A User Gait Adaptive Lower Limb Rehabilitation Exoskeleton Continuous Phase Control Method

Abstract

This invention belongs to the technical field of rehabilitation robots, specifically relating to a continuous phase control method for a lower limb rehabilitation exoskeleton that adapts to user gait. It employs a multi-task learning model, simultaneously outputting user identity probability and movement pattern probability to generate a probability map representing the similarity between the current gait and a certain movement pattern of a sample user in a sample database. A gait phase variable parameter prediction model is used to predict gait phase variable parameters, employing angular velocity data from the initial stage of the current gait cycle and the matched movement pattern to predict the gait phase variable parameters for the current cycle. Finally, a pre-constructed relational constraint model corresponding to the matched sample user and their movement pattern is used to obtain the desired joint angles from the real-time gait phase, and the exoskeleton is tracked and controlled via a PD controller. This invention can adaptively adjust the exoskeleton's motion assistance mode according to the user's gait characteristics, and is suitable for rehabilitation training of patients with lower limb motor dysfunction.

Description

Technical Field

[0001] This invention belongs to the technical field of rehabilitation robots, and more specifically, relates to a continuous phase control method for a lower limb rehabilitation exoskeleton that adapts to user gait. Background Technology

[0002] With the rapid development of robotics technology, lower limb rehabilitation exoskeletons have been widely used in the field of rehabilitation medicine. Traditional control methods for lower limb rehabilitation exoskeletons mostly employ finite state machine control strategies, achieving exoskeleton motion control through predefined gait cycles, fixed state switching conditions, and preset joint trajectories. However, this control method has poor adaptability when facing different walking states and the differentiated gait of different users, making it difficult to meet the personalized needs of different users.

[0003] In recent years, continuous phase control methods have gradually become a research hotspot. This method calculates multiple gait phase variables using the thigh swing angle and uses these variables as key inputs to the control system. The thigh angle and its integral are typically selected as phase variables. Since the phase variable curves can vary significantly under different motion states (such as different walking speeds and different terrains), they can be used to adjust the underlying control strategies of assistive robots such as exoskeletons and prostheses. The phase space curve consists of the mean-removed thigh angle integral and the thigh angle itself, where the mean thigh angle refers to the average value over the entire gait cycle. However, since the mean thigh angle for the current gait cycle cannot be accurately obtained before the cycle is complete, existing methods typically use a fixed value derived from prior knowledge to replace this parameter.

[0004] Compared to traditional finite state machine control, continuous phase control is better able to adapt to complex walking environments and can accurately plan the desired output joint angles by estimating the user's real-time gait phase variables. However, existing continuous phase control methods often rely on fixed prior knowledge when predicting phase variable parameters, lacking consideration for individual user differences, resulting in insufficient control accuracy and adaptability.

[0005] Therefore, how to design a continuous phase control method for lower limb rehabilitation exoskeleton that can adaptively adjust the assist mode according to the user's gait characteristics has become an urgent problem to be solved in the current technical field. Summary of the Invention

[0006] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a continuous phase control method for a lower limb rehabilitation exoskeleton that is user gait adaptive. The purpose is to design a continuous phase control method for a lower limb rehabilitation exoskeleton that can adaptively adjust the assistance mode according to the user's gait characteristics.

[0007] To achieve the above objectives, according to one aspect of the present invention, a method for continuous phase control of a user gait-adaptive lower limb rehabilitation exoskeleton is provided, comprising:

[0008] S1. Collect the thigh angle data of the healthy leg of the target user during the current gait cycle, calculate the corresponding gait data, input the gait data into the constructed multi-task learning model, and output the user identity probability vector and the movement pattern probability vector to generate a normalized probability map, which represents the relative similarity between the target user's current gait cycle and each movement pattern of each sample user; according to the probability map, match the sample user and its movement pattern that are most similar to the target user's current gait cycle.

[0009] S2. Obtain the initial thigh angle data of the healthy leg in the current gait cycle, the K thigh joint angular velocities of the healthy leg obtained by K measurements within the preset early time window of the swing phase, and the matched motion pattern. Input the trained gait phase variable parameter prediction model to obtain the gait phase variable parameters of the current gait cycle, so as to calculate the real-time gait phase.

