy-specific indices. Within the same system framework, it uniformly outputs multidimensional quantitative results that take into account both motion output accuracy (direction error, coupling index) and differences in internal neural synergy state (synergy-specific indices). This provides a more comprehensive report on state changes for clinical use and provides a good data compatibility foundation for subsequent linkage with other rehabilitation robots or digital therapy platforms. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a diagram of the upper limb collaborative assessment and training control architecture for collaborative activation according to the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 An upper limb coordination assessment and training control system based on co-activation is disclosed. The system preferably comprises a data acquisition end, an analysis end, and an interaction end. The data acquisition end includes at least a multi-channel surface electromyography sensor and an end-effector force sensor; the analysis end performs signal preprocessing, co-decomposition, template calling, musculoskeletal mapping, and target generation; the interaction end outputs the target vector, current vector, and status prompts; the system is capable of forming a closed-loop control link under isometric force application tasks in a fixed posture.

[0025] like Figure 1 As shown, the upper limb coordination assessment and training control system based on co-activation specifically includes: a data acquisition module, a signal preprocessing module, a coordination extraction module, a reference coordination template library, an individualized musculoskeletal mapping module, a habitual force estimation module, a target force generation module, a dual-mode control module, and a feedback output and stage assessment module. The main functions of each module are shown in Table 1 below: Table 1 System Functional Modules Co-activation upper limb coordination assessment and training control architecture, such as Figure 1 As shown, the system has two driving modes: the first is the cooperative vector driving mode (W-based), which generates a progressive target direction based on the habitual force vector and the theoretical cooperative force vector, and is suitable for solving the stage where the target direction deviation is large; the second is the cooperative activation driving mode (C-based), which generates a second target force vector based on the difference between the reference cooperative activation and the current cooperative activation, and is suitable for the state where the direction error is small but the cooperative ratio still has deviation, as shown in Table 2.

[0026] Table 2 Comparison of W-based and C-based methods in, Represents the vector normalization operator. This represents the projection operation that maps the co-activation difference vector to the end force space.

[0027] The individualized musculoskeletal mapping module maps the reference synergistic structure onto the joint geometry and muscle parameters corresponding to the target object, thereby generating a theoretical synergistic force vector that matches the target object, which serves as input to the target generation module. Unlike schemes that only use a population average model or directly use an ideal trajectory as a fixed target, this invention improves the biomechanical rationality of the target vector through individualized mapping, and enhances the accessibility and smooth transition of the initial control stage through a weighted fusion and progressive update mechanism of the habitual force vector and the theoretical synergistic force vector.

[0028] The process of constructing an individualized musculoskeletal model may include: ① Establishing a basic limb musculoskeletal model based on a preset task posture. This basic limb musculoskeletal model can be an upper limb model from an open-source musculoskeletal modeling platform, or a custom upper limb rigid body-muscle line model or other equivalent musculoskeletal model; ② Geometrically scaling the skeletal segments, muscle attachment points, and muscle path points in the basic limb musculoskeletal model according to the target object's upper arm length, forearm length, hand length, and joint posture parameters; ③ Based on the collected surface electromyography (EMG) signals, distal force data, and optional kinematic data, using the EMG envelope as the muscle activation input and the distal force as the model output constraint, constructing an error function between the predicted distal force and the measured distal force, and constraining and optimizing the muscle force scaling factor, lever arm correction coefficient, or equivalent mapping parameters from muscle activation to distal force in the scaled model, thereby obtaining an individualized musculoskeletal mapping model for theoretical synergistic force vector calculation. For muscle biomechanical parameters such as optimal muscle fiber length, tendon relaxation length, and feather angle, the default values ​​after scaling of the basic model can be used; however, if multi-posture kinematic data and sufficient calibration data are available, these parameters can also be used as optional parameters in constraint optimization.

[0029] The individualized musculoskeletal mapping module is used to convert muscle weights in the co-structure matrix into the end-effector force direction of the target object under a preset task posture. The basic limb musculoskeletal model can adopt open-source upper limb musculoskeletal models such as OpenSim, or a custom upper limb rigid body-muscle line model or other equivalent models. The model includes at least skeletal segments, joint postures, muscle paths, muscle biomechanical parameters, and the mapping relationship between muscle unit activation and end-effector force.

