Gait motion generative model and method for various gait speeds and ground slopes

Abstract

The present invention relates to a virtual walking motion data generation model and a method for generating virtual walking motion data for various walking speeds and ground slopes, and more specifically, to a virtual walking motion data generation model comprising a conditional modified autoencoder that receives walking motion data as input data and learns using walking speed and ground slope as condition values ​​to derive walking motion generation data.

Description

Technology Field

[0001] The present invention relates to a model and method for generating virtual walking motion data for various walking speeds and ground slopes. The present invention is a result of a national research project titled “Development and Verification of Algorithms for Predicting Walking Instability and Improving Stability Using Soft RoboWear” (Project Unique Number: CT230026, Project Number: 20231193, Ministry: Seoul Metropolitan Government, Project Management Agency: Seoul Business Agency, Research Project Name: Campus Town Technology Matching Support Project, Project Executing Agency: Chung-Ang University). Background Technology

[0002] Human gait is divided into several stages, and lower-limb assistive wearable robots are controlled differently depending on the gait stage. Therefore, active research and development are underway to develop models that predict human gait stages using thigh or calf angles (referred to as "gait motion data") and to utilize these models for the control of lower-limb assistive wearable robots. To develop a model that predicts gait stages, gait motion data from the wearable robot user is required.

[0003] However, since human walking motion data varies significantly depending on walking speed and ground slope, a large number of walking motion data measured at various walking speeds and ground slopes are required, and collecting such data requires a lot of time and cost.

[0004] Walking motion data is time-series data, and representative methods for generating time-series data include regression and correlation-based methods.

[0005] Regression methods generate new walking motion data for walking speed and ground slope by utilizing linear regression or polynomial regression and walking motion data acquired from experiments regarding walking speed and ground slope. The linear regression method generates new time-series data using two samples. Therefore, since the generated data varies depending on which experimentally acquired data is used as a sample, it is difficult to guarantee the quality of the generated data. Additionally, there is a weakness in that the generated data may contain high-frequency errors. The polynomial regression method suffers from the same problem.

[0006] The correlation-based method generates signals by utilizing the correlation between the signal values ​​of the previous time step and the signal values ​​of the current time step. However, since this correlation changes depending on walking speed and ground slope, it is very difficult to find correlations for walking speed and ground slope, which were not used in the experiment, using limited experimental data.

[0007] Furthermore, the gait cycles obtained from the experiment vary from sample to sample. Due to these differences in gait cycles, it is impossible to generate virtual data by directly applying existing regression and correlation methods. Prior art literature

[0008] Japanese Published Patent 2024-022278, Korean Registered Patent 10-2474407, Korean Published Patent 10-2023-0115181, Japanese Published Patent 2018-042672 The problem to be solved

[0009] Accordingly, the present invention has been devised to solve the aforementioned conventional problems, and according to an embodiment of the present invention, the purpose is to provide a model and method for generating virtual walking motion data for various walking speeds and ground slopes, which can virtually generate walking motion data for walking speeds and ground slopes that were not used in the experiment.

[0010] According to an embodiment of the present invention, the purpose is to provide a virtual walking motion data generation model and a method for generating various walking speeds and ground slopes, using a walking motion data generation algorithm that combines a conditional variational autoencoder, initial-final difference (IFD), time nondimensionalization, and loss function weight optimization to solve the problems of existing data generation methods.

[0011] According to an embodiment of the present invention, data for multiple ground slopes and walking speeds can be generated using walking motion data for a small number of ground slopes and walking speeds, and the generated data can be utilized to control a lower limb assistance wearable robot at various ground slopes, and additionally, when a pedestrian changes their walking speed, lower limb assistance control can be performed in accordance with the changed motion. The purpose is to provide a model and method for generating virtual walking motion data for various walking speeds and ground slopes.

[0012] Meanwhile, the technical problems to be solved by the present invention are not limited to those mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art to which the present invention belongs from the description below. means of solving the problem

[0013] The first objective of the present invention can be achieved as a virtual walking motion data generation model for various walking speeds and ground slopes, characterized by including a conditional modified autoencoder that receives walking motion data as input data and learns using walking speed and ground slope as condition values ​​to derive walking motion generation data.

[0014] In addition, a loss function is applied, and the difference between the first and last values ​​of the gait cycle of the gait motion generation data (IFD) is added to the loss function for training, thereby maintaining the characteristics of gait data where the first and last values ​​are similar.

