Lower limb exoskeleton gait trajectory planning method and device based on self-oscillation

Abstract

The present invention provides a method and device for planning gait trajectories for a lower-limb exoskeleton based on self-oscillation. The method comprises continuously collecting joint angle information and plantar pressure information from a plurality of walking groups of a wearer, preliminarily dividing the walking process into multiple gait phases based on the plantar pressure information, using the preliminarily acquired gait phases and corresponding joint angle information as a data set, building a convolutional neural network (CNN) model, using the data set as training data, and continuously training the CNN model to accurately acquire the starting point of each gait phase. Each gait phase is accurately acquired from the starting point, establishing an adaptive oscillator model, inputting the data set into the adaptive oscillator model, outputting a tracking phase, and performing gait trajectory planning for the lower-limb exoskeleton based on the gait phases acquired by the CNN convolutional neural network model. The predicted output trajectory is dynamically adjusted in real time online. The present invention can adapt to the requirement of non-constant walking speed and achieve dynamic gait trajectory planning.

Description

Technical Field

[0001] The present invention relates to the technical field of gait trajectory planning for lower limb exoskeleton robots, and in particular to a method and device for gait trajectory planning for lower limb exoskeleton robots based on self-oscillation. Background Art

[0002] The elderly have weakened bodily functions, their muscles degenerate and atrophy, and are susceptible to diseases such as osteoporosis, hemiplegia, and stroke. However, there is currently a severe shortage of rehabilitation therapists who can provide rehabilitation training for patients with movement disorders. According to forecasts, as the aging population continues to deepen, the demand for rehabilitation-assisting exoskeleton robots will continue to increase, and the role of exoskeleton robots in the field of medical rehabilitation will become increasingly important.

[0003] However, most exoskeletons currently adopt predefined trajectories, and predefined gait trajectories often cannot adapt to the requirement that the walking speed is not constant when people walk. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a gait trajectory planning method and device for a lower limb exoskeleton based on self-oscillation, which uses an adaptive oscillator to track the wearer's gait, and divides the gait phases through a CNN convolutional neural network. Finally, the predicted output trajectory is dynamically adjusted in real time online according to the wearer's movement intention.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] The lower limb exoskeleton gait trajectory planning method based on self-oscillation includes:

[0007] Continuously collect joint angle information and plantar pressure information of the wearer during several walking processes;

[0008] Based on the plantar pressure information, the walking process is preliminarily divided into four gait stages: heel strike, flat foot, toe strike, and swing. The preliminarily obtained gait stages and corresponding joint angle information are used as the data set.

[0009] Building a CNN convolutional neural network model, using the data set as training data, and continuously training the CNN convolutional neural network model to accurately obtain the starting point of each gait phase, and accurately obtain each gait phase through the starting point;

[0010] An adaptive oscillator model is established, the dataset is input into the adaptive oscillator model, the tracking phase is output, and the gait phase obtained by the CNN convolutional neural network model is combined to perform lower limb exoskeleton gait trajectory planning, and the predicted output trajectory is dynamically adjusted in real time online.

[0011] Furthermore, the joint angle information includes the flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint. The joint angle information is coupled with the angular velocity of the hip joint in the sagittal plane and the coronal plane to estimate the zero position of the human body, and through coordinate axis conversion, the measured flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint in the IMU coordinate system are converted from the IMU coordinate system to the flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint in the human coordinate system.

[0012] Furthermore, the construction method of the CNN convolutional neural network model is:

[0013] Establish convolutional layer, fully connected layer, convolutional transpose layer and softmax layer;

[0014] The Paddlepaddle deep learning framework is used to train the CNN convolutional neural network model. The cross-entropy function is used to calculate the loss value of multiple classifications. The Adam optimizer is selected to reduce the loss. Multiple sets of data are iterated and the CNN convolutional neural network model with the smallest loss value is saved until the end of the iteration.

[0015] The flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint in the human coordinate system are input into the CNN convolutional neural network model, and the starting point of each gait stage is output.

[0016] Furthermore, the gait stage is judged according to the prediction results of the CNN convolutional neural network model at the two previous moments. If the prediction results at the two previous moments change, it indicates that the gait stage has changed. The moment when the gait stage changes is obtained as the starting point of the gait stage.

