Gait phase prediction and assistance distribution method and system for lower extremity exoskeleton

Abstract

This invention discloses a method and system for gait phase prediction and assist distribution for lower limb exoskeletons, belonging to the field of exoskeleton robot control technology. It extracts spatiotemporal features and predicts gait phases from human gait temporal motion data using a spatial graph convolutional network and a directed acyclic graph model of the lower limb skeleton. The lower limb skeleton's directed acyclic graph model captures spatial features of the human gait phase, achieving spatiotemporal fusion prediction of the gait phase. Furthermore, a gait pattern mapping relation library is used to perform continuous matching of gait patterns, thereby analyzing accurate and reasonable predicted gait patterns. In addition, the predicted gait patterns are used to target the predicted assist torque sequence of the lower limb exoskeleton, achieving a precise and low-energy-consumption target active assist torque sequence. This allows for precise output distribution of assist through the exoskeleton motor driver, expanding the applicability of lower limb exoskeletons in complex walking scenarios and improving walking efficiency.

Description

Technical Field

[0001] This invention belongs to the field of exoskeleton robot control technology, specifically relating to a gait phase prediction and assist distribution method and system for lower limb exoskeletons. Background Technology

[0002] Lower limb exoskeleton robots, by providing auxiliary torques to key joints such as the hip and knee during human walking, can effectively reduce the metabolic costs for wearers and have broad application prospects in fields such as medical rehabilitation, elderly and mobility assistance, and military load-bearing. Existing active assistive exoskeletons generally employ control strategies based on gait event recognition, that is, detecting events such as heel contact and toe lift through foot pressure switches or inertial sensors, thereby triggering a predefined assist curve. A common premise of these methods is the assumption that the movement of both lower limbs during human walking is completely symmetrical about the sagittal plane; therefore, the assist strategy typically uses the same amplitude and phase for both lower limbs. However, research shows that unconventional straight-line gaits such as curved walking, tilted walking, and even reverse walking account for as much as 8%-50% of daily walking. In these cases, the kinematic characteristics of both the inner and outer lower limbs exhibit significant asymmetry: the outer lower limb changes direction by extending its swing, while the inner lower limb swings significantly less compared to conventional straight-line walking. If lower limb exoskeletons continue to use symmetrical assist strategies, not only will they fail to effectively reduce metabolic costs, but they may also interfere with the wearer's natural movement due to misaligned assist timing, increasing the risk of falls. In addition, existing gait phase detection methods mostly use thresholding or traditional machine learning classifiers, which are insufficient in modeling the spatial dependencies between joints and make it difficult to predict the phase of future multiple frames in advance, resulting in inherent delays in exoskeleton control.

[0003] As mentioned above, how to provide a gait phase prediction and assist allocation method and system for lower limb exoskeletons that can achieve spatiotemporal fusion prediction of gait phase, analyze gait patterns, and achieve precise assist allocation has become an urgent research topic in this field. Summary of the Invention

[0004] The purpose of this invention is to provide a gait phase prediction and assist allocation method and system for lower limb exoskeletons, in order to solve the above-mentioned problems existing in the prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a gait phase prediction and assist allocation method for lower limb exoskeletons, comprising: A directed acyclic graph model of the lower limb skeleton is constructed based on the connection relationship of the lower limb joints. Human gait temporal motion data is obtained. Based on the spatial graph convolutional network and the directed acyclic graph model of the lower limb skeleton, spatiotemporal feature extraction and gait phase prediction are performed on the human gait temporal motion data to obtain gait phase sequence prediction results. A preset gait pattern mapping relationship library is obtained. The current gait phase sequence is extracted from the human gait temporal motion data. The current gait phase sequence and the gait phase sequence prediction result are concatenated into a complete gait phase sequence. The complete gait phase sequence is input into the gait pattern mapping relationship library to calculate the pattern matching degree of each gait pattern from the gait pattern mapping relationship library. Based on the pattern matching degree of each gait pattern, the corresponding gait pattern is selected as the current gait pattern. The predicted gait pattern is inferred based on the current gait pattern. Step frequency is extracted based on the gait phase sequence prediction results to obtain the predicted step frequency. The lower limb exoskeleton assist torque prediction sequence is calculated based on the gait phase sequence prediction results and the predicted step frequency. The lower limb exoskeleton assist torque prediction sequence is then activated based on the predicted gait pattern to obtain the target active assist torque sequence. Based on the target active assist torque sequence, a corresponding target active assist torque command is generated and sent to the exoskeleton motor driver to control the motor to output the corresponding active assist torque, thereby achieving assist distribution.

[0006] In one possible design, a directed acyclic graph (DAG) model of the lower limb skeleton is constructed based on the connectivity of the lower limb joints. Temporal motion data of human gait is acquired. Based on a spatial graph convolutional network and the DAG model of the lower limb skeleton, spatiotemporal feature extraction and gait phase prediction are performed on the temporal motion data of the human gait to obtain gait phase sequence prediction results, including: A preset sampling window is obtained. Within the sampling window, motion data of multiple target human body parts are sampled to obtain temporal motion data of each target human body part. The temporal motion data of each target human body part is then integrated to form original human gait temporal motion data. The original human gait temporal motion data is subjected to low-pass filtering, and the original human gait temporal motion data after low-pass filtering is standardized according to the gait cycle to obtain human gait temporal motion data. Based on the connection relationship of the human lower limb joints, several key lower limb skeletal joints are selected as graph model nodes. According to the connection relationship of the human lower limb joints, directed edges of the graph model are formed. Each graph model node is integrated into a graph model point set, and each graph model directed edge is integrated into a graph model edge set. Based on the graph model point set and the graph model edge set, a directed acyclic graph model of the lower limb skeleton is constructed. The human gait temporal motion data is used as point features. Based on the directed acyclic graph model of the lower limb skeleton, the human lower limb skeleton skeletal vector is extracted. The human lower limb skeleton skeletal vector is used as edge features. The point features and the edge features are integrated to form human gait feature data. Obtain a preset gait phase prediction model, wherein the gait phase prediction model includes an input layer, a spatial graph convolutional layer, a temporal convolutional layer, an efficient channel attention layer, a fully connected layer, and an output layer connected in sequence; The human gait feature data is used as input and input to the input layer of the gait phase prediction model; The human gait feature data is obtained from the input layer of the gait phase prediction model through the spatial graph convolutional layer of the gait phase prediction model, and spatial feature extraction is performed on the human gait feature data through the spatial graph convolutional layer to obtain human gait spatial enhancement features. The human gait spatial enhancement features are obtained from the spatial graph convolutional layer of the gait phase prediction model through the temporal convolutional layer, and the temporal features of the human gait spatial enhancement features are extracted through the temporal convolutional layer to obtain the spatiotemporal coupling features of human gait. The spatiotemporal coupling features of human gait are obtained from the temporal convolutional layer of the gait phase prediction model through the efficient channel attention layer, and attention extraction and weighted calculation are performed on the spatiotemporal coupling features of human gait through the efficient channel attention layer to obtain attention-enhanced human gait phase features. The attention-enhanced human gait phase features are obtained from the efficient channel attention layer of the gait phase prediction model through the fully connected layer of the gait phase prediction model, and feature mapping and gait phase probability output are performed on the attention-enhanced human gait phase features through the fully connected layer to obtain the human gait phase probability distribution. The human gait phase probability distribution is obtained from the fully connected layer of the gait phase prediction model through the output layer, and the human gait prediction phase corresponding to each sampling time is extracted from the human gait phase probability distribution through the output layer. The human gait prediction phases corresponding to each sampling time are integrated into a phase sequence in chronological order to form a gait phase sequence prediction result.

