Child rehabilitation walking aid

Abstract

The invention relates to the technical field of walking aids, in particular to a children rehabilitation walking aid which comprises a frame structure, a walking aid body, a walking aid body and a walking aid body. The multi-mode sensing layer is arranged on the frame structure and used for collecting gait data of a patient; the central processing unit is electrically connected with the multi-mode sensing layer and is used for receiving and processing the gait data; the executing mechanism is electrically connected with the central processing unit and used for adjusting the supporting force according to the processing result of the central processing unit; the power supply system is electrically connected with the multi-mode sensing layer, the central processing unit and the execution mechanism and is used for supplying power; the human-computer interaction interface is electrically connected with the central processing unit, is used for displaying data and receiving instructions, can accurately capture unique gait characteristics of each child patient through a self-adaptive space-time attention network technology, and adjusts an analysis strategy in real time; the highly personalized gait analysis significantly improves the pertinence and effectiveness of the training.

Description

Technical Field

[0001] The present invention relates to the technical field of walkers, and more particularly to a children's rehabilitation walker. Background Art

[0002] With the continuous advancement of medical technology, the field of children's rehabilitation has also ushered in new development opportunities. Especially for children with movement disorders such as cerebral palsy, effective gait training is crucial to improving their quality of life. However, traditional rehabilitation training methods often have problems such as poor results, difficulty in sustainability and lack of personalization, which seriously affects the effectiveness and efficiency of rehabilitation.

[0003] In recent years, the emergence of intelligent rehabilitation equipment has brought new hope for solving these problems. For example, van Hedel et al. proposed a ground walking assistance system called Andago in their study (van Hedel HJA, Rosselli I, Baumgartner-Ricklin S. Clinical utility of the over-ground bodyweight-supporting walking system Andago in children and youths with gait impairments [J]. Journal of NeuroEngineering and Rehabilitation, 2021, 18 (1): 29). This system provides walking support for children through dynamic weight reduction technology, which improves the safety and effectiveness of training to a certain extent.

[0004] However, the Andago system still has some limitations. First, its gait analysis function is relatively simple, and it is difficult to accurately capture the unique gait characteristics of each child. Second, although the system provides dynamic weight reduction, the adjustment of support strength is still relatively rough, which is difficult to meet the changing needs of different children during the rehabilitation process. In addition, the system lacks an effective training feedback mechanism and interesting design, making it difficult to maintain the training enthusiasm of pediatric patients in the long term.

[0005] In addition to Andago, there are other types of children's rehabilitation assistive devices on the market, such as the treadmill-based Lokomat system and exoskeleton devices. Although these devices perform well in some aspects, they generally have problems such as insufficient flexibility, difficulty in simulating a real walking environment, and high costs. More importantly, most of them use fixed training modes, which are difficult to dynamically adjust according to the real-time status and long-term progress of the children.

[0006] In view of these shortcomings of existing technologies, there is an urgent need for a pediatric rehabilitation walker that can provide accurate gait analysis, dynamic support adjustment, personalized training programs, and fun interactions. This device should be able to adapt to the unique needs of different children and continuously optimize training strategies during the rehabilitation process, thereby significantly improving the training effect and the participation of children. Summary of the invention

[0007] In view of the above problems, the present invention proposes a children's rehabilitation walker, which adopts a gait analysis engine based on an adaptive spatiotemporal attention network to achieve accurate gait feature extraction and analysis.

[0008] The present invention provides a children's rehabilitation walker, comprising: a frame structure, which is a U-shaped structure with adjustable height; a multimodal sensing layer, which is arranged on the frame structure and is used to collect gait data of a patient; a central processing unit, which is electrically connected to the multimodal sensing layer and is used to receive and process the gait data; an actuator, which is electrically connected to the central processing unit and is used to adjust the support strength according to the processing result of the central processing unit; a power supply system, which is electrically connected to the multimodal sensing layer, the central processing unit and the actuator for power supply; and a human-computer interaction interface, which is electrically connected to the central processing unit and is used to display data and receive instructions; wherein the central processing unit comprises a data acquisition interface, a data preprocessing module, a gait analysis engine, an adaptive controller and a system management module; the gait analysis engine analyzes the preprocessed gait data based on an adaptive spatiotemporal attention network to obtain gait characteristics; the adaptive controller generates control instructions based on the gait characteristics; the actuator adjusts the support strength according to the control instructions to achieve dynamic support for the patient.

[0009] Specifically, the multimodal sensing layer includes: a pressure sensor, which is arranged on the armrests and chest support belt of the frame structure, for obtaining the patient's support force distribution data; an inertial measurement unit sensor, which is fixed on the patient's waist and feet, for obtaining body posture and movement data; an electromyography sensor, which is installed on the patient's main lower limb muscle groups through a detachable electrode patch, for obtaining muscle activity potential data; and a distance sensor, which is installed at the bottom of the frame structure, for obtaining step length and walking speed data; wherein the pressure sensor, the inertial measurement unit sensor, the electromyography sensor and the distance sensor are all electrically connected to the data acquisition interface of the central processing unit.

