Flexible exoskeleton device for relieving Parkinson's freezing gait and control method thereof

Abstract

The invention discloses a flexible exoskeleton device for relieving Parkinson's freezing gait and a control method thereof.The exoskeleton device comprises a driving device, a flexible supporting and wearing structure, a multi-source sensing module and a control circuit module, and tension directly acts on the corresponding position of quadriceps femoris muscle groups through a flexible cable driven by a motor; the knee lifting action during muscle contraction is simulated, and the hip joint buckling action is completed. The invention provides an exoskeleton control method based on joint torque estimation under the condition of unknown tasks, and the method comprises the steps: predicting the joint torque of a human body based on a deep neural network, and mapping a joint torque estimation value into an auxiliary torque required by an exoskeleton through continuous functions such as zooming, delaying, filtering and the like; and the driving motor is controlled by feedback to complete tracking of the auxiliary torque. According to the control method, the aperiodic and unstructured human motion tasks can be dynamically adapted without pre-defining task types or gait phases, and individualized assistance can be provided for Parkinson patients.

Description

Technical Field

[0001] The present invention relates to the technical field of exoskeleton robots, and in particular to a flexible exoskeleton device for alleviating Parkinson's freezing gait and a control method thereof. Background Art

[0002] Parkinson's disease (PD) is one of the most common neurodegenerative diseases. The primary manifestations of PD are motor dysfunction, including slow movements (bradykinesia), resting tremor, rigidity, gait disturbances, and postural instability. Among the many motor symptoms, gait disturbances are particularly prominent due to their significant impact on patients' quality of life and ability to move independently. As the disease progresses, patients often experience freezing of gait (FOG), characterized by transient difficulty initiating or maintaining walking, which can lead to falls, disability, and related complications. The treatment of FOG in PD is a complex medical problem. Drug treatments have drawbacks such as reduced efficacy and the induction of dyskinesias. Surgical treatments, such as implanting a deep brain stimulator (DBS) to target the subthalamic nucleus (STN), are expensive and have variable efficacy. While strategies utilizing auditory, visual, or vibrotactile cues have shown some effectiveness in treating FOG, they face challenges such as limited efficacy, wide inter-individual variability, environmental limitations, and uncertain long-term efficacy.

[0003] In light of these challenges, the rapid development of wearable robotics has seen significant progress in theoretical research, technological development, and clinical application. These devices have been successfully applied to enhance motor function or rehabilitation in patients with neurological disorders such as stroke, spinal cord injury, and cerebral palsy, achieving positive results. Existing lower-limb exoskeleton robots primarily utilize rigid mechanical structures, directly providing support, assistance, and walking functions to the human body through rigid connections. Their advantages lie in their ability to precisely control position and efficiently transmit external forces, thereby assisting or enhancing the wearer's motor skills. However, rigid exoskeletons are generally heavy, increasing the wearer's energy consumption. Furthermore, the rigid structure can restrict joint mobility, altering the wearer's natural gait pattern. Flexible lower-limb exoskeletons are wearable robotic devices based on flexible materials and mechanical structures. They aim to reduce kinematic constraints, improve wearer comfort, reduce energy consumption, and restore lower-limb motor function. Flexible lower-limb exoskeletons typically use Bowden cables to transmit force generated by an electromechanical drive unit from the proximal end (e.g., the waist) to a distal attachment point on the exoskeleton suit, thereby assisting lower-limb movement. Mechanical assistance through wearable robots is expected to improve the biomechanical characteristics of walking in Parkinson's freezing of gait patients, thereby alleviating the onset of freezing of gait. Summary of the Invention

[0004] To address these issues, the present invention proposes a flexible exoskeleton device and control method for alleviating Parkinson's freezing of gait. This exoskeleton eliminates the need for task classification or manual adjustment, directly mapping estimated joint torques into exoskeleton assistive torques. This adapts to non-periodic, unstructured transitional tasks. Furthermore, through human-machine collaborative exoskeleton assistance, Parkinson's patients receive external rhythmic cues, compensating for the lack of internal rhythms in their basal ganglia, improving the biomechanical properties of their gait, and thus reducing the occurrence of freezing of gait.

