Muscle coordination-based wearable exoskeleton device control method
By assessing the lower limb muscle coordination patterns of children with cerebral palsy in real time and generating auxiliary torque using a Gaussian mixture model, the problem that existing exoskeleton devices cannot improve abnormal muscle coordination is solved, and individualized adaptive rehabilitation training effects are achieved.
Patent Information
- Application Number
- CN202610927213.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-25
- Publication Date
- 2026-08-25
AI Technical Summary
Existing wearable exoskeleton control methods are mainly based on joint kinematics, which cannot assess and improve the degree of abnormality in muscle coordination in children with cerebral palsy, nor can they guide the remodeling of the child's neuromuscular control towards normal neuromuscular control.
By acquiring electromyographic signals of key muscle groups in both lower limbs of the exoskeleton device wearer in real time, the likelihood probability of muscle synergy patterns is evaluated using a pre-trained Gaussian mixture model, the overall anomaly degree E is calculated, and an auxiliary torque is generated based on this to drive joint movement.
It quantifies the degree of abnormality at the neuromuscular control level, provides a basis for the generation of auxiliary torque, realizes adaptive rehabilitation training with decreasing auxiliary force, and ensures individualized bilateral auxiliary intensity distribution.
Smart Images
Figure CN122624271A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of wearable exoskeleton device control technology, specifically, it relates to a control method for wearable exoskeleton devices based on muscle coordination. Background Technology
[0002] Congenital cerebral palsy primarily manifests as central motor disorders and postural abnormalities. The core problem for these children lies not only in weakened muscle strength or limited joint range of motion, but also in the abnormal control of muscles by the central nervous system. Muscle coordination in children with cerebral palsy often exhibits significant abnormalities, mainly manifested as a reduced number of coordinators, deviations from normal muscle control within the coordinator vectors, and disordered activation timing. Wearable exoskeletons can be used in the rehabilitation training of these children; however, current exoskeleton control methods are mainly based on joint kinematics, such as using predefined trajectory tracking to control the child's lower limbs to complete training movements. These methods primarily focus on the motor execution level and cannot assess or improve the degree of abnormality in muscle coordination, nor can they guide the reshaping of the child's neuromuscular control towards normal neuromuscular control, thereby failing to achieve the goals of rehabilitation training. Summary of the Invention
[0003] The purpose of this invention is to provide a control method for wearable exoskeleton devices based on muscle coordination, so as to solve the technical problems existing in the prior art.
[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for controlling a wearable exoskeleton device based on muscle coordination includes the following steps: Step S1: Real-time acquisition of electromyographic signals of key muscle groups in both lower limbs of the exoskeleton device wearer, including rectus femoris, biceps femoris, tibialis anterior, and gastrocnemius. Step S2: Preprocess the electromyographic signals and extract the normalized muscle activation of each muscle group; Step S3: Input the normalized muscle activation of each muscle group into the pre-trained Gaussian mixture model, calculate the likelihood probability that the normalized muscle activation vector belongs to the normal muscle synergy mode, and determine the best matching Gaussian component. Step S4: Calculate the overall anomaly degree E based on the likelihood probability; Step S5: Based on the comprehensive anomaly degree E and the corresponding auxiliary strength coefficient... This generates auxiliary torques for each joint; Step S6: Drive the movement of each joint actuator with the auxiliary torque.
[0005] Further, in step S2, the method for extracting the normalized muscle activation of each muscle group is as follows: First, the original surface electromyography (EMG) signal is preprocessed to extract the envelope of the EMG signal; then, the normalized muscle activation is calculated using the following formula. : ; in, Let be the envelope value of the electromyographic signal of the m-th muscle group at time t. This represents the maximum electromyographic amplitude measured during maximal voluntary contraction of this muscle group.
[0006] Furthermore, in step S3, the construction of the pre-trained Gaussian mixture model includes the following steps: Step S3.1: Collect surface electromyography (EMG) signals of key muscle groups in both lower limbs of multiple normal children during walking, obtain normalized muscle activation vectors after preprocessing, and construct a training sample set; Step S3.2: Construct a Gaussian mixture model, which consists of n Gaussian components, each of which includes a mean vector. A covariance matrix and a weighting coefficient ,and ; Step S3.3: Train the Gaussian mixture model using the expectation-maximization algorithm. After training, fix the parameters of each Gaussian component in the controller of the wearable exoskeleton device.
