Walking-aid exoskeleton system control method based on terrain-physiology strong coupling perception

Abstract

The embodiment of the invention provides a walking-aid exoskeleton system control method based on terrain-physiology strong coupling perception, and belongs to the technical field of operation, and the method specifically comprises the steps: collecting time sequence features of muscle tension, plantar pressure and main joint angles of a patient, and carrying out the feature fusion of the time sequence features based on a cross-modal attention model, fused physiological features are obtained; acquiring topographic features, capturing local topographic features influencing the physiological state of the patient in the topographic features by using the fused physiological features to obtain physiological strong-coupling topographic features, and updating the fused physiological features according to the physiological strong-coupling topographic features; the updated fused physiological features are used as driving conditions, a noise prediction network is designed to gradually execute noise prediction and denoising processes from the diffusion moment # imgabs0 # to 1, and finally predicted multi-source physiological signals at the next moment are obtained; and performing feedback control on the exoskeleton system according to the predicted multi-source physiological signal at the next moment. By means of the scheme, the real-time performance and adaptability of exoskeleton system control are improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of operation technology, and in particular to a control method for a walking-assisting exoskeleton system based on terrain-physiological strong coupling perception. Background Art

[0002] At present, for patients with impaired limb function who need to be guided to recover, the use of walking exoskeleton scaffolds to assist movement is very important. It not only enables patients with mobility impairment to take care of themselves, but also can be used to guide patients to perform functional recovery training. Therefore, walking exoskeletons are crucial for patients with mobility impairment. In actual use, walking exoskeletons will face a variety of complex terrain environments. How to perceive and plan a reasonable gait to safely pass through these terrains is an important issue.

[0003] At present, walking exoskeleton devices have been widely used in patients with mobility impairments or patients who need limb function recovery to help them achieve self-care and functional rehabilitation training. However, in actual applications, patients will face various complex and unpredictable terrain environments, such as stairs, ramps, sand, etc. These terrain differences put higher demands on walking exoskeletons, requiring the equipment to be able to sense different terrains and adjust gait in real time to ensure the safety of patients and the effectiveness of the equipment. Therefore, how to design a walking exoskeleton system that can sense the surrounding environment in real time and dynamically adjust gait planning is an important problem that needs to be solved at present.

[0004] It can be seen that there is an urgent need for a control method for a walking-assisting exoskeleton system based on terrain-physiological strong coupling perception with high real-time control and adaptability. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a method for controlling a walking-assisting exoskeleton system based on terrain-physiological strong coupling perception, which at least partially solves the problems of poor control real-time performance and adaptability in the prior art.

[0006] The embodiment of the present invention provides a method for controlling a walking-assisting exoskeleton system based on terrain-physiological strong coupling perception, comprising: Step 1: collect the time series features of the patient's muscle tension, plantar pressure and main joint angles and fuse them based on the cross-modal attention model to obtain fused physiological features; Step 2, obtaining terrain features and using fused physiological features to capture local terrain representations that affect the patient's physiological state in the terrain features, obtaining physiologically strongly coupled terrain features, and updating the fused physiological features accordingly; Step 3: Using the updated fused physiological features as the driving condition, the noise prediction network is designed to gradually execute from the diffusion moment The noise prediction and denoising process from 1 to 1 finally obtains the predicted multi-source physiological signal at the next moment; Step 4: feedback control the exoskeleton system based on the predicted multi-source physiological signals at the next moment.

[0007] According to a specific implementation of the embodiment of the present invention, step 1 specifically includes: Step 1.1: Use electromyography sensors to collect muscle tension of specific muscle groups in the patient's legs ; Step 1.2: Use the plantar pressure pad to collect the plantar pressure of each area of ​​the patient's foot and calculate the pressure center based on it ; in, and Indicates the coordinates of the pressure center in the cushion surface coordinate system at the nth moment; Step 1.3: Use joint angle sensors to collect the main joint angles of the patient's legs ; Step 1.4, use a one-dimensional convolutional network to extract the temporal features of muscle tension, plantar pressure, and main joint angles, respectively. , and ; In step 1.5, the temporal features of muscle tension, plantar pressure, and main joint angles are fused based on the cross-modal attention model.