[0005] The technical solution to achieve the purpose of the present invention is as follows: In a first aspect, the present invention provides a flexible exoskeleton device for alleviating Parkinson's freezing gait, comprising:

[0006] The drive device, through the flexible cable driven by the motor, applies tension directly to the corresponding position of the quadriceps muscle group, simulating the knee lifting action during muscle contraction and completing the hip flexion action;

[0007] Flexible support and wearable structure, used to ensure the fit of the device to the human body and to distribute the load of the device on the human body;

[0008] Multi-source sensing module, used to extract gait dynamics characteristics in real time and provide raw biomechanical data input for joint torque estimation model;

[0009] The control circuit module is used to analyze the sensor data stream and run the neural network model for joint torque prediction, driving the motor to provide the auxiliary torque generated by the model.

[0010] Specifically, the driving device includes a motor, a flexible cable, a guide block, a motor bracket, a connecting rod, a spring, and an encoder; the motor is fixed to the motor bracket by bolts, the end of the motor shaft is directly rigidly connected to the encoder, and the flexible cable is wrapped around the motor shaft and fixed on its surface through holes; the guide block is arranged on a double connecting rod, and the springs on both sides are nested on the connecting rod; one end of the flexible cable of the driving device is fixed to the front side of the thigh strap.

[0011] Specifically, the flexible support and wearing structure includes a waist belt, thigh straps, hanging shoulder straps, and a fabric rope loop; the waist belt is worn from the upper edge of the human pelvis to above the sacrum, and driving devices are installed on both sides of the front end of the waist belt; the thigh straps are worn on the middle of the human thigh; the hanging shoulder straps are worn on the upper body and are connected and fixed to the waist belt buckle; the fabric rope loop wraps the flexible cable, and the two ends are fixed to the bottom of the driving device and the front side of the thigh straps.

[0012] Specifically, the multi-source sensing module includes an inertial measurement unit (IMU), a foot pressure sensing insole, and a tension and compression sensor. The IMU is integrated into the support structure of the human hip joint and thighs; the foot pressure sensing insole is embedded with a six-axis IMU. The tension sensor is placed at the anchor point where the thigh straps connect to the flexible cable to directly measure the real-time tension of the cable.

[0013] Specifically, the control circuit module, which includes a main processor, coprocessor, voltage regulator, and power supply, is fixed to the back of the waistband and is responsible for handling motor control and deep learning inference tasks. The main controller executes the exoskeleton control loop at a 60Hz frequency, managing communication with the motor, tension sensor, and coprocessor. The coprocessor uses a trained temporal convolutional neural network (TCN) to infer the estimated bio-torque of the hip joint in real time and returns the estimated joint torque value to the main processor. The voltage regulator circuit is used to stably step down the high voltage output of the parallel power supply system to the low voltage required by the main processor and coprocessor.

[0014] In a second aspect, the present invention proposes an exoskeleton control method based on joint torque estimation under task unknown conditions, comprising the following steps:

[0015] Step S1: Acquire the IMU data fixed on the human hip joint and both sides of the thigh, the vertical force of the bipedal pressure sensing insoles, and the distribution center position of the vertical force. All sensor data are synchronized through the bus and transmitted to the coprocessor at a fixed frequency to form a multi-channel time series input;

[0016] Step S2: Based on the current moment, intercept the time series data of a fixed length in the past to form a network input window; independently perform normalization on each feature channel in the input window;

[0017] Step S3: The normalized data enters a temporal feature extraction network (TCN) composed of multiple layers of causal dilated convolutional modules. The network extracts multi-scale spatiotemporal features layer by layer and denormalizes the TCN output into biological joint torque estimates.

[0018] Step S4: converting the joint torque value estimated in real time by the deep neural network into the exoskeleton assist torque through torque scaling, time delay, and low-pass filtering operations;

[0019]

[0020] Among them, τ exo Expressed as the exoskeleton assist torque, τ est Expressed as the estimated value of the joint torque, LPF is a second-order Butterworth low-pass filter, α and Δt are the scale factor and delay time respectively, L -1represents the inverse Laplace transform, s represents the Laplace complex variable, t represents the time variable, ω s and ω c denote the resonant frequency and cutoff frequency respectively;

[0021] Step S5: A control method based on a feedback control architecture and an iterative learning mechanism is used to track the exoskeleton assist torque.