[0007] Furthermore, in step S3, the normalized muscle activation vector at the current moment... Likelihood probability of belonging to normal muscle coordination pattern Calculate using the following formula: ; Best Match Component Index for:
[0008] In the formula, Let be the probability density function of the nth Gaussian component.
[0009] Further, in step S4, the overall anomaly degree E is calculated using the following formula:
[0010] In the formula, The current likelihood probability, and These are the maximum and minimum values of the log-likelihood of the Gaussian mixture model on the training set, respectively.
[0011] Furthermore, in step S5, the auxiliary torque of each joint... Calculate using the following formula: ; In the formula, As the reference auxiliary gain coefficient, This is the auxiliary strength coefficient for the corresponding side. Let be the torque arm coefficient of the i-th muscle with respect to the j-th joint; The activation correction for the i-th muscle is calculated using the following formula: ; In the formula, The Gaussian component index that best matches the current activation mode. For the first The mean of the Gaussian components in the i-th muscle. represents the current actual activation level of the i-th muscle.
[0012] Furthermore, the auxiliary strength coefficient on the corresponding side The calculation method is as follows:
[0013] In the formula, The auxiliary strength coefficient on the left side. The auxiliary strength coefficient on the right side. The overall motor function score for the left lower limb was determined. The overall motor ability score of the right lower limb is determined.
[0014] Furthermore, the comprehensive motor ability score of the left lower limb Comprehensive motor ability score of the right lower limb The calculation method is as follows:
[0015] In the formula, The overall motor ability score for the left or right lower limb is determined. This represents the actual range of motion of the j-th joint on that side during the calibration movement. This represents the reference range of motion for the corresponding joint in a normal child of the same age on the same side. This refers to the number of joints on that side involved in the assessment.
[0016] Furthermore, the reference auxiliary gain coefficient The value is determined based on the rated output torque of each joint actuator in the wearable exoskeleton device, and its range is from 10% to 30% of the rated torque of the actuator.
[0017] Furthermore, the aforementioned The possible values are as follows: (1) Targeting the rectus femoris muscle and knee joint, The value ranges from +1.5 to +2.2; (2) Targeting the biceps femoris and knee joint, The value ranges from -0.9 to -1.4; (3) Targeting the tibialis anterior muscle and ankle joint, The value ranges from +1.0 to +1.6; (4) Targeting the gastrocnemius muscle and ankle joint, The value ranges from -1.3 to -2.0; (5) Targeting the rectus femoris and hip joint, The value ranges from +1.6 to +2.5; (6) Targeting the biceps femoris and hip joint, The value range is from -1.1 to -1.8; Positive values indicate that the muscle contraction produces flexion or dorsiflexion torque, while negative values indicate extension or plantarflexion torque.
[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention uses a pre-trained Gaussian mixture model to evaluate the likelihood probability of the wearer's muscle coordination patterns in real time and calculates the comprehensive abnormality E based on this. Unlike existing exoskeleton control methods that only focus on joint angles or torque trajectories, this method directly quantifies the degree of abnormality at the neuromuscular control level, assesses motor dysfunction, and provides a more fundamental decision-making basis for the generation of auxiliary torque.
[0019] This invention directly incorporates the comprehensive abnormality degree E into the auxiliary torque calculation formula, making the magnitude of the auxiliary torque proportional to the degree of deviation of the muscle coordination pattern from the normal. The more the wearer's muscle coordination pattern deviates from the normal, the stronger the auxiliary torque. As the muscle coordination pattern gradually improves during training, the abnormality degree E decreases, and the auxiliary torque automatically decreases, achieving adaptive rehabilitation with decreasing assistance.
[0020] This invention assesses the comprehensive motor ability scores of both lower limbs and allocates the auxiliary intensity coefficients to the left and right sides accordingly. The side with weaker motor ability receives a larger auxiliary intensity coefficient, ensuring a reasonable distribution of auxiliary intensity on both sides and avoiding excessive assistance to the side with stronger motor ability. This makes bilateral training more in line with individualized rehabilitation needs.