[0030] The system first performs geometric scaling on the basic limb musculoskeletal model based on the target object's upper arm length, forearm length, hand length, and preset joint posture. Let the first joint in the basic model be... The length of each limb segment is The measured length of the corresponding limb segment of the target object is Then the scale factor of this limb segment can be expressed as: The system updates the corresponding limb length, muscle attachment point, muscle path point, and muscle lever arm relationship under the preset posture based on the scale factor, thereby obtaining the scaled basic limb musculoskeletal model.

[0031] After geometric scaling, the system performs finite parameter calibration on the scaled model based on the acquired surface electromyography (SEMG) signals and distal force data. Specifically, the SEMG signals are filtered, rectified, envelope extracted, and normalized before being used as muscle activation inputs, while the distal force data serves as model output constraints. Considering that isometric force data under a single posture or a small number of postures is usually insufficient to stably identify all muscle anatomical parameters, this invention preferably calibrates the muscle force scaling factor, muscle force gain, lever arm correction coefficient, or equivalent mapping parameters from muscle activation to distal force, rather than requiring the unique identification of all muscle mechanical parameters solely based on SEMG signals.

[0032] In one embodiment, let the first The sampling window of the first Normalized electromyographic activation of a muscle mass is After geometric scaling, the basic musculoskeletal model is in the preset task pose. The calculated first The force vector at the end of the activation of a muscle unit is The muscle force scaling factor to be calibrated is Then the predicted end force can be expressed as: in, The number of muscles involved in the modeling. The muscle force scale factor. Used to characterize differences in muscle force output, electromyographic-mechanical gain, or equivalent force arm between the target object and the baseline model.

[0033] To ensure that the predicted end force matches the measured end force, the following parameter calibration objective function can be constructed: in, For the first The measured end force of each sampling window Represents the vector normalization operator. To prevent constants with a denominator of zero, and These are the weights of the direction error term and the magnitude error term, respectively. The regularization coefficient is . This is the initial scale factor in the basic model or swarm model. Since this invention is primarily used to generate the target force direction, in a preferred embodiment, the weight of the direction error term can be set. Weight greater than the amplitude error term This improves the stability of the theoretical synergistic force direction. The optimization can employ constrained least squares, nonlinear least squares, or other equivalent optimization algorithms.

[0034] After completing individualized calibration, for the first The theoretical synergistic force vector of a synergistic unit can be obtained by weighting the muscle weights in the synergistic structure matrix with the individualized unit activation end force vector: in, Indicates the first The collaboration in the first Weight on a muscle mass Indicates the calibrated first Muscle strength scale factor This represents the geometrically scaled base musculoskeletal model calculated under the preset task pose. The force vector at the end of the muscle unit activation point. This represents the normalization operator. Thus, the system can convert high-dimensional cooperative structures into low-dimensional end-force directions, providing input for target force generation in cooperative vector-driven mode and fine-tuning control in cooperative activation-driven mode.

[0035] It should be noted that the OpenSim upper limb model is only one preferred implementation of the basic musculoskeletal model, and this invention is not limited to any specific open-source software or specific musculoskeletal model format. Any upper limb musculoskeletal model that can provide muscle pathways, muscle biomechanical parameters, and the mapping relationship between joint posture and end-effector force can be used as the basic musculoskeletal model in this invention.

[0036] During the initialization phase, the system acquires baseline data of the target object under a preset upper limb posture to obtain electromyography, distal force, and optional kinematic parameters. Based on the baseline acquisition results, the system extracts the current coordination structure matrix and coordination activation coefficient matrix, and calls the reference coordination template associated with the corresponding task direction to complete the individualized model initialization, obtaining the theoretical coordination force vector, habitual force vector, and initial error index. Compared with single-target or fixed-process schemes, this invention achieves staged closed-loop regulation from direction control to coordination ratio control through a hierarchical target generation mechanism of first and second modes.

[0037] The process of condition detection and mode switching is as follows: In W-based mode, the system can evaluate whether the switching conditions are met at fixed intervals or after several interactions. Preferably, the mode switching criteria include at least two conditions: First, the system calculates whether the angle error between the current end force direction and the first target force vector is less than a preset angle threshold, which can be 3-10 degrees, preferably 4-6 degrees; Second, the system further calculates whether the synergy specificity index between the main synergies has entered a plateau period to determine whether the current synergy ratio tends to stabilize. The synergy specificity index can be defined as: in, This represents the integral value, root mean square value, or weighted area of ​​task-specific coactivation within the evaluation window. This represents the integral, root mean square, or weighted area of ​​object-specific coactivation within the same window. This indicates the correlation between activation coefficients. To prevent tiny constants with a denominator of zero.