[0022] In one embodiment, step S3 includes the following steps:

[0023] Step S31: Each layer of the causal dilation convolution module fills the historical blank area on the left side of the time axis to ensure that the convolution kernel can cover enough historical time steps;

[0024]

[0025] Where y conv (t) represents the convolution output at time step t, W(k) represents the weight of the convolution kernel at the kth position, K is the size of the convolution kernel, x(tD·k) is the value of the input signal at time step (tD·k) (expanded sampling), and D represents the delay step of the input signal in the time dimension;

[0026] Step S32: dilate the convolution and trim the redundant data on the right side of the convolution output. The output trimming length should be equal to the padding on the left side to keep the sequence length consistent.

[0027] Step S33: After convolution and cropping, the features are nonlinearly transformed using the ReLU function without destroying causality;

[0028] Step S34: Add a Dropout layer after the activation function to randomly set some neurons to zero to ensure that the output depends only on historical inputs and prevent the model from overfitting;

[0029] Step S35: In each module, the 1x1 convolution is used to align the dimensions, and the convolution path output is added to the residual path output for connection. The activation function is used to alleviate the gradient vanishing problem.

[0030]

[0031] Where y represents the output of the activation function, H represents the convolution output, and x represents the residual input.

[0032] In one embodiment, step S5 is specifically as follows:

[0033] By changing the desired motor speed To adjust the force of the motor shaft on the load and achieve stable tracking of the exoskeleton torque, The desired torque, measured torque τ and motor angle θm Sure.

[0034]

[0035] Where Δθ m,des It is expressed as the desired motor angle change, and T is the control period;

[0036] An algorithm combining iterative learning compensation with proportional-damping control is used to achieve auxiliary torque tracking. The task of the motor tracking the desired torque is mainly completed by feedback control. Iterative learning is introduced as feedforward compensation. During multiple iterations, the system gradually adjusts the control signal to compensate for the error in each execution.

[0037]

[0038] Where, K represents the expected adjustment of the system to the angle (or position) at the n+1th iteration of the iterative learning compensation algorithm in the i-th control cycle. p is the proportional gain, e τ =τ-τ des is the torque error, τ is the measured exoskeleton torque, τ des is the desired exoskeleton torque, K d is the damping gain, To measure the motor speed. m,des (i,n) is the expected adjustment of the system to the angle (or position) at the i-th control cycle and the n-th iteration, i is the time index or the number of control cycles, the time elapsed in the step, n is the current step, K L is the iterative learning gain, D is the estimate of the delay between the command and the realization of the motor position change. β∈[0,1] is the weighting coefficient of the learning trajectory, which is used to add a "forgetting" mechanism in the learning process. flt (i,n) is the filtered torque error trajectory at the i-th control cycle and the n-th iteration, expressed as:

[0039] e flt (i,n)=(1-μ)e flt (i,n-1)+μe τ (i,n)

[0040] Among them, e τ (i,n) is the exoskeleton torque error, and μ∈[0,1] is the weighting coefficient of the learning error.

[0041] In a third aspect, the present invention proposes an electronic device comprising a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions are executed by the processor to complete the steps of the method described in the second aspect.

[0042] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the second aspect.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] (1) The present invention provides a flexible hip joint exoskeleton device based on flexible materials and mechanical structure design. The device transmits the force generated by the electromechanical drive unit from the proximal end (such as the waist) to the attachment point at the distal end of the exoskeleton suit, helping the patient to complete hip flexion movements, thereby assisting the lower limb movements.

[0045] (2) The present invention proposes an exoskeleton control method under task-unknown conditions through a network model based on instantaneous joint torque estimation. This method can dynamically adapt to non-periodic and unstructured human motion tasks without pre-defining the task type or gait phase, and can provide personalized assistance for Parkinson's patients.

[0046] (3) The present invention provides Parkinson's patients with external rhythmic cues through the assistance of a human-machine collaborative exoskeleton, compensates for the lack of internal rhythm in the basal ganglia of Parkinson's patients, improves the biomechanical characteristics of the patients' gait, and thus reduces the occurrence of freezing gait. The application of flexible exoskeletons opens up new approaches to solving medical problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0048] Figure 1 This is an assembly diagram of the flexible exoskeleton structure of Example 1 of the present invention;

[0049] Figure 2 This is an assembly diagram of the exoskeleton drive device of Example 1 of the present invention;

[0050] Figure 3 The exoskeleton control circuit module of embodiment 1 of the present invention;

[0051] Figure 4 This is a flow chart of an exoskeleton control method based on joint torque estimation under unknown task conditions proposed in Example 2 of the present invention.