[0021] After the Gaussian mixture model of this invention is trained offline using normal infant gait electromyography data, the parameters are fixedly deployed in the controller. When used online, only probability density calculation and weighted summation need to be performed, without iterative optimization. The inference calculation time at a single moment is extremely short, which meets the requirements of real-time control. Attached Figure Description
[0022] Figure 1 This is a flowchart of Embodiment 1 of the present invention.
[0023] Figure 2 This is a system block diagram of the controller and sensor in this invention.
[0024] Figure 3 This is a schematic diagram of the main structure of the wearable exoskeleton device.
[0025] The corresponding names of the attached figures are as follows: 1-Controller, 2-Hip joint actuator, 3-Knee actuator, 4-Ankle actuator. Detailed Implementation
[0026] To enable those skilled in the art to have a clearer understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described below are merely for illustrative purposes and to facilitate understanding. The technical solutions provided by the present invention are not limited to those provided in the following embodiments, nor should they limit the scope of protection of the present invention.
[0027] In the description of this invention, it should be noted that the term "congenital cerebral palsy" refers to a central motor and postural developmental disorder syndrome caused by non-progressive damage to the brain of a developing fetus or infant; "normalized muscle activation" refers to the percentage value obtained by normalizing the amplitude of the electromyographic signal envelope relative to the amplitude of the electromyographic signal during the maximum voluntary contraction of the muscle; "Gaussian mixture model" refers to a probability density model composed of a linear combination of multiple Gaussian distributions; and "likelihood probability" refers to the probability density value of the current observation data under a given model. Example
[0028] like Figure 1 As shown, this embodiment provides a control method for a wearable exoskeleton device based on muscle coordination. This method is executed by the controller of the wearable exoskeleton device and specifically includes the following steps: 1. Step S1: Real-time acquisition of electromyographic signals of key muscle groups in both lower limbs of the exoskeleton device wearer. The controller uses eight electromyography (EMG) sensors to collect raw digital EMG signals from the bilateral rectus femoris, biceps femoris, tibialis anterior, and gastrocnemius muscles in real time. The sampling frequency is, for example, 1000Hz. A total of eight EMG sensors are arranged bilaterally, four on each side, corresponding to the four key muscle groups mentioned above. The raw signals are pre-amplified differentially, hardware-guided bandpass filtered (passband range 20 to 500Hz), and then converted from analog to digital before being input to the controller as digital signals.
[0029] 2. Step S2: Extract normalized muscle activation values for each muscle group. The controller performs the following preprocessing steps on the raw digital electromyography signals acquired in step S1 to extract the normalized muscle activation of each muscle group: Preprocessing example: (1) Perform digital bandpass filtering on the original digital electromyography (EMG) signal with a passband range of 20 to 500 Hz to further eliminate low-frequency motion artifacts and high-frequency electromagnetic interference, and improve the signal-to-noise ratio; (2) Subtract the DC component from the digitally filtered signal to make the signal mean zero, and eliminate the influence of baseline drift on subsequent calculations; (3) Take the absolute value of the mean-removed signal to convert the bipolar signal into a unipolar signal and obtain the instantaneous amplitude information of the signal; (4) Perform low-pass filtering on the full-wave rectified signal with a cutoff frequency of 6 Hz to obtain the envelope of the EMG signal. This envelope reflects the change in the activation intensity of the muscle on the time axis and is denoted as the EMG signal envelope value of the m-th muscle group at time t. It should be noted that those skilled in the art can also use other existing and mature preprocessing methods.
[0030] Based on the above preprocessing, the normalized muscle activation level is calculated using the following formula. :
[0031] in, Let be the envelope value of the electromyographic signal of the m-th muscle group at time t. The maximum electromyographic amplitude measured during maximal voluntary contraction of this muscle group, after normalization. The value ranges from 0% to 100%.