[0010] S3. Input the real-time gait phase into the pre-constructed relational constraint model corresponding to the sample user and its movement pattern obtained by matching. The constraint model includes relational constraints between the gait phase and the angle data of the thigh, knee and / or ankle joint of the affected leg, respectively, to obtain the expected angle data of the thigh, knee and / or ankle joint of the affected leg, which is used to perform continuous phase control of the lower limb rehabilitation exoskeleton of the affected leg.

[0011] Furthermore, the gait data is calculated as follows:

[0012]

[0013] In the formula, express Data on the angle of the healthy leg thigh over time. It represents the interval between the current time and the start time of its current gait cycle; This represents the mean value of the healthy leg thigh angle during the current gait cycle; , Constituting users In sports mode The following gait data.

[0014] Furthermore, the multi-task learning model is constructed as follows:

[0015] Collect thigh angle data of the healthy leg of each sample user in M ​​continuous walking modes to calculate gait data of the corresponding continuous movement mode; use gait data as input data, user identity and movement mode as output labels to construct a training sample set, and train a multi-task learning model.

[0016] The network architecture of the multi-task learning model is based on the Shared-Bottom model, which includes a shared bottom-level feature extraction network and two independent task branch towers. The bottom-level feature extraction network includes a cascaded first convolutional unit, a feature compression layer, a second convolutional unit, a first feature transpose layer, a bidirectional LSTM, a second feature transpose layer, and a global adaptive average pooling layer. Each convolutional unit includes a cascaded Conv1D layer, a ReLU activation layer, and a MaxPool1D layer. The two independent task branch towers are tower A for motion pattern classification and tower B for identity recognition.

[0017] Furthermore, the gait phase variable parameter prediction model is a bagged decision tree.

[0018] Furthermore, the relational constraint model corresponding to each motion mode of each sample user is constructed as follows:

[0019] The initial thigh angle data of the healthy leg for multiple gait cycles of the sample user in this movement mode, the K thigh joint angular velocities of the healthy leg obtained by K measurements within the preset early time window of the swing phase, and the movement mode are input into a trained gait phase variable parameter prediction model to obtain the gait phase variable parameter values ​​for the corresponding gait cycle; the parameter values ​​are used as the mean of the healthy leg thigh angle data for the current gait cycle to calculate the gait phase for the corresponding gait cycle;

[0020] The relationship constraints between the gait phase of each sample user in each movement mode and the angle data of the thigh, knee and / or ankle joint of the affected leg are obtained by constructing multiple discrete points and fitting them. The construction method of each discrete point is as follows: based on the gait phase of the sample user in any gait cycle in the movement mode, the gait phase value at a certain moment is determined; the angle data of the thigh, knee and / or ankle joint of the affected leg of the sample user in the movement mode at that certain moment are collected; the gait phase value and the angle data of the thigh, knee and / or ankle joint of the affected leg constitute a discrete point.

[0021] Furthermore, calculate the real-time gait phase. The method is as follows:

[0022]

[0023] In the formula, , , express t Data on the angle of the healthy leg thigh over time. t This represents the interval between the current time and the start time of its current gait cycle; normalization coefficient. , , , Indicates the current gait cycle x Maximum value Indicates the current gait cycle x Minimum value, Indicates the current gait cycle y Maximum value Indicates the current gait cycle y Minimum value.

[0024] Furthermore, before executing S3, the method also includes:

[0025] The real-time gait phase is passed through a Kalman filter-based smoother to obtain a smoothed gait phase, and S3 is executed using the smoothed gait phase.

[0026] According to another aspect of the present invention, an electronic device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method described above.

[0027] According to another aspect of the invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is run by a processor, it controls the device where the storage medium is located to perform the steps of the method described above.

[0028] According to another aspect of the present invention, a computer program product is provided, comprising a computer program or instructions, characterized in that, when the computer program or instructions are executed by a processor, they implement the steps of the method described above.