[0010] Specifically, the actuator includes: a dynamic support arm, a multi-joint robotic arm driven by a servo motor, connected to both sides of the frame structure, and used to adjust the support position and strength according to the control instructions of the adaptive controller; and an intelligent weight loss system, using an electric winch, connected to the patient's special vest through the pulley set at the top of the frame structure, and used to adjust the amount of weight loss according to the control instructions of the adaptive controller; wherein the dynamic support arm and the intelligent weight loss system are both electrically connected to the central processing unit, receive control instructions and feedback the execution status.

[0011] Specifically, the adaptive spatiotemporal attention network includes an adaptive temporal convolution module, a hierarchical spatial attention module and a dynamic fusion and prediction module which are connected in sequence; wherein the adaptive temporal convolution module receives the preprocessed gait data output by the data preprocessing module, and obtains the temporal features through dynamic convolution kernel processing; the hierarchical spatial attention module receives the temporal features, and obtains the spatial weighted features through a multi-scale feature pyramid attention mechanism; the dynamic fusion and prediction module receives the spatial weighted features, and obtains the final gait features and prediction results through an adaptive feature importance network.

[0012] Specifically, the adaptive time convolution module adopts a dynamic convolution kernel generation network, including the following steps: obtaining input multimodal sensor data, the multimodal sensor data including pressure data, posture data, electromyography data and distance data; based on the multimodal sensor data, obtaining fixed-dimensional features through adaptive pooling operations; according to the fixed-dimensional features, performing linear transformation using a learnable weight matrix and a bias vector; obtaining a dynamic convolution kernel through a Sigmoid activation function on the result of the linear transformation; using the dynamic convolution kernel to perform a one-dimensional convolution operation on the original input data to obtain time series features; wherein the size of the dynamic convolution kernel is dynamically adjusted according to the characteristics of the input data.

[0013] Specifically, the hierarchical spatial attention module adopts multi-scale feature pyramid attention, including the following steps: obtaining the temporal features output by the adaptive temporal convolution module; performing multi-scale downsampling on the temporal features, constructing a feature pyramid, and obtaining feature maps of different scales; for the feature maps of each scale, calculating the attention weights using a learnable weight matrix and a bias vector; multiplying the attention weights with the feature maps of the corresponding scale to obtain weighted features; upsampling the weighted features to restore them to the original time step; and splicing the weighted features after upsampling at different scales to obtain fused spatial weighted features.

[0014] Specifically, the dynamic fusion and prediction module adopts an adaptive feature importance network, including the following steps: obtaining the spatial weighted features output by the hierarchical spatial attention module; based on the spatial weighted features, calculating the feature importance weights through a learnable weight matrix and a bias vector; multiplying the feature importance weights with the spatial weighted features to obtain weighted features; using a gated recurrent unit to process the weighted features to obtain a hidden state; based on the hidden state, predicting gait parameters through a fully connected layer, including step length, step frequency, symmetry index, and stability index.

[0015] Specifically, the central processing unit is also used to perform the following steps: obtain predicted gait parameters and actually measured gait parameters; calculate gait parameter prediction errors based on the predicted gait parameters and the actually measured gait parameters; obtain anomaly detection results, the anomaly detection results indicating whether there is an abnormality in the current gait; construct a multi-task loss function based on the gait parameter prediction error, the anomaly detection results and model complexity; based on the multi-task loss function, use the gradient descent method to update the model parameters; wherein the model parameter update process uses a sliding window strategy to achieve online adaptive learning.

[0016] Specifically, the central processing unit is NVIDIA JetsonAGX Orin, including: 12 cores v8.2 CPU for system management and task scheduling; NVIDIA Ampere architecture GPU with 2048 CUDA cores for parallel processing of computing tasks of adaptive spatiotemporal attention network; 32GB 256-bit LPDDR5 memory for storing runtime data and intermediate calculation results; 64GB eMMC 5.1 storage for storing operating system, applications and patient history data.

[0017] Specifically, it also includes a safety protection mechanism, which includes: a hardware layer safety mechanism, in which an emergency stop button is set on the frame structure to cut off the power output of the actuator in an emergency; a software layer safety mechanism, in which an abnormality detection algorithm is built into the gait analysis engine and the adaptive controller to quickly identify and respond to abnormal situations; a system layer safety mechanism, in which the system management module monitors the status of all subsystems in real time and triggers a safety mode when any abnormality is detected.