[0032] After the above processing, the normalized muscle activation values of eight muscle groups are obtained at each time step, forming an eight-dimensional normalized muscle activation value vector. Where T represents the transpose, and each component of this vector corresponds to the normalized muscle activation of the left rectus femoris, left biceps femoris, left tibialis anterior, left gastrocnemius, right rectus femoris, right biceps femoris, right tibialis anterior, and right gastrocnemius.
[0033] 3. Step S3: Input the normalized muscle activation values of each muscle group into the pre-trained Gaussian mixture model and perform likelihood calculation. The Gaussian mixture model is stored in the controller, and its construction and training process is completed offline before the exoskeleton device leaves the factory, as detailed below: Step S3.1: Collect surface electromyography (EMG) signals of key muscle groups in the bilateral lower limbs of multiple normal children during walking; select normal children aged 3 to 6 years with no history of neuromuscular diseases as subjects, place EMG sensors on the rectus femoris, biceps femoris, tibialis anterior, and gastrocnemius muscles of the subjects' bilateral lower limbs, guide the subjects to walk on flat ground at a natural pace, and simultaneously collect eight channels of EMG signals at a sampling frequency of 1000 Hz; process the collected raw EMG signals according to the preprocessing procedure described in step S2 to obtain the normalized muscle activation vector at each time point, and aggregate the vectors at all time points to construct a training sample set.
[0034] Step S3.2: Construct a Gaussian mixture model: The Gaussian mixture model is defined by N Gaussian components, each defined by the following three parameters: a mean vector... , representing the average activation level of each muscle in the typical muscle synergistic activation pattern represented by this component; an 8×8 covariance matrix. This represents the degree of variability in muscle activation within this component and the correlation between muscles; a weighting coefficient. This indicates the proportion of the component in the mixed model, and all weight coefficients satisfy the constraints. and >0; where the number of Gaussian components N is determined according to the Bayesian information criterion, and is usually 4 to 6. In this embodiment, N=5.
[0035] Step S3.3: Train the parameters of the Gaussian mixture model using the Expectation-Maximization (EM) algorithm. The Expectation-Maximization algorithm is an iterative optimization algorithm that includes the following two alternating steps: Expectation step: Based on the current model parameters, calculate the posterior probability of each training sample belonging to each Gaussian component, i.e., the responsibility of the nth component for the kth sample. :
[0036] In the formula, Let be the normalized muscle activation vector of the k-th training sample. It is a multidimensional Gaussian probability density function; Maximization step: Update the parameters of each Gaussian component based on the posterior probability calculated in the expectation step.
[0037]
[0038]
[0039] In the formula, K is the total number of training samples; Iteratively execute the above steps until the log-likelihood function value converges. After convergence, calculate the mean vector of each Gaussian component. Covariance matrix and weighting coefficients Save this as a knowledge template for normal muscle coordination patterns; simultaneously, record the log-likelihood values of all samples in the training set and obtain their maximum values. and minimum value All the above parameters and recorded values are stored in the controller's program memory for online inference.
[0040] When used online, for each time step, the normalized muscle activation vector... The controller calculates the likelihood probability that it belongs to the Gaussian mixture model. :
[0041] In the formula, The complete mathematical expression for the probability density function of the nth Gaussian component is:
[0042] in, For vector dimensions, Let be the determinant of the covariance matrix. It is the inverse of the covariance matrix; At the same time, determine the Gaussian component index that maximizes the weighted probability density. :
[0043] The index This indicates the mean vector of the most similar normal cooperating module matched by the current muscle activation pattern. This will serve as a reference benchmark for calculating muscle activation corrections in subsequent steps.
[0044] 4. Step S4: Calculate the overall anomaly degree E based on the likelihood probability; During the online inference phase, the overall anomaly degree E is calculated using the following formula:
[0045] In the formula, The current likelihood probability, and These are the maximum and minimum values of the log-likelihood of the Gaussian mixture model on the training set, which have been recorded and stored in advance during the training phase.
[0046] Based on the above, the current log-likelihood value is linearly mapped to the [0,1] interval. The closer the current muscle activation pattern is to the muscle coordination pattern of a normal toddler, the higher the likelihood probability. The higher the probability, the closer the overall abnormality E is to 0; conversely, the more severe the abnormality of the muscle coordination pattern, the lower the likelihood probability, and the closer E is to 1.