[0007] In one possible design, a preset gait pattern mapping database is obtained; the current gait phase sequence is extracted from the human gait temporal motion data; the current gait phase sequence and the gait phase sequence prediction result are concatenated to form a complete gait phase sequence; the complete gait phase sequence is input into the gait pattern mapping database to calculate the pattern matching degree of each gait pattern from the database; and the corresponding gait pattern is selected as the current gait pattern based on the pattern matching degree of each gait pattern, including: The gait phase at each sampling moment is identified from the human gait time sequence motion data, and the gait phase at each sampling moment is integrated into a gait phase sequence within the sampling window according to the time order of each sampling moment, so as to use the gait phase sequence within the sampling window as the current gait phase sequence. The current gait phase sequence and the gait phase sequence prediction result are concatenated in chronological order to form a complete gait phase sequence, wherein the complete gait phase sequence includes multiple consecutive gait cycles; The proportion of gait phase in each gait cycle is calculated in the complete gait phase sequence, and the temporal parameters of each key gait event are identified based on the complete gait phase sequence, wherein the key gait event is a limb response event made by the human body to adapt to the gait phase. The gait phase ratio in each gait cycle and the timing parameters of each key gait event are integrated into the current key gait event features; Obtain a preset gait pattern mapping relationship library, wherein the gait pattern mapping relationship library is used to store the mapping relationship between event features and gait patterns corresponding to each key gait event type; The current key gait event features are input into the gait pattern mapping relationship library. The feature similarity between the current key gait event features and the event features of each key gait event type stored in the gait pattern mapping relationship library is calculated. The feature similarity between the current key gait event features and the event features of each key gait event type stored in the gait pattern mapping relationship library is used as the pattern matching degree of the gait pattern corresponding to each key gait event type. Obtain the preset gait pattern prior constraints, take the highest pattern matching degree among the pattern matching degrees of each gait pattern as the maximum likelihood, calculate the posterior probability of each gait pattern using Bayes' theorem, and select the gait pattern with the highest posterior probability as the current gait pattern.

[0008] In one possible design, inferring the predicted gait pattern based on the current gait pattern includes: The gait phase sequence prediction results are input into the gait pattern mapping relationship library to calculate the pattern matching degree of each gait pattern from the gait pattern mapping relationship library, and the corresponding gait pattern is selected as the initial predicted gait pattern based on the pattern matching degree of each gait pattern. The initial predicted gait pattern and the current gait pattern are compared to obtain the pattern comparison result; If the pattern comparison result shows that the initial predicted gait pattern and the current gait pattern are the same, then the initial predicted gait pattern is considered to be a stable gait pattern, and the initial predicted gait pattern is used as the predicted gait pattern. If the pattern comparison result shows that the initial predicted gait pattern and the current gait pattern are different, then the initial predicted gait pattern is considered to be a changing gait pattern. Based on the current gait phase sequence and the prediction result of the gait phase sequence, the corresponding gait pattern change time is calculated. The gait pattern change time, the current gait pattern and the initial predicted gait pattern are integrated to generate a predicted gait pattern.

[0009] In one possible design, gait frequency is extracted based on the gait phase sequence prediction results to obtain a predicted gait frequency. Then, a lower limb exoskeleton assist torque prediction sequence is calculated based on the gait phase sequence prediction results and the predicted gait frequency, including: Multiple key predicted gait events are extracted from the gait phase sequence prediction results. The time interval between adjacent key predicted gait events of the same type is calculated based on each key predicted gait event, so as to generate a predicted step frequency based on the time interval between adjacent key predicted gait events of the same type. A preset torque curve of key human joints is obtained. Based on the predicted step frequency, the torque curve of key human joints is decomposed into Fourier decomposition to obtain the corresponding sine harmonic coefficient and cosine harmonic coefficient through multiple harmonics. The periodic average joint torque is calculated according to the torque curve of key human joints, and the periodic average joint torque is used as the DC component of the torque of key human joints. Based on the gait phase sequence prediction results, the predicted gait phase at each prediction time is extracted. Based on the predicted step frequency and the predicted gait phase at each prediction time, the predicted value of the lower limb exoskeleton assist torque at each prediction time is calculated using the following formula (1): (1) in, The predicted time in the gait phase sequence prediction result. This refers to the harmonic order of the torque curve of the key joints of the human body. The predicted time in the predicted gait phase The corresponding predicted gait phase, For the predicted step frequency, This refers to the DC component of the torque at the key joints of the human body. For the first The cosine harmonic coefficients obtained from the second harmonic. For the first The sinusoidal harmonic coefficients obtained from the subharmonics This refers to the total harmonic order of the torque curves of the key joints of the human body. The predicted time in the predicted gait phase The corresponding predicted value of the lower limb exoskeleton assist torque; According to the time sequence of each prediction moment, the predicted values ​​of the lower limb exoskeleton assist torque at each prediction moment are arranged and integrated to form a lower limb exoskeleton assist torque prediction sequence.

[0010] In one possible design, the predicted lower limb exoskeleton assist torque prediction sequence is target-activated based on the predicted gait pattern to obtain a target active assist torque sequence, including: Obtain a preset lower limb exoskeleton activation coefficient reference table, wherein the lower limb exoskeleton activation coefficient reference table includes multiple sub-tables, and each sub-table corresponds to a gait pattern; The gait phase sequence prediction results are input into the lower limb exoskeleton activation coefficient reference table, so that the corresponding sub-table is selected as the target activation sub-table according to the predicted gait pattern. Using the target activation sub-table, the gait phase sequence prediction results at each prediction time in the lower limb exoskeleton assist torque prediction sequence are matched with the lower limb exoskeleton assist torque activation coefficients to obtain the lower limb exoskeleton assist torque activation coefficients at each prediction time. The target active assist torque is obtained by using the activation coefficient of the lower limb exoskeleton assist torque at each prediction time to activate the corresponding lower limb exoskeleton assist torque prediction sequence, and then integrating the target active assist torque at each prediction time into a target active assist torque sequence in chronological order.

[0011] In one possible design, based on the target active assist torque sequence, a corresponding target active assist torque command is generated, and the target active assist torque command is sent to the exoskeleton motor driver. The exoskeleton motor driver then controls the motor to output the corresponding active assist torque to achieve assist distribution, including: Obtain a preset maximum assist torque threshold, and use the maximum assist torque threshold to perform torque peak absolute value limiting processing on the target active assist torque sequence to obtain a limited target active assist torque sequence; The active assist torque sequence of the limited target is smoothed to obtain a stable active assist torque sequence. A preset lower limb exoskeleton dynamics feedforward model is obtained, and the feedforward torque compensation value of the exoskeleton motor driver is calculated based on the lower limb exoskeleton dynamics feedforward model to obtain the lower limb exoskeleton feedforward torque compensation value. The feedforward torque compensation value of the lower limb exoskeleton is used to compensate the active assist torque sequence of the stable target to obtain the final active assist torque sequence of the target. Based on the final target active assist torque sequence, a control voltage signal for the lower limb exoskeleton is generated, and the control voltage signal is used as the target active assist torque command. According to each predicted moment, the target active assist torque command is sent to the exoskeleton motor driver, so that the exoskeleton motor driver can control the motor to output the corresponding electromagnetic torque as the active assist torque to achieve assist distribution.