[0015] In addition, it may be characterized by further including a loss function weight optimization unit that optimizes the weights of the loss function to improve the similarity between the actual walking motion data and the walking motion generation data.

[0016] And the above weight optimization unit may be characterized by arbitrarily assigning weight values, generating walking motion data for the same ground slope and walking speed using data on ground slope and walking speed measured in an experiment, quantifying the similarity between the generated data and the data measured in the experiment, and obtaining the optimal weight value showing the highest data similarity through the same process for various weight values, and using it when generating data on ground slope and walking speed that was not used in the experiment.

[0017] In addition, it may be characterized by further including a data preprocessing unit that preprocesses data before receiving walking motion data from an experiment as input data for a conditional deformation autoencoder.

[0018] And the data preprocessing unit may be characterized by including: a walk cycle-based splitting unit that splits time series data using a walk cycle; a time nondimensionalization unit that nondimensionalizes data time using the walk cycle; and a time interval correction unit that obtains data at the same nondimensional time interval through linear interpolation.

[0019] In addition, the above-mentioned gait cycle-based division unit divides the input gait motion data to be used for learning according to the gait cycle, the time nondimensionalization unit nondimensionalizes the time of the gait motion data by the gait cycle to change the time of the data from 0 to 1, and the time interval correction unit obtains data values ​​at a constant nondimensional time interval using linear interpolation, and through this process, the data can be characterized in that they have the same time interval under all ground slope and gait speed conditions.

[0020] And it may further include a data post-processing unit that post-processes walking motion generation data generated from the above-mentioned conditional deformation autoencoder.

[0021] In addition, the data post-processing unit may be characterized by including a physical time conversion unit that converts the time of the walking motion generation data, expressed as dimensionless time generated by the conditional deformation autoencoder, into physical time information by multiplying it by the walking cycle.

[0022] In addition, the data post-processing unit may be characterized by performing signal smoothing through a filter to remove high-frequency noise components.

[0023] The second objective of the present invention can be achieved as a method for generating virtual walking motion data for various walking speeds and ground slopes, characterized by comprising: a preprocessing step of dividing walking motion data to be used for learning according to a walking cycle, nondimensionalizing data time using the walking cycle, and obtaining data at the same nondimensional time interval through linear interpolation; a step of a conditional deformable autoencoder receiving walking motion data as input data and learning using walking speed and ground slope as condition values ​​to derive walking motion generated data; and a postprocessing step of converting the time of the walking motion generated data, expressed as nondimensional time generated by the conditional deformable autoencoder, into physical time information by multiplying it by the walking cycle.

[0024] In addition, the loss function according to the following mathematical formula 1 is applied, and the difference between the first and last values ​​of the gait cycle of the gait motion generation data (IFD) is added to the loss function for training, thereby maintaining the characteristics of gait data where the first and last values ​​are similar.

[0025] [Mathematical Formula 1]

[0026]

[0027] In the above mathematical formula 1, where N is the number of samples, M is the number of (dimensionless) time intervals, β is the weight, and y is virtual walking motion data generated by the decoder, is the actual measured walking motion data used as input to the encoder, σ is the standard deviation value output from the encoder, and μ is the average value output from the encoder.

[0028] And the above loss function weight optimization unit may further include a step of optimizing the weights of the loss function to improve the similarity between the actual walking motion data and the walking motion generation data.

[0029] In addition, the similarity of the data can be quantified using DA of Equation 2 and DS of Equation 3 below, and for various β values, the DA and DS values ​​are calculated, and the β having the minimum DA and DS is selected as the optimal β.

[0030] [Mathematical Formula 2]

[0031]

[0032] [Mathematical Formula 3]

[0033]

[0034] In the above mathematical formulas 2 and 3, here represents the average value of the experimental values ​​at the j-th time interval, and is the average of the generated data values ​​in the j-th time interval, is the variance of the experimental values ​​at the j-th time interval, is the variance of the generated data values ​​at the j-th time interval.

[0035] The third objective of the present invention can be achieved as a walking phase prediction system characterized by predicting walking phases by including a virtual walking motion data generation model for various walking speeds and ground slopes according to the first objective mentioned above.