[0038] If the following conditions are met: If the indicator continues for L evaluation windows, it is determined that it has entered a plateau period. Indicates the first Synergistic specificity indicators corresponding to each evaluation window Where L is the plateau threshold and L is the number of consecutive windows. When both the direction error condition and the cooperative-specific stability condition are met, it is determined that the coarse adjustment target of the first mode has reached the preset requirements, and a switch from the W-based mode to the C-based mode is triggered. If the conditions are not met, the W-based mode continues to be maintained.

[0039] Given the inherent redundancy in human motion control, when a target object performs an equal-length force application task in a preset posture, its end-effector force vector is usually difficult to be completely coaxial with the target direction, thus generating residual components in a plane orthogonal to the target direction. This invention defines the orthogonal components outside the target direction as coupling forces, and uses them as important indicators for evaluating output accuracy and coordination misalignment. The coupling index characterizes the degree of coupling between the orthogonal components of the target direction and the principal direction components, and can be defined as an index based on amplitude ratio, root mean square ratio, peak ratio, correlation quantity, or a combination thereof. In a preferred embodiment, the coupling index can be defined as: in, The force component applied in the direction of the target. These are the force components in the orthogonal directions. This indicates the correlation between force curves; , representing 6 orthogonal directions. The overall coupling index of a single test can be expressed as the average value or weighted combination of the coupling indices of each orthogonal direction.

[0040] The feedback output module can display the current force direction, target force direction, and their angular error using vector diagrams, arrows, color status, text prompts, or voice prompts. It can also simultaneously output coupling and coordination-specific indicators to generate a quantitative evaluation result that considers both motion output accuracy and coordination state differences. After completing a preset cycle of interactive tasks, the system can re-record the target object's electromyography and distal force data, compare and analyze them with the initial evaluation results or reference template, and generate a state change report, such as the direction error curve, coupling indicator curve, and coordination-specific indicator trend.

[0041] Example 2 like Figure 1 As shown, a method for upper limb coordination assessment and training control based on co-activation includes the following steps: S1, acquire surface electromyography signals and distal force data of the target object when performing a preset isometric force application task involving multiple task directions; Collect multi-channel surface electromyography, distal three-dimensional / six-dimensional force, task tags, posture parameters, and optional joint kinematics and limb length information.

[0042] In one embodiment, the target subject performs an isometric force application task while seated, with the upper limbs maintaining a preset posture and the wrist fixed and gripping the force sensor handle. During the baseline acquisition phase, the maximum isometric contraction force can be recorded, and the subject performs force application tasks in preset directions such as up, down, forward, backward, left, and right under visual cues. Each direction can last for 3-8 seconds, preferably about 5 seconds; each direction can be repeated 3-8 times, preferably about 5 times; a rest interval can be set between trials.

[0043] Surface electromyography (EMG) signals of major upper limb muscle groups are acquired simultaneously, including but not limited to the anterior deltoid, middle deltoid, posterior deltoid, biceps brachii, long head of triceps brachii, lateral head of triceps brachii, pectoralis major, brachioradialis, pronator teres, flexor carpi radialis, palmaris longus, flexor carpi ulnaris, and flexor digitorum superficialis. The EMG sampling frequency can be 1000-2000 Hz, and the force signal sampling frequency can be 100-1000 Hz. EMG preprocessing may include 20-450 Hz bandpass filtering, 50 / 60 Hz notch filtering, full-wave rectification, 3-10 Hz low-pass envelope extraction, and intra-session normalization. Force signals may undergo low-pass filtering and steady-state window extraction.

[0044] S2, based on the end force data under the condition of no target feedback, estimate the habitual force vector of the target object; and extract the current co-activation coefficient based on the surface electromyography signal; Habitual force refers to the direction of the end force formed by a target object naturally applying force during the baseline assessment phase without target vector guidance, and is used to characterize its current natural force application pattern. Habitual force is initially assessed during the baseline assessment phase, and its general process is shown in Table 3.