[0029] In summary, compared with the prior art, the technical solutions conceived by this invention have the following main advantages:

[0030] 1. This invention proposes a continuous phase control method for a lower limb rehabilitation exoskeleton that adapts to user gait. First, a multi-task learning model is introduced to achieve personalization and movement pattern recognition, simultaneously outputting the user identity probability P(i) and movement pattern probability P(m), generating a probability map map[i,m] representing the similarity between the current gait and a certain movement pattern of a sample user in the sample database. Traditional methods use the same set of parameters for all users and movement patterns, ignoring individual differences and contextual changes. The multi-task learning proposed in this invention can dynamically match the most similar gait pattern, thereby achieving personalized parameter recommendation and improving the targeting of control. Furthermore, this invention proposes to use a gait phase variable parameter prediction model (preferably a bagged decision tree) to predict gait phase variable parameters. Using the angular velocity data at the beginning of the current gait cycle and the matched movement pattern, the gait phase variable parameters for the current cycle are predicted, replacing the original fixed values. Since gait phase variable parameters change in each step, fixed values ​​cannot reflect real-time gait characteristics. Finally, using a pre-constructed relational constraint model corresponding to the matched sample users and their movement patterns, the desired joint angles are obtained from the real-time gait phase, and the exoskeleton is tracked and controlled via a PD controller. Therefore, this invention can adaptively adjust the exoskeleton's motion assistance mode according to the user's gait characteristics, making it suitable for rehabilitation training of patients with lower limb motor dysfunction. Attached Figure Description

[0031] Figure 1 The flowchart illustrates a continuous phase control method for a user gait-adaptive lower limb rehabilitation exoskeleton, as provided in an embodiment of the present invention.

[0032] Figure 2 The probability diagram provided for the embodiments of the present invention.

[0033] Figure 3 This is a framework diagram of a multi-task learning model provided in an embodiment of the present invention.

[0034] Figure 4 A comparison chart of predicted and actual values ​​of continuous gait phase variable parameters provided in an embodiment of the present invention.

[0035] Figure 5 This is a comparison chart of the actual and predicted gait phase plane curves of different subjects provided in an embodiment of the present invention.

[0036] Figure 6 A comparison of the Euclidean distance between the gait phase plane predicted curves and the actual curves obtained by different estimation methods provided in the embodiments of the present invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0038] Example 1

[0039] A method for continuous phase control of a user gait-adaptive lower limb rehabilitation exoskeleton, such as Figure 1 As shown, it includes:

[0040] S1. Collect the thigh angle data of the healthy leg of the target user during the current gait cycle, calculate the corresponding gait data, input the gait data into the constructed multi-task learning model, and output the user identity probability vector and the movement pattern probability vector to generate a normalized probability map, which represents the relative similarity between the target user's current gait cycle and each movement pattern of each sample user; according to the probability map, match the sample user and its movement pattern that are most similar to the target user's current gait cycle.

[0041] S2. Obtain the initial thigh angle data of the healthy leg in the current gait cycle, the K thigh joint angular velocities of the healthy leg obtained by K measurements within the preset early time window of the swing phase, and the matched motion pattern. Input the trained gait phase variable parameter prediction model to obtain the gait phase variable parameters of the current gait cycle, so as to calculate the real-time gait phase.

[0042] S3. Input the real-time gait phase into the pre-constructed relational constraint model corresponding to the sample user and its movement pattern obtained by matching. The constraint model includes relational constraints between the gait phase and the angle data of the thigh, knee and / or ankle joint of the affected leg, respectively, to obtain the expected angle data of the thigh, knee and / or ankle joint of the affected leg, which is used to perform continuous phase control of the lower limb rehabilitation exoskeleton of the affected leg.

[0043] This embodiment addresses the fact that existing continuous phase control methods rely on fixed prior knowledge (such as the mean thigh angle) when predicting phase variable parameters, lacking adaptability to individual user differences and different movement patterns (such as speed and slope), resulting in insufficient control accuracy and poor generalization ability. Specifically, existing methods use a fixed mean thigh angle (derived from historical data or average values) to calculate phase variables, but this mean changes dynamically during the actual gait cycle and is greatly affected by individual user differences and walking patterns. Therefore, directly using a fixed value leads to phase estimation bias, which in turn affects the trajectory tracking performance of the exoskeleton and the user experience. Therefore, this embodiment proposes a new method.

[0044] The key technical means employed in this embodiment include:

[0045] (1) Multi-task learning model: used to simultaneously identify user identity and movement pattern (such as walking on flat ground, uphill, downhill, brisk walking, etc.).

[0046] (2) Gait phase variable parameter prediction model: used to predict gait phase variable parameters of the current gait cycle (e.g., mean dynamic thigh angle). ), replacing the original (i.e., traditional) fixed value.