[0052] Figure 5 This is a flow chart of an auxiliary torque tracking algorithm proposed in Example 2 of the present invention. DETAILED DESCRIPTION

[0053] In order to make the technical solution, purpose and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and technical solutions. The accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, and are not used to limit the scope of the present invention.

[0054] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood broadly. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediary; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0055] Example 1

[0056] The present invention discloses a flexible exoskeleton device for alleviating Parkinson's freezing gait, which can help Parkinson's patients maintain optimized gait parameters and thus avoid the occurrence of freezing gait. The structural assembly diagram of the exoskeleton device is shown in FIG. Figure 1 As shown, specifically including:

[0057] Drive device 1, through a motor-driven flexible cable, applies tension directly to the corresponding position of the quadriceps femoris muscle group, simulating the knee lift movement during muscle contraction and completing the hip flexion movement to help patients break the gait freezing state;

[0058] Flexible support and wearable structure ensure that the device fits snugly and disperses the load on the body, balancing support and comfort.

[0059] Multi-source sensing module, used to extract gait dynamics characteristics in real time and provide raw biomechanical data input for joint torque estimation model;

[0060] The control circuit module 4 is used to analyze the sensor data stream and run the neural network model for joint torque prediction, driving the motor to provide the auxiliary torque generated by the model.

[0061] For details, see Figure 1 The complete structure of the exoskeleton device is as follows: the flexible support and wearing structure includes a waist belt 202, thigh straps 204, hanging shoulder straps 201, and a fabric rope loop 203; the waist belt 202 is worn from the upper edge of the human pelvis to above the sacrum, and the drive device 1 is installed on both sides of the front end of the waist belt 202; the thigh straps 204 are worn in the middle of the human thigh, and the inertial measurement unit 301 is placed on the outside; the hanging shoulder straps 201 are worn on the upper body of the human body and are connected and fixed to the waist belt buckle to share the load of the exoskeleton device; the flexible cable 102 is wrapped in the fabric rope loop 203, and its two ends are fixed to the bottom of the drive device 1 and the front side of the thigh straps 204.

[0062] For details, see Figure 1 and Figure 2 The internal structure of the driving device includes a motor 101, a flexible cable 102, a guide block 103, a motor bracket 104, a connecting rod 105, a spring 106, and an encoder 107; the motor 101 is fixed to the motor bracket 104 by bolts, and the end of the motor shaft is directly rigidly connected to the encoder 107 through a bearing, and the flexible cable 102 is wrapped around the motor shaft and fixed on its surface through holes; the fixing blocks on both sides fix the double connecting rod on the motor bracket 104, and the guide block 103 is placed in the middle of the connecting rod 105. At the same time, the springs 106 on both sides are nested on the connecting rod 105 to provide elastic constraints for the guide block 103; the other end of the flexible cable 102 passes through the fixing block and is fixed to the thigh strap 204 through a metal buckle.

[0063] For details, see Figure 1 The sensor module and Figure 3 The multi-source sensing module includes an inertial measurement unit (IMU) 301, a foot pressure sensing insole, and a tension sensor 302. The IMU 301 is integrated into the hip joint and the outer thighs of both legs; the foot pressure sensing insole is embedded with a six-axis IMU. The tension sensor 302 is placed at the anchor point where the thigh straps connect to the flexible cable, directly measuring the real-time tension of the cable. The control circuit module 4, which is fixed to the back of the waist belt 202, includes a main processor, a coprocessor, and a power supply. The control circuit module 4 includes a main processor 401, a coprocessor 402, a voltage regulator 403, and a power supply 404. The main controller 401 executes the exoskeleton control loop at a frequency of 60Hz and manages communication with the motor, tension sensor, and coprocessor 402; the coprocessor 402 uses a trained temporal convolutional neural network to infer the bio-torque estimation value of the hip joint in real time, and returns the estimated joint torque estimation value result to the main processor 401; the voltage stabilizing circuit 403 is used to stably step down the parallel output high voltage of the power supply 404 to the low voltage required by the main processor and coprocessor.