[0045] Table 3 Calculation process of habituation The direction vector of the habitual force can be expressed as: in, Indicates the first The second test was in The normalized force vector in each direction, This indicates the number of trials. If the angle between the updated direction of habitual force and the direction of habitual force in the previous cycle is less than the angle threshold... If the previous cycle's habitual force vector is used, then the updated habitual force vector is used as the starting point for the next cycle; otherwise, the updated habitual force vector is used as the starting point for the next cycle. A temperature of 2-8 degrees Celsius is acceptable, with 3-5 degrees Celsius being preferred.

[0046] In the collaborative extraction stage, the preprocessed electromyography matrix is ​​processed using a nonnegative matrix factorization algorithm. Decomposition yields the cooperative structure matrix W and the cooperative activation coefficient matrix C, such that... ,in, , , , For the number of muscle channels, For time points, For the number of collaborations.

[0047] To determine the appropriate number of collaborations, this invention can calculate two criteria based on the variance explained rate (VAF), and determine the final number of collaborations when both criteria are met simultaneously: Criterion 1, the global VAF is greater than a preset threshold, which can be 0.85–0.95, preferably about 0.90; Criterion 2, under the premise of satisfying criterion 1, when adding a collaboration, the increment of VAF is less than a preset threshold, which can be 0.01–0.08, preferably about 0.05.

[0048] in, This represents the average level of muscle activation. This indicates the degree of muscle activation reconstruction resulting from the decomposition; the higher the value, the better the reconstruction quality.

[0049] Reference collaboration templates can be obtained by averaging across objects, estimating cluster centers, or updating based on historical data. In a preferred embodiment, the template library further records common collaboration (ES), task-specific collaboration (CS), and object-specific collaboration (SS) labels to calculate collaboration-specific indices and use them for subsequent mode switching.

[0050] To achieve automatic annotation of extracted collaborations in real time, the system performs similarity matching between the extracted collaboration structure vector within the current evaluation window and collaboration templates in the reference collaboration template library. Specifically, for the current collaboration vector... and reference template vector After normalization, the cosine similarity between the two is calculated: The system determines the matching relationship between the current collaborative vector and the reference template based on the maximum similarity principle or by using the Hungarian algorithm. If the maximum similarity between the current collaborative vector and the task-specific collaborative template is greater than the first similarity threshold... If the maximum similarity between the current collaboration vector and the object-specific collaboration template is greater than the second similarity threshold, then it is marked as a task control specific collaboration (CS). Or its maximum similarity with the task control-specific co-template is lower than the third similarity threshold. If the activation amplitude exceeds a preset activation threshold, it is marked as an object-specific cooperative (SS). The similarity can also be achieved using the Pearson correlation coefficient instead of cosine similarity.

[0051] S3, obtain the theoretical cooperative force vector and reference cooperative activation coefficient corresponding to each task direction; The general model is scaled based on upper arm length, forearm length, joint posture, and calibration parameters, and the maximum isometric force of muscles, tendon relaxation length, feather angle, or lever arm parameters are calibrated based on collected electromyography, distal force, and optional kinematic data. For the first... For each reference coordination, the theoretical coordination force vector can be expressed as: in, Indicates the first The collaboration in the first Weight on a muscle mass Indicates the first The end force vector corresponding to the activation of a muscle unit. This represents the normalization operation. This allows the abstract cooperative structure to be transformed into a visible and controllable end-target force direction.

[0052] S4. In the first mode, based on a preset difficulty coefficient, the habitual force vector and the theoretical collaborative force vector are weighted and combined to generate a first target force vector. In the cooperative vector-driven mode, the system is accustomed to force vectors. Synergistic force vector with theory The first target force vector is generated by weighted combination: in, Indicates the first The first target force vector corresponding to each task direction; This represents the difficulty coefficient; the larger the value, the closer the first target force vector is to the theoretical collaborative force vector. Indicates the first The habitual force vector of each task direction; Indicates the first The theoretical synergistic force vector corresponding to each task direction. By adjusting The value of can ensure that the control does not rely excessively on the reference target direction in the early stages.

[0053] The feedback output module displays a visual comparison between the current force direction and the first target force direction on the display terminal, allowing the target object to adjust the applied force direction based on the feedback information. The system monitors changes in direction error and key coordination indicators in real time and dynamically updates the system based on error change trends. When the direction error continues to decrease, it can be increased. The update step size can be 0.02-0.15, preferably 0.05-0.10, to gradually converge the first target force vector to the theoretical cooperative force vector; conversely, if the direction error increases or the control stability decreases, and is accompanied by enhanced object-specific cooperative activation, then the step size should be reduced. To reduce the target deflection magnitude. This is achieved through iterative updates. The system can improve the consistency between the current force direction and the reference direction.