[0012] Secondly, the present invention provides a gait phase prediction and assist distribution system for lower limb exoskeletons, comprising: The gait phase prediction unit is used to construct a directed acyclic graph model of the lower limb skeleton based on the connection relationship of the lower limb joints, acquire human gait temporal motion data, and perform spatiotemporal feature extraction and gait phase prediction on the human gait temporal motion data based on the spatial graph convolutional network and the lower limb skeleton directed acyclic graph model to obtain gait phase sequence prediction results. The gait pattern prediction unit is used to acquire a preset gait pattern mapping relationship library, extract the current gait phase sequence from the human gait temporal motion data, concatenate the current gait phase sequence and the gait phase sequence prediction result into a complete gait phase sequence, input the complete gait phase sequence into the gait pattern mapping relationship library, calculate the pattern matching degree of each gait pattern from the gait pattern mapping relationship library, select the corresponding gait pattern as the current gait pattern based on the pattern matching degree of each gait pattern, and infer the predicted gait pattern based on the current gait pattern. The assist torque calculation unit is used to extract the step frequency based on the gait phase sequence prediction result to obtain the predicted step frequency, calculate the lower limb exoskeleton assist torque prediction sequence based on the gait phase sequence prediction result and the predicted step frequency, and perform target activation on the lower limb exoskeleton assist torque prediction sequence based on the predicted gait pattern to obtain the target active assist torque sequence. The assist distribution execution unit is used to generate a corresponding target active assist torque command based on the target active assist torque sequence, and send the target active assist torque command to the exoskeleton motor driver so as to control the motor to output the corresponding active assist torque and realize assist distribution.

[0013] Thirdly, the present invention provides an electronic device comprising a memory, a processor, and a transceiver connected in sequence, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the gait phase prediction and assist allocation method for lower limb exoskeleton as described in the first aspect or any possible design of the first aspect.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, perform the gait phase prediction and assist allocation method for lower limb exoskeleton as described in the first aspect or any possible design of the first aspect.

[0015] Fifthly, the present invention provides a computer program product containing instructions that, when the instructions are executed on a computer, cause the computer to perform the gait phase prediction and assist allocation method for a lower limb exoskeleton as described in the first aspect or any possible design of the first aspect.

[0016] Beneficial Effects: This invention provides a method and system for gait phase prediction and assist allocation for lower limb exoskeletons, including: First, constructing a directed acyclic graph (DAG) model of the lower limb skeleton based on the connection relationships of the human lower limb joints, acquiring human gait temporal motion data, and performing spatiotemporal feature extraction and gait phase prediction on the human gait temporal motion data based on a spatial graph convolutional network and the lower limb skeleton DAG model to obtain gait phase sequence prediction results; Second, acquiring a preset gait pattern mapping relationship library, extracting the current gait phase sequence from the human gait temporal motion data, concatenating the current gait phase sequence and the gait phase sequence prediction results into a complete gait phase sequence, and inputting the complete gait phase sequence into the gait pattern mapping relationship library to calculate the gait phase prediction results from the gait pattern mapping relationship library. The pattern matching degree of each gait pattern is used to select the corresponding gait pattern as the current gait pattern, and the predicted gait pattern is inferred based on the current gait pattern. Then, the step frequency is extracted based on the gait phase sequence prediction result to obtain the predicted step frequency. The lower limb exoskeleton assist torque prediction sequence is calculated based on the gait phase sequence prediction result and the predicted step frequency. The lower limb exoskeleton assist torque prediction sequence is then target-activated based on the predicted gait pattern to obtain the target active assist torque sequence. Finally, based on the target active assist torque sequence, the corresponding target active assist torque command is generated and sent to the exoskeleton motor driver to control the motor to output the corresponding active assist torque, thereby realizing assist distribution. By employing a spatial graph convolutional network and a lower limb skeleton directed acyclic graph model, spatiotemporal feature extraction and gait phase prediction are performed on human gait temporal motion data. The lower limb skeleton directed acyclic graph model is used to capture spatial features of human gait phase and achieve spatiotemporal fusion prediction of gait phase. Furthermore, a gait pattern mapping relation library is used to perform continuous matching of gait patterns, thereby analyzing accurate and reasonable predicted gait patterns. In addition, the predicted gait patterns are used to target activation of the lower limb exoskeleton assist torque prediction sequence, thereby achieving a precise and low-energy active assist torque sequence. This allows for precise output distribution of assist through the exoskeleton motor driver, expanding the applicability of the lower limb exoskeleton in complex walking scenarios and improving walking assistance efficiency. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the gait phase prediction and assist allocation method for lower limb exoskeletons provided in an embodiment of the present invention. Figure 2 A schematic diagram of the model structure of the gait phase prediction model provided in an embodiment of the present invention; Figure 3This is a functional structure diagram of a gait phase prediction and assist distribution system for a lower limb exoskeleton provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0019] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0020] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0021] Example: like Figure 1 As shown, the first aspect of this embodiment provides a gait phase prediction and assist allocation method for lower limb exoskeletons, which may include, but is not limited to, the following steps: S1. Based on the connection relationship of the lower limb joints, a directed acyclic graph model of the lower limb skeleton is constructed to obtain human gait temporal motion data. Based on the spatial graph convolutional network and the lower limb skeleton directed acyclic graph model, spatiotemporal feature extraction and gait phase prediction are performed on the human gait temporal motion data to obtain gait phase sequence prediction results. In one possible implementation, in steps S11-S111, a directed acyclic graph model of the lower limb skeleton is constructed based on the connection relationships of the lower limb joints, and human gait temporal motion data is obtained. Based on the spatial graph convolutional network and the lower limb skeleton directed acyclic graph model, spatiotemporal feature extraction and gait phase prediction are performed on the human gait temporal motion data to obtain gait phase sequence prediction results, including: S11. Obtain a preset sampling window, and sample motion data of multiple target human body parts within the sampling window to obtain temporal motion data of each target human body part, and integrate the temporal motion data of each target human body part to form original human gait temporal motion data. S12. Perform low-pass filtering on the original human gait temporal motion data, and standardize the original human gait temporal motion data after low-pass filtering according to the gait cycle to obtain human gait temporal motion data. S13. Based on the connection relationship of the human lower limb joints, select multiple key lower limb skeletal joints as graph model nodes, form directed edges of the graph model according to the connection relationship of the human lower limb joints, integrate each graph model node into a graph model point set, integrate each graph model directed edge into a graph model edge set, and construct a directed acyclic graph model of the lower limb skeleton based on the graph model point set and the graph model edge set. S14. Using the human gait temporal motion data as point features, extract the human lower limb skeleton skeletal vector based on the directed acyclic graph model of the lower limb skeleton, and use the human lower limb skeleton skeletal vector as edge features. Integrate the point features and the edge features to form human gait feature data. S15. Obtain a preset gait phase prediction model, where, for example... Figure 2 As shown, the gait phase prediction model includes an input layer, a spatial graph convolutional layer, a temporal convolutional layer, an efficient channel attention layer, a fully connected layer, and an output layer connected in sequence. S16. The human gait feature data is used as input and input to the input layer of the gait phase prediction model; S17. Obtain the human gait feature data from the input layer of the gait phase prediction model through the spatial graph convolutional layer of the gait phase prediction model, and extract spatial features from the human gait feature data through the spatial graph convolutional layer to obtain human gait spatial enhancement features; S18. Obtain the human gait spatial enhancement features from the spatial graph convolutional layer of the gait phase prediction model through the temporal convolutional layer of the gait phase prediction model, and extract the temporal features of the human gait spatial enhancement features through the temporal convolutional layer to obtain the human gait spatiotemporal coupling features. S19. The human gait spatiotemporal coupling features are obtained from the temporal convolutional layer of the gait phase prediction model through the efficient channel attention layer of the gait phase prediction model, and attention extraction and weighted calculation are performed on the human gait spatiotemporal coupling features through the efficient channel attention layer to obtain attention-enhanced human gait phase features. S110. The attention-enhanced human gait phase features are obtained from the efficient channel attention layer of the gait phase prediction model through the fully connected layer of the gait phase prediction model, and the attention-enhanced human gait phase features are feature-mapped and gait phase probability outputs are performed on the attention-enhanced human gait phase features through the fully connected layer to obtain the human gait phase probability distribution. S111. The human gait phase probability distribution is obtained from the fully connected layer of the gait phase prediction model through the output layer of the gait phase prediction model, and the human gait prediction phase corresponding to each sampling time is extracted from the human gait phase probability distribution through the output layer, so as to integrate the human gait prediction phase corresponding to each sampling time into a phase sequence in chronological order, and form a gait phase sequence prediction result.

