Exoskeleton equipment control method, device, equipment, medium and program product

Abstract

The invention relates to an exoskeleton equipment control method and device, equipment, a medium and a program product. The method comprises the following steps: in response to a signal acquisition instruction, acquiring an original electromyographic signal generated when an operator operates exoskeleton equipment through a surface electrode placed on the muscle skin surface of the operator, determining a target electric signal envelope according to the original electromyographic signal, and determining a torque prediction model according to the target electric signal envelope and the torque prediction model. Target joint torque data is determined, the target joint torque data is adopted to control the exoskeleton equipment to move, and the torque prediction model is obtained by training a variational auto-encoder model through the sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals. By adopting the method, the flexibility of exoskeleton equipment control can be improved.

Description

Technical Field

[0001] The present application relates to the field of device control technology, and in particular to an exoskeleton device control method, apparatus, device, medium and program product. Background Art

[0002] In the existing high-voltage direct current transmission technology, a converter station is required to convert AC power into DC power for long-distance transmission, and then the DC power is converted back into AC power at the transmission terminal for use in the power grid. Therefore, in order to ensure the reliability of converter station equipment, maintenance methods for converter station equipment have emerged.

[0003] In order to ensure the safety of operators, existing maintenance methods generally require operators to control exoskeleton equipment to perform maintenance on converter station equipment.

[0004] However, in existing maintenance methods, the exoskeleton device requires manual setting of the start-up conditions, which reduces the flexibility of the exoskeleton device control. Summary of the invention

[0005] Based on this, it is necessary to provide an exoskeleton device control method, device, equipment, medium and program product that can improve the control flexibility of the exoskeleton device in response to the above technical problems.

[0006] In a first aspect, the present application provides an exoskeleton device control method, comprising:

[0007] In response to the signal collection instruction, the original electromyographic signal generated by the operator when operating the exoskeleton device is collected through the surface electrodes placed on the muscle skin surface of the operator;

[0008] Determine the target electrical signal envelope according to the original electromyographic signal;

[0009] According to the target electrical signal envelope and the torque prediction model, the target joint torque data is determined, and the target joint torque data is used to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training a variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0010] In one embodiment, determining a target electrical signal envelope according to an original electromyographic signal includes:

[0011] The original electromyographic signal is filtered to obtain the target electromyographic signal; the target electromyographic signal is input into the envelope extraction model to obtain the target electrical signal envelope; wherein the envelope extraction model is constructed based on the convolutional neural network and the capsule network.

[0012] In one embodiment, filtering the original electromyographic signal to obtain the target electromyographic signal includes:

[0013] The original electromyographic signal is preliminarily filtered to obtain a candidate electromyographic signal; according to a preset signal cutoff frequency, the high-frequency signal in the candidate electromyographic signal is filtered to obtain a target electromyographic signal; wherein the high-frequency signal is a signal whose frequency is higher than the signal cutoff frequency.

[0014] In one embodiment, inputting the target electromyographic signal into an envelope extraction model to obtain the target electromyographic signal envelope includes:

[0015] The target electromyographic signal is input into the convolutional neural network to obtain the local features of the target electromyographic signal; the local features are input into the capsule network to obtain the target electrical signal envelope.

[0016] In one embodiment, the torque prediction model includes an encoder and a decoder; determining target joint torque data according to the target electrical signal envelope and the torque prediction model includes:

[0017] The target electrical signal envelope is input into the encoder, and the electromyographic signal characteristics corresponding to the target electrical signal envelope are obtained after the encoder encodes the target electrical signal envelope; the electromyographic signal characteristics are input into the decoder, and the target joint torque data are obtained after the decoder decodes the electromyographic signal characteristics.

[0018] In one embodiment, a variational autoencoder model is trained using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals, including:

[0019] Obtain sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals; determine the sample electrical signal envelope based on the sample electromyographic signals; input the sample electrical signal envelope into the variational autoencoder model to obtain reconstructed joint torque data; determine the model reconstruction loss and the model divergence loss based on the sample joint torque data and the reconstructed joint torque data; use the model reconstruction loss and the model divergence loss to train the variational autoencoder model to obtain a torque prediction model.

[0020] In a second aspect, the present application also provides an exoskeleton device control device, comprising:

[0021] A signal acquisition module, for responding to a signal acquisition instruction, and collecting raw electromyographic signals generated when the operator operates the exoskeleton device through surface electrodes placed on the muscle and skin surface of the operator;

[0022] An envelope determination module is used to determine the target electric signal envelope according to the original electromyographic signal;

[0023] The torque determination module is used to determine the target joint torque data according to the target electrical signal envelope and the torque prediction model, and use the target joint torque data to control the movement of the exoskeleton device.