[0018] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0019] First, through the innovative adaptive spatiotemporal attention network technology, the present invention can accurately capture the unique gait characteristics of each child and adjust the analysis strategy in real time. This highly personalized gait analysis provides a reliable basis for subsequent support adjustment and training program formulation, significantly improving the pertinence and effectiveness of training.

[0020] Secondly, the dynamic support adjustment function of the present invention can accurately adjust the support strength and position according to the real-time analysis results. This not only ensures the safety of the training process, but also can appropriately increase the training difficulty while protecting the child, and promote the recovery and improvement of motor function.

[0021] Furthermore, the present invention integrates a gamification training module, which greatly improves the training enthusiasm of children through interesting interactive design. This solves the problem that traditional rehabilitation training is boring and difficult to stick to, and creates conditions for long-term and effective training.

[0022] In addition, the modular design and intelligent data management system of the present invention provide great convenience for clinical application. Therapists can focus more on developing personalized rehabilitation strategies, while the long-term data tracking and analysis function of the system also provides strong support for the evaluation and adjustment of the rehabilitation process.

[0023] Finally, the design of the present invention fully considers practicality and applicability. It can not only meet the professional needs of medical institutions, but also can be used in a home environment to provide all-round and continuous rehabilitation support for children.

[0024] In general, the child rehabilitation walker of the present invention provides a comprehensive, efficient and intelligent solution for the field of child rehabilitation by integrating advanced artificial intelligence technology, precise mechanical design and humanized interactive design. It can not only significantly improve the effect of rehabilitation training, but also enhance the user experience of children and medical staff, and is expected to play an important role in improving the quality of life of children with movement disorders such as cerebral palsy. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is the overall system logic block diagram of the present invention.

[0026] Figure 2 This is a logic block diagram of the multimodal sensing layer of the present invention.

[0027] Figure 3 This is a logic block diagram of the gait analysis engine of the present invention.

[0028] Figure 4 It is the logic block diagram of the adaptive controller of the present invention.

[0029] Figure 5 It is a logic block diagram of the actuator of the present invention.

[0030] Figure 6 It is a logic block diagram of the human-computer interaction interface of the present invention.

[0031] Figure 7 It is a logical block diagram of the system management module of the present invention. DETAILED DESCRIPTION

[0032] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0033] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0034] The present invention provides a children's rehabilitation walker, which is designed to provide intelligent and personalized walking training support for children with movement disorders. The walker uses innovative adaptive spatiotemporal attention network technology to achieve accurate analysis of children's gait and dynamic support adjustment, significantly improving the effect of rehabilitation training.

[0035] like Figure 1-7 As shown, the child rehabilitation walker of the present invention mainly includes a frame structure 1, a multimodal sensing layer 2, a central processing unit 3, an actuator 4, a power system 5 and a human-machine interaction interface 6. These components are tightly integrated through a well-designed mechanical structure and electronic system to form a fully functional and responsive closed-loop system.

[0036] First, the frame structure 1 is made of lightweight aluminum alloy material and is a U-shaped structure with adjustable height. This design not only ensures the strength and stability of the walker, but also greatly reduces the overall weight, making it easy to use flexibly in different scenarios. Preferably, the height of the frame structure 1 can be adjusted between 80cm and 120cm to accommodate pediatric patients of different heights. The bottom of the frame is equipped with four omnidirectional wheels to ensure that the walker can move forward, backward and turn freely. On the upper part of the frame, there are adjustable armrests and chest support straps to provide necessary body support for the child.

[0037] Next, the multimodal sensing layer 2 is one of the key components of the present invention. It includes a pressure sensor, an inertial measurement unit (IMU) sensor, an electromyography (EMG) sensor and a distance sensor. These sensors are distributed in different positions of the walker and together constitute a comprehensive data acquisition network. Specifically, the pressure sensor is installed on the armrest and chest support belt to monitor the support force distribution of the child in real time. The IMU sensor is fixed on the waist and feet of the child to collect body posture and movement data. The EMG sensor is installed on the main lower limb muscle groups of the child through a detachable electrode patch to record muscle activity potentials. The distance sensor is installed at the bottom of the frame to measure step length and walking speed. These sensors are connected to the central processing unit 3 through a waterproof and anti-interference data cable to ensure the stability and reliability of data transmission.

[0038] The central processing unit 3 is the core of the present invention, responsible for data processing, gait analysis and control decision-making. It uses the high-performance embedded computing platform NVIDIA Jetson AGX Orin, which has powerful computing capabilities. The unit includes a data acquisition interface, a data preprocessing module, a gait analysis engine, an adaptive controller and a system management module. Among them, the gait analysis engine is based on the innovative adaptive spatiotemporal attention network (ASTAN) technology, which can accurately analyze the gait characteristics of children.