[0022] In the specific implementation process, the motion data sampling can be performed by fixing inertial measurement units (IMUs) to multiple target body parts of the exoskeleton wearer. These target body parts include at least the sternum, pelvis, left thigh, left calf, right thigh, and right calf. The sampling frequency of the IMU can be preset to 100Hz. During motion data sampling, triaxial angles, triaxial accelerations, and triaxial angular velocities of each target body part in the sagittal plane can be simultaneously collected as corresponding temporal motion data, thus forming raw human gait temporal motion data. The collected raw human gait temporal motion data is then subjected to low-pass filtering. A second-order Butterworth low-pass filter can be used to remove corresponding high-frequency noise, followed by zero-phase forward and reverse filtering to eliminate waveform distortion. Furthermore, the filtered raw human gait temporal motion data is segmented by gait cycle (using the heel contact (IC) event as the starting point of the gait cycle), and the data within each gait cycle is standardized into multiple sampling points of fixed length using linear interpolation, forming the human gait temporal motion data.

[0023] Based on the natural connections and kinematic dependencies between the joints of the human lower limbs, a directed acyclic graph (DAG) model of the lower limb skeleton is constructed. At least five key lower limb skeletal joints—the waist, left thigh, left calf, right thigh, and right calf—are defined as the set of graphical points in the DAG model. The set of graphical edges is defined according to the physiological connections of bones from proximal joints to distal joints. Specifically, the waist to each thigh is an outgoing edge, and each thigh to the corresponding calf is an outgoing edge. In addition, reverse edges from distal joints to proximal joints can be established to characterize the bidirectional dependencies between joints. Each graphical node represents a key lower limb skeletal joint, and each directed edge represents a bone connecting two joints. The direction of the directed edge represents the motion transmission relationship and kinematic dependency order between the key lower limb skeletal joints.

[0024] In specific application scenarios, in the gait phase prediction model, the spatial feature extraction of the human gait feature data through the spatial graph convolutional layer requires aggregating upstream and downstream joint information along directed edges using the convolutional kernel of the spatial graph convolutional layer to obtain the biomechanical dependence between joints, thereby aggregating the feature information of neighboring nodes and generating a feature map containing spatial structure information as the human gait spatial enhancement feature. Then, the temporal feature extraction of the human gait spatial enhancement feature through the temporal convolutional layer requires the use of a one-dimensional temporal convolutional network to extract the dynamic change law of gait along the time axis. Convolution operations are performed on the feature maps of multiple consecutive frames along the time axis to extract the dynamic change features of gait movement on the time scale; a deep motion feature map containing spatiotemporal coupling information is generated as the human gait spatiotemporal coupling feature.

[0025] Accordingly, an efficient channel attention layer is embedded after the temporal convolutional layer of the gait phase prediction model to adaptively recalibrate the importance of each feature channel and focus on the most critical motion patterns. Specifically: First, global average pooling is performed on the input human gait spatiotemporal coupling features to compress the high-dimensional feature map into a one-dimensional channel description vector. Then, adaptive one-dimensional convolution is used to perform a convolution operation on the channel description vector (the convolution kernel size is determined by a nonlinear mapping function based on the number of channels), thereby outputting the channel attention weights for each channel. After normalization by the Sigmoid activation function, these weights are multiplied channel-by-channel with the human gait spatiotemporal coupling features to obtain the output feature map after channel attention weighting, which serves as the attention-enhanced human gait phase feature. This channel feature weighting processing mechanism enables adaptive recalibration of the importance of different feature channels, thereby obtaining a motion feature channel most relevant to gait phase changes.

[0026] The fully connected layer of the gait phase prediction model performs feature mapping on the high-dimensional attention-enhanced human gait phase features to obtain the predicted probability of the gait phase corresponding to each gait phase category. Then, the predicted probability of each gait phase is calculated using the Softmax function, forming a human gait phase probability distribution. Finally, the output layer of the gait phase prediction model selects the gait phase with the highest probability value as the gait phase prediction result for each sampling time. The gait phase prediction results from each sampling time are integrated to form a gait phase sequence prediction result. It should be noted that the gait phase categories in this embodiment include at least four types: heel strike phase, toe strike phase, heel lift-off phase, and foot swing phase. Ultimately, the gait phase prediction model forms a prediction process that takes human gait temporal motion data within a fixed time window (sampling window) starting from the current time as input and a continuous multi-frame gait phase sequence within a future time window as output, achieving advanced prediction of human gait.

[0027] Each element in the gait phase sequence prediction result is a four-classified phase set (including heel strike phase, toe strike phase, heel lift-off phase, and foot swing phase). Furthermore, the gait phase prediction model requires sampling gait motion datasets from multiple healthy subjects at various walking speeds as training samples. These training samples are divided into training, validation, and test sets according to a preset training ratio. The cross-entropy loss function is used as the model optimization objective, and the Adam optimizer is selected for gradient updates. During training, graph learning freezes the initial graph structure in the first five iterations to ensure training stability, and then gradually updates the graph's topology to adapt to the action recognition task, resulting in a trained gait phase prediction model. The trained gait phase prediction model is then validated on the test set, and evaluation metrics such as prediction accuracy, precision, and recall are calculated. The validated gait phase prediction model can then be used for real-time feature analysis and gait phase prediction.

[0028] The gait phase prediction model in this embodiment applies a directed acyclic graph (DAG) model of the lower limb skeleton to gait phase prediction of the lower limb exoskeleton. By constructing the DAG model, it explicitly models the motion transmission relationship from the waist through the thigh and calf to the ankle. It also utilizes an efficient channel attention module to adaptively focus on key motion feature channels (spatial features), achieving prediction of a future time period (up to [percentage missing]). The model provides advance prediction of continuous gait phases within a given timeframe. In digital twin simulation experiments, the gait phase prediction model in this embodiment achieves a prediction accuracy of up to 97.89%, significantly outperforming traditional temporal models such as CNN, RNN, LSTM, and TCN, thus providing sufficient reaction lead for the exoskeleton controller.