[0024] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0025] In response to the signal collection instruction, the original electromyographic signal generated by the operator when operating the exoskeleton device is collected through the surface electrodes placed on the muscle skin surface of the operator;

[0026] Determine the target electrical signal envelope according to the original electromyographic signal;

[0027] According to the target electrical signal envelope and the torque prediction model, the target joint torque data is determined, and the target joint torque data is used to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training a variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0028] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0029] In response to the signal collection instruction, the original electromyographic signal generated by the operator when operating the exoskeleton device is collected through the surface electrodes placed on the muscle skin surface of the operator;

[0030] Determine the target electrical signal envelope according to the original electromyographic signal;

[0031] According to the target electrical signal envelope and the torque prediction model, the target joint torque data is determined, and the target joint torque data is used to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training a variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0032] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0033] In response to the signal collection instruction, the original electromyographic signal generated by the operator when operating the exoskeleton device is collected through the surface electrodes placed on the muscle skin surface of the operator;

[0034] Determine the target electrical signal envelope according to the original electromyographic signal;

[0035] According to the target electrical signal envelope and the torque prediction model, the target joint torque data is determined, and the target joint torque data is used to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training a variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0036] The above-mentioned exoskeleton device control method, apparatus, device, medium and program product, in response to the signal acquisition instruction, collects the original electromyographic signal generated by the operator when operating the exoskeleton device through the surface electrodes placed on the muscle skin surface of the operator, and determines the target electrical signal envelope according to the original electromyographic signal, and then determines the target joint torque data according to the target electrical signal envelope and the torque prediction model, and uses the target joint torque data to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training the variational autoencoder model using sample electromyographic signals and sample joint torque data. Compared with the related art, which requires manual setting of the start-up conditions of the exoskeleton device, the above-mentioned method uses a trained torque prediction model to predict the target joint torque data according to the target electrical signal envelope corresponding to the original electromyographic signal, and then uses the target joint torque data to control the movement of the exoskeleton device, which can improve the flexibility of the exoskeleton device control. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 is a flow chart of an exoskeleton device control method in one embodiment;

[0039] Figure 2 is a schematic diagram of a process flow of a target electrical signal envelope in one embodiment;

[0040] Figure 3 A schematic diagram of a process for determining a target electromyographic signal in one embodiment;

[0041] Figure 4 A schematic diagram of a process for determining target joint torque data in one embodiment;

[0042] Figure 5 A schematic diagram of a process for training a torque prediction model in one embodiment;

[0043] Figure 6 A schematic diagram of a flow chart of an exoskeleton device control method in another embodiment;

[0044] Figure 7 is a structural block diagram of an exoskeleton device control device in one embodiment;

[0045] Figure 8 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] In the existing high-voltage direct current transmission technology, a converter station is required to convert AC power into DC power for long-distance transmission, and then the DC power is converted back into AC power at the transmission terminal for use in the power grid. Therefore, in order to ensure the reliability of converter station equipment, maintenance methods for converter station equipment have emerged.

[0048] In order to ensure the safety of operators, the existing maintenance methods generally require operators to control the exoskeleton equipment to perform maintenance on the converter station equipment. However, in the existing maintenance methods, the exoskeleton equipment needs to manually set the start conditions, which reduces the flexibility of the exoskeleton equipment control.

[0049] Based on this, in an exemplary embodiment, a method for controlling an exoskeleton device is provided, and the method is described by taking the method applied to a control device associated with an exoskeleton device as an example, wherein the control device may be independent of the exoskeleton device or may be deployed in the exoskeleton device, and this embodiment does not limit this. Figure 1 As shown, the specific steps include:

[0050] S101, in response to a signal collection instruction, collecting raw electromyographic signals generated when an operator operates an exoskeleton device through surface electrodes placed on the surface of the operator's muscles and skin.

[0051] The signal acquisition instruction is used to represent an instruction for acquiring electromyographic signals. For example, the signal acquisition instruction may be issued according to a fixed signal acquisition interval, or may be issued when an operator is detected operating the exoskeleton device, which is not limited in this embodiment. The original electromyographic signal is a surface electromyographic signal that has been amplified.

[0052] It is understandable that when the operator's joints move, the muscles associated with the joints will be stretched or contracted. Since the stretching or contraction of muscles is caused by changes in muscle fibers, when muscle fibers are activated, bioelectric ions in the cells will flow, resulting in a corresponding temporary change in muscle potential. This change in potential during muscle activity is the electromyographic signal. The electromyographic signal can reflect the degree of muscle activity to a certain extent. When the degree of muscle activation is high, the corresponding electromyographic signal will be larger, and vice versa.

[0053] Based on this, in order to better capture the electromyographic signal, the electromyographic signal of the operator can be obtained through one or more needle electrodes inserted through the skin and into the muscle. However, since the method of using needle electrodes to obtain electromyographic signals is relatively complicated, the surface electromyographic signal generated by the operator when operating the exoskeleton device can be obtained through surface electrodes placed on the surface of the operator's muscle skin.

[0054] It is understandable that the strength, size and frequency of the surface electromyographic signal are correlated with the strength of the muscle when it contracts. That is to say, for a certain movement state, a certain surface electromyographic signal can be obtained corresponding to it. Therefore, the change of the corresponding elbow joint can be determined according to the change of the surface electromyographic signal, and then the exoskeleton device can be controlled.