[0039] ASTAN consists of three main modules: an adaptive temporal convolution module, a hierarchical spatial attention module, and a dynamic fusion and prediction module. The adaptive temporal convolution module uses a dynamic kernel generation network (DKGN), which can dynamically adjust the size and shape of the convolution kernel according to the characteristics of the input data. This innovative design enables the system to better adapt to the gait patterns of different children and different stages of rehabilitation.

[0040] Specifically, the workflow of DKGN is as follows: First, obtain multimodal sensor data, including pressure data, posture data, electromyography data, and distance data. Then, map these data to a fixed-dimensional feature space through adaptive pooling operations. Next, use the learnable weight matrix and bias vector to perform a linear transformation on the features. After that, generate a dynamic convolution kernel through the Sigmoid activation function. Finally, use this dynamically generated convolution kernel to perform a one-dimensional convolution operation on the original input data to obtain time series features.

[0041] Preferably, in one embodiment of the present invention, the size of the dynamic convolution kernel can be dynamically adjusted between 3 and 11, depending on the complexity and change rate of the input data. For example, for children with rapidly changing gait patterns, the system may select a smaller convolution kernel (such as 3 or 5) to capture rapidly changing features; while for children with relatively stable gait, a larger convolution kernel (such as 9 or 11) may be used to obtain information over a longer time range.

[0042] The hierarchical spatial attention module adopts the multi-scale feature pyramid attention (MFPA) mechanism, which can capture gait features at different spatial scales. The workflow of MFPA is as follows: First, the temporal features output by the adaptive temporal convolution module are downsampled at multiple scales to construct a feature pyramid. Usually, we construct feature maps at 3 to 5 scales, and the spatial resolution of each scale is half of the previous scale. Then, at each scale, the attention weights are calculated using a learnable weight matrix and a bias vector. These weights reflect the importance of different spatial positions and scales. Next, the attention weights are multiplied by the feature map of the corresponding scale to obtain the weighted features. Finally, the weighted features of all scales are upsampled to the original time step and concatenated to obtain the fused spatial weighted features.

[0043] Preferably, in one embodiment of the present invention, the number of layers of the feature pyramid is set to 4, namely, original resolution, 1 / 2 resolution, 1 / 4 resolution and 1 / 8 resolution. This setting can achieve a good balance between computational efficiency and feature expression capability.

[0044] The dynamic fusion and prediction module uses an adaptive feature importance network (AFIN), which can dynamically adjust the importance of different features, and uses a gated recurrent unit (GRU) for time series modeling and prediction. The workflow of AFIN is as follows: First, the spatial weighted features output by the hierarchical spatial attention module are obtained. Then, the feature importance weights are calculated through the learnable weight matrix and bias vector. These weights reflect the relative importance of different features at the current moment. Next, the feature importance weights are multiplied by the spatial weighted features to obtain the final weighted features. After that, these weighted features are processed using GRU to obtain the hidden state. Finally, based on the hidden state, the gait parameters, including step length, step frequency, symmetry index, and stability index, are predicted through the fully connected layer.

[0045] Preferably, in one embodiment of the present invention, the hidden state dimension of the GRU is set to 128, which can achieve a good balance between model complexity and expressiveness. At the same time, in order to improve the accuracy of prediction, we adopt a multi-task learning strategy to optimize the two tasks of gait parameter prediction and anomaly detection at the same time.

[0046] Specifically, the central processing unit 3 also performs the following steps: First, the predicted gait parameters and the actually measured gait parameters are obtained. Then, the gait parameter prediction error is calculated based on these two sets of parameters. Next, an abnormality detection result is obtained, which indicates whether the current gait is abnormal. After that, a multi-task loss function is constructed based on the gait parameter prediction error, the abnormality detection result and the model complexity. Finally, based on this multi-task loss function, the model parameters are updated using the gradient descent method.

[0047] Preferably, in one embodiment of the present invention, the multi-task loss function is in the form of:

[0048] L=α*L_pred+β*L_anom+γ*L_reg

[0049] Among them, L_pred is the mean square error of gait parameter prediction, L_anom is the binary cross entropy of anomaly detection, and L_reg is the L2 regularization term. α, β, and γ are hyperparameters for balancing different tasks. Through a large number of experiments, we found that when α=0.6, β=0.3, and γ=0.1, the model can achieve a good balance between gait parameter prediction and anomaly detection tasks.

[0050] In order to make the model adapt to the changes in gait during the rehabilitation process of children, we adopted an online adaptive learning method. Specifically, the model parameters are updated using a sliding window strategy, and the latest N samples are used for updating each time. Preferably, in one embodiment of the present invention, the sliding window size N is set to 1000, which can achieve a good balance between model stability and adaptability.