[0029] S2. Obtain a preset gait pattern mapping relationship library, extract the current gait phase sequence from the human gait temporal motion data, concatenate the current gait phase sequence and the gait phase sequence prediction result into a complete gait phase sequence, input the complete gait phase sequence into the gait pattern mapping relationship library, calculate the pattern matching degree of each gait pattern from the gait pattern mapping relationship library, select the corresponding gait pattern as the current gait pattern based on the pattern matching degree of each gait pattern, and infer the predicted gait pattern based on the current gait pattern; In one possible implementation, steps S21-S27 involve obtaining a preset gait pattern mapping database, extracting the current gait phase sequence from the human gait temporal motion data, concatenating the current gait phase sequence and the gait phase sequence prediction result into a complete gait phase sequence, inputting the complete gait phase sequence into the gait pattern mapping database, calculating the pattern matching degree of each gait pattern from the database, and selecting the corresponding gait pattern as the current gait pattern based on the pattern matching degree of each gait pattern. S21. Identify the gait phase at each sampling moment from the human gait time sequence motion data, and integrate the gait phases at each sampling moment into a gait phase sequence within the sampling window according to the time order of each sampling moment, so as to use the gait phase sequence within the sampling window as the current gait phase sequence. S22. The current gait phase sequence and the gait phase sequence prediction result are concatenated in chronological order to form a complete gait phase sequence, wherein the complete gait phase sequence includes multiple consecutive gait cycles; S23. Calculate the gait phase ratio in each gait cycle in the complete gait phase sequence, and identify the temporal parameters of each key gait event based on the complete gait phase sequence, wherein the key gait event is a limb response event made by the human body to adapt to the gait phase. S24. Integrate the gait phase ratio in each gait cycle and the timing parameters of each key gait event into the current key gait event features; S25. Obtain a preset gait pattern mapping relationship library, wherein the gait pattern mapping relationship library is used to store the mapping relationship between event features and gait patterns corresponding to each key gait event type; S26. Input the current key gait event features into the gait pattern mapping relationship library, and calculate the feature similarity between the current key gait event features and the event features of each key gait event type stored in the gait pattern mapping relationship library, so as to use the feature similarity between the current key gait event features and the event features of each key gait event type stored in the gait pattern mapping relationship library as the pattern matching degree of the gait pattern corresponding to each key gait event type; S27. Obtain the preset gait pattern prior constraints, take the highest pattern matching degree among the pattern matching degrees of each gait pattern as the maximum likelihood, calculate the posterior probability of each gait pattern using Bayes' formula, and select the gait pattern with the highest posterior probability as the current gait pattern.

[0030] It should be noted that in the gait phase prediction and assist allocation method provided in this embodiment, the temporal parameters of key gait events are extracted from the complete gait phase sequence. The key gait events include at least heel strike events, toe strike events, heel lift-off events, and foot swing events. Correspondingly, the temporal parameters of the key gait events include at least the time interval between two adjacent heel strike events (i.e., gait cycle duration), the time interval between heel strike events and heel lift-off events (i.e., foot strike duration), the alignment relationship of bilateral gait events (e.g., the time offset between the left heel strike event and the right heel lift-off event), and the duration ratio of the left and right foot swing events. These temporal parameters of key gait events all have temporal attributes and can quantitatively characterize the degree of symmetry / asymmetry of human walking, thus they can well cope with different gait patterns.

[0031] In specific application scenarios, the gait pattern mapping database is a mapping database established through offline experiments: First, healthy subjects need to be recruited, and their gait event features under multiple gait patterns (including at least straight walking, curved walking, tilted walking, and reverse walking) are collected simultaneously; for each gait cycle, the typical temporal correspondence of gait event features under each gait pattern is statistically analyzed; and the typical temporal correspondence of gait event features under each gait pattern is encoded into a mapping database in the form of "gait pattern - gait time feature" relationship pairs.

[0032] Accordingly, the current key gait event features are input into the gait pattern mapping database, and the similarity between the current key gait event features and the time features of the key gait event type corresponding to each gait pattern is calculated using weighted Euclidean distance or likelihood function, so as to determine the current gait pattern based on the similarity calculation results.

[0033] Among them, the preset gait pattern prior constraints are used to constrain the variation law of bilateral gait patterns. For example, when the gait pattern of the right lower limb is curvilinear walking, the left lower limb may be curvilinear walking, a transitional gait from straight walking to curvilinear walking, or a transitional gait from curvilinear walking to straight walking, but it cannot be reverse walking.

[0034] In one possible implementation, steps S28-S211, inferring the predicted gait pattern based on the current gait pattern, includes: S28. Input the gait phase sequence prediction result into the gait pattern mapping relationship library to calculate the pattern matching degree of each gait pattern from the gait pattern mapping relationship library, and select the corresponding gait pattern as the initial predicted gait pattern based on the pattern matching degree of each gait pattern. S29. Compare the initial predicted gait pattern with the current gait pattern to obtain a pattern comparison result; S210. If the pattern comparison result is that the initial predicted gait pattern and the current gait pattern are the same, then the initial predicted gait pattern is considered to be a stable gait pattern, and the initial predicted gait pattern is used as the predicted gait pattern. S211. If the pattern comparison result shows that the initial predicted gait pattern and the current gait pattern are different, then the initial predicted gait pattern is considered to be a changing gait pattern. Based on the current gait phase sequence and the prediction result of the gait phase sequence, the corresponding gait pattern change time is calculated. The gait pattern change time, the current gait pattern and the initial predicted gait pattern are integrated to generate a predicted gait pattern.

[0035] S3. Based on the gait phase sequence prediction results, extract the step frequency to obtain the predicted step frequency. Calculate the lower limb exoskeleton assist torque prediction sequence based on the gait phase sequence prediction results and the predicted step frequency. Then, based on the predicted gait pattern, perform target activation on the lower limb exoskeleton assist torque prediction sequence to obtain the target active assist torque sequence. In one possible implementation, steps S31-S34 involve extracting the gait frequency based on the gait phase sequence prediction result to obtain a predicted gait frequency, and calculating a lower limb exoskeleton assist torque prediction sequence based on the gait phase sequence prediction result and the predicted gait frequency, including: S31. Extract multiple key predicted gait events from the gait phase sequence prediction results, calculate the time interval between adjacent key predicted gait events of the same type based on each key predicted gait event, and generate a predicted step frequency based on the time interval between adjacent key predicted gait events of the same type. S32. Obtain the preset torque curve of key human joints, perform Fourier decomposition on the torque curve of key human joints based on the predicted step frequency, obtain the corresponding sine harmonic coefficient and cosine harmonic coefficient through multiple harmonics, and calculate the periodic average joint torque according to the torque curve of key human joints, so as to use the periodic average joint torque as the DC component of the torque of key human joints. S33. Based on the gait phase sequence prediction results, extract the predicted gait phase at each prediction time. Based on the predicted step frequency and the predicted gait phase at each prediction time, calculate the predicted value of the lower limb exoskeleton assist torque at each prediction time using the following formula (1): (1) in, The predicted time in the gait phase sequence prediction result. This refers to the harmonic order of the torque curve of the key joints of the human body. The predicted time in the predicted gait phase The corresponding predicted gait phase, For the predicted step frequency, This refers to the DC component of the torque at the key joints of the human body. For the first The cosine harmonic coefficients obtained from the second harmonic. For the first The sinusoidal harmonic coefficients obtained from the subharmonics This refers to the total harmonic order of the torque curves of the key joints of the human body. The predicted time in the predicted gait phase The corresponding predicted value of the lower limb exoskeleton assist torque; S34. Arrange and integrate the predicted values ​​of the lower limb exoskeleton assist torque at each prediction time according to the time sequence of each prediction time to form a lower limb exoskeleton assist torque prediction sequence.

[0036] It should be noted that in the preset gait phase prediction and assist allocation method provided in this embodiment, the torque curve of the key human joints is constructed based on the natural motion variation law of the torque of each key human joint. The torque curve of the key human joints is periodic. Therefore, it can be approximated by multiple harmonics using a finite number of Fourier series (preferably, the harmonic order is selected as 3-5), thereby extracting the sine harmonic coefficients and cosine harmonic coefficients obtained from multiple harmonics. The sine harmonic coefficients and cosine harmonic coefficients can be regarded as related to the predicted step frequency. The quadratic polynomial fitting function.

[0037] In one possible implementation, to ensure the adaptability of gait patterns, corresponding human key joint torque curves are generated for each gait pattern to obtain the sine harmonic coefficients and cosine harmonic coefficients corresponding to each gait pattern. When calculating the lower limb exoskeleton assist torque, the corresponding human key joint torque curves, sine harmonic coefficients, and cosine harmonic coefficients can be selected according to the predicted gait pattern to calculate the lower limb exoskeleton assist torque prediction sequence under the corresponding gait pattern.