[0055] For example, for elbow flexion and extension, when the elbow is bent, the biceps brachii is compressed, and when the elbow is extended, the biceps brachii is stretched. During the entire elbow flexion and extension process, the muscle force of the biceps brachii will change with the rotation of the joint, and the corresponding joint torque of the elbow will also change. In this process, the surface electromyographic signal of the biceps brachii will also change synchronously.

[0056] In one possible implementation, the surface electromyographic signal generated when the operator operates the exoskeleton device can be collected by placing surface electrodes on the surface of the operator's muscle skin according to a preset sampling frequency within a fixed sampling period. For example, the sampling frequency can be set to 1000Hz and the collection time can be set to 60s, at which time 60,000 data points (surface electromyographic signal) can be obtained. Among them, the amplitude of the surface electromyographic signal is about 0~1.5mv, the effective frequency range is about 0-500Hz, and the main energy is concentrated in the range of 50-150Hz.

[0057] Due to the influence of the external environment, the surface electromyography signals directly obtained will have greater noise interference. In order to solve this problem, the collected surface electromyography signals can be filtered. For example, a Butterworth bandpass filter can be used to filter out low-frequency interference signals with a frequency lower than 10Hz due to electrode and muscle movement, as well as invalid signals higher than 500Hz. Then, a notch filter can be used to filter out the power frequency interference caused by 50Hz AC in the surface electromyography signal.

[0058] Furthermore, since the strength of the surface electromyographic signal is very weak, after acquiring the surface electromyographic signal, the surface electromyographic signal can be amplified by a surface electromyographic signal sensor to obtain the original electromyographic signal. For example, a biovoltage amplifier circuit can be used to amplify the surface electromyographic signal by 1000 times to obtain the original electromyographic signal.

[0059] S102, determining a target electric signal envelope according to the original electromyographic signal.

[0060] Among them, the target electrical signal envelope is a signal envelope obtained based on the original electromyographic signal. Furthermore, the signal envelope of the electromyographic signal can reflect the change of the signal amplitude over time, that is, the target electrical signal envelope can reflect the contraction and relaxation of the muscle.

[0061] In one possible implementation, since the main energy of the surface electromyographic signal is concentrated in the range of 50-150Hz, the high-frequency signal in the original electromyographic signal can be filtered out first to obtain the electromyographic signal to be processed; then, the electromyographic signal to be processed can be input into the envelope extraction device to obtain the target electrical signal envelope.

[0062] For example, the electromyographic signal to be processed may be input into a 1 Hz first-order Butterworth low-pass filter, and the electromyographic signal to be processed may be processed by the 1 Hz first-order Butterworth low-pass filter to obtain the target electric signal envelope. For example, the envelope data may be {0.5, 0.6, 0.8…}.

[0063] S103, determining target joint torque data according to the target electrical signal envelope and the torque prediction model, and using the target joint torque data to control the movement of the exoskeleton device.

[0064] Among them, the torque prediction model refers to a neural network model that can predict the joint torque data corresponding to the electrical signal envelope; further, the torque prediction model is obtained by training the variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals. The variational autoencoder is a neural network model for semi-supervised learning, which combines the ideas of autoencoders and variational inference, and can learn the potential representation of data and generate new data samples with diversity. The target joint torque data is the joint torque data predicted based on the target electrical signal envelope.

[0065] In one embodiment, the target electrical signal envelope can first be linearly normalized and nonlinearly normalized to be mapped to the range of 0-100. For example, the normalized envelope data can be {50, 60, 80...}; then, the processed target electrical signal envelope is input into a trained torque prediction model, and the torque prediction model outputs the target joint torque data based on the processed target electrical signal envelope and model parameters.

[0066] Furthermore, after determining the target joint torque data, the target joint torque data can be used to control the movement of the exoskeleton device, thereby assisting the operator in performing dangerous tasks such as maintenance of converter station equipment.

[0067] In addition, the above-mentioned method for determining the target joint torque data can be applied not only to exoskeleton devices, but also to other fields that require electromyographic signal analysis and processing, such as medical rehabilitation, sports science, etc. In this embodiment, there is no limitation on this.

[0068] In the above exoskeleton device control method, in response to the signal acquisition instruction, the surface electrodes placed on the muscle skin surface of the operator are used to collect the original electromyographic signals generated when the operator operates the exoskeleton device, and the target electrical signal envelope is determined based on the original electromyographic signals, and then the target joint torque data is determined based on the target electrical signal envelope and the torque prediction model, and the target joint torque data is used to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training the variational autoencoder model using sample electromyographic signals and sample joint torque data. Compared with the related art, which requires manual setting of the start-up conditions of the exoskeleton device, the above method uses a trained torque prediction model to predict the target joint torque data based on the target electrical signal envelope corresponding to the original electromyographic signal, and then uses the target joint torque data to control the movement of the exoskeleton device, which can improve the flexibility of exoskeleton device control.

[0069] In order to ensure the accuracy of the target electrical signal envelope, based on the above embodiment, in this embodiment, an optional method for determining the target electrical signal envelope is provided, such as Figure 2 As shown, the specific steps include:

[0070] S201, filtering the original electromyographic signal to obtain a target electromyographic signal.