[0047] (3) Virtual constraints: The pre-constructed relation constraint model corresponding to the matched sample users and their motion patterns is adopted. The desired joint angle is obtained from the real-time gait phase, and the exoskeleton is tracked and controlled by the PD controller. The relation constraint model is the desired joint trajectory generated based on the predicted phase variables of the sample users.

[0048] In this embodiment, a multi-task learning model first achieves personalization and pattern recognition. By sharing a low-level feature extraction network, it simultaneously outputs the user identity probability P(i) and the movement pattern probability P(m), generating a probability map map[i,m] representing the similarity between the current gait and a certain movement pattern of a sample user in the sample library. Traditional methods use the same set of parameters for all users and movement patterns, ignoring individual differences and contextual changes. The multi-task learning proposed in this embodiment can dynamically match the most similar gait patterns, thereby achieving personalized parameter recommendations and improving the targeting of control. Secondly, this embodiment proposes to use a gait phase variable parameter prediction model (preferably a bagged decision tree) to predict gait phase variable parameters. Using the angular velocity data at the beginning of the current gait cycle and the matched movement pattern, it predicts the gait phase variable parameters for the current cycle, replacing the original fixed values. Gait phase variable parameters change in each step, and fixed values ​​cannot reflect real-time gait characteristics. Among them, the bagged decision tree has strong regression prediction capabilities and can accurately predict the mean of the entire cycle based on early gait data, thereby improving the accuracy of the phase variables.

[0049] Regarding the matching of users and movement patterns, for new users (relative to the sample users during the training of the multi-task learning model), their gait data for any step cycle is used. 、 The input is fed into a pre-trained multi-task learning model, and the output is the user identity probability. and motion pattern probability Generate a normalized probability graph ,like Figure 2As shown, this indicates the similarity between the user's gait and existing gait patterns in the sample database. In this way, the system can automatically match the most similar movement pattern to the user's current movement.

[0050] Furthermore, the method in this embodiment can also solve the problems of slow response and unnatural switching in complex, non-steady-state environments (such as speed / slope changes) of traditional control methods. The corresponding technical means to solve this problem is: a unified control framework based on continuous phase variables, that is, a relational constraint model. This invention completely abandons discrete state switching. The phase variable is a continuous, monotonically increasing signal that directly and without delay reflects the real-time progress of the user's gait. Regardless of whether the user is walking fast, slow, or uphill, the system uses the same set of control laws, only the speed at which the desired trajectory and phase variable are generated are different. The control process is completely synchronized with the user's intention. The faster the user's hips move, the faster the phase variable changes, and the faster the exoskeleton's movement rhythm naturally becomes, achieving natural adaptation without delay or switching. The system responds to the user's "movement rhythm" itself, rather than laboriously "identifying" which discrete movement state the user is in.

[0051] Meanwhile, the method in this embodiment can also solve the problem of significant gait differences between different users or the same user in different states, making it difficult for a single control strategy to take into account all these differences. The corresponding technical means to solve this problem is a personalized adaptation mechanism of "multi-task learning model + probabilistic graph matching". When a new user's gait data is input, the probability graph map[i, m] output by the model is actually a "gait feature space" constructed from a large number of prior samples, which finds the "neighbor" (i.e., the most likely user ID and movement pattern) that is most similar to the current gait. The system then uses the ideal parameters corresponding to this "neighbor" (predicted by the gait phase variable parameter prediction model) to control the exoskeleton, which is equivalent to dynamically selecting a set of quasi-optimal control parameters for the new user.

[0052] As a preferred implementation method, the gait data is calculated as follows:

[0053]

[0054] In the formula, express t Data on the angle of the healthy leg thigh over time. t It represents the interval between the current time and the start time of its current gait cycle; This represents the mean value of the healthy leg thigh angle during the current gait cycle; , Constituting users i In sports mode m The following gait data.

[0055] As a preferred implementation method, the multi-task learning model can be constructed as follows:

[0056] Collect thigh angle data of the healthy leg of each sample user in M ​​continuous walking modes to calculate gait data of the corresponding continuous movement mode; use gait data as input data, user identity and movement mode as output labels to construct a training sample set, and train a multi-task learning model.