[0038] In one possible implementation, steps S35-S38 involve target activation of the lower limb exoskeleton assist torque prediction sequence based on the predicted gait pattern to obtain a target active assist torque sequence, including: S35. Obtain a preset lower limb exoskeleton activation coefficient reference table, wherein the lower limb exoskeleton activation coefficient reference table includes multiple sub-tables, and each sub-table corresponds to a gait pattern; S36. Input the gait phase sequence prediction result into the lower limb exoskeleton activation coefficient reference table, so as to select the corresponding sub-table as the target activation sub-table according to the predicted gait pattern; S37. Using the target activation sub-table, the lower limb exoskeleton assist torque activation coefficient is matched for the gait phase sequence prediction results at each prediction time in the lower limb exoskeleton assist torque prediction sequence, so as to obtain the lower limb exoskeleton assist torque activation coefficient at each prediction time. S38. Activate the target corresponding to the lower limb exoskeleton assist torque prediction sequence by using the activation coefficient of the lower limb exoskeleton assist torque at each prediction time to obtain the target active assist torque at each prediction time, and integrate the target active assist torque at each prediction time into a target active assist torque sequence in chronological order.

[0039] It should be noted that in the gait phase prediction and assist allocation method provided in this embodiment, the lower limb exoskeleton activation coefficient reference table includes multiple sub-tables, each sub-table corresponding to a gait pattern. For example, the lower limb exoskeleton activation coefficient reference table includes a straight walking sub-table, a curved walking sub-table, an inclined walking sub-table, and a reverse walking sub-table; and each sub-table includes different correlations between gait phases and lower limb exoskeleton assist torque activation coefficients. Specifically, taking the straight walking sub-table as an example, the lower limb exoskeleton assist torque activation coefficient corresponding to the heel-to-ground phase is 0.5, the lower limb exoskeleton assist torque activation coefficient corresponding to the toe-to-ground phase is 1, and the lower limb exoskeleton assist torque activation coefficient corresponding to the heel-off phase is 1. The activation coefficient for the lower limb exoskeleton assist torque is 0.5, and the activation coefficient corresponding to the foot swing phase is 0. When the activation coefficient for the lower limb exoskeleton assist torque is 1, it means that the target active assist torque for that gait phase is fully provided by the lower limb exoskeleton as active assistance. When the activation coefficient for the lower limb exoskeleton assist torque is 0.5, it means that the target active assist torque for that gait phase is halved. When the activation coefficient for the lower limb exoskeleton assist torque is 0, it means that the target active assist torque for that gait phase is not provided by the lower limb exoskeleton, and the active assistance is 0 (only the lower limb exoskeleton needs to follow the movement based on inertia). This setting and matching of the activation coefficient for the lower limb exoskeleton assist torque achieves on-demand assistance, which not only reduces the metabolic cost of the human body but also reduces the energy consumption of the lower limb exoskeleton.

[0040] In addition, in possible application scenarios, each sub-table may also include independent lower limb exoskeleton assist torque activation coefficients corresponding to both lower limbs. Especially when the gait mode is curvilinear walking, since the gait phases of both lower limbs are not symmetrical, the lower limb exoskeleton assist torque activation coefficients at each prediction time are also set differently to ensure that the lower limb exoskeleton assistance is more in line with the specific walking scenario.

[0041] S4. Based on the target active assist torque sequence, generate a corresponding target active assist torque command, and send the target active assist torque command to the exoskeleton motor driver to control the motor to output the corresponding active assist torque and realize assist distribution.

[0042] In one possible implementation, steps S41-S46 involve generating a corresponding target active assist torque command based on the target active assist torque sequence, and sending the target active assist torque command to the exoskeleton motor driver to control the motor to output the corresponding active assist torque, thereby achieving assist distribution. This includes: S41. Obtain a preset maximum assist torque threshold, and use the maximum assist torque threshold to perform torque peak absolute value limiting processing on the target active assist torque sequence to obtain a limited target active assist torque sequence; S42. Smooth the active assist torque sequence of the limited target to obtain a stable active assist torque sequence; S43. Obtain a preset lower limb exoskeleton dynamics feedforward model, and calculate the feedforward torque compensation value of the exoskeleton motor driver based on the lower limb exoskeleton dynamics feedforward model to obtain the lower limb exoskeleton feedforward torque compensation value. S44. The feedforward torque compensation value of the lower limb exoskeleton is used to compensate the torque of the stable target active assist torque sequence to obtain the final target active assist torque sequence; S45. Based on the final target active assist torque sequence, generate a control voltage signal for the lower limb exoskeleton, and use the control voltage signal as the target active assist torque command; S46. According to each predicted time, the target active assist torque command is sent to the exoskeleton motor driver, so as to use the exoskeleton motor driver to control the motor to output the corresponding electromagnetic torque as the active assist torque, thereby realizing the assist distribution.

[0043] It should be noted that the smoothing of the active assist torque sequence of the limited target can preferably be achieved by using a first-order low-pass filter. The filter coefficient of the filter can be adjusted and set according to the actual response requirements, so as to obtain a smoothed and stable active assist torque sequence of the target.

[0044] After obtaining the smoothed and stable target active assist torque sequence, in order to accurately reproduce the active assist torque during output, it is necessary to compensate for the gravity and inertial disturbance torque generated by the motion of the lower limb exoskeleton linkage itself. Therefore, it is necessary to call the pre-set exoskeleton dynamics feedforward model (which can be obtained by recursive least squares method), introduce rotational inertia term, gravity term and friction term (including Coulomb friction and viscous friction of the lower limb exoskeleton), so as to form the corresponding final target active assist torque sequence.

[0045] After receiving the target active assist torque command, the motor driver drives the motors of each key joint to output the corresponding electromagnetic torque. After being amplified by the reducer, the motors of each key joint apply the active assist torque to the thigh of the human body through the mechanical linkage.

[0046] After completing the assist output at one prediction moment, it is carried over to the next prediction moment, ensuring that the electromagnetic torque switches accordingly, completing the assist for a period of time (all prediction moments); immediately after walking is completed, the gait phase prediction and assist allocation method of steps S1-S4 is repeated to achieve low latency. Continuous gait phase prediction and real-time assist distribution ensure the continuous and smooth operation of the lower limb exoskeleton.

[0047] like Figure 3 As shown, the second aspect of this embodiment provides a hardware system for implementing the gait phase prediction and assist allocation method for lower limb exoskeletons described in the first aspect of the embodiment, including: The gait phase prediction unit is used to construct a directed acyclic graph model of the lower limb skeleton based on the connection relationship of the lower limb joints, acquire human gait temporal motion data, and perform spatiotemporal feature extraction and gait phase prediction on the human gait temporal motion data based on the spatial graph convolutional network and the lower limb skeleton directed acyclic graph model to obtain gait phase sequence prediction results. The gait pattern prediction unit is used to acquire a preset gait pattern mapping relationship library, extract the current gait phase sequence from the human gait temporal motion data, concatenate the current gait phase sequence and the gait phase sequence prediction result into a complete gait phase sequence, input the complete gait phase sequence into the gait pattern mapping relationship library, calculate the pattern matching degree of each gait pattern from the gait pattern mapping relationship library, select the corresponding gait pattern as the current gait pattern based on the pattern matching degree of each gait pattern, and infer the predicted gait pattern based on the current gait pattern. The assist torque calculation unit is used to extract the step frequency based on the gait phase sequence prediction result to obtain the predicted step frequency, calculate the lower limb exoskeleton assist torque prediction sequence based on the gait phase sequence prediction result and the predicted step frequency, and perform target activation on the lower limb exoskeleton assist torque prediction sequence based on the predicted gait pattern to obtain the target active assist torque sequence. The assist distribution execution unit is used to generate a corresponding target active assist torque command based on the target active assist torque sequence, and send the target active assist torque command to the exoskeleton motor driver so as to control the motor to output the corresponding active assist torque and realize assist distribution.