[0064] Example 2

[0065] In this example, a method for controlling an exoskeleton based on joint torque estimation under unknown task conditions is disclosed, including a joint torque prediction model and an auxiliary torque tracking algorithm. The execution flow chart is as follows: Figure 4 and Figure 5 As shown, the steps of the method include:

[0066] Step S1: A multimodal sensing module is constructed using six-axis IMU sensors deployed at the hip joint and lateral thigh, a high-density piezoresistive plantar pressure sensing insole (with a built-in six-axis IMU to compensate for motion artifacts), and anchor tension sensors to collect human kinematic parameters. IMU data is transmitted via the CAN bus at a constant sampling rate, while vertical ground reaction force (vGRF) and center of plantar pressure (COP) data are transmitted synchronously via Bluetooth 5.0. The system uses an STM32 as the master control unit to execute the control algorithm in a real-time loop. A temporal convolutional network (TCN) is deployed on the NVIDIA Jetson Xavier NX motherboard for joint torque estimation. Communication between the main processor and coprocessor is performed using the universal asynchronous receiver / transmitter protocol. Data from the IMUs (including gyroscope and accelerometer data) fixed to the hip joint and thigh, as well as the vertical force and the center of vertical force distribution of the bipedal pressure sensing insoles, are acquired. All sensor data is synchronized via the bus and transmitted to the coprocessor at a fixed frequency, forming a multi-channel time series input.

[0067] Step S2: Using the current moment as a benchmark, set the window length based on the gait cycle duration (usually 0.8-1.2 seconds), intercept the past fixed-length time series data to form the network input window. Each feature channel within the input window is independently normalized.

[0068] Step S3: The normalized data enters the temporal convolutional neural network (TCN) which is composed of multiple layers of causal dilated convolution modules. Figure 4 , specifically including the following steps:

[0069] Step S31: Each layer of the causal dilation convolution module fills the historical blank area on the left side of the time axis to ensure that the convolution kernel can cover enough historical time steps;

[0070]

[0071] Where y conv (t) represents the convolution output at time step t, W(k) represents the weight of the convolution kernel at the kth position, K is the size of the convolution kernel, x(tD·k) is the value of the input signal at time step (tD·k) (expanded sampling), and D represents the delay step of the input signal in the time dimension;

[0072] Step S32: dilate the convolution and trim the redundant data on the right side of the convolution output. The output trimming length should be equal to the padding on the left side to keep the sequence length consistent.

[0073] Step S33: After convolution and cropping, the features are nonlinearly transformed using the ReLU function without destroying causality;

[0074] Step S34: Add a Dropout layer after the activation function to randomly set some neurons to zero to ensure that the output depends only on historical inputs and prevent the model from overfitting;

[0075] Step S35: In each module, the 1x1 convolution is used to align the dimensions, and the convolution path output is added to the residual path output for connection. The activation function is used to alleviate the gradient vanishing problem.

[0076]

[0077] Where y represents the output of the activation function, H represents the convolution output, and x represents the residual input.

[0078] The network extracts multi-scale spatiotemporal features layer by layer and denormalizes the TCN output into biological joint torque estimates.

[0079] Step S4: The joint torque values ​​estimated in real time by the deep neural network are converted into exoskeleton assist torque through torque scaling, time delay, and low-pass filtering. A second-order Butterworth filter is used to bandpass filter the joint torque estimates, selectively retaining the effective frequency band of the biomechanical signal and filtering out high-frequency noise and low-frequency drift.

[0080]

[0081] Among them, τ exo Expressed as the exoskeleton assist torque, τ est Expressed as the estimated value of the joint torque, LPF is a second-order Butterworth low-pass filter, α and Δt are the scale factor and delay time respectively, L -1 represents the inverse Laplace transform, s represents the Laplace complex variable, t represents the time variable, ω s and ω c represent the resonant frequency and cutoff frequency respectively.

[0082] Step S5: By changing the desired motor speed To adjust the force of the motor shaft on the load and achieve stable tracking of the exoskeleton auxiliary torque, The desired torque, measured torque τ and motor angle θ m OK, you can refer to Figure 5 .

[0083]

[0084] Where Δθ m,des It is expressed as the desired motor angle change, and T is the control period;

[0085] An algorithm combining iterative learning compensation with proportional-damping control is used to achieve auxiliary torque tracking. The task of the motor tracking the desired torque is mainly completed by feedback control. Iterative learning is introduced as feedforward compensation. During multiple iterations, the system gradually adjusts the control signal to compensate for the error in each execution.