[0071] The target electromyographic signal is used to represent the electromyographic signal obtained after filtering the original electromyographic signal.

[0072] Optionally, since the main energy of the surface electromyographic signal is concentrated in the range of 50-150Hz, the high-frequency signal in the original electromyographic signal can be filtered out first, and then the filtered original electromyographic signal can be converted into the same direction using full-wave rectification technology, that is, the negative part of the signal is mapped to the positive part, so that all values ​​of the signal are in the same direction, thereby obtaining the target electromyographic signal.

[0073] Exemplarily, full-wave rectification can be achieved by taking the absolute value of the electromyographic signal, that is, s''(t)=|s'(t)|, wherein s'(t) is the original electromyographic signal after filtering; s''(t) is the target electromyographic signal.

[0074] S202, inputting the target electromyographic signal into an envelope extraction model to obtain a target electrical signal envelope.

[0075] The envelope extraction model refers to a neural network model that can predict the envelope of an electrical signal. In this embodiment, the envelope extraction model is constructed based on a convolutional neural network and a capsule network.

[0076] In one possible implementation, the target electromyographic signal may be directly input into a trained envelope extraction model, and the envelope extraction model outputs the target electromyographic signal envelope according to the target electromyographic signal and model parameters.

[0077] In another possible implementation, the target electromyographic signal can be input into a convolutional neural network to obtain local features of the target electromyographic signal; the local features are input into a capsule network to obtain a target electrical signal envelope. The local features are signal features of the target electromyographic signal locally. The convolutional neural network is used to automatically extract local features of the surface electromyographic signal, while the capsule network is used to integrate these local features and generate global features to obtain the target electrical signal envelope. Furthermore, in this embodiment, the convolutional neural network can be an active sketch network.

[0078] Specifically, the target EMG signal can be input into the active sketch network. The active sketch network uses a convolutional layer to extract the local features of the target EMG signal through a window of size 3x3. Subsequently, a bias is added to the output result of the convolutional layer, and a nonlinear activation function is passed to obtain the local features of the target EMG signal.

[0079] Among them, the nonlinear activation function can be the linear rectification function ReLU; the formula corresponding to the convolution layer is , I is the input target EMG signal; m is the parameter corresponding to the target EMG signal; K is the convolution kernel; n is the parameter corresponding to the convolution kernel; s(i,j) is the feature map corresponding to the output local feature.

[0080] Furthermore, the local features can be input into the capsule network, which integrates the local features to obtain the output features. The activation function of the capsule network is: ,s j is the vector input of the capsule network, i.e., the local feature; v j It is the output feature after the activation function; then, the output feature of the capsule network can be deconvolved to reconstruct the envelope of the original electromyographic signal, that is, the target electrical signal envelope.

[0081] In an embodiment of the present application, after filtering the original electromyographic signal, the target electromyographic signal obtained is input into an envelope extraction model to obtain the target electrical signal envelope, thereby improving the accuracy of envelope extraction.

[0082] In order to ensure the accuracy of the target electromyographic signal, based on the above embodiment, in this embodiment, an optional method for determining the target electromyographic signal is provided, such as Figure 3 As shown, the specific steps include:

[0083] S301, performing preliminary filtering processing on the original electromyographic signal to obtain a candidate electromyographic signal.

[0084] The candidate electromyographic signal refers to the electromyographic signal after preliminary filtering.

[0085] It is understandable that in the process of preprocessing the original EMG signal, the non-random components in the EMG signal, such as environmental noise, equipment noise, etc., must be filtered out first in order to better capture and analyze the random components of the EMG signal, that is, the muscle activity electrical signal.

[0086] In one feasible implementation, in order to effectively remove the non-random components in the surface EMG signal, a 410 Hz first-order Butterworth high-pass filter can be used to perform preliminary filtering on the original EMG signal to obtain a candidate EMG signal. This process can remove approximately 99% of the signal power and effectively eliminate the low-frequency interference noise in the EMG signal.

[0087] Among them, the specific filtering formula of the Butterworth high-pass filter is: , H is the frequency response of the filter, j is the imaginary unit, is the frequency, is the cut-off frequency which is 410Hz.

[0088] S302, filtering high-frequency signals in the candidate electromyographic signals according to a preset signal cutoff frequency to obtain target electromyographic signals.

[0089] Among them, the signal cutoff frequency is used to characterize a numerical value that can measure the signal frequency. In this embodiment, the signal cutoff frequency can be determined based on artificial experience or through a large number of experiments. This application does not impose any restrictions on this. A high-frequency signal is a signal whose frequency is higher than the signal cutoff frequency.

[0090] In one embodiment, the high-frequency signal in the candidate electromyographic signal that is higher than the signal cutoff frequency can be filtered according to the preset signal cutoff frequency to obtain the target electromyographic signal. For example, a low-pass filter can be used to remove the high-frequency noise in the candidate electromyographic signal that is higher than the signal cutoff frequency.