[0057] In practice, sensors (such as inertial measurement units attached to the thigh, calf, and foot) can be used to collect data from each sample user (a total of N users). M Angle data of the thigh, knee, and ankle joints of the unaffected leg in a continuous walking pattern , , The x and y values ​​calculated using the gait data calculation formula above will be used as input to the model. Sample user ID and motion pattern category ( i,m () as labels for sample data.

[0058] This implementation constructs a multi-task (motion pattern, person in motion) learning model with a shared underlying layer, used for simultaneous motion pattern classification and user identification. The model input is... x, y The output includes user identification results and motion pattern classification results. i, m The model obtains the user's gait characteristics based on the input, and then automatically identifies the user's movement pattern (such as walking on flat ground, uphill, downhill, brisk walking, etc.) based on the gait characteristics, and determines the user's identity (such as different patients or healthy people).

[0059] Specifically, such as Figure 3 As shown, the network architecture of the aforementioned multi-task learning model is based on the Shared-Bottom model, comprising a shared bottom-level feature extraction network and two independent task branch towers. The bottom-level feature extraction network includes cascaded first convolutional units, feature compression layers, second convolutional units, first feature transpose layers, bidirectional LSTM, second feature transpose layers, and global adaptive average pooling layers; each convolutional unit includes cascaded Conv1D layers, ReLU activation, and MaxPool1D layers. The two independent task branch towers are Tower A for motion classification and Tower B for identity recognition.

[0060] For example, the overall structure of this network starts with a batch_size×2×100 input data (containing two features, 100 time steps, and the model input is a matrix of x and y). First, time-series features are extracted using a one-dimensional convolutional module. This involves passing through a Conv1D layer with 2 input channels, 32 output channels, and a kernel size of 5, followed by ReLU activation. Then, a MaxPool1D layer with a pooling window of 2 is used to compress the features before passing them through a second Conv1D layer (32 input channels, 64 output channels, and a kernel size of 3) and ReLU activation. This is followed by a second MaxPool1D layer (pool=2) to obtain a 64×25 feature map. Next, the features are transposed to 25×64 to match the input format of the LSTM and fed into a bidirectional LSTM (input dimension 64, hidden units 128, bidirectional concatenation yields 25×256). The output is then transposed to 256×25 and subjected to global adaptive average pooling (AdaptiveAvgPool1d) to obtain a 256×1 vector, which is then compressed into a 256-dimensional shared feature output. Finally, this shared feature is fed into two parallel fully connected branches, where Tower A... Tower B is used for motion classification (256→128→M classes), and Tower B is used for identity recognition (256→128→N classes), thus realizing dual-task learning based on shared features.

[0061] It should be noted that the multi-task learning network structure here is not limited to this embodiment, and can also be modified to the following common structures or others.

[0062] The multi-task learning model learns commonalities in all gait data by sharing the underlying feature extraction layer (e.g., the periodicity of gait, the basic relationship between thigh angle and integral), and then learns user-specific features and pattern-specific features on the top-level branches respectively.

[0063] As a preferred implementation, the relational constraint model corresponding to each motion mode of each sample user is constructed as follows:

[0064] The initial thigh angle data of the healthy leg for multiple gait cycles of the sample user in this movement mode, the K thigh joint angular velocities of the healthy leg obtained by K measurements within the preset early time window of the swing phase, and the movement mode are input into a trained gait phase variable parameter prediction model to obtain the gait phase variable parameter values ​​for the corresponding gait cycle; the parameter values ​​are used as the mean of the healthy leg thigh angle data for the current gait cycle to calculate the gait phase for the corresponding gait cycle;

[0065] The relationship constraints between the gait phase of each sample user in each movement mode and the angle data of the thigh, knee and / or ankle joint of the affected leg are obtained by constructing multiple discrete points and fitting them. The construction method of each discrete point is as follows: based on the gait phase of the sample user in any gait cycle in the movement mode, the gait phase value at a certain moment is determined; the angle data of the thigh, knee and / or ankle joint of the affected leg of the sample user in the movement mode at that certain moment are collected; the gait phase value and the angle data of the thigh, knee and / or ankle joint of the affected leg constitute a discrete point.

[0066] This preferred method can establish gait phase by collecting walking data from the healthy leg of sample users. Three desired joint angles for the affected leg The relationships between these constraints form a relational model (i.e., virtual constraints on joint angles), where 'a' represents the joint type. One or more relational constraints are selected based on actual needs. In this embodiment, virtual constraints... The construction method is as follows: fit the actual joint angles of all sampling points within the current gait cycle to the gait phase. cubic spline interpolation, This represents the desired joint angle for the corresponding joint.