[0086]

[0087] Where, K represents the expected adjustment of the system to the angle or position at the n+1th iteration of the iterative learning compensation algorithm in the i-th control cycle. p is the proportional gain, e τ =τ-τ des is the torque error, τ is the measured exoskeleton torque, τ des is the desired exoskeleton torque, K d is the damping gain, To measure the motor speed; Δθ m,des (i,n) is the expected adjustment of the system to the angle or position at the i-th control cycle and the n-th iteration, i is the time index or the number of control cycles, the time elapsed in the step, n is the current step, K L is the iterative learning gain, D s is an estimate of the delay between the command and the realization of the motor position change; β∈[0,1] is the weighting coefficient of the learning trajectory, which is used to add a "forgetting" mechanism in the learning process; e flt (i,n) is the filtered torque error trajectory at the i-th control cycle and the n-th iteration, expressed as:

[0088] e flt (i,n)=(1-μ)e flt (i,n-1)+μe τ (i,n)

[0089] Among them, e τ (i,n) is the exoskeleton torque error, and μ∈[0,1] is the weighting coefficient of the learning error.

[0090] This algorithm can maintain a relatively fast response speed when dealing with complex tasks, and can also reduce the system error through continuous iteration of learning compensation. It can effectively track the exoskeleton auxiliary torque generated based on the joint torque estimation value.

[0091] Example 3

[0092] In this embodiment, an electronic device is disclosed, including a memory and a processor, and computer instructions stored in the memory and executed on the processor, to complete the steps of the exoskeleton control method based on joint torque estimation disclosed in Example 2.

[0093] These computer program instructions can be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce computer-implemented processing, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes of Example 2.

[0094] The above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, it should be understood by those skilled in the art that any person skilled in the art can still modify the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or perform equivalent replacements on some of the technical features thereof. However, these modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A flexible exoskeleton device for alleviating Parkinson's freezing gait, characterized in that: Includes: Includes: The drive device, through the flexible cable driven by the motor, applies tension directly to the corresponding position of the quadriceps muscle group, simulating the knee lifting action during muscle contraction and completing the hip flexion action; Flexible support and wearable structure, used to ensure the fit of the device to the human body and to distribute the load of the device on the human body; Multi-source sensing module, used to extract gait dynamics characteristics in real time and provide raw biomechanical data input for joint torque estimation model; The control circuit module is used to analyze the sensor data stream and run the neural network model for joint torque prediction, driving the motor to provide the auxiliary torque generated by the model.

2. A flexible exoskeleton device for alleviating Parkinson's freezing of gait according to claim 1, characterized in that: The driving device includes a motor, a flexible cable, a guide block, a motor bracket, a connecting rod, a spring and an encoder; the motor is fixed to the motor bracket, the end of the motor shaft is directly rigidly connected to the encoder, and the flexible cable is wrapped around the motor shaft and fixed on its surface through holes; the guide block is arranged on the double connecting rod, and the springs on both sides are nested on the connecting rod; one end of the flexible cable of the driving device is fixed to the front side of the thigh strap.

3. The flexible exoskeleton device for alleviating Parkinson's freezing of gait according to claim 1, characterized in that: The flexible support and wearing structure includes a waist belt, thigh straps, hanging shoulder straps and a fabric rope loop; the waist belt is worn from the upper edge of the human pelvis to above the sacrum, and driving devices are installed on both sides of the front end of the waist belt; the thigh straps are worn on the middle of the human thigh; the hanging shoulder straps are worn on the upper body and are connected and fixed to the waist belt buckle; the fabric rope loop wraps the flexible cable, and the two ends are fixed to the bottom of the driving device and the front side of the thigh straps.

4. The flexible exoskeleton device for alleviating Parkinson's freezing of gait according to claim 1, characterized in that: The multi-source sensing module includes an inertial measurement unit, a foot pressure sensing insole, and a tension and compression sensor; the inertial measurement unit is integrated into the human hip joint and the outer sides of both thighs; the foot pressure sensing insole is embedded with a six-axis IMU, and the tension sensor is placed at the anchor point where the bilateral thigh straps are connected to the flexible cable to directly measure the real-time tension of the cable.

5. The flexible exoskeleton device for alleviating Parkinson's freezing of gait according to claim 1, characterized in that: The control circuit module includes a main processor, a coprocessor, a voltage stabilization circuit, and a power supply. The control circuit module is fixed to the back of the waist belt. The main controller executes the exoskeleton control loop at a frequency of 60Hz and manages communication with the motor, tension sensor, and coprocessor. The coprocessor uses a trained temporal convolutional neural network to infer the biotorque estimate of the hip joint in real time and returns the estimated joint torque value to the main processor. The voltage stabilization circuit is used to stably step down the parallel output high voltage of the power supply system to the low voltage required by the main processor and coprocessor.