[0008] According to a specific implementation of the embodiment of the present invention, step 1.5 specifically includes: Step 1.5.1, calculate the attention weights between the electromyographic signal, plantar pressure, and joint angle and perform feature alignment ; ; in, Indicates the electromyography-foot pressure alignment feature, represents the myoelectric-joint angle alignment feature, , , are all trainable parameters, is the temperature coefficient, T represents the transposition operation; Step 1.5.2: Fuse the obtained cross-modal alignment features and use them as the patient's fused physiological features ; ; in, and are all weight coefficients.

[0009] According to a specific implementation of the embodiment of the present invention, step 2 specifically includes: Step 2.1: Use LiDAR to scan the terrain contour to obtain the 3D point cloud information of the terrain, and then use the pre-trained Point Transformer model to extract the features of the 3D point cloud information as the terrain features. ; Step 2.2: Using fused physiological features Capturing terrain features There are local topographic representations that affect the patient's physiological state, and physiological strong coupling topographic features are obtained. ; Step 2.3: Using physiologically strongly coupled terrain features Update and integrate physiological features Multi-source physiological representation in.

[0010] According to a specific implementation of the embodiment of the present invention, step 2.2 specifically includes: Step 2.2.1: Terrain features and fusion of physiological characteristics Input the multi-layer linear mapping model to calculate the terrain query items and key items of holistic physiological perception ; Step 2.2.2, query items based on terrain and key items of holistic physiological perception Computing holistic physiological perception attention mask : ; in, , , and All are trainable parameters; Step 2.2.3: Use the multi-layer perception model to integrate physiological features Perform fine-grained decomposition to obtain The fused physiological features of different representation levels are then used to calculate the perceptual key items of different granularity features based on the multi-layer linear mapping model. ; Step 2.2.4, using terrain query items and perceptual key items of different granularity features Computing fine-grained physiologically aware attention masks : ; in, represents the concatenation operator, and All are trainable parameters; Step 2.2.5: Using holistic physiologically aware attention mask and fine-grained physiologically aware attention masks Capture terrain features separately Topographic features that affect the patient's overall physiological state and topographical features that affect different physiological details of patients : ; in, , , and All are trainable parameters; Step 2.2.6, Topographical features that will affect the patient's overall physiological state and topographical features that affect different physiological details of patients Connect and obtain physiological strong coupling terrain features .

[0011] According to a specific implementation of the embodiment of the present invention, step 2.3 specifically includes: Step 2.3.1, design a spatial block-by-space visual attention module to capture physiologically strongly coupled terrain features The nonlinear interaction between the spatial blocks across regions is enhanced : Step 2.3.2, using enhanced features Calculation of terrain-adaptive physiological information update parameters ; ; in, , , and All are trainable parameters; Step 2.3.3, Update parameters using terrain-adaptive physiological information Fusion of physiological characteristics To update: ; in, and represents the trainable parameters, Represents the updated fused physiological features.

[0012] According to a specific implementation of an embodiment of the present invention, step 2.3.1 specifically includes: Designing a spatial block-by-block visual attention module for physiologically strongly coupled terrain features All the regional spatial block features in are averaged and pooled to obtain spatial block features ; Use a one-dimensional convolutional layer and a sigmoid function to calculate the spatial block-by-block visual attention mask: ; in, Represents a one-dimensional variable kernel convolution layer, whose convolution kernel size The calculation process is as follows: ; In the formula, Get the nearest odd number. and is a hyperparameter, is the spatial block feature The characteristic dimension of Utilizing spatial block-wise visual attention masks Enhanced physiological strong coupling terrain features Context information, to obtain enhanced features : ; in, is the element-wise multiplication operator, For adjusting the spatial block-wise visual attention mask size so that it is The same size.