[0067] As a preferred implementation method, the real-time gait phase can be calculated. The method is as follows:

[0068]

[0069] In the formula, , , express t Data on the angle of the healthy leg thigh over time. t This represents the interval between the current time and the start time of its current gait cycle; normalization coefficient. , , , Indicates the current gait cycle x Maximum value Indicates the current gait cycle x Minimum value, Indicates the current gait cycle y Maximum value Indicates the current gait cycle y Minimum value.

[0070] Considering that the calculated gait phase may experience some data jitter or abrupt changes, leading to sudden changes in the controller output angle and potentially causing injury to the patient, this embodiment preferably uses the gait phase calculated in real time. The smoothed gait phase is obtained using a Kalman filter-based smoother. Then, the desired knee joint angle of the lower limb exoskeleton can be calculated using a virtual constraint model. and ankle angle The desired joint angle and angular velocity (the desired joint angle is derived by differentiating the angular velocity with respect to time) are then used as inputs to the PD controller.

[0071]

[0072] in and It is an adjustable parameter. and The measured joint angles and angular velocities of the exoskeleton are... Indicates input up to the number The current of the joint motor.

[0073] To better illustrate the effectiveness of this invention, examples are provided below.

[0074] Figure 4 The graph illustrates the correspondence between the predicted and actual measured values ​​of continuous gait phase variable parameters. The horizontal axis represents the actual measured values, and the vertical axis represents the model's predicted values. The scatter plot shows that the distribution of the scatter points is almost diagonal, indicating a high degree of consistency between the predicted and actual values. This demonstrates that the proposed method can effectively capture the true variation patterns of continuous gait phase variables with high prediction accuracy, providing reliable data support for exoskeleton control.

[0075] Figure 5 The results show a comparison of gait phase planes from different subjects. The horizontal and vertical axes represent the phase plane curves, which are composed of thigh angles and their integrals, indicating the trajectory of change within a gait cycle. The orange curve represents the gait phase plane trajectory obtained from actual measurements, while the brown curve represents the result predicted using this method. It can be seen that the two curves are highly consistent in overall shape and key inflection points, indicating that the proposed method can accurately reconstruct the individual's gait phase evolution process.

[0076] Figure 6 The paper presents a comparison of the Euclidean distance between the predicted gait phase plane curves and the actual curves obtained by different estimation methods. The results show that the proposed method significantly reduces the prediction error, decreasing the Euclidean distance by approximately 64.64% compared to traditional estimation methods based on prior knowledge. This indicates that the proposed method has higher accuracy and reliability in predicting gait phase variable parameters.

[0077] Example 2

[0078] This application also relates to an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0079] The electronic device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The memory can be used to store computer programs and / or modules. The processor performs various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory.

[0080] The relevant technical solutions are the same as above, and will not be repeated here.

[0081] Example 3

[0082] This application also relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0083] Specifically, the memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0084] The relevant technical solutions are the same as above, and will not be repeated here.

[0085] Example 4

[0086] This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of the method described in the above embodiments of this application.

[0087] The relevant technical solutions are the same as above, and will not be repeated here.