6. A control method for a flexible wearable exoskeleton device for alleviating Parkinson's freezing gait according to claim 1, characterized in that: include: Step S1: Acquire the IMU data fixed on the human hip joint and both sides of the thigh, the vertical force of the bipedal pressure sensing insoles, and the distribution center position of the vertical force. All sensor data are synchronized through the bus and transmitted to the coprocessor at a fixed frequency to form a multi-channel time series input; Step S2: Based on the current moment, intercept the time series data of a fixed length in the past to form a network input window; independently normalize each feature channel in the input window; Step S3: The normalized data enters a temporal feature extraction network (TCN) composed of multiple layers of causal dilated convolutional modules. The network extracts multi-scale spatiotemporal features layer by layer and denormalizes the TCN output into biological joint torque estimates. Step S4: converting the joint torque value estimated in real time by the deep neural network into the exoskeleton assist torque through torque scaling, time delay, and low-pass filtering operations; Among them, τ exo Expressed as the exoskeleton assist torque, τ est Expressed as the estimated value of the joint torque, LPF is a second-order Butterworth low-pass filter, α and Δt are the scale factor and delay time respectively, L -1 represents the inverse Laplace transform, s represents the Laplace complex variable, t represents the time variable, ω s and ω c denote the resonant frequency and cutoff frequency respectively; Step S5: A control method based on a feedback control architecture and an iterative learning mechanism is used to track the exoskeleton assist torque.

7. The method according to claim 6, characterized in that Step S3 includes the following steps: Step S31: Each layer of the causal dilation convolution module fills the historical blank area on the left side of the time axis to ensure that the convolution kernel can cover enough historical time steps; Where y conv (t) represents the convolution output at time step t, W(k) represents the weight of the convolution kernel at the kth position, K is the size of the convolution kernel, x(tD·k) is the value of the input signal at time step (tD·k), and D represents the delay step of the input signal in the time dimension; Step S32: dilate the convolution and trim the redundant data on the right side of the convolution output. The output trimming length should be equal to the left padding to keep the sequence length consistent. Step S33: After convolution and cropping, the features are nonlinearly transformed using the ReLU function without destroying causality; Step S34: Add a Dropout layer after the activation function to randomly set some neurons to zero to ensure that the output depends only on historical inputs and prevent the model from overfitting; Step S35: Align the dimensions in each module through 1x1 convolution, add the convolution path output and the residual path output and connect them, and use the activation function to alleviate the gradient vanishing problem; Where y represents the output of the activation function, H represents the convolution output, and x represents the residual input.

8. The method according to claim 6, characterized in that Step S5 includes the following steps: By changing the desired motor speed To adjust the force of the motor shaft on the load and achieve stable tracking of the exoskeleton torque, The desired torque, measured torque τ and motor angle θ m Sure: Where Δθ m,des It is expressed as the desired motor angle change, and T is the control period; An algorithm combining iterative learning compensation with proportional-damping control is used to achieve auxiliary torque tracking. The task of tracking the desired torque of the motor is mainly completed by feedback control. Iterative learning is introduced as feedforward compensation. During multiple iterations, the system gradually adjusts the control signal to compensate for the error in each execution. Where, K represents the expected adjustment of the system to the angle or position at the n+1th iteration of the iterative learning compensation algorithm in the i-th control cycle. p is the proportional gain, e τ =τ-τ des is the torque error, τ is the measured exoskeleton torque, τ des is the desired exoskeleton torque, K d is the damping gain, To measure the motor speed; Δθ m,des (i,n) is the expected adjustment of the system to the angle or position at the i-th control cycle and the n-th iteration, i is the time index or the number of control cycles, the time elapsed in the step, n is the current step, K L is the iterative learning gain, D s is an estimate of the delay between the command and the realization of the motor position change; β∈[0,1] is the weighting coefficient of the learning trajectory, e flt (i,n) is the filtered torque error trajectory at the i-th control cycle and the n-th iteration, expressed as: e flt (i,n)=(1-μ)e flt (i,n-1)+μe τ (i,n) Among them, e τ (i,n) is the exoskeleton torque error, and μ∈[0,1] is the weighting coefficient of the learning error.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 6 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 6 to 8 are implemented.

Images