[0047] 5. Step S5: Based on the comprehensive anomaly degree E and the corresponding auxiliary strength coefficient. Generate auxiliary torques for each joint. Based on the comprehensive anomaly degree E calculated in step S4 and the corresponding auxiliary strength coefficient, the controller generates auxiliary torques for each joint: (1) Obtaining the auxiliary strength coefficient: auxiliary strength coefficient The intensity is allocated based on the wearer's left and right lower limb motor abilities, with the side with weaker motor ability receiving a larger auxiliary strength coefficient. The specific allocation method is as follows:
[0048] In the formula, The auxiliary strength coefficient on the left side. The auxiliary strength coefficient on the right side. The overall motor function score for the left lower limb was determined. The overall motor ability score of the right lower limb; The method for obtaining the comprehensive motor ability score mentioned above is as follows: During the initial calibration phase of training, the wearer is guided to perform bilateral symmetrical calibration movements in sequence. These movements include hip flexion, knee extension, and ankle dorsiflexion, with each movement repeated multiple times. The maximum range of motion of each joint under effective motor intent is measured using the angle encoder built into each joint actuator. For each joint, its actual range of motion is calculated using the following formula:
[0049] in, Let be the maximum angle reached by the j-th joint during the calibration motion. To achieve the minimum angle, the joint angle data is provided by the angle encoder inside the corresponding driver; Then, calculate the overall motor ability score of the lower limb using the following formula. :
[0050] In the formula, The overall motor ability score for the left or right lower limb is determined. This represents the actual range of motion of the j-th joint on that side during the calibration movement. This represents the reference range of motion for the corresponding joint in a normal child of the same age on the same side. The number of joints involved in the assessment on this side (e.g.) =3, which refers to the three joints: hip, knee, and ankle. The value range is [0,1]. The smaller the value, the weaker the motor ability on that side, and the greater the gap with normal children of the same age.
[0051] (2) Calculation of muscle activation correction amount For each muscle i, its activation correction is calculated using the following formula. :
[0052] In the formula, The Gaussian component index that best matches the current activation mode (the best matching Gaussian component index determined in step S3). For the first The mean of the Gaussian components in the i-th muscle represents the expected activation level of that muscle under normal synergistic mode. The current actual activation level of the i-th muscle (obtained in step S2); The correction amount is determined based on the deviation between the mean vector of the Gaussian component that best matches the current activation mode in the Gaussian mixture model and the current activation level. When this occurs, it indicates that the muscle is currently underactivated, below the expected level in normal synergistic mode; when When the value is high, it indicates that the muscle is currently overactivated, exceeding the normal level; by mapping this correction amount to joint assist torque, the exoskeleton indirectly guides the wearer's muscle activation pattern toward a normal synergistic pattern.
[0053] (3) Determination of torque arm coefficient Torque arm coefficient This represents the torque exerted on the j-th joint by the contraction of the i-th muscle, derived from an anatomical model of the lower limb musculoskeletal system of infants and young children. For example, by performing magnetic resonance imaging or three-dimensional ultrasound scans on the lower limbs of several normal children aged 3 to 6 years, and after image segmentation and three-dimensional reconstruction, the sagittal lever arm lengths of the tendon attachment points of each key muscle group relative to the rotation centers of the hip, knee, and ankle joints are measured. The statistical distribution range of each lever arm length for all subjects is then taken as... The preset value range is given in this embodiment. The value range of the torque arm coefficient for each muscle-joint combination is shown in the table below: rectus femoris knee joint +1.5 to +2.2 biceps femoris knee joint -0.9 to -1.4 Tibialis anterior muscle ankle joint +1.0 to +1.6 Gastrocnemius muscle ankle joint -1.3 to -2.0 rectus femoris Hip joint +1.6 to +2.5 biceps femoris Hip joint -1.1 to -1.8 Positive values indicate that the muscle contraction generates a flexion or dorsiflexion torque, while negative values indicate an extension or plantar flexion torque. Preferably, to accommodate individual differences among infants of different body types, before the first wearing of the device or before each training session, the thigh and calf lengths manually measured and entered by the therapist can be compared with the reference lower limb lengths of the aforementioned standard infant model. Linear scaling is performed using the following scaling formula: ,in, To measure the wearer's lower limb length, The lower limb length is used as a reference for the standard model.