[0017] Furthermore, the method of establishing the adaptive oscillator model is:

[0018] The wearer's walking process is regarded as an approximately periodic signal and decomposed into multiple Fourier angle superpositions. The expression is:

[0019]

[0020] Where, is to reconstruct the joint angle, and are the Fourier decomposition coefficients, is the number of Fourier decomposition pairs, is the adaptive oscillator model phase;

[0021] Fourier decomposition coefficients and The modified differential equation is as follows:

[0022]

[0023]

[0024] Where, is a learning constant used to adjust and The speed of change, Fourier decomposition coefficient and is the amplitude of the periodic signal;

[0025] The starting point of the heel-strike gait phase is predicted by the CNN convolutional neural network model, and the phase of the adaptive oscillator model corresponding to the starting point is marked as , the gait of the cycle is divided as follows:

[0026]

[0027] Where, is the tracking phase during the wearer's walking process.

[0028] A lower limb exoskeleton gait trajectory planning device based on self-oscillation includes:

[0029] A data acquisition module is used to continuously collect joint angle information and plantar pressure information of the wearer during several walking processes;

[0030] The data preprocessing module is used to preliminarily divide the walking process into four gait stages: heel strike, flat foot, toe strike, and swing according to the plantar pressure information, and use the preliminarily obtained gait stages and corresponding joint angle information as the data set;

[0031] A data training module is used to build a CNN convolutional neural network model, use the data set as training data, and continuously train the CNN convolutional neural network model to accurately obtain the starting point of each gait phase, and accurately obtain each gait phase through the starting point;

[0032] The data tracking module is used to establish an adaptive oscillator model, input the data set into the adaptive oscillator model, output the tracking phase, combine the gait phase obtained by the CNN convolutional neural network model, perform lower limb exoskeleton gait trajectory planning, and dynamically adjust the predicted output trajectory in real time online.

[0033] A self-oscillating lower limb exoskeleton gait trajectory planning device comprising an inertial measurement unit, a force-sensitive resistor, a processor, and a memory;

[0034] The inertial measurement unit is used to continuously collect joint angle information of the wearer during several walking processes;

[0035] The force-sensitive resistor is used for plantar pressure information;

[0036] The processor is configured to execute the self-oscillating lower limb exoskeleton gait trajectory planning method when running the computer program;

[0037] The memory is used to store a computer program that can be executed on the processor.

[0038] Furthermore, the processor is a Raspberry Pi 4b.

[0039] A storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned self-oscillating lower limb exoskeleton gait trajectory planning method.

[0040] A computer program product includes a computer program, which, when executed by a processor, implements the above-mentioned self-oscillating lower limb exoskeleton gait trajectory planning method.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] The present invention combines a CNN convolutional neural network and an adaptive oscillator. The adaptive oscillator is a suitable strategy with the inherent characteristics of generating temporally continuous rhythmic motion, and can respond to non-periodic inputs to generate oscillatory motion with an organized pattern, which can be used for trajectory shaping of lower limb exoskeleton. However, the gait information generated by the adaptive oscillator is not periodic, which is not convenient for dividing the gait phase. The CNN convolutional neural network model is a better way to predict the starting time of each gait phase, and thus can accurately predict the four gait phases of heel landing, flat feet, toe landing, and swinging. Therefore, the present invention is based on a self-oscillating lower limb exoskeleton gait trajectory planning method, which collects joint angle information and plantar pressure information of several groups of wearers during walking, obtains a data set after pre-processing, and then combines the CNN convolutional neural network model and the adaptive oscillator model to obtain gait information when the walking speed is not constant, performs lower limb exoskeleton gait trajectory planning, and dynamically adjusts the predicted output trajectory in real time online. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0044] Figure 1 This is a flow chart of the lower limb exoskeleton gait trajectory planning method based on self-oscillation of the present invention.

[0045] Figure 2 Schematic diagram of the lower limb exoskeleton gait trajectory planning device based on self-oscillation of the present invention.

[0046] Figure 3Schematic diagram of the lower limb exoskeleton gait trajectory planning robot based on self-oscillation of the present invention. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0048] In the following description, numerous specific details are provided to provide a more thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without one or more of these details. In other instances, certain technical features well known in the art are not described to avoid confusion with the present invention.