[0048] The working process, working details and technical effects of the system provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0049] like Figure 4As shown, the third aspect of this embodiment provides an electronic device, including: a memory, a processor, and a transceiver that are sequentially and communicatively connected, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the gait phase prediction and assist allocation method for lower limb exoskeletons as described in the first aspect of the embodiment.

[0050] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0051] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard) transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the electronic device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0052] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0053] The fourth aspect of this embodiment provides a storage medium that stores instructions containing the gait phase prediction and assist allocation method for lower limb exoskeletons as described in the first aspect of this embodiment. That is, the storage medium stores instructions that, when executed on a computer, perform the gait phase prediction and assist allocation method for lower limb exoskeletons as described in the first aspect of this embodiment.

[0054] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0055] The working process, working details, and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0056] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the gait phase prediction and assist allocation method for lower limb exoskeletons as described in the first aspect of this embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0057] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for gait phase prediction and assist allocation for lower limb exoskeletons, characterized in that, include: A directed acyclic graph model of the lower limb skeleton is constructed based on the connection relationship of the lower limb joints. Human gait temporal motion data is obtained. Based on the spatial graph convolutional network and the directed acyclic graph model of the lower limb skeleton, spatiotemporal feature extraction and gait phase prediction are performed on the human gait temporal motion data to obtain gait phase sequence prediction results. A preset gait pattern mapping relationship library is obtained. The current gait phase sequence is extracted from the human gait temporal motion data. The current gait phase sequence and the gait phase sequence prediction result are concatenated into a complete gait phase sequence. The complete gait phase sequence is input into the gait pattern mapping relationship library to calculate the pattern matching degree of each gait pattern from the gait pattern mapping relationship library. Based on the pattern matching degree of each gait pattern, the corresponding gait pattern is selected as the current gait pattern. The predicted gait pattern is inferred based on the current gait pattern. Step frequency is extracted based on the gait phase sequence prediction results to obtain the predicted step frequency. The lower limb exoskeleton assist torque prediction sequence is calculated based on the gait phase sequence prediction results and the predicted step frequency. The lower limb exoskeleton assist torque prediction sequence is then activated based on the predicted gait pattern to obtain the target active assist torque sequence. Based on the target active assist torque sequence, a corresponding target active assist torque command is generated and sent to the exoskeleton motor driver to control the motor to output the corresponding active assist torque, thereby achieving assist distribution.

2. The gait phase prediction and assist allocation method for lower limb exoskeletons according to claim 1, characterized in that, A directed acyclic graph (DAG) model of the lower limb skeleton is constructed based on the connection relationships of the lower limb joints. Temporal motion data of human gait is obtained. Based on a spatial graph convolutional network and the DAG model of the lower limb skeleton, spatiotemporal feature extraction and gait phase prediction are performed on the temporal motion data of the human gait to obtain gait phase sequence prediction results, including: A preset sampling window is obtained. Within the sampling window, motion data of multiple target human body parts are sampled to obtain temporal motion data of each target human body part. The temporal motion data of each target human body part is then integrated to form original human gait temporal motion data. The original human gait temporal motion data is subjected to low-pass filtering, and the original human gait temporal motion data after low-pass filtering is standardized according to the gait cycle to obtain human gait temporal motion data. Based on the connection relationship of the human lower limb joints, several key lower limb skeletal joints are selected as graph model nodes. According to the connection relationship of the human lower limb joints, directed edges of the graph model are formed. Each graph model node is integrated into a graph model point set, and each graph model directed edge is integrated into a graph model edge set. Based on the graph model point set and the graph model edge set, a directed acyclic graph model of the lower limb skeleton is constructed. The human gait temporal motion data is used as point features. Based on the directed acyclic graph model of the lower limb skeleton, the human lower limb skeleton skeletal vector is extracted. The human lower limb skeleton skeletal vector is used as edge features. The point features and the edge features are integrated to form human gait feature data. Obtain a preset gait phase prediction model, wherein the gait phase prediction model includes an input layer, a spatial graph convolutional layer, a temporal convolutional layer, an efficient channel attention layer, a fully connected layer, and an output layer connected in sequence; The human gait feature data is used as input and input to the input layer of the gait phase prediction model; The human gait feature data is obtained from the input layer of the gait phase prediction model through the spatial graph convolutional layer of the gait phase prediction model, and spatial feature extraction is performed on the human gait feature data through the spatial graph convolutional layer to obtain human gait spatial enhancement features. The human gait spatial enhancement features are obtained from the spatial graph convolutional layer of the gait phase prediction model through the temporal convolutional layer, and the temporal features of the human gait spatial enhancement features are extracted through the temporal convolutional layer to obtain the spatiotemporal coupling features of human gait. The spatiotemporal coupling features of human gait are obtained from the temporal convolutional layer of the gait phase prediction model through the efficient channel attention layer, and attention extraction and weighted calculation are performed on the spatiotemporal coupling features of human gait through the efficient channel attention layer to obtain attention-enhanced human gait phase features. The attention-enhanced human gait phase features are obtained from the efficient channel attention layer of the gait phase prediction model through the fully connected layer of the gait phase prediction model, and feature mapping and gait phase probability output are performed on the attention-enhanced human gait phase features through the fully connected layer to obtain the human gait phase probability distribution. The human gait phase probability distribution is obtained from the fully connected layer of the gait phase prediction model through the output layer, and the human gait prediction phase corresponding to each sampling time is extracted from the human gait phase probability distribution through the output layer. The human gait prediction phases corresponding to each sampling time are integrated into a phase sequence in chronological order to form a gait phase sequence prediction result.

3. The gait phase prediction and assist allocation method for lower limb exoskeletons according to claim 1, characterized in that, A preset gait pattern mapping database is obtained; the current gait phase sequence is extracted from the human gait temporal motion data; the current gait phase sequence and the gait phase sequence prediction result are concatenated into a complete gait phase sequence; the complete gait phase sequence is input into the gait pattern mapping database to calculate the pattern matching degree of each gait pattern from the database; and the corresponding gait pattern is selected as the current gait pattern based on the pattern matching degree of each gait pattern, including: The gait phase at each sampling moment is identified from the human gait time sequence motion data, and the gait phase at each sampling moment is integrated into a gait phase sequence within the sampling window according to the time order of each sampling moment, so as to use the gait phase sequence within the sampling window as the current gait phase sequence. The current gait phase sequence and the gait phase sequence prediction result are concatenated in chronological order to form a complete gait phase sequence, wherein the complete gait phase sequence includes multiple consecutive gait cycles; The proportion of gait phase in each gait cycle is calculated in the complete gait phase sequence, and the temporal parameters of each key gait event are identified based on the complete gait phase sequence, wherein the key gait event is a limb response event made by the human body to adapt to the gait phase. The gait phase ratio in each gait cycle and the timing parameters of each key gait event are integrated into the current key gait event features; Obtain a preset gait pattern mapping relationship library, wherein the gait pattern mapping relationship library is used to store the mapping relationship between event features and gait patterns corresponding to each key gait event type; The current key gait event features are input into the gait pattern mapping relationship library. The feature similarity between the current key gait event features and the event features of each key gait event type stored in the gait pattern mapping relationship library is calculated. The feature similarity between the current key gait event features and the event features of each key gait event type stored in the gait pattern mapping relationship library is used as the pattern matching degree of the gait pattern corresponding to each key gait event type. Obtain the preset gait pattern prior constraints, take the highest pattern matching degree among the pattern matching degrees of each gait pattern as the maximum likelihood, calculate the posterior probability of each gait pattern using Bayes' theorem, and select the gait pattern with the highest posterior probability as the current gait pattern.