[0088] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements 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 user gait adaptive lower limb rehabilitation exoskeleton continuous phase control method, characterized in that, The method comprises the following steps: S1, collecting the healthy side thigh angle data of the target user in the current gait cycle, calculating the corresponding gait data, inputting the gait data into the constructed multi-task learning model, outputting the user identity probability vector and the motion mode probability vector, and generating the normalized probability graph to represent the relative similarity between the target user in the current gait cycle and each motion mode of each sample user; according to the probability graph, the most similar sample user and the motion mode of the target user in the current gait cycle are matched; S2, obtaining the initial thigh angle data of the healthy side leg in the current gait cycle, K healthy side thigh joint angular velocities corresponding to K measurements in the preset early time window of the swing phase, and the matched motion mode, inputting the trained gait phase variable parameter prediction model to obtain the gait phase variable parameter of the current gait cycle, and calculating the real-time gait phase; S3, inputting the real-time gait phase into the pre-constructed relationship constraint model corresponding to the matched sample user and the motion mode, the constraint model comprising the relationship constraint between the gait phase and the angle data of the thigh, knee joint and / or ankle joint of the affected side leg, obtaining the expected angle data of the thigh, knee joint and / or ankle joint of the affected side leg, and performing continuous phase control of the lower limb rehabilitation exoskeleton of the affected side leg; The multi-task learning model is constructed in the following manner: Collect the thigh angle data of the healthy side leg of each sample user in M continuous walking modes to calculate the gait data corresponding to the continuous motion mode; the gait data is used as the input data, and the user identity and the motion mode are used as the output label to construct the training sample set and train the multi-task learning model; The relationship constraint model corresponding to each motion mode of each sample user is constructed in the following manner: The initial thigh angle data of the healthy side leg corresponding to multiple gait cycles of the sample user in the motion mode, K healthy side thigh joint angular velocities corresponding to K measurements in the preset early time window of the swing phase, and the motion mode are input into the trained gait phase variable parameter prediction model to obtain the gait phase variable parameter value corresponding to the gait cycle; the parameter value is used as the mean value of the thigh angle data of the healthy side leg in the current gait cycle to calculate the gait phase of the corresponding gait cycle; The relationship constraint between the gait phase and the angle data of the thigh, knee joint and / or ankle joint of the affected side leg of each sample user in each motion mode is obtained by constructing multiple discrete points and fitting, wherein each discrete point is constructed in the following manner: based on the gait phase of a gait cycle of the sample user in the motion mode, the gait phase value at a certain moment is determined; the angle data of the thigh, knee joint and / or ankle joint of the affected side leg of the sample user at the certain moment in the motion mode is collected; the gait phase value and the angle data of the thigh, knee joint and / or ankle joint of the affected side leg constitute a discrete point.

2. The continuous phase control method of a lower extremity rehabilitation exoskeleton according to claim 1, wherein, The gait data is calculated in the following manner: In the formula, represents t leg angle data at the current time, t represents the interval between the current time and the start time of the current gait cycle in which the current time is located; represents the mean of the healthy side leg angle data of the current gait cycle; , constitute the user i gait data in the motion mode m .

3. The continuous phase control method of a lower extremity rehabilitation exoskeleton according to claim 1, wherein, The network architecture of the multi-task learning model is based on a Shared-Bottom model, and includes a shared bottom feature extraction network and two independent task branch towers; the bottom feature extraction network comprises cascaded first convolution units, a feature compression layer, second convolution units, a first feature transposition layer, a bidirectional LSTM, a second feature transposition layer, and a global adaptive average pooling layer; each convolution unit comprises cascaded Conv1D layers, ReLU activations, and MaxPool1D layers; and the two independent task branch towers are Tower A for motion mode classification and Tower B for identity recognition.

4. The continuous phase control method of a lower extremity rehabilitation exoskeleton according to claim 1, wherein, The gait phase variable parameter prediction model is a bagged decision tree.

5. The continuous phase control method of a lower extremity rehabilitation exoskeleton according to claim 1, wherein, Computing real-time gait phase in the following manner: In the formula, , , represents t leg angle data at the current time, t represents the interval between the current time and the start time of the current gait cycle in which the current time is located; Normalization coefficient , , , represents the maximum value of the thigh angle of the healthy leg in the current gait cycle x , represents the minimum value of the thigh angle of the healthy leg in the current gait cycle x , represents the maximum value of the thigh angle of the healthy leg in the current gait cycle y , represents the minimum value of the thigh angle of the healthy leg in the current gait cycle y , represents the gait phase variable parameter value of the current gait cycle, and is used as the average value of the thigh angle data of the healthy leg in the current gait cycle.

6. The continuous phase control method of the lower extremity rehabilitation exoskeleton according to any one of claims 1 to 5, characterized in that, Before performing S3, the method further comprises: The real-time gait phase is smoothed by a Kalman filter-based smoother to obtain a smoothed gait phase, and S3 is performed using the smoothed gait phase. 7.An electronic device comprising a memory and a processor, the memory storing a computer program, wherein, The processor implements the steps of the method of any one of claims 1 to 6 when executing the computer program.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium comprises a stored computer program, wherein the computer program, when executed by a processor, controls the device in which the storage medium is located to perform the steps of the method of any one of claims 1 to 6.

Images