[0054] (4) Generation of auxiliary torque Reference auxiliary gain coefficient The value is determined based on the rated output torque of each joint actuator, and its range is from 10% to 30% of the rated torque of the actuator. In this embodiment, 20% is used. Auxiliary torques of each joint The calculation formula for (j, taking the hip, knee, and ankle joints) is as follows:
[0055] In the formula, As the reference auxiliary gain coefficient, The auxiliary strength coefficient for the corresponding side (taken for the left side) Take the right side ), Let be the torque arm coefficient of the i-th muscle with respect to the j-th joint. The summation symbol represents the activation correction for the i-th muscle. This indicates the summation over all key muscle groups (rectus femoris, biceps femoris, tibialis anterior, and gastrocnemius).
[0056] Based on the above, the greater the overall abnormality E (the more severe the deviation of the muscle coordination pattern from the normal), the greater the auxiliary mapping coefficient and the stronger the auxiliary torque; the weaker the motor ability on the side, the greater the auxiliary intensity coefficient and the greater the auxiliary torque. When a muscle is underactivated, the auxiliary torque drives the joint to move in the direction that promotes the contraction of that muscle; when a muscle is overactivated, the auxiliary torque drives the joint to move in the direction that inhibits that muscle. In this way, the wearer's muscle coordination pattern is guided closer to the normal reference pattern.
[0057] 6. Step S6: Drive the movement of each joint actuator with auxiliary torque. The controller sends the auxiliary torque generated in step S5 to the corresponding hip joint actuator, knee joint actuator, and ankle joint actuator via the CAN bus. The torque loop inside each actuator receives the target torque command, uses the target torque as input, and uses the actual torque fed back by the torque sensor inside the actuator as feedback. After calculation by the proportional controller, it outputs the pulse width modulation duty cycle signal of the drive motor to drive the coordinated movement of each joint actuator.
[0058] During training, the controller executes steps S1 to S6 in a 1-millisecond control cycle, continuously monitoring the degree of muscle synergy abnormality and applying auxiliary torque. As the wearer's muscle synergy pattern gradually improves, the overall abnormality E gradually decreases, and the auxiliary torque will adaptively decrease, realizing the rehabilitation principle of decreasing assistance. Example
[0059] This embodiment provides a wearable exoskeleton device for executing the control method described in Embodiment 1; as Figure 3 As shown, the main body of the wearable exoskeleton device adopts a mature lower limb exoskeleton mechanical structure in the field, including conventional components such as body support, pelvic support, thigh support, calf support and foot support. On the left and right lower limb parts of the main body of the device, hip joint actuator 2, knee actuator 3 and ankle actuator 4 are symmetrically arranged. Each actuator includes a DC brushless servo motor and a harmonic reducer connected to it. The output shaft of the motor drives the corresponding joint to rotate after the torque is amplified by the reducer.
[0060] The hip joint actuator 2 is installed at the joint connecting the pelvic support and the thigh support, and is used to drive the wearer's hip joint to perform flexion and extension movements in the sagittal plane; the knee actuator 3 is installed at the joint connecting the thigh support and the lower leg support, and is used to drive the knee joint to perform flexion and extension movements in the sagittal plane; the ankle actuator 4 is installed at the joint connecting the lower leg support and the foot support, and is used to drive the ankle joint to perform dorsiflexion and plantarflexion movements. Each actuator integrates an angle encoder and a torque sensor. The angle encoder provides real-time feedback on the current joint angle, and the torque sensor provides real-time feedback on the output torque, forming a torque closed-loop control.
[0061] The above-described mechanical structures are mature in this field, and their detailed mechanical construction will not be elaborated further here. This embodiment focuses on illustrating the sensor configuration and controller function configuration in conjunction with the control method of the present invention.