[0049] It should be understood that the present invention can be implemented in different forms and should not be interpreted as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to make disclosure thorough and complete and to fully convey the scope of the present invention to those skilled in the art.

[0050] Example 1

[0051] Example 1 provides a lower limb exoskeleton gait trajectory planning method based on self-oscillation, such as Figure 1 As shown, the following steps are included:

[0052] Step S1, data collection: continuously collecting joint angle information and plantar pressure information of the wearer during several walking processes;

[0053] Step S2, data preprocessing: preliminarily divide the walking process into four gait stages based on the plantar pressure information: heel strike, flat foot, toe strike, and swing. The preliminarily obtained gait stages and corresponding joint angle information are used as a data set;

[0054] Step S3, data training: Build a CNN convolutional neural network model, use the dataset as training data, and continuously train the CNN convolutional neural network model to accurately obtain the starting point of each gait phase, and accurately obtain each gait phase through the starting point;

[0055] Step S4, data tracking: Establish an adaptive oscillator model, input the data set into the adaptive oscillator model, output the tracking phase, combine it with the gait phase obtained by the CNN convolutional neural network model, perform lower limb exoskeleton gait trajectory planning, and dynamically adjust the predicted output trajectory in real time online.

[0056] In this embodiment, the adaptive oscillator is a suitable strategy, which has the inherent characteristics of generating temporally continuous rhythmic motion, can respond to non-periodic inputs, and generate organized oscillatory motions, which can be used for trajectory shaping of the lower limb exoskeleton. However, the adaptive oscillator does not generate periodic gait information, which is not convenient for dividing the gait phase. The CNN convolutional neural network model is a better way to predict the starting time of each gait phase, and thus can accurately predict the four gait phases of heel landing, flat feet, toe landing, and swinging. Therefore, the self-oscillation-based lower limb exoskeleton gait trajectory planning method provided in this embodiment 1 collects joint angle information and plantar pressure information of several groups of walking processes of the wearer, obtains a data set after preprocessing, and then combines the CNN convolutional neural network model and the adaptive oscillator model to obtain gait information when the walking speed is not constant, performs lower limb exoskeleton gait trajectory planning, and dynamically adjusts the predicted output trajectory in real time online.

[0057] The following is a detailed description of the self-oscillation-based lower limb exoskeleton gait trajectory planning method provided in this embodiment.

[0058] In step S1, the joint angle information includes the flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint. The joint angle information is coupled with the angular velocity of the hip joint in the sagittal plane and the coronal plane to estimate the zero position of the human body, and through coordinate axis conversion, the measured flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint in the IMU coordinate system are converted from the IMU coordinate system to the flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint in the human coordinate system.

[0059] The hip flexion angle, which refers to the degree to which the hip bends forward, is used to assess the functional status of the hip joint. The hip extension angle, which refers to the degree to which the hip extends backward, is used to assess the functional status of the hip joint. By establishing a hip joint motion model, the range of motion of the hip joint in the sagittal and coronal planes is obtained, and the extreme motion values ​​in each plane are calculated, thereby estimating the human body's zero position, which refers to the position of the person when standing upright.

[0060] The knee flexion angle refers to the degree to which the knee bends forward and is used to assess the functional status of the knee joint. The knee extension angle refers to the degree to which the knee extends backward and is used to assess the functional status of the knee joint.

[0061] Specifically, the extraction architecture of the zero position is as follows:

[0062] Experimental observations show that during the swing phase, the angular velocity of the hip joint approaches zero in the frontal plane, as the thigh approaches its zero position (i.e., the thigh reaches its distal end). In the sagittal plane, if the human thigh is modeled as a simple pendulum, then the upright thigh position corresponds to the maximum angular velocity of the hip joint. However, actual thigh motion cannot be completely considered an idealized model, so we construct a function:

[0063]

[0064] Where, , is the proportionality coefficient, 、 are the angular velocities in the sagittal and coronal planes, It is a judgment standard. This is a cosine function. When its value reaches the maximum value, it means that the current human body is at zero position. This time is recorded as , the corresponding hip joint angle is .

[0065] During the swing phase, the swing leg moves from back to front, which means Greater than zero. During this period, the time of zero position in the p-th gait cycle can be obtained by IMU ( ) is detected and marked as (p = 1, 2, 3, ...). In order to filter out the normal fluctuations that occur during walking, we use one of the The estimated zero position is the square of the estimated zero positions during the gait cycle:

[0066] .