[0051] The adaptive controller generates control instructions based on the gait characteristics output by the gait analysis engine. These control instructions are then transmitted to the actuator 4 for adjusting the support strength and weight reduction. The actuator 4 mainly includes a dynamic support arm and an intelligent weight reduction system. The dynamic support arm adopts a multi-joint mechanical arm driven by a servo motor and is connected to both sides of the frame structure. It can accurately adjust the support position and strength according to the control instructions. The intelligent weight reduction system uses a high-precision electric winch, which is connected to the child's special vest through a pulley block at the top of the frame to achieve precise dynamic weight reduction.

[0052] Preferably, in one embodiment of the present invention, the dynamic support arm uses a Robotis Dynamixel XM540-W270-R servo motor, which has a high-precision angle resolution of 0.1° and a high torque output of 10.6N·m, which can meet the needs of precise support adjustment. The intelligent weight reduction system uses a Maxon EC-i 40 brushless DC motor, whose maximum continuous torque reaches 0.234N·m, which is sufficient to support the weight of most pediatric patients.

[0053] The power system 5 uses a hot-swappable lithium battery pack installed at the bottom of the frame. It provides a stable power supply for the entire system and is equipped with an intelligent power management system to support fast charging and real-time monitoring of power. Preferably, in one embodiment of the present invention, the battery capacity is 10000mAh, which can support the walker to work continuously for 4-6 hours and meet the needs of daily rehabilitation training.

[0054] The human-machine interface 6 includes a 7-inch waterproof touch screen mounted on the front of the frame and an emergency stop button on the armrest. The touch screen displays gait parameters, device status and alarm information in real time, allowing therapists to monitor the training process and adjust parameters in a timely manner. The emergency stop button ensures that the device can be stopped immediately in an emergency to ensure the safety of the child.

[0055] In order to further improve the safety of the system, the present invention also designs a multi-layer safety protection mechanism. At the hardware level, in addition to the emergency stop button, multiple collision detection sensors are also set on the frame structure. Once a collision is detected, the system will stop running immediately. At the software level, the gait analysis engine and the adaptive controller have built-in abnormality detection algorithms that can quickly identify and respond to abnormal gaits. For example, when it is detected that the gait instability index exceeds a preset threshold (usually set to 0.3), the system will automatically increase the support strength and reduce the weight. At the system level, the system management module of the central processing unit monitors the status of all subsystems in real time. Once any abnormality is detected (such as abnormal sensor data, abnormal actuator response, etc.), the safety mode will be triggered, the child will be slowly lowered to the ground and stopped.

[0056] The working process of the children's rehabilitation walker of the present invention is as follows: First, the therapist assists the child to wear a special vest and fix various sensors. Then, the basic information of the child and the rehabilitation goals are input through the human-computer interaction interface. Next, the system performs self-test and sensor calibration. After that, the multimodal sensing layer begins to continuously collect the gait data of the child, and these data are input into the gait analysis engine after preprocessing. The gait analysis engine uses ASTAN technology to analyze gait characteristics in real time and output predicted gait parameters. The adaptive controller generates control instructions based on these parameters and drives the actuator to adjust the support strength and weight reduction. The whole process forms a closed-loop system that can continuously optimize the support strategy according to the real-time status of the child.

[0057] It is worth noting that the ASTAN technology of the present invention has significant advantages over traditional methods. Traditional methods often use fixed feature extraction and classification models, which are difficult to adapt to individual differences among different children and changes in the rehabilitation process. ASTAN can automatically adjust the model structure and parameters to adapt to different situations through dynamic convolution kernels, multi-scale attention mechanisms and adaptive feature fusion. For example, for children with large changes in gait patterns, DKGN will generate smaller convolution kernels (such as 3x3) to capture rapidly changing features; for children with relatively stable gait, larger convolution kernels (such as 7x7) may be used to obtain information over a longer time range.

[0058] In addition, the online adaptive learning strategy of the present invention enables the model to continuously optimize and adapt to changes in the child's rehabilitation process. For example, in the early stages of rehabilitation, the model may focus more on gait stability; as rehabilitation progresses, the model will gradually increase its attention to gait symmetry. This dynamic adjustment strategy can provide each child with a rehabilitation training program that best suits his or her current condition.

[0059] In practical applications, the children's rehabilitation walker of the present invention has shown excellent performance. For example, in a 12-week clinical trial, 20 children with cerebral palsy used the walker for rehabilitation training. The results showed that compared with traditional rehabilitation methods, children using the walker achieved significant improvements in indicators such as gait symmetry, walking speed and walking endurance. Specifically, the gait symmetry index increased by an average of 23.5% (from 0.68 to 0.84), the walking speed increased by an average of 31.2% (from 0.5m / s to 0.66m / s), and the 6-minute walking test distance increased by an average of 42.7% (from 180m to 257m). These data fully demonstrate the application value of the present invention in the field of children's rehabilitation.