4. The gait phase prediction and assist allocation method for lower limb exoskeletons according to claim 3, characterized in that, Inferring a predicted gait pattern based on the current gait pattern includes: The gait phase sequence prediction results are input into the gait pattern mapping relationship library to calculate the pattern matching degree of each gait pattern from the gait pattern mapping relationship library, and the corresponding gait pattern is selected as the initial predicted gait pattern based on the pattern matching degree of each gait pattern. The initial predicted gait pattern and the current gait pattern are compared to obtain the pattern comparison result; If the pattern comparison result shows that the initial predicted gait pattern and the current gait pattern are the same, then the initial predicted gait pattern is considered to be a stable gait pattern, and the initial predicted gait pattern is used as the predicted gait pattern. If the pattern comparison result shows that the initial predicted gait pattern and the current gait pattern are different, then the initial predicted gait pattern is considered to be a changing gait pattern. Based on the current gait phase sequence and the prediction result of the gait phase sequence, the corresponding gait pattern change time is calculated. The gait pattern change time, the current gait pattern and the initial predicted gait pattern are integrated to generate a predicted gait pattern.

5. The gait phase prediction and assist allocation method for lower limb exoskeletons according to claim 1, characterized in that, Based on the gait phase sequence prediction results, gait frequency is extracted to obtain the predicted gait frequency. Then, based on the gait phase sequence prediction results and the predicted gait frequency, a lower limb exoskeleton assist torque prediction sequence is calculated, including: Multiple key predicted gait events are extracted from the gait phase sequence prediction results. The time interval between adjacent key predicted gait events of the same type is calculated based on each key predicted gait event, so as to generate a predicted step frequency based on the time interval between adjacent key predicted gait events of the same type. A preset torque curve of key human joints is obtained. Based on the predicted step frequency, the torque curve of key human joints is decomposed into Fourier decomposition to obtain the corresponding sine harmonic coefficient and cosine harmonic coefficient through multiple harmonics. The periodic average joint torque is calculated according to the torque curve of key human joints, and the periodic average joint torque is used as the DC component of the torque of key human joints. Based on the gait phase sequence prediction results, the predicted gait phase at each prediction time is extracted. Based on the predicted step frequency and the predicted gait phase at each prediction time, the predicted value of the lower limb exoskeleton assist torque at each prediction time is calculated using the following formula (1): （1） in, The predicted time in the gait phase sequence prediction result. This refers to the harmonic order of the torque curve of the key joints of the human body. The predicted time in the predicted gait phase The corresponding predicted gait phase, For the predicted step frequency, This refers to the DC component of the torque at the key joints of the human body. For the first The cosine harmonic coefficients obtained from the second harmonic. For the first The sinusoidal harmonic coefficients obtained from the subharmonics This refers to the total harmonic order of the torque curves of the key joints of the human body. The predicted time in the predicted gait phase The corresponding predicted value of the lower limb exoskeleton assist torque; According to the time sequence of each prediction moment, the predicted values ​​of the lower limb exoskeleton assist torque at each prediction moment are arranged and integrated to form a lower limb exoskeleton assist torque prediction sequence.

6. The gait phase prediction and assist allocation method for lower limb exoskeletons according to claim 5, characterized in that, Based on the predicted gait pattern, the lower limb exoskeleton assist torque prediction sequence is target-activated to obtain a target active assist torque sequence, including: Obtain a preset lower limb exoskeleton activation coefficient reference table, wherein the lower limb exoskeleton activation coefficient reference table includes multiple sub-tables, and each sub-table corresponds to a gait pattern; The gait phase sequence prediction results are input into the lower limb exoskeleton activation coefficient reference table, so that the corresponding sub-table is selected as the target activation sub-table according to the predicted gait pattern. Using the target activation sub-table, the gait phase sequence prediction results at each prediction time in the lower limb exoskeleton assist torque prediction sequence are matched with the lower limb exoskeleton assist torque activation coefficients to obtain the lower limb exoskeleton assist torque activation coefficients at each prediction time. The target active assist torque is obtained by using the activation coefficient of the lower limb exoskeleton assist torque at each prediction time to activate the corresponding lower limb exoskeleton assist torque prediction sequence, and then integrating the target active assist torque at each prediction time into a target active assist torque sequence in chronological order.

7. The gait phase prediction and assist allocation method for lower limb exoskeletons according to claim 1, characterized in that, Based on the target active assist torque sequence, a corresponding target active assist torque command is generated, and the target active assist torque command is sent to the exoskeleton motor driver. The exoskeleton motor driver then controls the motor to output the corresponding active assist torque to achieve assist distribution, including: Obtain a preset maximum assist torque threshold, and use the maximum assist torque threshold to perform torque peak absolute value limiting processing on the target active assist torque sequence to obtain a limited target active assist torque sequence; The active assist torque sequence of the limited target is smoothed to obtain a stable active assist torque sequence. A preset lower limb exoskeleton dynamics feedforward model is obtained, and the feedforward torque compensation value of the exoskeleton motor driver is calculated based on the lower limb exoskeleton dynamics feedforward model to obtain the lower limb exoskeleton feedforward torque compensation value. The feedforward torque compensation value of the lower limb exoskeleton is used to compensate the active assist torque sequence of the stable target to obtain the final active assist torque sequence of the target. Based on the final target active assist torque sequence, a control voltage signal for the lower limb exoskeleton is generated, and the control voltage signal is used as the target active assist torque command. According to each predicted moment, the target active assist torque command is sent to the exoskeleton motor driver, so that the exoskeleton motor driver can control the motor to output the corresponding electromagnetic torque as the active assist torque to achieve assist distribution.

8. A gait phase prediction and assist distribution system for lower limb exoskeletons, characterized in that, The method for gait phase prediction and assist allocation for lower limb exoskeletons as described in any one of claims 1 to 7 includes: The gait phase prediction unit is used to construct a directed acyclic graph model of the lower limb skeleton based on the connection relationship of the lower limb joints, acquire human gait temporal motion data, and perform spatiotemporal feature extraction and gait phase prediction on the human gait temporal motion data based on the spatial graph convolutional network and the lower limb skeleton directed acyclic graph model to obtain gait phase sequence prediction results. The gait pattern prediction unit is used to acquire a preset gait pattern mapping relationship library, extract the current gait phase sequence from the human gait temporal motion data, concatenate the current gait phase sequence and the gait phase sequence prediction result into a complete gait phase sequence, input the complete gait phase sequence into the gait pattern mapping relationship library, calculate the pattern matching degree of each gait pattern from the gait pattern mapping relationship library, select the corresponding gait pattern as the current gait pattern based on the pattern matching degree of each gait pattern, and infer the predicted gait pattern based on the current gait pattern. The assist torque calculation unit is used to extract the step frequency based on the gait phase sequence prediction result to obtain the predicted step frequency, calculate the lower limb exoskeleton assist torque prediction sequence based on the gait phase sequence prediction result and the predicted step frequency, and perform target activation on the lower limb exoskeleton assist torque prediction sequence based on the predicted gait pattern to obtain the target active assist torque sequence. The assist distribution execution unit is used to generate a corresponding target active assist torque command based on the target active assist torque sequence, and send the target active assist torque command to the exoskeleton motor driver so as to control the motor to output the corresponding active assist torque and realize assist distribution.

9. An electronic device, characterized in that, The device includes a memory, a processor, and a transceiver that are sequentially and communicatively connected. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the gait phase prediction and assist allocation method for lower limb exoskeletons as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or the instructions are executed by the computer, they implement the gait phase prediction and assist allocation method for lower limb exoskeletons as described in any one of claims 1 to 7.

Images