[0067] Then in Hip joint angle in the sagittal plane at moment It can be obtained by the following formula:

[0068]

[0069] Where, for The angle measured by the original IMU at that moment.

[0070] In this embodiment, inertial measurement units (IMUs) are placed on two hip joints and two knee joints in the human lower limb exoskeleton, requiring a total of four inertial measurement units. The data acquired by the inertial measurement units is measured in the IMU coordinate system, so coordinate conversion is required. Coordinate conversion is a description of the position of a spatial entity, and is the process of transforming from one coordinate system to another. It is achieved by establishing a one-to-one correspondence between the two coordinate systems. This embodiment refers to converting the flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint measured by the IMU coordinate system into the hip joint angle and knee joint angle in the human coordinate system.

[0071] In step S2, the specific method for preliminarily dividing the walking process into four gait stages: heel landing, flat foot landing, toe landing, and swinging, based on the plantar pressure information is as follows:

[0072] Multiple pressure sensors are placed on both the left and right feet, located on the soles or heels. These sensors collect pressure changes during walking and then assess gait. The changes in pressure on both feet are used to initially identify four gait stages: heel strike, flat foot, toe strike, and sway. In this embodiment, four to eight pressure sensors are placed on each foot.

[0073] The prediction results of the CNN convolutional neural network model are four gait stages: 0, 1, 2, and 3 (corresponding to heel landing, flat feet, toe landing, and swaying). The four gait stages are divided here to produce the data results of the CNN convolutional neural network model. The training input data of the CNN convolutional neural network model are four joint angles, and the training output data are four gait stages.

[0074] In step S3, the initially obtained gait phase and corresponding joint angle information are used as training data for the CNN convolutional neural network to train the CNN convolutional neural network model and obtain a suitable CNN convolutional neural network model. The method for constructing the CNN convolutional neural network model is as follows:

[0075] Step S301: Establish a convolutional layer, a fully connected layer, a convolutional transpose layer, and a softmax layer.

[0076] The convolution layer is used to extract input data features. The expression of the convolution layer is:

[0077] (1)

[0078] Where, is the output tensor of a layer, σ is an activation function, is the value of the convolution kernel, is the input tensor, * is the convolution operation, is the bias tensor;

[0079] The function of the fully connected layer is classification. As an activation function, the expression of the convolutional layer is:

[0080] (2)

[0081] (3)

[0082] Where, is a nonlinear activation function, are the weights between neurons, is the input tensor, is the bias tensor between neurons.

[0083] Convolutional transpose layer: When the extracted two-dimensional feature tensor is obtained, a convolution-transpose network is used to restore it to the dimension of the phase image.

[0084] Softmax layer: Convert all output results into probability values ​​between 0 and 1, with the sum of probabilities being 1. Select the largest index value of the tensor as the output. Suppose there is an array V, Vi represents the i-th element in V, then the softmax value of this element is the ratio of the exponent of the element to the sum of the exponents of all elements.

[0085] (4).

[0086] Step S202: Use the Paddlepaddle deep learning framework to train the CNN convolutional neural network model and use the cross entropy function to calculate the loss value of multiple classifications. Select the Adam optimizer to reduce the loss. In one embodiment of the present invention, the learning rate is 0.001, and multiple sets of data are iterated. The CNN convolutional neural network model with the smallest loss value is saved until the iteration is completed. In one embodiment of the present invention, the number of iterations is 500 to 1000.

[0087] It should be noted that the CNN convolutional neural network model only predicts the gait phase based on the joint angle data. The detection of the heel-strike gait characteristic event still needs to include the gait phase switching judgment condition.

[0088] The gait phase is determined based on the prediction results of the CNN convolutional neural network model at two moments before and after. If the prediction results at the two moments before and after change, it indicates that the gait phase has changed. The moment when the gait phase changes is obtained as the starting point of the gait phase.

[0089] Specifically, the gait stage is judged based on the output of the CNN convolutional neural network model at the previous moment and the output of the CNN convolutional neural network model at the current moment. When the prediction result of the CNN convolutional neural network model changes, it indicates that the gait stage has changed. The moment when the gait stage changes is obtained as the starting point of the gait stage.