[0013] According to a specific implementation of the embodiment of the present invention, step 3 specifically includes: Step 3.1, design the noise prediction network for the diffusion moment A random noise sample The noise prediction network first uses the sample mapping layer to Map to the denoising domain to obtain the mapping sample ; Step 3.2, stacking The noise parsing module gradually extracts mapping samples The features of the algorithm are extracted by using residual connections to stabilize the feature extraction process. Noise parsing module, finally get the mapping sample The feature is input into the linear mapping layer to obtain the noise prediction result; Step 3.3, after obtaining the noise prediction network, gradually execute from the diffusion moment The noise prediction and denoising process to 1 finally obtains the predicted multi-source physiological signal at the next moment , wherein the predicted multi-source physiological signal at the next moment Including the predicted muscle tension, the predicted pressure center and the predicted main joint angles, the expression of the denoising process is: ; in, is a predefined variance parameter, and , represents random noise, when hour, It obeys Gaussian distribution, that is ,otherwise , is a random noise sample, subject to distributed.

[0014] According to a specific implementation of the embodiment of the present invention, step 3.2 specifically includes: Step 3.2.1, the process of calculating the output characteristics of the first noise analysis module is: Updated fusion physiological features and diffusion moment Aggregate information: ; in, represents a multi-layer perceptron, represents the diffusion time embedding layer, Represents aggregate features; In the aggregation feature Extracting mapping samples under the guidance of Self-attention features: ; in, represents the self-attention module, and Both represent linear mapping layers, represents the extracted self-attention features; In the aggregation feature Under the guidance of Features: ; in, represents a feed-forward neural network, represents the output features of the first noise parsing module; Step 3.2.2, repeat step 3.2.1 to calculate the output characteristics of the subsequent noise analysis module until Noise parsing module, finally get the mapping sample The feature is input into the linear mapping layer to obtain the noise prediction result.

[0015] According to a specific implementation of the embodiment of the present invention, step 4 specifically includes: Step 4.1, calculate the impedance control torque based on the predicted pressure center and the predicted major joint angles ; Where M is the inertia matrix, which represents the response of the system to acceleration, Indicates Frame pressure center prediction results, is the damping matrix, which is a function of the plantar pressure and is used to adjust the damping characteristics of the system. is the stiffness matrix, a function of plantar pressure, used to adjust the stiffness of the system, Indicates Frame joint angle prediction results, and Based on The calculated joint angular velocity and angular acceleration are , and are the desired joint angle, angular velocity, and angular acceleration, respectively; Step 4.2: Calculate the electromyographic signal control torque based on the predicted muscle tension ; in, represents the prediction result of muscle tension signal, represents the gain function, which is used to convert the electromyographic signal into the driving torque. represents the stiffness coefficient, which indicates the response strength driven by the electromyographic signal; Step 4.3, weighted combination of impedance control torque and electromyographic signal control torque is performed to obtain joint torque and feedback control is performed on the exoskeleton system based on the torque. The expression of the joint torque is: ; in, , is the weight coefficient.

[0016] The control scheme of the walking-assisting exoskeleton system based on terrain-physiological strong coupling perception in the embodiment of the present invention includes: step 1, collecting the time series characteristics of the patient's muscle tension, plantar pressure and main joint angles and fusing them based on the cross-modal attention model to obtain fused physiological characteristics; step 2, obtaining terrain characteristics and using the fused physiological characteristics to capture the local terrain representation that affects the patient's physiological state in the terrain characteristics, obtaining physiological strong coupling terrain characteristics, and updating the fused physiological characteristics accordingly; step 3, using the updated fused physiological characteristics as the driving condition, designing a noise prediction network to gradually execute from the diffusion moment The noise prediction and denoising process of step 1 is finally performed to obtain the predicted multi-source physiological signal at the next moment; step 4, feedback control is performed on the exoskeleton system according to the predicted multi-source physiological signal at the next moment.

[0017] The beneficial effects of the embodiments of the present invention are as follows: through the scheme of the present invention, the strong coupling relationship between human physiological state and terrain environment is used to predict the multi-source physiological signals of the patient's legs (i.e. muscle signals, plantar force signals and main leg joint angles) and feedback control of the exoskeleton device is implemented based on the signals, thereby reducing the response delay of the system. This design can effectively solve the problem of response delay and movement adaptability of patients when walking on different terrains, and at the same time improve the adaptability of the walking exoskeleton system to complex terrains and the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flowchart of a method for controlling a walking-assisting exoskeleton system based on terrain-physiological strong coupling perception is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0021] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0022] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present invention, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0023] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention. The drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0024] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0025] An embodiment of the present invention provides a method for controlling a walking-assisting exoskeleton system based on terrain-physiological strong coupling perception, and the method can be applied to the process of controlling an exoskeleton system for assisted walking in medical or life scenarios.