[0054] S5, real-time acquisition of the current end force direction of the target object, calculation of the angle error between the current end force direction and the first target force vector, and calculation of the cooperative specificity index based on the current cooperative activation coefficient; The process of condition detection and mode switching is as follows: In W-based mode, the system can evaluate whether the switching conditions are met at fixed intervals or after several interactions. Preferably, the mode switching criteria include at least two conditions: First, the system calculates whether the angle error between the current end force direction and the first target force vector is less than a preset angle threshold, which can be 3-10 degrees, preferably 4-6 degrees; Second, the system further calculates whether the synergy specificity index between the main synergies has entered a plateau period to determine whether the current synergy ratio tends to stabilize. The synergy specificity index can be defined as: in, This represents the integral value, root mean square value, or weighted area of ​​task-specific coactivation within the evaluation window. This represents the integral, root mean square, or weighted area of ​​object-specific coactivation within the same window. This indicates the correlation between activation coefficients. To prevent tiny constants with a denominator of zero.

[0055] If the following conditions are met: If the indicator continues for L evaluation windows, it is determined that it has entered a plateau period. Indicates the first Synergistic specificity indicators corresponding to each evaluation window Where L is the plateau threshold and L is the number of consecutive windows. When both the direction error condition and the cooperative-specific stability condition are met, it is determined that the coarse adjustment target of the first mode has reached the preset requirements, and a switch from the W-based mode to the C-based mode is triggered. If the conditions are not met, the W-based mode continues to be maintained.

[0056] S6, when the included angle error is less than the preset angle threshold, and the synergistic specificity index meets the plateau condition within a preset number of consecutive evaluation windows, the control mode is switched from the first mode to the second mode. The dual-mode target generation and switching module preferably performs mode switching based on at least two conditions: First, the angle error between the current force direction and the first target force vector is within a certain range. The angle threshold was less than the limit within each trial; secondly, the synergistic specificity index was within a continuous range. The system enters a plateau phase within the evaluation window. After completing the above-mentioned cooperative matching and classification, the system calculates the cooperative specificity index based on the cooperative activation coefficients of those labeled as task-control specific cooperatives (CS) and those labeled as object-specific cooperatives (SS). in, This represents the integral value, root mean square value, or weighted area of ​​task-specific coactivation within the evaluation window. This represents the integral, root mean square, or weighted area of ​​object-specific coactivation within the same window. This indicates the correlation between activation coefficients. To prevent small constants with a denominator of zero. If the following conditions are met... If the condition persists for L windows, it is determined that the system has entered a plateau phase.

[0057] S7, in the second mode, calculate the difference vector between the reference co-activation coefficient and the current co-activation coefficient, project the difference vector onto the end force space spanned by the theoretical co-activation force vector to obtain the force correction vector, and generate the second target force vector based on the superposition and normalization of the current end force vector and the force correction vector.

[0058] When the system switches to the co-activation driven mode, the system calculates the reference co-activation coefficient vector. With the current co-activation coefficient vector Difference vectors between : Let the theoretical cooperative force vector set be Arrange them into a theoretical synergy matrix by column: The mapping from the co-activation difference vector to the end force space can be expressed as: in, For force correction vector, This is the collaborative weight matrix; when all collaborative weights are the same, The identity matrix can be taken; This is used to correct the gain coefficient, which is used to control the magnitude of the force correction. Expanded, it can be expressed as: in, For the first The activation differences of each collaboration, This corresponds to the collaborative weights. Subsequently, the system uses the current end force vector... With force correction vector Generate the second target force vector: in, This represents the vector normalization operator. The above calculation allows the deviation of the co-activation ratio to be converted into a correction direction in the end force space, which can then be used for fine target adjustment in co-activation driven modes.

[0059] To improve the stability of the second target force vector output, the system can calculate... or The current synergistic activation coefficients are first smoothed using moving mean square or exponential smoothing, with a smoothing window of 50-300 ms, preferably 80-200 ms. When the difference between the synergistic specificity index and the reference template narrows to a preset range and remains stable, the system determines that the synergistic ratio control of the second mode has met the preset requirements. In one embodiment, the theoretical synergistic force vector set... It can be pre-calculated and stored in the reference collaborative template library for subsequent real-time table lookup and retrieval.