[0090] For example, when the prediction result of the CNN convolutional neural network model swings from 3 to 0 heel strike, it is the moment of heel strike, and this time is recorded as the starting point of the heel strike gait phase.

[0091] In step S4, the method for establishing the adaptive oscillator model is:

[0092] Human walking can be regarded as a rhythmic movement, so it can be regarded as an approximately periodic signal, which can be decomposed into multiple Fourier angle superpositions, as shown in the following formula:

[0093]

[0094] Where, is to reconstruct the joint angle, and are the Fourier decomposition coefficients, is the number of Fourier decomposition pairs, is the adaptive oscillator model phase.

[0095] According to the adaptive oscillator model parameter experiment conducted by S3, it is measured , the joint angle can be reconstructed well. In the actual walking process of the human body, the various parameters of the above formula are always changing, so it is necessary to introduce angle errors to correct the parameters of the adaptive oscillator model as follows:

[0096]

[0097]

[0098]

[0099] Where, represents the angle error, is the input joint angle, is the adaptive oscillator model phase, is the oscillator frequency, is the coupling strength.

[0100] The above formulas are ultimately the adaptive oscillator phase A correction is to track an adaptive oscillator on its cycle time during walking, by adjusting the phase of the adaptive oscillator By analyzing the Fourier decomposition coefficients, it is divided into a period of 0~2π, which can determine the walking phase of a person in real time. and The modified differential equation is as follows:

[0101]

[0102]

[0103] Where, is a learning constant used to adjust and The speed of change is determined by experiment. Fourier decomposition coefficients and It is the amplitude of the periodic signal, which represents the change of a person's stride during walking. It is an adaptive tracking of the periodic space of a person during walking. The stride of a person is constantly changing during walking.

[0104] As above, using As the output phase of the fast-converging adaptive oscillator, accurate gait phase estimation can be achieved. is a continuously increasing variable. The starting point of the heel-strike gait phase is predicted by the CNN convolutional neural network model, and the phase of the adaptive oscillator model corresponding to the starting point is marked as , the gait of the cycle is divided as follows:

[0105]

[0106] Where, It is the tracking phase during the wearer's walking process, which is between 0 and 2π. The algorithm adjusts the input trajectory of the exoskeleton according to this phase to achieve online planning of the lower limb exoskeleton gait trajectory.

[0107] In summary, this embodiment proposes a gait trajectory planning method for lower limb exoskeleton based on self-oscillation, which includes the construction of a sensor system, the architecture and training of a CNN convolutional neural network model, and the collection of joint angle information through inertial measurement units (IMUs). After data processing, the information is passed to the adaptive oscillator model and the CNN convolutional neural network model. The adaptive oscillator model outputs To carry out online planning of the exoskeleton's gait trajectory, compared with other existing technologies, it can achieve variable-speed lower limb exoskeleton assistance, and can also dynamically adjust the predicted output trajectory online.

[0108] Example 2

[0109] Example 2 provides a lower limb exoskeleton gait trajectory planning device based on self-oscillation, such as Figure 2 Shown, including:

[0110] The data acquisition module is used for data acquisition: continuously collecting joint angle information and plantar pressure information of the wearer during several walking processes;

[0111] The data preprocessing module is used for data preprocessing: the walking process is preliminarily divided into four gait stages: heel strike, flat foot, toe strike, and swing according to the plantar pressure information. The preliminarily obtained gait stages and corresponding joint angle information are used as the data set;

[0112] Data training module, used for data training: building a CNN convolutional neural network model, using the data set as training data, and continuously training the CNN convolutional neural network model to accurately obtain the starting point of each gait phase, and accurately obtain each gait phase through the starting point;

[0113] The data tracking module is used for data tracking: establishing an adaptive oscillator model, inputting the data set into the adaptive oscillator model, outputting the tracking phase, combining the gait phase obtained by the CNN convolutional neural network model, performing lower limb exoskeleton gait trajectory planning, and dynamically adjusting the predicted output trajectory in real time online.