[0026] See also Figure 1 , is a flow chart of a method for controlling a walking-assisting exoskeleton system based on terrain-physiological strong coupling perception provided by an embodiment of the present invention. Figure 1 As shown, the method mainly comprises the following steps: Step 1: collect the time series features of the patient's muscle tension, plantar pressure and main joint angles and fuse them based on the cross-modal attention model to obtain fused physiological features; In specific implementation, at the multi-source physiological signal fusion characterization stage, the process can be as follows: The muscle tension of specific muscle groups in the patient's legs is collected through electromyographic sensors , used to detect muscle contraction in real time.

[0027] The plantar pressure pad is used to collect the force conditions of each area of ​​the sole of the foot and calculate the pressure center . and Represents the coordinates of the pressure center in the cushion surface coordinate system at the nth moment.

[0028] The angles of the main leg joints (knee joint, hip joint, etc.) are collected through joint angle sensors .

[0029] The time series features of muscle tension, plantar pressure and joint angle are extracted through a one-dimensional convolutional network, respectively. , and .

[0030] The above features are fused based on the cross-modal attention model. First, the attention weights between the electromyographic signal, plantar pressure, and joint angle are calculated and feature alignment is performed: ; ; in , , are all trainable parameters, is the temperature coefficient, and T represents the transposition operation. Then, the cross-modal alignment features obtained by fusion are used as the fused physiological features of the patient. : ; in and are all weight coefficients.

[0031] Step 2, obtaining terrain features and using fused physiological features to capture local terrain representations that affect the patient's physiological state in the terrain features, obtaining physiologically strongly coupled terrain features, and updating the fused physiological features accordingly; In specific implementation, the process of updating the fused physiological features based on terrain-physiological strong coupling perception is as follows: Use LiDAR to scan the terrain contour and obtain the 3D point cloud information of the terrain. Then use the pre-trained PointTransformer model to extract the features of the 3D point cloud signal and use it as the terrain feature. .

[0032] Using fusion physiological features capture There are local topographic representations that affect the patient's physiological state. The specific process is: a) and Input the multi-layer linear mapping model to calculate the terrain query items and key items of holistic physiological perception .

[0033] b) Calculate the holistic physiological perception attention mask : ; in, , , and These are all trainable parameters.

[0034] c) Using multi-layer perception model Perform fine-grained decomposition to obtain The fused physiological features with different representation levels are then used to calculate the perceptual key items of different granularity features based on the multi-layer linear mapping model, which is expressed as .

[0035] d) Using terrain query items and perceptual key items of different granularity features Computing fine-grained physiologically aware attention masks : ; in, represents the concatenation operator, and are all trainable parameters.

[0036] e) Using holistic physiological perception attention mask and fine-grained physiologically aware attention masks Capture terrain features separately Topographic features that affect the patient's overall physiological state and topographical features that affect different physiological details of patients : ; ; in, , , and are all trainable parameters. and By connecting, we get the physiological strong coupling terrain characteristics, denoted as .

[0037] use renew Multi-source physiological representation in order to enhance its terrain adaptability. The specific process is as follows: a) Design a spatial block-by-block visual attention module to capture The nonlinear interaction relationship between spatial blocks across regions is used to enhance the understanding of terrain context information. The specific process is as follows. First, All the regional spatial block features in are averaged and pooled to obtain spatial block features. Then, a one-dimensional convolutional layer and a Sigmoid function are used to calculate the spatial block visual attention mask: ; in Represents a one-dimensional variable kernel convolution layer, whose convolution kernel size The calculation process is as follows: ; In the formula, Get the nearest odd number. and is a hyperparameter, is the spatial block feature feature dimension. It can adapt to the feature distribution of regional spatial blocks, thereby effectively learning the contextual information relationship across spatial blocks. Based on the above process, the visual attention mask of each spatial block is obtained. Next, use Enhancement Context information: ; in is the element-wise multiplication operator; For adjustment size so that it is The same size. Indicates the enhanced .