[0060] Finally, the feedback output module outputs at least one or more of the following: current force direction, target force direction, angle error, coupling force index, and synergy-specific index. The coupling force index can be defined as the ratio, correlation quantity, or weighted combination of the residual force component in the orthogonal plane of the target direction to the main force component in the target direction. The phased evaluation module records the direction error, coupling index, synergy-specific index, template matching degree, and their changing trends in different periods, and generates exportable phased evaluation results or state change reports.

[0061] This invention can be deployed in devices that perform upper limb coordinated state assessment and target vector guidance in the form of isometric force application or similar tasks. It can run as a standalone software system on a device that includes a multi-channel surface electromyography acquisition device, an end effector force sensor, and a display terminal, or it can be embedded in rehabilitation robots, exoskeletons, digital therapy platforms, and other upper limb interactive devices as the core control module for target force generation and state assessment.

[0062] At the application level, this invention can be used for collaborative state assessment, target vector generation, feedback interaction, construction of health sample benchmarks, algorithm verification, and the establishment of human-computer interaction experimental platforms; stroke, traumatic brain injury, and spinal cord injury are preferred application targets. It should be noted that the focus of this invention is on the system structure, data processing flow, and control logic, rather than disease diagnosis or treatment steps that are directly implemented on the human body.

[0063] Example 3 This embodiment provides a parameter operation example of an upper limb coordination assessment and training control system based on co-activation. It should be noted that this embodiment is used to illustrate system parameter configuration, target force generation process, and mode switching logic, and does not limit the specific parameter values ​​of the invention, nor does it imply a limitation on the treatment effect for specific diseases.

[0064] In this embodiment, the target subject performs an isometric force application task while in a preset seated posture, with the upper limbs maintaining a fixed posture and the wrist gripping the end-effector force sensor handle. The system collects surface electromyography (EMG) signals and end-effector force data from the target subject in multiple nominal force application directions, and estimates the habitual force vector corresponding to each task direction under conditions without target feedback. Simultaneously, the system calls upon a reference coordinating template library and calculates the theoretical coordinating force vector for the corresponding task direction based on an individualized musculoskeletal mapping model.

[0065] In the cooperative vector-driven mode, the system sets the initial difficulty coefficient. The value is 0.20-0.40, preferably 0.30; when the angle error between the current force direction of the target object and the first target force vector continues to decrease in several consecutive tasks, the system increases the value in steps of 0.02-0.15. The preferred step size is 0.05-0.10, allowing the first target force vector to gradually transition from the habitual force vector to the theoretical cooperative force vector. If the included angle error increases, or if the control stability decreases accompanied by enhanced object-specific cooperative activation, the system performance will decrease. This reduces the magnitude of target direction adjustment.

[0066] During the mode switching determination process, the system continuously calculates the angle error between the current end force direction and the first target force vector, and calculates the synergy specificity index. When the angle error is within a certain range... Within each trial, the angle is less than the preset threshold, and the synergistic specificity index is within a continuous range. When the plateau condition is met within an evaluation window, the system switches the control mode from cooperative vector-driven mode to cooperative activation-driven mode. The angle threshold can be 3-10 degrees, preferably 4-6 degrees. 3-10 is acceptable, with 3-5 being the preferred value; 3-10 is acceptable, with 3-5 being the preferred value.

[0067] In the co-activation driven mode, the system calculates the reference co-activation coefficient vector. With the current co-activation coefficient vector Difference vectors between And generate force correction vectors based on theoretical synergistic force matrix. The system then bases its decisions on the current end force vector. With force correction vector Generate the second target force vector To avoid abrupt changes in the target force direction, the system can be configured with a correction gain coefficient. and to Amplitude limiting or smoothing processing is performed. The smoothing processing can use moving average, moving root mean square, or exponential smoothing, and the smoothing window can be 50-300 ms, preferably 80-200 ms.

[0068] In this embodiment, the phased evaluation results output by the system include one or more of the following: current force direction, target force direction, angle error, coupling force index, synergy specificity index, difficulty coefficient change trend, and mode switching status. These phased evaluation results can be used to describe the state changes of the target object in different training cycles, and can also be used for subsequent device linkage or parameter adjustment, but are not directly used as disease diagnosis conclusions or treatment effect judgments.