[0036] The fourth objective of the present invention can be achieved as a lower limb assistance wearable robot control system, which is a system for controlling a lower limb assistance wearable robot, comprising a virtual walking motion data generation model for various walking speeds and ground slopes according to the first objective mentioned above, and controlling the lower limb assistance wearable robot at various ground slopes using the generated walking motion data, and controlling the lower limb assistance in accordance with the changed motion when the pedestrian changes the walking speed. Effects of the invention

[0037] According to the virtual walking motion data generation model and generation method for various walking speeds and ground slopes according to an embodiment of the present invention, it has the effect of virtually generating walking motion data for walking speeds and ground slopes that were not used in the experiment.

[0038] According to the virtual walking motion data generation model and generation method for various walking speeds and ground slopes according to an embodiment of the present invention, a walking motion data generation algorithm that combines a conditional variational autoencoder, an initial-final difference (IFD), time nondimensionalization, and loss function weight optimization is used to solve the problems of existing data generation methods.

[0039] In addition, according to the virtual walking motion data generation model and generation method for various walking speeds and ground slopes according to the embodiment of the present invention, data for multiple ground slopes and walking speeds can be generated by using walking motion data for a small number of ground slopes and walking speeds, and the generated data can be utilized to control a lower limb assistance wearable robot at various ground slopes, and also has the effect of enabling lower limb assistance control in accordance with the changed motion when a pedestrian changes their walking speed.

[0040] Meanwhile, the effects obtainable from the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art to which the present invention belongs from the description below. Brief explanation of the drawing

[0041] The following drawings attached to this specification illustrate preferred embodiments of the present invention and serve to further enhance understanding of the technical concept of the present invention together with the detailed description of the invention; therefore, the present invention should not be interpreted as being limited only to the matters described in such drawings. FIG. 1 is a block diagram of a conditional variational autoencoder according to an embodiment of the present invention, FIG. 2 is a structural diagram of a virtual walking motion data generation model for various walking speeds and ground slopes according to an embodiment of the present invention, FIG. 3 is a loss function applied to an embodiment of the present invention, Figure 4 shows DA and DS, which quantify data relevance according to an embodiment of the present invention. Specific details for implementing the invention

[0042] The above objects, other objects, features, and advantages of the present invention will be easily understood through the following preferred embodiments associated with the accompanying drawings. However, the present invention is not limited to the embodiments described herein and may be embodied in other forms. Rather, the embodiments introduced herein are provided to ensure that the disclosed content is thorough and complete, and to ensure that the spirit of the invention is sufficiently conveyed to a person skilled in the art.

[0043] In this specification, when a component is described as being on another component, it means that it may be formed directly on the other component or that a third component may be interposed between them. Also, in the drawings, the thicknesses of the components are exaggerated for the effective description of the technical content.

[0044] The embodiments described herein will be explained with reference to cross-sectional and / or plan views, which are exemplary illustrations of the invention. In the drawings, the thicknesses of films and regions are exaggerated for effective explanation of the technical content. Accordingly, the shapes of the exemplary drawings may be modified by manufacturing techniques and / or tolerances, etc. Accordingly, the embodiments of the invention are not limited to the specific shapes depicted but include variations in shape produced according to the manufacturing process. For example, a region depicted as a right angle may be rounded or have a certain curvature. Accordingly, the regions illustrated in the drawings have properties, and the shapes of the regions illustrated in the drawings are intended to illustrate specific shapes of the regions of the device and are not intended to limit the scope of the invention. Although terms such as first, second, etc., have been used to describe various components in the various embodiments of this specification, these components should not be limited by such terms. These terms are used merely to distinguish one component from another. The embodiments described and illustrated herein also include their complementary embodiments.

[0045] The terms used herein are for describing the embodiments and are not intended to limit the invention. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text. As used herein, 'comprises' and / or 'comprising' do not exclude the presence or addition of one or more other components to the mentioned components.

[0046] In describing the specific embodiments below, various specific details have been included to explain the invention more specifically and to aid understanding. However, a reader with sufficient knowledge in the art to understand the invention will recognize that it can be used without these various specific details. In some cases, it is noted in advance that commonly known aspects that are not significantly related to the invention have been omitted to prevent unnecessary confusion in describing the invention.