[0091] The low-pass filter selects Butterworth filter and Chebyshev filter. Among them, Butterworth filter has the flattest response in the passband, that is, the frequency response curve is smooth, while Chebyshev filter allows some ripples in the passband in exchange for a steeper cutoff slope. Furthermore, the filtering formula of Butterworth filter is, , H is the frequency response of the filter, j is the imaginary unit, is the frequency, is the set signal cutoff frequency, n is the order of the filter, and the order needs to be selected according to the filtering requirements.

[0092] For example, since the frequency of electromyographic signals is generally distributed in the range of 0-500 Hz, and most of the effective information is concentrated in the range of 20-150 Hz, a Butterworth filter with a signal cutoff frequency set to 150 Hz can be used to filter high-frequency signals above 150 Hz in the candidate electromyographic signals to obtain the target electromyographic signals.

[0093] In an embodiment of the present application, a candidate electromyographic signal is obtained by performing preliminary filtering processing on the original electromyographic signal, and the high-frequency signal in the candidate electromyographic signal is filtered according to a preset signal cutoff frequency to obtain a target electromyographic signal, thereby ensuring the accuracy of the target electromyographic signal.

[0094] In order to ensure the accuracy of the target joint torque data, based on the above embodiment, in this embodiment, the torque prediction model includes an encoder and a decoder; further, an optional method for determining the target joint torque data is provided, such as Figure 4 As shown, the specific steps include:

[0095] S401, inputting the target electrical signal envelope into the encoder, and obtaining the electromyographic signal feature corresponding to the target electrical signal envelope output by the encoder after encoding the target electrical signal envelope.

[0096] Among them, the electromyographic signal feature is the hidden feature extracted by the encoder from the target electrical signal envelope.

[0097] In one possible implementation, the target electrical signal envelope can be input into an encoder within a torque prediction model, which encodes the target electrical signal envelope and extracts hidden features of the encoded sequence; then, the extracted hidden features are output as electromyographic signal features corresponding to the target electrical signal envelope.

[0098] Exemplarily, since the distribution of the target electrical signal envelope x in the variational autoencoder can be defined as a Gaussian distribution, the following formula (1) can be referred to to determine the electromyographic signal feature corresponding to the target electrical signal envelope according to the target electrical signal envelope.

[0099] (1)

[0100] Among them, Encoder is the encoder; , N(0,1) is called the standard normal distribution; and is a function parameterized by the encoder neural network and can be viewed as a function of the input data x and , z is the output latent variable, that is, the electromyographic signal feature.

[0101] S402, inputting the electromyographic signal features into a decoder, and obtaining target joint torque data output by the decoder after decoding the electromyographic signal features.

[0102] In one embodiment, the electromyographic signal feature can be input into a decoder, and the decoder decodes the electromyographic signal feature to obtain the target joint torque data. For example, the following formula (2) can be referred to, and the electromyographic signal feature can be decoded to obtain the target joint torque data. Wherein, Decoder is a decoder; is the target joint torque data.

[0103] (2)

[0104] In an embodiment of the present application, the target electrical signal envelope is input into the encoder to obtain the electromyographic signal characteristics corresponding to the target electrical signal envelope, and the electromyographic signal characteristics are input into the decoder to obtain the target joint torque data, thereby ensuring the accuracy of the target joint torque data.

[0105] In order to ensure the accuracy of the torque prediction model, based on the above embodiment, in this embodiment, an optional method for training the torque prediction model is provided, such as Figure 5 As shown, the specific steps include:

[0106] S501, obtaining sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0107] The sample electromyographic signal is a standard electromyographic signal used to train the torque prediction model; and the sample joint torque data is the standard torque data corresponding to the standard electromyographic signal.

[0108] It is understandable that in order for the torque prediction model to learn the data features of the torque data, it is necessary to obtain standard sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0109] S502: Determine a sample electrical signal envelope according to the sample electromyographic signal.

[0110] The sample electrical signal envelope is the signal envelope corresponding to the sample electromyographic signal.

[0111] In one possible implementation, referring to the above steps S201-S202, the sample electromyographic signal is filtered and the filtered sample electromyographic signal is input into the envelope extraction model to obtain the sample electrical signal envelope.

[0112] S503, inputting the sample electrical signal envelope into the variational autoencoder model to obtain reconstructed joint torque data.

[0113] The reconstructed joint torque data is the joint torque data obtained after the variational autoencoder reconstructs the sample electrical signal envelope.

[0114] In one possible implementation, the sample electrical signal envelope may be input into a variational autoencoder model, and the variational autoencoder model outputs reconstructed joint torque data based on the sample electrical signal envelope and model parameters.

[0115] S504: Determine a model reconstruction loss and a model divergence loss according to the sample joint torque data and the reconstructed joint torque data.

[0116] Among them, the model reconstruction loss is used to measure the difference between the model's output and the original input; the model divergence loss is used to measure the difference between the model's potential representation and the prior distribution.

[0117] Optionally, the model reconstruction loss and the model divergence loss may be determined based on the difference between the sample joint torque data and the reconstructed joint torque data.