[0069] The methods of the present invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or a non-transitory machine-readable medium and subsequently stored on a local recording medium, downloaded via a network. Thus, the methods described herein can be processed by software stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code that, when accessed and executed by the computer, processor, or hardware, implements the methods described herein. Furthermore, when a general-purpose computer accesses the code used to implement the methods shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for performing the methods shown herein.

[0070] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A method for upper limb coordination assessment and training control based on co-activation, characterized in that, Includes the following steps: Acquire surface electromyography signals and distal force data of the target object when performing a preset isometric force application task involving multiple task directions; Based on the end force data under targetless feedback conditions, the habitual force vector of the target object is estimated; and the current co-activation coefficient is extracted based on the surface electromyography signal. Obtain the reference collaboration and reference collaboration activation coefficients corresponding to each task direction; Based on the individualized musculoskeletal model and the aforementioned reference coordination, the theoretical coordination force vector of the target object is estimated; In the first mode, based on a preset difficulty coefficient, the habitual force vector and the theoretical collaborative force vector are weighted and combined to generate a first target force vector; The current end force direction of the target object is acquired in real time, the angle error between the current end force direction and the first target force vector is calculated, and the cooperative specificity index is calculated based on the current cooperative activation coefficient. When the included angle error is less than a preset angle threshold, and the synergistic specificity index meets the plateau condition within a preset number of consecutive evaluation windows, the control mode will be switched from the first mode to the second mode. In the second mode, the difference vector between the reference co-activation coefficient and the current co-activation coefficient is calculated, and the difference vector is projected onto the end force space spanned by the theoretical co-activation force vector to obtain the force correction vector. The second target force vector is generated by superposition and normalization of the current end force vector and the force correction vector.

2. The method for upper limb coordination assessment and training control based on co-activation according to claim 1, characterized in that, The steps for obtaining the theoretical synergistic force vector include: A basic limb musculoskeletal model is selected, which includes skeletal segments, joint degrees of freedom, muscle attachment points, muscle path points, muscle biomechanical parameters, and the mapping relationship between joint posture and end force. Based on the target object's upper arm length, forearm length, hand length, and preset task posture, the scale factor of each skeletal segment is calculated, and the skeletal segments, muscle attachment points, and muscle path points in the basic limb musculoskeletal model are geometrically scaled based on the scale factor to obtain the scaled basic limb musculoskeletal model. Based on the surface electromyography (EMG) signals and distal force data collected from the target object in a preset isometric force application task, the surface EMG envelope is used as the muscle activation input and the measured distal force is used as the output constraint. An error function between the predicted distal force and the measured distal force is constructed. One or more of the muscle force scale factor, lever arm correction coefficient, or equivalent mapping parameters from muscle activation to distal force in the scaled basal musculoskeletal model are constrained and optimized to obtain an individualized musculoskeletal mapping model. The individualized musculoskeletal mapping model is used to calculate the end force vector corresponding to the activation of each muscle unit. The theoretical collaborative force vector for the corresponding task direction is obtained by weighting and summing the collaborative weights of each muscle in the reference collaborative structure matrix with the end force vector when the corresponding muscle unit is activated, and then normalizing the sum.

3. The method for upper limb coordination assessment and training control based on co-activation according to claim 2, characterized in that, When performing decomposition using a nonnegative matrix factorization algorithm, the number of collaborations in the collaborative structure matrix is ​​determined based on the following dual criteria: Calculate the global variance explained rate when reconstructing the surface electromyography signal from the cooperative structure matrix and the current cooperative activation coefficients, wherein the global variance explained rate is greater than a first preset threshold. Furthermore, under the premise of satisfying the aforementioned conditions, when a collaboration is added, the increment of the global variance explanation rate is less than the second preset threshold.

4. The method for upper limb coordination assessment and training control based on co-activation according to claim 1, characterized in that, The process of obtaining the habitual force vector includes: For each task direction, the steady-state portion of the end force data of each test trial is extracted and normalized. The normalized force vector of each trial is averaged by a spherical method to obtain the initial habitual force vector. In each training cycle, the habituation force vector is re-estimated. If the angle between the updated habituation force vector and the habituation force vector of the previous cycle is less than the angle threshold, then the habituation force vector is considered to be the first to be considered. If the previous cycle's habitual force vector is used, then the updated habitual force vector will be used as the starting point for the generation of the target force in the next cycle; otherwise, the updated habitual force vector will be used as the starting point for the generation of the target force in the next cycle.