[0048] Hereinafter, the configuration and function of a virtual walking motion data generation model for various walking speeds and ground slopes according to an embodiment of the present invention will be described. FIG. 1 is a block diagram of a conditional variational autoencoder according to an embodiment of the present invention. FIG. 2 is a structural diagram of a virtual walking motion data generation model for various walking speeds and ground slopes according to an embodiment of the present invention.

[0049] A virtual walking motion data generation model for various walking speeds and ground slopes according to an embodiment of the present invention may be configured to include, overall, a data ionization unit, a conditional variational autoencoder, and a data post-processing unit.

[0050] The conditional transformation autoencoder receives walking motion data as input data and learns by using walking speed and ground slope as condition values ​​to derive walking motion generation data. That is, as illustrated in Fig. 1, the generative model algorithm receives walking motion data as input data and trains the conditional transformation autoencoder using walking speed and ground slope as condition values.

[0051] A virtual walking motion data generation model for various walking speeds and ground slopes according to an embodiment of the present invention applies a loss function, and the difference between the first and last values ​​of the walking cycle of the walking motion generation data (IFD) is added to the loss function for training, thereby maintaining walking data characteristics in which the first and last values ​​are similar.

[0052] IFD is used to account for the characteristics of gait data where the same behavior is repeated. Since humans repeat nearly constant walking motions at fixed ground slopes and walking speeds, it is more effective to generate multiple data points corresponding to a single walking cycle and then connect them, rather than generating walking motion data consisting of multiple connected walking cycles all at once. As mentioned earlier, since nearly constant walking motions are repeated, the starting and ending data points of a walking cycle must have nearly identical values. By adding the difference between the first and last values ​​of the generated data (IFD) to the loss function during training, the characteristic of gait data where the first and last values ​​are similar is maintained.

[0053] In addition, in an embodiment of the present invention, a loss function weight optimization unit is included to optimize the weights of the loss function in order to improve the similarity between actual walking motion data and the walking motion generation data.

[0054] The similarity between the actual data and the generated data is enhanced by adding a loss function weight optimization process. The loss function of a typical conditional variational autoencoder is simply expressed as the sum of reconstruction loss and regularization loss, and the quality of the generated data is determined by the weights for these two types of loss values. To find the optimal weight values, the following process is followed. First, weight values ​​are arbitrarily assigned. Walking motion data for the same ground slope and walking speed is generated using data on ground slope and walking speed measured in experiments. The similarity between this generated data and the data measured in experiments is quantified. The same process is applied to various weight values. The optimal weight value exhibiting the highest data similarity is found and used when generating data for ground slope and walking speed that were not used in the experiments.

[0055] In addition, the data preprocessing unit of the virtual walking motion data generation model for various walking speeds and ground slopes according to an embodiment of the present invention is configured to preprocess data before receiving walking motion data from an experiment as input data for a conditional modified autoencoder.

[0056] This data preprocessing unit may be configured to include a walk cycle-based segmentation unit that segments time series data using a walk cycle, a time nondimensionalization unit that nondimensionalizes data time using the walk cycle, and a time interval correction unit that obtains data at the same nondimensional time interval through linear interpolation.

[0057] The gait cycle-based segmentation unit divides the input gait motion data to be used for learning according to the gait cycle, the time nondimensionalization unit nondimensionalizes the time of the gait motion data by the gait cycle to change the time of the data from 0 to 1, and the time interval correction unit obtains data values ​​at a constant nondimensional time interval using linear interpolation, and through this process, the data have the same time interval under all ground slope and walking speed conditions.

[0058] Additionally, the data post-processing unit post-processes the gait motion generation data generated by the conditional deformation autoencoder, and this data post-processing unit includes a physical time conversion unit that converts the time of the gait motion generation data, expressed as dimensionless time generated by the conditional deformation autoencoder, into physical time information by multiplying it by the gait cycle.

[0059] In other words, time nondimensionalization techniques are used to enable the generation of data that considers changes in the gait cycle. The length of the gait cycle varies depending on the ground slope and walking speed. Therefore, to generate time-series data itself, the length of the output (generated data) of the generative model must vary according to the ground slope and walking speed. However, it is difficult to design multi-perceptron or convolutional neural network algorithms in which the output length changes according to conditions.