[0038] b) Utilization Calculation of terrain-adaptive physiological information update parameters : ; in, , , and These are all trainable parameters.

[0039] c) Utilization right To update: ; in, and represents the trainable parameters, Represents the updated fused physiological features.

[0040] Step 3: Using the updated fused physiological features as the driving condition, the noise prediction network is designed to gradually execute from the diffusion moment The noise prediction and denoising process from 1 to 1 finally obtains the predicted multi-source physiological signal at the next moment; In specific implementation, the process of multi-source physiological signal prediction driven by physiological features is as follows: This stage As a driving condition, the inverse diffusion model is guided to predict multi-source physiological signals.

[0041] First, a noise prediction network is designed. The specific process is as follows. A random noise sample The noise prediction network first uses the sample mapping layer to Map to the denoising domain to obtain the mapping sample ; Then, stack The noise analysis module gradually extracts The features of the noise parsing module are calculated as follows: 1) Yes and diffusion moment Aggregate information: ; in, represents a multi-layer perceptron, represents the diffusion time embedding layer, Represents an aggregate feature.

[0042] 2) In the aggregation feature Extracting mapping samples under the guidance of Self-attention features: ; in, represents the self-attention module, and Both represent linear mapping layers, Represents the extracted self-attention features.

[0043] 3) In the aggregation feature Under the guidance of Features: ; in, represents a feed-forward neural network, represents the output features of the first noise parsing module. Noise analysis module, and finally get The feature is input into the linear mapping layer to obtain the noise prediction result.

[0044] After obtaining the noise prediction results, we step by step execute The noise prediction and denoising process from 1 to 1 is as follows: ; in is a predefined variance parameter, and , represents random noise, when hour, It obeys Gaussian distribution, that is ,otherwise , is a random noise sample, subject to Distribution. Based on the above denoising process, we finally get The results include predictions for muscle tension, center of pressure, and joint angles.

[0045] Step 4: feedback control the exoskeleton system based on the predicted multi-source physiological signals at the next moment.

[0046] In specific implementation, the process of exoskeleton device feedback control based on multi-source physiological signals is as follows: Feedback control uses a hybrid method that integrates impedance control and electromyographic signal control to respond to the patient's movement needs in real time. Specifically, combining plantar pressure and joint angle, the torque calculation formula of impedance control is as follows: ; Where M is the inertia matrix, which represents the response of the system to acceleration, Indicates Frame pressure center prediction results, is the damping matrix, which is a function of the plantar pressure and is used to adjust the damping characteristics of the system. is the stiffness matrix, a function of plantar pressure, used to adjust the stiffness of the system, Indicates Frame joint angle prediction results, and Based on The calculated joint angular velocity and angular acceleration are , and are the desired joint angle, angular velocity, and angular acceleration, respectively.

[0047] The torque calculation formula for electromyographic signal control is as follows: ; in, represents the prediction result of muscle tension signal, represents the gain function, which is used to convert the electromyographic signal into the driving torque. Represents the stiffness coefficient, which indicates the response strength driven by the electromyographic signal.

[0048] Finally, the joint torque output is a weighted combination of the two control torques: ; in, , is the weight coefficient, The total driving output torque is used to achieve real-time control of the walking-assisting exoskeleton device.

[0049] The control method of the walking exoskeleton system based on terrain-physiology strong coupling perception provided in this embodiment predicts the multi-source physiological signals of the patient's legs (i.e., muscle signals, plantar force signals, and main leg joint angles) by utilizing the strong coupling relationship between human physiological state and terrain environment, and implements feedback control of the exoskeleton device based on the signals, thereby reducing the response delay of the system. This design can effectively solve the response delay and movement adaptability problems of patients when walking on different terrains, and at the same time improve the adaptability of the walking exoskeleton system to complex terrains and the user experience.

[0050] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof.