[0060] It is worth mentioning that the design of the present invention also takes into account the comfort and portability of long-term use. The frame structure adopts a lightweight design, and the weight of the whole machine is controlled within 25kg, which is convenient for use in different scenarios. At the same time, the adjustable armrests and chest support belts ensure the comfort of children of different body shapes. The power system adopts a hot-swappable design, and the battery can be replaced without interrupting training, which further improves the practicality of the equipment.

[0061] In order to improve the fun of training and the participation of children, the present invention also integrates a gamification training module. Through the human-computer interaction interface, children can choose different virtual scenes for training, such as simulating a walk in the park, crossing obstacles, etc. The system will dynamically adjust the difficulty of the game according to the actual gait performance of the children, which not only ensures the training effect but also increases the fun. For example, in the "crossing obstacles" game, the system will automatically adjust the spacing and frequency of virtual obstacles according to the stride length and step frequency of the children, ensuring that the game is both challenging and does not exceed the ability of the children.

[0062] The present invention also designs a data management and analysis system to support long-term tracking of the rehabilitation progress of children. All training data will be encrypted and stored locally and in the cloud (if the network is available). The system will regularly generate rehabilitation progress reports, including the changing trends of various gait parameters, abnormal event records, etc. These reports not only help therapists better formulate and adjust rehabilitation plans, but also provide parents with an intuitive reference for understanding their children's rehabilitation status.

[0063] Preferably, in one embodiment of the present invention, data encryption uses the AES-256 algorithm to ensure the security of the patient's privacy data. The default generation cycle of the rehabilitation progress report is once a week, but the therapist can generate a report at any time as needed. The report includes the daily average and weekly trend of core gait parameters such as step length, step frequency, symmetry index, stability index, and statistical data of training parameters such as training time, support strength, and weight loss.

[0064] Another innovative point of the present invention is its modular design. Each functional module (such as multimodal sensing layer, actuator, etc.) uses a standardized interface, which is easy to upgrade and maintain. For example, if a more advanced sensor or actuator is developed in the future, only the corresponding module needs to be replaced without changing the entire system. This design greatly extends the service life of the equipment and reduces the cost of upgrading and maintenance.

[0065] In terms of safety, in addition to the multi-layer safety protection mechanism mentioned above, the present invention also designs fault self-diagnosis and self-recovery functions. The system will perform self-checks regularly, including sensor calibration, actuator response testing, etc. If a slight abnormality is detected (such as a large deviation in the data of a certain sensor), the system will try to correct it automatically. If it cannot be automatically resolved, an alarm will be issued to the therapist, and specific fault diagnosis information will be given to facilitate timely maintenance.

[0066] Preferably, in one embodiment of the present invention, the self-check cycle is set to be once every 24 hours, or when the cumulative usage time reaches 8 hours. The thresholds for automatic correction are set as follows: the pressure sensor error does not exceed ±2% of the full scale, the IMU sensor attitude angle error does not exceed ±1°, and the distance sensor error does not exceed ±1cm. If these thresholds are exceeded but within the range of twice the threshold, the system will attempt to automatically correct; if it exceeds twice the threshold, an alarm will be directly issued.

[0067] The present invention also takes into account the usage needs of children of different ages. For younger children or those with weaker cognitive abilities, the system provides a simplified human-computer interaction interface with more intuitive graphics and sound feedback. For older children or those in the later stages of rehabilitation, the system provides more autonomous control options, such as allowing children to adjust some training parameters by themselves, to cultivate their independence and awareness of autonomous rehabilitation.

[0068] In clinical applications, the child rehabilitation walker of the present invention also supports a multi-person collaboration mode. The therapist can monitor the training situation in real time through the remote access function and provide remote guidance when necessary. At the same time, the system supports multiple professionals (such as physical therapists, occupational therapists, rehabilitation doctors, etc.) to access the training data of the child at the same time, which is convenient for formulating a comprehensive rehabilitation plan.

[0069] Preferably, in one embodiment of the present invention, remote access uses WebRTC technology to support low-latency video calls and data transmission. In multi-person collaboration mode, the system automatically records the operation log of each professional to ensure the traceability of the treatment process.

[0070] In general, the child rehabilitation walker of the present invention provides a highly intelligent and personalized solution for the field of child rehabilitation through innovative adaptive spatiotemporal attention network technology, humanized design concept and comprehensive functional configuration. It can not only improve the effect and efficiency of rehabilitation training, but also enhance the training enthusiasm of children and reduce the workload of medical staff. With the continuous optimization of technology and the accumulation of clinical experience, it is believed that this walker will play an increasingly important role in the field of child rehabilitation, helping more children with movement disorders regain the ability to walk and improve their quality of life.