[0060] To address these issues, a time nondimensionalization technique is applied. Specifically, the gait motion data to be used for training is divided according to the gait cycle. When the time of the gait motion data is nondimensionalized by the gait cycle, the time of the data varies from 0 to 1. Subsequently, data values ​​at a constant nondimensionalized time interval are obtained using linear interpolation. Through this process, the data have the same time interval under all ground slope and gait speed conditions (i.e., the length of the output is fixed).

[0061] This modified data is used for training a conditional modified autoencoder.

[0062] The output of a conditional deformation autoencoder generates data expressed in dimensionless time. By multiplying the time of this generated data by the gait cycle, gait motion data containing physical time information can be produced.

[0063] And the above data post-processing unit performs signal smoothing through a filter to remove high-frequency noise components.

[0065] A method (algorithm) for generating virtual walking motion data for various walking speeds and ground slopes according to an embodiment of the present invention will be described. First, FIG. 2 illustrates a structural diagram of a model for generating virtual walking motion data for various walking speeds and ground slopes according to an embodiment of the present invention.

[0066] First, the experimental gait motion data to be used for training is divided according to the gait cycle. Then, by non-dimensionalizing the time of the gait motion data by the gait cycle, the time of the data changes from 0 to 1. After that, data values ​​at constant dimensionless time intervals are obtained using linear interpolation.

[0067] Then, walking motion data is received as input data, and walking speed and ground slope are used as condition values ​​to train a conditional modified autoencoder to derive walking motion generation data.

[0068] The difference between the initial and final values ​​of the gait cycle of the gait motion generation data (IFD) is added to the above loss function for training, thereby maintaining the characteristics of gait data where the initial and final values ​​are similar.

[0069] Figure 3 illustrates the loss function (Equation 1) applied to an embodiment of the present invention.

[0070] In Fig. 3, where N is the number of samples, M is the number of (dimensionless) time intervals, and β is the weight,

[0072] At this time, we must find the optimal value of β, the weight of the regularization loss of the loss function, so that the generated data follows the distribution of the actual data well.

[0073] The influence of the weight β is determined by comparing the similarity between the actual walking motion data obtained from the experiment and the virtual data generated.

[0074] FIG. 4 shows DA and DS, which quantify data similarity according to an embodiment of the present invention. Data similarity can be quantified using DA and DS defined as shown in FIG. 4.

[0075] μ in Fig. 4 exp,j represents the average value of the experimental values ​​at the j-th time interval (for example, if there are 10 experimental samples, μ exp,2 represents the average value of the 10 experimental values ​​measured in the second time interval.). μ s,j is the average of the generated data values ​​in the j-th time interval, σ exp,j σ is the variance of the experimental values ​​at the j-th time interval. s,j represents the variance of the generated data values ​​at the j-th time interval.

[0076] Calculate the DA and DS values ​​for various β values, and select the β with the minimum DA and DS as the optimal β.

[0077] After that, virtual walking motion data is generated using the optimal β value for ground slope and walking speed conditions that were not used in the experiment.

[0078] Then, a post-processing step is performed to convert the time of the gait motion generation data, expressed as dimensionless time generated by the conditional deformation autoencoder, into physical time information by multiplying it by the gait cycle.

[0079] In addition, data generated by a conditional deformation autoencoder may contain high-frequency noise components, unlike experimental data. To remove this, signal smoothing is performed using a Sacitzky-Golay filter.

[0081] In addition, the device and method described above are not limited to the configurations and methods of the embodiments described above; rather, all or part of each embodiment may be selectively combined to allow for various modifications to be made.