[0118] S505, using model reconstruction loss and model divergence loss to train the variational autoencoder model to obtain a torque prediction model.

[0119] It can be understood that the variational autoencoder model can be optimized based on the model reconstruction loss and the model divergence loss until the model reconstruction loss and the model divergence loss of the variational autoencoder model are minimized at the same time. At this time, the torque prediction model can be obtained.

[0120] In another possible implementation, for the encoder and decoder parts of the variational autoencoder model, a neural network with 100 neurons can be used as the hidden layer, and the training iteration number of the variational autoencoder model is set to 10,000 times. The optimizer uses a deep learning optimization algorithm (Adaptive Moment Estimation, Adam) with a learning rate of 0.001 to train the torque prediction model.

[0121] In an embodiment of the present application, a variational autoencoder model is trained using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals to obtain a torque prediction model, which enables the torque prediction model to learn the potential representation of the torque data so that the torque prediction model can reconstruct the torque data corresponding to the electrical signal envelope.

[0122] Figure 6 FIG. 2 is a flow chart of an exoskeleton device control method in another embodiment. Based on the above embodiment, this embodiment provides an optional example of an exoskeleton device control method. Figure 6 The specific implementation process is as follows:

[0123] S601, obtaining sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0124] S602: Determine a sample electrical signal envelope according to the sample electromyographic signal.

[0125] S603, inputting the sample electrical signal envelope into the variational autoencoder model to obtain reconstructed joint torque data.

[0126] S604, determining a model reconstruction loss and a model divergence loss according to the sample joint torque data and the reconstructed joint torque data.

[0127] S605, using model reconstruction loss and model divergence loss to train the variational autoencoder model to obtain a torque prediction model.

[0128] S606, in response to the signal collection instruction, collecting the original electromyographic signal generated when the operator operates the exoskeleton device through the surface electrodes placed on the muscle skin surface of the operator.

[0129] S607, performing preliminary filtering processing on the original electromyographic signal to obtain a candidate electromyographic signal.

[0130] S608, filtering the high-frequency signal in the candidate electromyographic signal according to the preset signal cutoff frequency to obtain the target electromyographic signal.

[0131] The high-frequency signal is a signal whose frequency is higher than the signal cutoff frequency.

[0132] S609, input the target electromyographic signal into a convolutional neural network to obtain local features of the target electromyographic signal.

[0133] S610, inputting the local features into the capsule network to obtain the target electrical signal envelope.

[0134] S611, inputting the target electrical signal envelope into the encoder of the torque prediction model, and obtaining the electromyographic signal characteristics corresponding to the target electrical signal envelope output by the encoder after encoding the target electrical signal envelope.

[0135] S612, input the electromyographic signal features into the decoder of the torque prediction model to obtain the target joint torque data output by the decoder after decoding the electromyographic signal features.

[0136] S613, using the target joint torque data to control the movement of the exoskeleton device.

[0137] The specific process of S601-S613 above can refer to the description of the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0138] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0139] Based on the same inventive concept, the embodiment of the present application also provides an exoskeleton device control device for implementing the exoskeleton device control method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more exoskeleton device control device embodiments provided below can refer to the limitations of the exoskeleton device control method above, and will not be repeated here.

[0140] In an exemplary embodiment, Figure 7 As shown, an exoskeleton device control device 1 is provided, comprising: a signal acquisition module 10, an envelope determination module 20 and a torque determination module 30, wherein:

[0141] The signal acquisition module 10 is used to respond to the signal acquisition instruction and collect the original electromyographic signal generated by the operator when operating the exoskeleton device through the surface electrodes placed on the muscle skin surface of the operator;

[0142] An envelope determination module 20 is used to determine the target electrical signal envelope according to the original electromyographic signal;

[0143] The torque determination module 30 is used to determine the target joint torque data based on the target electrical signal envelope and the torque prediction model, and use the target joint torque data to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training a variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0144] In an exemplary embodiment, the envelope determination module 20 includes:

[0145] A signal filtering unit is used to filter the original electromyographic signal to obtain a target electromyographic signal;

[0146] The envelope determination unit is used to input the target electromyographic signal into the envelope extraction model to obtain the target electrical signal envelope; wherein the envelope extraction model is constructed based on a convolutional neural network and a capsule network.

[0147] In an exemplary embodiment, the signal filtering unit is specifically used for:

[0148] The original electromyographic signal is preliminarily filtered to obtain a candidate electromyographic signal; according to a preset signal cutoff frequency, the high-frequency signal in the candidate electromyographic signal is filtered to obtain a target electromyographic signal; wherein the high-frequency signal is a signal whose frequency is higher than the signal cutoff frequency.

[0149] In an exemplary embodiment, the envelope determination unit is specifically configured to:

[0150] The target electromyographic signal is input into the convolutional neural network to obtain the local features of the target electromyographic signal; the local features are input into the capsule network to obtain the target electrical signal envelope.