[0062] Controller 1 is fixedly mounted on the back panel of the device body and communicates with all drivers via a CAN bus to issue control commands and transmit sensor data. Preferably, the core processor of controller 1 is an ARM Cortex-M7 series microcontroller. Controller 1 integrates a program memory and a random access memory. The program memory stores the program code of the control method provided in Example 1 and the parameters of the pre-trained Gaussian mixture model (including the mean vector, covariance matrix, weight coefficients of each Gaussian component, and the maximum and minimum values of the log-likelihood of the training set). The random access memory is used for caching and calculating runtime data. Furthermore, controller 1 is also equipped with a touch screen as a human-machine interface to display the current training status and receive user input commands, and has a built-in rechargeable lithium battery to power the entire system.
[0063] The electromyography (EMG) sensors use Ag / AgCl surface electrodes, which are respectively arranged on the inner side of the flexible binding strap of the main body of the device, corresponding to four key muscle groups of the wearer's two lower limbs: rectus femoris (front of the thigh, responsible for knee extension and hip flexion), biceps femoris (back of the thigh, responsible for knee flexion and hip extension), tibialis anterior (front of the lower leg, responsible for ankle dorsiflexion), and gastrocnemius (back of the lower leg, responsible for ankle plantarflexion). A total of eight EMG sensors are arranged on both sides. Each EMG sensor collects the surface EMG signal of the corresponding muscle group. The raw signal is pre-amplified differentially, hardware bandpass filtered, and analog-to-digital converted before outputting the raw digital EMG signal to controller 1.
[0064] like Figure 2 As shown, controller 1 is functionally divided into the following units: a signal preprocessing unit, used to execute the electromyographic signal preprocessing process described in step S2 of embodiment 1, including digital bandpass filtering, mean removal processing, full-wave rectification, low-pass filtering for envelope calculation, and normalization processing; a co-analysis unit, storing pre-trained Gaussian mixture model parameters, used to execute the likelihood probability calculation and best-matching component determination in step S3 of embodiment 1; an anomaly assessment unit, used to execute the comprehensive anomaly E calculation described in step S4 of embodiment 1; an auxiliary torque generation unit, used to execute the auxiliary torque calculation described in step S5 of embodiment 1, including motor ability score calculation, auxiliary intensity coefficient allocation, muscle activation correction calculation, and torque synthesis; and a control execution unit, used to execute the drive command issuance described in step S6 of embodiment 1. The specific working methods of the above functional units have been described in detail in embodiment 1. Each functional unit is stored in the program memory of controller 1 in the form of program code and is called and executed by the central processing unit of controller 1.
[0065] With the above hardware configuration, the wearable exoskeleton device in this embodiment can collect electromyographic signals of the wearer's bilateral lower limbs in real time. After signal preprocessing, the signals are input into a pre-trained Gaussian mixture model for muscle coordination abnormality assessment. Based on the assessment results and the torque arm coefficient of each joint actuator, auxiliary torque is generated to drive the movement of each joint, thereby realizing rehabilitation training based on muscle coordination pattern analysis.
[0066] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.
Claims
1. A control method for a wearable exoskeleton device based on muscle coordination, characterized in that, Includes the following steps: Step S1: Real-time acquisition of electromyographic signals of key muscle groups in both lower limbs of the exoskeleton device wearer, including rectus femoris, biceps femoris, tibialis anterior, and gastrocnemius. Step S2: Preprocess the electromyographic signals and extract the normalized muscle activation of each muscle group; Step S3: Input the normalized muscle activation of each muscle group into the pre-trained Gaussian mixture model, calculate the likelihood probability that the normalized muscle activation vector belongs to the normal muscle synergy mode, and determine the best matching Gaussian component. Step S4: Calculate the overall anomaly degree E based on the likelihood probability; Step S5: Based on the comprehensive anomaly degree E and the corresponding auxiliary strength coefficient E on the side This generates auxiliary torques for each joint; Step S6: Drive the movement of each joint actuator with the auxiliary torque.
2. The control method for a wearable exoskeleton device based on muscle coordination according to claim 1, characterized in that: In step S2, the method for extracting the normalized muscle activation of each muscle group is as follows: First, the original surface electromyography (EMG) signal is preprocessed to extract the envelope of the EMG signal; then, the normalized muscle activation is calculated using the following formula. : ; in, Let be the envelope value of the electromyographic signal of the m-th muscle group at time t. This represents the maximum electromyographic amplitude measured during maximal voluntary contraction of this muscle group.