Claims

Claim 1 A virtual walking motion data generation model comprising: a conditional deformable autoencoder that receives walking motion data as input data and learns using walking speed and ground slope as condition values ​​to derive walking motion generation data; and a data preprocessing unit that preprocesses data before receiving walking motion data obtained from experiments as input data to the conditional deformable autoencoder; wherein the data preprocessing unit comprises: a walking cycle-based segmentation unit that segments time-series data using a walking cycle; and a time-nondimensionalization unit that nondimensionalizes data time using the walking cycle. A virtual walking motion data generation model for various walking speeds and ground slopes, comprising: a time interval correction unit that obtains data at the same dimensionless time interval through linear interpolation; a loss function weight optimization unit that applies a loss function, adds the difference between the first and last values ​​of the walking cycle of the walking motion generation data (IFD) to the loss function for training to maintain walking data characteristics where the first and last values ​​are similar, and optimizes the weights of the loss function to improve the similarity between the actual walking motion data and the walking motion generation data; wherein the weight optimization unit arbitrarily assigns weight values, generates walking motion data for the same ground slope and walking speed using data on ground slope and walking speed measured in an experiment, quantifies the similarity between the generated data and the data measured in the experiment, obtains the optimal weight value showing the highest data similarity by going through the same process for various weight values, and uses it when generating data on ground slope and walking speed not used in the experiment. Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 delete Claim 6 delete Claim 7 A virtual walking motion data generation model for various walking speeds and ground slopes, characterized in that, in claim 1, the walking cycle-based segmentation unit divides the input walking motion data to be used for learning according to the walking cycle, the time nondimensionalization unit nondimensionalizes the time of the walking motion data by the walking cycle to change the time of the data from 0 to 1, and the time interval correction unit obtains data values ​​at a constant nondimensional time interval using linear interpolation, and through this process, the data have the same time interval under all ground slope and walking speed conditions. Claim 8 A virtual walking motion data generation model for various walking speeds and ground slopes, characterized by further including, in claim 7, a data post-processing unit for post-processing walking motion generation data generated from the conditional deformation autoencoder. Claim 9 A virtual walking motion data generation model for various walking speeds and ground slopes, characterized in that, in claim 8, the data post-processing unit includes a physical time conversion unit that converts the time of walking motion generation data expressed as dimensionless time generated by the conditional deformation autoencoder into physical time information by multiplying it by a walking cycle. Claim 10 In claim 9, the data post-processing unit is characterized by performing signal smoothing through a filter to remove high-frequency noise components, thereby forming a virtual walking motion data generation model for various walking speeds and ground slopes. Claim 11 A method for generating virtual walking motion data for various walking speeds and ground slopes using a virtual walking motion data generation model according to claim 1, comprising: a preprocessing step of dividing walking motion data to be used for learning according to a walking cycle, nondimensionalizing data time using the walking cycle, and obtaining data at the same nondimensional time interval through linear interpolation; a step in which a conditional deformation autoencoder receives walking motion data as input data and learns using walking speed and ground slope as condition values ​​to derive walking motion generation data; and a postprocessing step of converting the time of the walking motion generation data expressed as nondimensional time generated by the conditional deformation autoencoder into physical time information by multiplying it by the walking cycle. Claim 12 A method for generating virtual walking motion data for various walking speeds and ground slopes, characterized in that, in claim 11, a loss function according to the following mathematical formula 1 is applied, and the difference between the initial and final values ​​of the walking cycle of the generated walking motion data (IFD) is added to the loss function for training to maintain walking data characteristics where the initial and final values ​​are similar: [Mathematical Formula 1] In the above mathematical formula 1, where N is the number of samples, M is the number of (dimensionless) time intervals, β is the weight, and y is virtual walking motion data generated by the decoder, is the actual measured walking motion data used as input to the encoder, σ is the standard deviation value output from the encoder, and μ is the average value output from the encoder. Claim 13 A method for generating virtual walking motion data for various walking speeds and ground slopes, characterized in that, in claim 12, the above loss function weight optimization unit further includes the step of optimizing the weights of the loss function to improve the similarity between the actual walking motion data and the walking motion generation data. Claim 14 A method for generating virtual walking motion data for various walking speeds and ground slopes according to claim 13, characterized in that the similarity of the data is quantified using DA of Equation 2 and DS of Equation 3 below, DA and DS values ​​are calculated for various β values, and the β having the minimum DA and DS is selected as the optimal β: [Equation 2] [Mathematical Formula 3] In the above mathematical formulas 2 and 3, here represents the average value of the experimental values ​​at the j-th time interval, and is the average of the generated data values ​​in the j-th time interval, is the variance of the experimental values ​​at the j-th time interval, is the variance of the generated data values ​​at the j-th time interval. Claim 15 A walking phase prediction system characterized by predicting walking phases, including a virtual walking motion data generation model for various walking speeds and ground slopes according to claim 1. Claim 16 A lower limb assistance wearable robot control system comprising: a virtual walking motion data generation model for various walking speeds and ground slopes according to claim 1; wherein the lower limb assistance wearable robot is controlled at various ground slopes using the generated walking motion generation data, and lower limb assistance control is performed in accordance with the changed motion when a pedestrian changes the walking speed.

Images