[0151] In an exemplary embodiment, the torque prediction model includes an encoder and a decoder; the torque determination module 30 is specifically used to:

[0152] The target electrical signal envelope is input into the encoder, and the electromyographic signal characteristics corresponding to the target electrical signal envelope are obtained after the encoder encodes the target electrical signal envelope; the electromyographic signal characteristics are input into the decoder, and the target joint torque data are obtained after the decoder decodes the electromyographic signal characteristics.

[0153] In an exemplary embodiment, the exoskeleton device control apparatus 1 further includes a model training module, wherein the model training module is specifically used to:

[0154] Obtain sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals; determine the sample electrical signal envelope based on the sample electromyographic signals; input the sample electrical signal envelope into the variational autoencoder model to obtain reconstructed joint torque data; determine the model reconstruction loss and the model divergence loss based on the sample joint torque data and the reconstructed joint torque data; use the model reconstruction loss and the model divergence loss to train the variational autoencoder model to obtain a torque prediction model.

[0155] Each module in the above exoskeleton device control device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0156] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 8As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store electromyographic signal data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for controlling an exoskeleton device is implemented.

[0157] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0158] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0159] In response to the signal collection instruction, the original electromyographic signal generated by the operator when operating the exoskeleton device is collected through the surface electrodes placed on the muscle skin surface of the operator;

[0160] Determine the target electrical signal envelope according to the original electromyographic signal;

[0161] According to the target electrical signal envelope and the torque prediction model, the target joint torque data is determined, and the target joint torque data is used to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training a variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0162] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0163] The original electromyographic signal is filtered to obtain the target electromyographic signal; the target electromyographic signal is input into the envelope extraction model to obtain the target electrical signal envelope; wherein the envelope extraction model is constructed based on the convolutional neural network and the capsule network.

[0164] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0165] The original electromyographic signal is preliminarily filtered to obtain a candidate electromyographic signal; according to a preset signal cutoff frequency, the high-frequency signal in the candidate electromyographic signal is filtered to obtain a target electromyographic signal; wherein the high-frequency signal is a signal whose frequency is higher than the signal cutoff frequency.

[0166] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0167] The target electromyographic signal is input into the convolutional neural network to obtain the local features of the target electromyographic signal; the local features are input into the capsule network to obtain the target electrical signal envelope.

[0168] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0169] The target electrical signal envelope is input into the encoder, and the electromyographic signal characteristics corresponding to the target electrical signal envelope are obtained after the encoder encodes the target electrical signal envelope; the electromyographic signal characteristics are input into the decoder, and the target joint torque data are obtained after the decoder decodes the electromyographic signal characteristics.

[0170] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0171] Obtain sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals; determine the sample electrical signal envelope based on the sample electromyographic signals; input the sample electrical signal envelope into the variational autoencoder model to obtain reconstructed joint torque data; determine the model reconstruction loss and the model divergence loss based on the sample joint torque data and the reconstructed joint torque data; use the model reconstruction loss and the model divergence loss to train the variational autoencoder model to obtain a torque prediction model.

[0172] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0173] In response to the signal collection instruction, the original electromyographic signal generated by the operator when operating the exoskeleton device is collected through the surface electrodes placed on the muscle skin surface of the operator;

[0174] Determine the target electrical signal envelope according to the original electromyographic signal;

[0175] According to the target electrical signal envelope and the torque prediction model, the target joint torque data is determined, and the target joint torque data is used to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training a variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0176] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0177] The original electromyographic signal is filtered to obtain the target electromyographic signal; the target electromyographic signal is input into the envelope extraction model to obtain the target electrical signal envelope; wherein the envelope extraction model is constructed based on the convolutional neural network and the capsule network.

[0178] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0179] The original electromyographic signal is preliminarily filtered to obtain a candidate electromyographic signal; according to a preset signal cutoff frequency, the high-frequency signal in the candidate electromyographic signal is filtered to obtain a target electromyographic signal; wherein the high-frequency signal is a signal whose frequency is higher than the signal cutoff frequency.

[0180] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0181] The target electromyographic signal is input into the convolutional neural network to obtain the local features of the target electromyographic signal; the local features are input into the capsule network to obtain the target electrical signal envelope.

[0182] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0183] The target electrical signal envelope is input into the encoder, and the electromyographic signal characteristics corresponding to the target electrical signal envelope are obtained after the encoder encodes the target electrical signal envelope; the electromyographic signal characteristics are input into the decoder, and the target joint torque data are obtained after the decoder decodes the electromyographic signal characteristics.

[0184] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0185] Obtain sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals; determine the sample electrical signal envelope based on the sample electromyographic signals; input the sample electrical signal envelope into the variational autoencoder model to obtain reconstructed joint torque data; determine the model reconstruction loss and the model divergence loss based on the sample joint torque data and the reconstructed joint torque data; use the model reconstruction loss and the model divergence loss to train the variational autoencoder model to obtain a torque prediction model.