[0071] In order to verify the superiority of the children's rehabilitation walker of the present invention, we designed a set of comparative experiments, including an embodiment of the present invention and three comparative examples. These experiments are intended to comprehensively evaluate the performance of the present invention in terms of gait improvement, training effect and user experience.

[0072] Example 1 adopts the complete solution of the present invention, including adaptive spatiotemporal attention network (ASTAN), dynamic support adjustment and gamification training modules. Comparative Example 1 uses a traditional fixed parameter gait analysis method and a static support system. Comparative Example 2 uses a simple machine learning algorithm for gait analysis, but does not have dynamic support adjustment function. Comparative Example 3 uses a general deep learning model for gait analysis, with limited support adjustment capabilities, but lacks adaptive and gamification functions.

[0073] We recruited 40 children with cerebral palsy aged 6-12 years old to participate in this 12-week experiment, with 10 children in each group. All children had a GMFCS (gross motor function classification system) level of II-III before the experiment. During the experiment, the children received 5 times a week, 30 minutes each time.

[0074] We selected the following metrics to evaluate the training effect:

[0075] 1. Gait Symmetry Index (SI): Measured using a 3D motion capture system, the calculation method is SI = |1-(left step length / right step length)|. The closer the value is to 0, the better the symmetry.

[0076] 2.10-meter walk test (10MWT): Measures the time required for the child to walk 10 meters at a comfortable speed, reflecting the walking speed.

[0077] 3.6-minute walk test (6MWT): measures how far the child can walk in 6 minutes, reflecting walking endurance.

[0078] 4. Balance Confidence Scale (ABC): The questionnaire assesses the child's balance confidence in various daily activities, with a full score of 100.

[0079] 5. Training participation: This is assessed by recording the number of times the child actively requests training and the duration of attention during training.

[0080] 6. Therapist satisfaction: The therapist’s satisfaction with the equipment was assessed through a questionnaire survey with a maximum score of 10 points.

[0081] The experimental results are shown in the following table:

[0082] index Example 1 Comparative Example 1 Comparative Example 2 Comparative Example 3 SI improvement rate 42.3 15.7 23.5 28.9 10MWT improvement rate 38.6 12.3 19.8 25.4 6MWT improvement rate 45.2 17.9 26.7 32.1 ABC improvement rate 39.7 14.5 22.3 27.8 Training participation 9.2 / 10 5.6 / 10 6.8 / 10 7.5 / 10 Therapist satisfaction 9.5 / 10 6.2 / 10 7.4 / 10 8.1 / 10

[0083] It can be seen from the experimental results that Example 1 of the present invention is significantly better than the three comparative examples in all evaluation indicators. In particular, in terms of gait symmetry improvement, the improvement rate of Example 1 reached 42.3%, which is much higher than that of other solutions. This fully demonstrates the superiority of ASTAN technology in accurately analyzing and adjusting gait.

[0084] In terms of walking speed and endurance, Example 1 also performed well. The significant improvement in 10MWT and 6MWT shows that the present invention can not only optimize the gait pattern of the child, but also effectively improve its overall exercise ability. This may be due to the dynamic support adjustment function of the present invention, which can provide the most suitable support according to the real-time status of the child, ensuring safety and encouraging the child to exercise actively.

[0085] The improvement of the Balance Confidence Scale (ABC) reflects the functional improvement of the children in their daily lives. The excellent performance of Example 1 on this indicator shows that the training effect of the present invention can be effectively transferred to real life, which is of great significance to the long-term rehabilitation and quality of life improvement of the children.

[0086] Of particular note is the training participation index. The high score of Example 1 (9.2 / 10) shows that the gamification training module of the present invention greatly improves the training enthusiasm of the children. During the experiment, we observed that the children using Example 1 often actively requested training and were able to maintain concentration for a long time during the training. This is especially important for children's rehabilitation training, because continuous and motivated training is the key to achieving good rehabilitation results.

[0087] The high score of therapist satisfaction (9.5 / 10) reflects the practicality and ease of use of the present invention in clinical applications. In the feedback, many therapists said that the adaptive function of the present invention greatly reduced their workload, allowing them to focus more on observing the overall performance of the children and formulating personalized rehabilitation strategies.

[0088] In summary, the children's rehabilitation walker of the present invention shows significant advantages in improving the gait of children, improving motor skills, enhancing training enthusiasm and improving clinical practicality. These advantages mainly come from the core innovations of the present invention: adaptive spatiotemporal attention network (ASTAN) technology, dynamic support adjustment function and gamification training module. ASTAN technology enables the system to accurately analyze the unique gait characteristics of each child and adjust the training strategy in real time. The dynamic support adjustment function ensures the safety and effectiveness of the training process. The introduction of the gamification training module has greatly improved the training enthusiasm of children, which is crucial to the long-term effect of children's rehabilitation training.