[0114] Example 3

[0115] Embodiment 3 provides a self-oscillating lower limb exoskeleton gait trajectory planning device, comprising an inertial measurement unit, a force-sensitive resistor, a processor, and a memory;

[0116] The inertial measurement unit is used to continuously collect joint angle information of the wearer during several walking processes;

[0117] Force-sensitive resistors are used for plantar pressure information;

[0118] The processor is configured to execute the above-mentioned self-oscillating lower limb exoskeleton gait trajectory planning method when running the computer program;

[0119] The memory is used to store computer programs that can be executed on the processor.

[0120] Specifically, the self-oscillating lower-limb exoskeleton gait trajectory planning device includes four inertial measurement units (IMUs), eight force-sensitive resistors (FSRs), and a Raspberry Pi 4b processor. Elastic bandages are used to secure the IMUs, with one placed at each joint to continuously collect joint angle information during the wearer's walking process. Eight FSR sensors are attached to the bottom of the slippers—four on the left and four on the right—to collect plantar pressure information. The FSR data acquisition and transmission module is attached to the top of the shoe. The Raspberry Pi 4b collects and records the wearer's gait data.

[0121] Example 4

[0122] Embodiment 4 provides a storage medium having a computer program stored therein. When the computer program is executed by a processor, the above-mentioned self-oscillating lower limb exoskeleton gait trajectory planning method is implemented.

[0123] Example 5

[0124] Example 5 provides a self-oscillating lower limb exoskeleton gait trajectory planning robot, such as Figure 3 As shown, it is used to execute the self-oscillating lower limb exoskeleton gait trajectory planning method as described in the above technical solution;

[0125] The self-oscillating lower limb exoskeleton gait trajectory planning robot includes a backboard control module (Ⅰ), a waist structure (Ⅱ), a knee joint mechanism (Ⅲ), a calf part (Ⅳ) and an ankle joint mechanism (V); the backboard control module (Ⅰ) integrates a processor, a battery, a voltage divider, and a TTL to USB serial port module; the waist structure (Ⅱ) is designed with an adjustable structure, connected to the backboard, and is equipped with two hip joint motors; the knee joint mechanism (Ⅲ) is driven by a flexible rope to move the two knee joint motors upward to connect the waist and calf; the calf part (Ⅳ) is equipped with a height-adjustable mechanism, and the ankle joint mechanism (V) is connected to the calf.

[0126] The voltage divider board is used to divide the voltage between the motor and the processor. The IMU communicates via serial communication. Using a TTL to USB serial port module, the serial port can be converted to a USB interface and connected to the Raspberry Pi, making it easier to process.

[0127] Example 6

[0128] Example 6 provides a computer program product, including a computer program, which, when executed by a processor, performs the above-mentioned self-oscillating lower limb exoskeleton gait trajectory planning method.

[0129] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media used in the embodiments provided herein may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0130] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0131] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A gait trajectory planning method for lower limb exoskeleton based on self-oscillation, characterized in that: include: Continuously collect joint angle information and plantar pressure information of the wearer during several walking processes; Based on the plantar pressure information, the walking process is preliminarily divided into four gait stages: heel strike, flat foot, toe strike, and swing. The preliminarily obtained gait stages and corresponding joint angle information are used as the data set. Build a CNN convolutional neural network model, use the data set as training data, and continuously train the CNN convolutional neural network model to accurately obtain the starting point of each gait phase. Accurately obtain each gait phase through the starting point: judge the gait phase according to the prediction results of the CNN convolutional neural network model at two moments before and after. If the prediction results of the two moments before and after change, it indicates that the gait phase has changed. The moment when the gait phase changes is obtained as the starting point of the gait phase; Establish an adaptive oscillator model, input the data set into the adaptive oscillator model, output the tracking phase, combine the gait phase obtained by the CNN convolutional neural network model, perform lower limb exoskeleton gait trajectory planning, and dynamically adjust the predicted output trajectory in real time online. The method to build the adaptive oscillator model is: The wearer's walking process is regarded as an approximately periodic signal and decomposed into multiple Fourier angle superpositions. The expression is: Where, is to reconstruct the joint angle, and are the Fourier decomposition coefficients, is the number of Fourier decomposition pairs, is the adaptive oscillator model phase; Fourier decomposition coefficients and The modified differential equation is as follows: Where, is a learning constant used to adjust and The speed of change, Fourier decomposition coefficient and is the amplitude of the periodic signal; The starting point of the heel-strike gait phase is predicted by the CNN convolutional neural network model, and the phase of the adaptive oscillator model corresponding to the starting point is marked as , the gait of the cycle is divided as follows: Where, is the tracking phase during the wearer's walking process.