3. The control method for a wearable exoskeleton device based on muscle coordination according to claim 2, characterized in that: In step S3, the construction of the pre-trained Gaussian mixture model includes the following steps: Step S3.1: Collect surface electromyography (EMG) signals of key muscle groups in both lower limbs of multiple normal children during walking, obtain normalized muscle activation vectors after preprocessing, and construct a training sample set; Step S3.2: Construct a Gaussian mixture model, which consists of n Gaussian components, each of which includes a mean vector. A covariance matrix and a weighting coefficient ,and ; Step S3.3: Train the Gaussian mixture model using the expectation-maximization algorithm. After training, fix the parameters of each Gaussian component in the controller of the wearable exoskeleton device.
4. The control method for a wearable exoskeleton device based on muscle coordination according to claim 3, characterized in that: In step S3, the normalized muscle activation vector at the current moment Likelihood probability of belonging to normal muscle coordination pattern Calculate using the following formula: ; Best Match Component Index for: ; In the formula, Let be the probability density function of the nth Gaussian component.
5. The control method for a wearable exoskeleton device based on muscle coordination according to claim 4, characterized in that: In step S4, the overall anomaly degree E is calculated using the following formula: In the formula, The current likelihood probability is... and These are the maximum and minimum values of the log-likelihood of the Gaussian mixture model on the training set, respectively.
6. The control method for a wearable exoskeleton device based on muscle coordination according to claim 5, characterized in that: In step S5, the auxiliary torque of each joint Calculate using the following formula: In the formula, As the reference auxiliary gain coefficient, This is the auxiliary strength coefficient for the corresponding side. Let be the torque arm coefficient of the i-th muscle with respect to the j-th joint; The activation correction for the i-th muscle is calculated using the following formula: ； In the formula, The Gaussian component index that best matches the current activation mode. For the first The mean of the Gaussian components in the i-th muscle. Let be the current actual activation level of the i-th muscle.
7. The control method for a wearable exoskeleton device based on muscle coordination according to claim 6, characterized in that: The auxiliary strength coefficient on the corresponding side The calculation method is as follows: ； In the formula, The auxiliary strength coefficient on the left side. The auxiliary strength coefficient on the right side. The overall motor function score for the left lower limb was determined. The overall motor ability score of the right lower limb is determined.
8. The control method for a wearable exoskeleton device based on muscle coordination according to claim 7, characterized in that: The comprehensive motor ability score of the left lower limb Comprehensive motor ability score of the right lower limb The calculation method is as follows: ； In the formula, A comprehensive motor ability score for the left or right lower limb. This represents the actual range of motion of the j-th joint on that side during the calibration movement. This represents the reference range of motion for the corresponding joint in a normal child of the same age on that side. This refers to the number of joints on that side involved in the assessment.
9. The control method for a wearable exoskeleton device based on muscle coordination according to claim 6, characterized in that: The reference auxiliary gain coefficient The value is determined based on the rated output torque of each joint actuator in the wearable exoskeleton device, and its range is from 10% to 30% of the rated torque of the actuator.
10. The control method for a wearable exoskeleton device based on muscle coordination according to claim 6, characterized in that: The The possible values are as follows: (1) Targeting the rectus femoris muscle and knee joint, The value ranges from +1.5 to +2.2; (2) Targeting the biceps femoris and knee joint, The value ranges from -0.9 to -1.4; (3) Targeting the tibialis anterior muscle and ankle joint, The value ranges from +1.0 to +1.6; (4) Targeting the gastrocnemius muscle and ankle joint, The value ranges from -1.3 to -2.0; (5) Targeting the rectus femoris and hip joint, The value ranges from +1.6 to +2.5; (6) Targeting the biceps femoris and hip joint, The value range is from -1.1 to -1.8; Positive values indicate that the muscle contraction produces flexion or dorsiflexion torque, while negative values indicate extension or plantarflexion torque.