[0051] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A control method for a walking-assisting exoskeleton system based on terrain-physiological strong coupling perception, characterized in that: include: Step 1: collect the time series features of the patient's muscle tension, plantar pressure and main joint angles and fuse them based on the cross-modal attention model to obtain fused physiological features; Step 2, obtaining terrain features and using fused physiological features to capture local terrain representations that affect the patient's physiological state in the terrain features, obtaining physiologically strongly coupled terrain features, and updating the fused physiological features accordingly; Step 3: Using the updated fused physiological features as the driving condition, the noise prediction network is designed to gradually execute from the diffusion moment The noise prediction and denoising process from 1 to 1 finally obtains the predicted multi-source physiological signal at the next moment; Step 4: feedback control the exoskeleton system based on the predicted multi-source physiological signals at the next moment.

2. The method according to claim 1, characterized in that: The step 1 specifically includes: Step 1.1: Use electromyography sensors to collect muscle tension of specific muscle groups in the patient's legs ; Step 1.2: Use the plantar pressure pad to collect the plantar pressure of each area of ​​the patient's foot and calculate the pressure center based on it ； in and Indicates the coordinates of the pressure center in the cushion surface coordinate system at the nth moment; Step 1.3: Use joint angle sensors to collect the main joint angles of the patient's legs ; Step 1.4, use a one-dimensional convolutional network to extract the temporal features of muscle tension, plantar pressure, and main joint angles, respectively. , and ; In step 1.5, the temporal features of muscle tension, plantar pressure, and main joint angles are fused based on the cross-modal attention model.

3. The method according to claim 2, characterized in that The step 1.5 specifically includes: Step 1.5.1, calculate the attention weights between the electromyographic signal, plantar pressure, and joint angle and perform feature alignment ； ； in, Indicates the electromyography-foot pressure alignment feature, represents the myoelectric-joint angle alignment feature, , , are all trainable parameters, is the temperature coefficient, T represents the transposition operation; Step 1.5.2: Fuse the obtained cross-modal alignment features and use them as the patient's fused physiological features ; ； in, and are all weight coefficients.

4. The method according to claim 3, characterized in that: The step 2 specifically includes: Step 2.1: Use LiDAR to scan the terrain contour to obtain the 3D point cloud information of the terrain, and then use the pre-trained Point Transformer model to extract the features of the 3D point cloud information as the terrain features. ; Step 2.2: Using fusion physiological features Capturing terrain features There are local topographic representations that affect the patient's physiological state, and physiological strong coupling topographic features are obtained. ; Step 2.3: Using physiologically strongly coupled terrain features Update and integrate physiological features Multi-source physiological representation in.

5. The method according to claim 4, characterized in that The step 2.2 specifically includes: Step 2.2.1: Terrain features and fusion of physiological characteristics Input the multi-layer linear mapping model to calculate the terrain query items and key items of holistic physiological perception ; Step 2.2.2, query items based on terrain and key items of holistic physiological perception Computing holistic physiological perception attention mask : ； in, , , and All are trainable parameters; Step 2.2.3: Use the multi-layer perception model to integrate physiological features Perform fine-grained decomposition to obtain The fused physiological features of different representation levels are then used to calculate the perceptual key items of different granularity features based on the multi-layer linear mapping model. ; Step 2.2.4, using terrain query items and key items of perception of different granularity features Computing fine-grained physiologically aware attention masks : ； in, represents the concatenation operator, and All are trainable parameters; Step 2.2.5: Using holistic physiologically aware attention mask and fine-grained physiologically aware attention masks Capture terrain features separately Topographic features that affect the patient's overall physiological state and topographical features that affect different physiological details of patients : ； ； in, , , and All are trainable parameters; Step 2.2.6, Topographical features that will affect the patient's overall physiological state and topographical features that affect different physiological details of patients Connect and obtain physiological strong coupling terrain features .