2. The method for planning gait trajectories of a lower limb exoskeleton based on self-oscillation according to claim 1, characterized in that: The joint angle information includes the flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint. The joint angle information is coupled with the angular velocity of the hip joint in the sagittal plane and the coronal plane to estimate the zero position of the human body, and through coordinate axis conversion, the measured flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint in the IMU coordinate system are converted from the IMU coordinate system to the flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint in the human coordinate system.

3. The method for planning gait trajectories of a lower limb exoskeleton based on self-oscillation according to claim 1, characterized in that: The construction method of CNN convolutional neural network model is: Establish convolutional layer, fully connected layer, convolutional transpose layer and softmax layer; The Paddlepaddle deep learning framework is used to train the CNN convolutional neural network model. The cross-entropy function is used to calculate the loss value of multiple classifications. The Adam optimizer is selected to reduce the loss. Multiple sets of data are iterated and the CNN convolutional neural network model with the smallest loss value is saved until the end of the iteration. The flexion angle and extension angle of the hip joint and the flexion angle and extension angle of the knee joint in the human coordinate system are input into the CNN convolutional neural network model, and the starting point of each gait stage is output.

4. A lower limb exoskeleton gait trajectory planning device based on self-oscillation, characterized in that: include: A data acquisition module is used to continuously collect joint angle information and plantar pressure information of the wearer during several walking processes; The data preprocessing module is used to preliminarily divide the walking process into four gait stages: heel strike, flat foot, toe strike, and swing according to the plantar pressure information, and use the preliminarily obtained gait stages and corresponding joint angle information as the data set; A data training module is used to build a CNN convolutional neural network model, use the data set as training data, and continuously train the CNN convolutional neural network model to accurately obtain the starting point of each gait phase. Each gait phase is accurately obtained through the starting point: the gait phase is judged according to the prediction results of the CNN convolutional neural network model at two moments before and after. If the prediction results of the two moments before and after change, it indicates that the gait phase has changed. The moment when the gait phase changes is obtained as the starting point of the gait phase; The data tracking module is used to establish an adaptive oscillator model, input the data set into the adaptive oscillator model, output the tracking phase, combine the gait phase obtained by the CNN convolutional neural network model, perform lower limb exoskeleton gait trajectory planning, and dynamically adjust the predicted output trajectory in real time online. The method to build the adaptive oscillator model is: The wearer's walking process is regarded as an approximately periodic signal and decomposed into multiple Fourier angle superpositions. The expression is: Where, is to reconstruct the joint angle, and are the Fourier decomposition coefficients, is the number of Fourier decomposition pairs, is the adaptive oscillator model phase; Fourier decomposition coefficients and The modified differential equation is as follows: Where, is a learning constant used to adjust and The speed of change, Fourier decomposition coefficient and is the amplitude of the periodic signal; The starting point of the heel-strike gait phase is predicted by the CNN convolutional neural network model, and the phase of the adaptive oscillator model corresponding to the starting point is marked as , the gait of the cycle is divided as follows: Where, is the tracking phase during the wearer's walking process.

5. A self-oscillating lower limb exoskeleton gait trajectory planning device, characterized by: Includes an inertial measurement unit, a force-sensitive resistor, a processor, and a memory; The inertial measurement unit is used to continuously collect joint angle information of the wearer during several walking processes; The force-sensitive resistor is used for plantar pressure information; The processor is configured to execute the self-oscillating lower limb exoskeleton gait trajectory planning method according to any one of claims 1 to 3 when running a computer program; The memory is used to store a computer program that can be executed on the processor.

6. The self-oscillating lower limb exoskeleton gait trajectory planning device according to claim 5, characterized in that: The processor is Raspberry Pi 4b.

7. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the self-oscillating lower limb exoskeleton gait trajectory planning method according to any one of claims 1 to 3 is implemented.

8. A computer program product comprising a computer program, characterized in that: When the computer program is executed by a processor, the self-oscillating lower limb exoskeleton gait trajectory planning method according to any one of claims 1 to 3 is implemented.

Images