5. The method for upper limb coordination assessment and training control based on co-activation according to claim 1, characterized in that, The formula for calculating the first target force vector is: in, Let be the first target force vector corresponding to the j-th task direction. Represents the vector normalization operator. This represents the difficulty coefficient corresponding to the j-th task direction. Let represent the habitual force vector in the j-th task direction. This represents the theoretical synergistic force vector corresponding to the j-th task direction.

6. The method for upper limb coordination assessment and training control based on co-activation according to claim 1, characterized in that, The calculation process of the synergistic specificity index includes: The extracted collaborative structure vector within the current evaluation window is matched with collaborative templates in the reference collaborative template library based on similarity. The corresponding collaborative activation coefficients are then categorized into task-specific and object-specific collaborative activation coefficients based on the matching results. The similarity matching includes calculating the cosine similarity or Pearson correlation coefficient between the current collaborative structure vector and the reference collaborative template, and determining the category label of the current collaborative structure vector based on a preset similarity threshold. The collaborative specificity index is calculated using the following formula. : in, This indicates the amplitude characteristics of the task-specific activation coefficient within the evaluation window. This represents the magnitude characteristics of the object-specific activation coefficients within the same evaluation window. This indicates the correlation between activation coefficients.

7. The method for upper limb coordination assessment and training control based on co-activation according to claim 6, characterized in that, The calculation process for the plateau period conditions includes: When satisfied And after L evaluation windows, the synergistic specificity index is determined to have entered a plateau period, wherein This represents the synergistic specificity index corresponding to the nth evaluation window. This is the threshold for the plateau period.

8. A co-activation-based upper limb coordination assessment and training control system, comprising the method described in any one of claims 1-7, characterized in that, include: The data acquisition and collaborative extraction module is used to acquire surface electromyography (EMG) signals and end force data of the target object when performing a preset isometric force application task involving multiple task directions, and to extract the current collaborative activation coefficient based on the surface EMG signals. The habituation force estimation module is used to calculate and output the habituation force vector corresponding to each task direction based on the end force data under the condition of no target feedback. The target force generation module is used to obtain the reference coordination and reference coordination activation coefficients corresponding to each task direction; it estimates the theoretical coordination force vector based on the individualized musculoskeletal model. In the first mode, the habitual force vector and the theoretical collaborative force vector are weighted and combined based on a preset difficulty coefficient to generate a first target force vector; In the second mode, a second target force vector is generated by projecting the difference vector between the reference co-activation coefficient and the current co-activation coefficient onto the end force space. The dual-mode control module is used to calculate the angle error between the current end force direction and the first target force vector in real time, and to calculate the cooperative specificity index based on the current cooperative activation coefficient. And when the included angle error is less than a preset angle threshold and the collaborative specificity index meets the continuous window plateau condition, the control mode is switched from the first mode to the second mode.

9. A co-activation-based upper limb coordination assessment and training control system according to claim 8, characterized in that, The system also includes: The individualized musculoskeletal mapping module is used to perform cooperative decomposition of the surface electromyography signal to obtain a cooperative structure matrix, select a basic limb musculoskeletal model including skeletal segments, joint degrees of freedom, muscle paths, muscle mechanical parameters and the mapping relationship between joint posture and end force, and geometrically scale the basic limb musculoskeletal model according to the target object's upper arm length, forearm length, hand length and preset task posture. The individualized musculoskeletal mapping module is also used to constrain and calibrate one or more of the muscle force scale factor, lever arm correction coefficient, or equivalent mapping parameters from muscle activation to end force in the scaled basal musculoskeletal model based on the surface electromyography signals and end force data collected by the target object in a preset isometric force application task, so as to obtain an individualized musculoskeletal mapping model. The individualized musculoskeletal mapping module is also used to calculate the end force vector corresponding to the activation of each muscle unit based on the individualized musculoskeletal mapping model, and to perform weighted summation and normalization of the muscle weights in the reference synergistic structure matrix with the corresponding end force vectors to obtain the theoretical synergistic force vector.

10. A co-activation-based upper limb coordination assessment and training control system according to claim 8, characterized in that, The system also includes: The phase evaluation module is used to calculate and output coupling indices in the corresponding orthogonal directions based on the end force data. : in, The force component applied in the direction of the target. These are the force components in the orthogonal directions. This indicates the correlation between force curves, and d represents the orthogonal direction.

Images