6. The method according to claim 5, characterized in that Step 2.3 specifically includes: Step 2.3.1, design a spatial block-by-space visual attention module to capture physiologically strongly coupled terrain features The nonlinear interaction between the spatial blocks across regions is enhanced ; Step 2.3.2, using enhanced features Calculation of terrain-adaptive physiological information update parameters ; ； in, , , and All are trainable parameters; Step 2.3.3, Update parameters using terrain-adaptive physiological information Fusion of physiological characteristics To update: ； in, and represents the trainable parameters, Represents the updated fused physiological features.

7. The method according to claim 6, characterized in that The step 2.3.1 specifically includes Designing a spatial block-by-block visual attention module to focus on physiologically strongly coupled terrain features All the regional spatial block features in are averaged and pooled to obtain spatial block features ; Use a one-dimensional convolutional layer and a sigmoid function to calculate the spatial block-by-block visual attention mask: ； in, Represents a one-dimensional variable kernel convolution layer, whose convolution kernel size The calculation process is as follows: ； In the formula, Get the nearest odd number. and is a hyperparameter, is the spatial block feature The characteristic dimension of Utilizing spatial block-wise visual attention masks Enhanced physiological strong coupling terrain features Context information, to obtain enhanced features : ； in, is the element-wise multiplication operator, For adjusting the spatial block-wise visual attention mask size so that it is The same size.

8. The method according to claim 7, characterized in that The step 3 specifically includes: Step 3.1, design the noise prediction network for the diffusion moment A random noise sample The noise prediction network first uses the sample mapping layer to Map to the denoising domain to obtain the mapping sample ; Step 3.2, stacking The noise parsing module gradually extracts mapping samples The features of the algorithm are extracted by using residual connections to stabilize the feature extraction process. Noise parsing module, finally get the mapping sample The feature is input into the linear mapping layer to obtain the noise prediction result; Step 3.3, after obtaining the noise prediction network, gradually execute from the diffusion moment The noise prediction and denoising process to 1 finally obtains the predicted multi-source physiological signal at the next moment , wherein the predicted multi-source physiological signal at the next moment Including the predicted muscle tension, the predicted pressure center and the predicted main joint angles, the expression of the denoising process is: ； in, is a predefined variance parameter, and , represents random noise, when hour, It obeys Gaussian distribution, that is ,otherwise , is a random noise sample, subject to distributed.

9. The method according to claim 8, characterized in that The step 3.2 specifically includes: Step 3.2.1, the process of calculating the output characteristics of the first noise analysis module is: Updated fusion physiological features and diffusion moment Aggregate information: ； in, represents a multi-layer perceptron, represents the diffusion time embedding layer, Represents aggregate features; In the aggregation feature Extracting mapping samples under the guidance of Self-attention features: ； in, represents the self-attention module, and Both represent linear mapping layers, represents the extracted self-attention features; In the aggregation feature Under the guidance of Features: ； in, represents a feed-forward neural network, represents the output features of the first noise parsing module; Step 3.2.2, repeat step 3.2.1 to calculate the output characteristics of the subsequent noise analysis module until Noise parsing module, finally get the mapping sample The feature is input into the linear mapping layer to obtain the noise prediction result.

10. The method according to claim 9, characterized in that The step 4 specifically includes: Step 4.1, calculate the impedance control torque based on the predicted pressure center and the predicted major joint angles ； Where M is the inertia matrix, which represents the response of the system to acceleration, Indicates Frame pressure center prediction results, is the damping matrix, which is a function of the plantar pressure and is used to adjust the damping characteristics of the system. is the stiffness matrix, which is a function of the plantar pressure and is used to adjust the stiffness of the system. Indicates Frame joint angle prediction results, and Based on The calculated joint angular velocity and angular acceleration, , and are the desired joint angle, angular velocity, and angular acceleration, respectively; Step 4.2: Calculate the electromyographic signal control torque based on the predicted muscle tension ； in, represents the prediction result of muscle tension signal, represents the gain function, which is used to convert the electromyographic signal into the driving torque. represents the stiffness coefficient, which indicates the response strength driven by the electromyographic signal; Step 4.3, weighted combination of impedance control torque and electromyographic signal control torque is performed to obtain joint torque and feedback control is performed on the exoskeleton system based on the torque. The expression of the joint torque is: ； in, , is the weight coefficient.

Images