[0186] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0187] In response to the signal collection instruction, the original electromyographic signal generated by the operator when operating the exoskeleton device is collected through the surface electrodes placed on the muscle skin surface of the operator;

[0188] Determine the target electrical signal envelope according to the original electromyographic signal;

[0189] According to the target electrical signal envelope and the torque prediction model, the target joint torque data is determined, and the target joint torque data is used to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training a variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

[0190] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0191] The original electromyographic signal is filtered to obtain the target electromyographic signal; the target electromyographic signal is input into the envelope extraction model to obtain the target electrical signal envelope; wherein the envelope extraction model is constructed based on the convolutional neural network and the capsule network.

[0192] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0193] The original electromyographic signal is preliminarily filtered to obtain a candidate electromyographic signal; according to a preset signal cutoff frequency, the high-frequency signal in the candidate electromyographic signal is filtered to obtain a target electromyographic signal; wherein the high-frequency signal is a signal whose frequency is higher than the signal cutoff frequency.

[0194] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0195] The target electromyographic signal is input into the convolutional neural network to obtain the local features of the target electromyographic signal; the local features are input into the capsule network to obtain the target electrical signal envelope.

[0196] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0197] The target electrical signal envelope is input into the encoder, and the electromyographic signal characteristics corresponding to the target electrical signal envelope are obtained after the encoder encodes the target electrical signal envelope; the electromyographic signal characteristics are input into the decoder, and the target joint torque data are obtained after the decoder decodes the electromyographic signal characteristics.

[0198] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0199] Obtain sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals; determine the sample electrical signal envelope based on the sample electromyographic signals; input the sample electrical signal envelope into the variational autoencoder model to obtain reconstructed joint torque data; determine the model reconstruction loss and the model divergence loss based on the sample joint torque data and the reconstructed joint torque data; use the model reconstruction loss and the model divergence loss to train the variational autoencoder model to obtain a torque prediction model.

[0200] It should be noted that the data involved in this application (including but not limited to electromyographic signal data, etc.) are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0201] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0202] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0203] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for controlling an exoskeleton device, characterized in that: The method comprises: In response to the signal collection instruction, the original electromyographic signal generated by the operator when operating the exoskeleton device is collected through the surface electrodes placed on the muscle skin surface of the operator; Determining a target electrical signal envelope according to the original electromyographic signal; According to the target electrical signal envelope and the torque prediction model, the target joint torque data is determined, and the movement of the exoskeleton device is controlled by using the target joint torque data; wherein the torque prediction model is obtained by training a variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

2. The method according to claim 1, characterized in that: Determining the target electrical signal envelope according to the original electromyographic signal includes: Filtering the original electromyographic signal to obtain a target electromyographic signal; The target electromyographic signal is input into an envelope extraction model to obtain a target electrical signal envelope; wherein the envelope extraction model is constructed based on a convolutional neural network and a capsule network.

3. The method according to claim 2, characterized in that The filtering process of the original electromyographic signal to obtain a target electromyographic signal includes: The original electromyographic signal is subjected to preliminary filtering to obtain a candidate electromyographic signal; According to a preset signal cutoff frequency, the high-frequency signal in the candidate electromyographic signal is filtered to obtain a target electromyographic signal; wherein the high-frequency signal is a signal whose signal frequency is higher than the signal cutoff frequency.

4. The method according to claim 2, characterized in that: The step of inputting the target electromyographic signal into an envelope extraction model to obtain a target electrical signal envelope comprises: Inputting the target electromyographic signal into the convolutional neural network to obtain local features of the target electromyographic signal; The local features are input into the capsule network to obtain the target electrical signal envelope.

5. The method according to claim 1, characterized in that The torque prediction model includes an encoder and a decoder; Determining target joint torque data according to the target electrical signal envelope and the torque prediction model includes: Inputting the target electrical signal envelope into the encoder to obtain the electromyographic signal feature corresponding to the target electrical signal envelope output by the encoder after encoding the target electrical signal envelope; The electromyographic signal feature is input into the decoder to obtain target joint torque data outputted by the decoder after decoding the electromyographic signal feature.

6. The method according to claim 1, characterized in that The method of using the sample electromyographic signal and the sample joint torque data associated with the sample electromyographic signal to train the variational autoencoder model includes: Acquire sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals; Determining a sample electrical signal envelope according to the sample electromyographic signal; Inputting the sample electrical signal envelope into a variational autoencoder model to obtain reconstructed joint torque data; Determining a model reconstruction loss and a model divergence loss according to the sample joint torque data and the reconstructed joint torque data; The variational autoencoder model is trained using the model reconstruction loss and the model divergence loss to obtain the torque prediction model.

7. An exoskeleton device control device, characterized in that: The device comprises: A signal acquisition module, for responding to a signal acquisition instruction, and collecting raw electromyographic signals generated when the operator operates the exoskeleton device through surface electrodes placed on the muscle and skin surface of the operator; An envelope determination module, used to determine the target electrical signal envelope according to the original electromyographic signal; A torque determination module is used to determine target joint torque data based on the target electrical signal envelope and the torque prediction model, and use the target joint torque data to control the movement of the exoskeleton device; wherein the torque prediction model is obtained by training a variational autoencoder model using sample electromyographic signals and sample joint torque data associated with the sample electromyographic signals.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images