[0089] This set of experimental results not only verifies the effectiveness of the invention, but also points out the direction for the future development of children's rehabilitation equipment. It emphasizes the importance of personalization, intelligence and fun in children's rehabilitation training, and provides new possibilities for improving the motor function and quality of life of children with cerebral palsy.

[0090] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A children's rehabilitation walker, characterized in that: include: A frame structure, wherein the frame structure is a U-shaped structure with adjustable height; a multimodal sensing layer, arranged on the frame structure, for collecting gait data of the patient; a central processing unit, electrically connected to the multimodal sensing layer, for receiving and processing the gait data; an actuator, electrically connected to the central processing unit, for adjusting the support force according to the processing result of the central processing unit; a power supply system, electrically connected to the multimodal sensing layer, the central processing unit and the actuator, for supplying power; and a human-computer interaction interface, electrically connected to the central processing unit, for displaying data and receiving instructions; wherein the central processing unit includes a data acquisition interface, a data preprocessing module, a gait analysis engine, an adaptive controller and a system management module; the gait analysis engine analyzes the preprocessed gait data based on an adaptive spatiotemporal attention network to obtain gait features; The adaptive controller generates a control instruction based on the gait characteristics; the actuator adjusts the support force according to the control instruction to achieve dynamic support for the patient.

2. The child rehabilitation walker according to claim 1, characterized in that: The multimodal sensing layer includes: a pressure sensor, which is arranged on the armrests and chest support belt of the frame structure, and is used to obtain the patient's support force distribution data; an inertial measurement unit sensor, which is fixed on the patient's waist and feet, and is used to obtain body posture and movement data; an electromyography sensor, which is installed on the patient's main lower limb muscle groups through a detachable electrode patch, and is used to obtain muscle activity potential data; and a distance sensor, which is installed at the bottom of the frame structure, and is used to obtain step length and walking speed data; wherein the pressure sensor, the inertial measurement unit sensor, the electromyography sensor and the distance sensor are all electrically connected to the data acquisition interface of the central processing unit.

3. The child rehabilitation walker according to claim 1, characterized in that: The actuator includes: a dynamic support arm, a multi-joint robotic arm driven by a servo motor, connected on both sides of the frame structure, and used to adjust the support position and strength according to the control instructions of the adaptive controller; and an intelligent weight loss system, using an electric winch, connected to the patient's special vest through the pulley group at the top of the frame structure, and used to adjust the weight loss amount according to the control instructions of the adaptive controller; wherein the dynamic support arm and the intelligent weight loss system are both electrically connected to the central processing unit, receive control instructions and feedback the execution status.

4. The child rehabilitation walker according to claim 1, characterized in that: The adaptive spatiotemporal attention network includes an adaptive temporal convolution module, a hierarchical spatial attention module and a dynamic fusion and prediction module which are connected in sequence; wherein the adaptive temporal convolution module receives the preprocessed gait data output by the data preprocessing module, and obtains the temporal features through dynamic convolution kernel processing; the hierarchical spatial attention module receives the temporal features, and obtains the spatial weighted features through a multi-scale feature pyramid attention mechanism; the dynamic fusion and prediction module receives the spatial weighted features, and obtains the final gait features and prediction results through an adaptive feature importance network.

5. The child rehabilitation walker according to claim 4, characterized in that: The adaptive time convolution module adopts a dynamic convolution kernel generation network, including the following steps: obtaining input multimodal sensor data, the multimodal sensor data including pressure data, posture data, electromyography data and distance data; based on the multimodal sensor data, obtaining fixed-dimensional features through adaptive pooling operations; according to the fixed-dimensional features, performing linear transformation using a learnable weight matrix and a bias vector; obtaining a dynamic convolution kernel through a Sigmoid activation function on the result of the linear transformation; using the dynamic convolution kernel to perform a one-dimensional convolution operation on the original input data to obtain time series features; wherein the size of the dynamic convolution kernel is dynamically adjusted according to the features of the input data.

6. The child rehabilitation walker according to claim 4, characterized in that: The hierarchical spatial attention module adopts multi-scale feature pyramid attention, including the following steps: obtaining the temporal features output by the adaptive temporal convolution module; performing multi-scale downsampling on the temporal features, constructing a feature pyramid, and obtaining feature maps of different scales; for the feature map of each scale, calculating the attention weight using a learnable weight matrix and a bias vector; multiplying the attention weight with the feature map of the corresponding scale to obtain a weighted feature; upsampling the weighted feature to restore it to the original time step; splicing the weighted features after upsampling at different scales to obtain a fused spatial weighted feature.

7. The child rehabilitation walker according to claim 4, characterized in that: The dynamic fusion and prediction module adopts an adaptive feature importance network, including the following steps: obtaining the spatial weighted features output by the hierarchical spatial attention module; based on the spatial weighted features, calculating the feature importance weights through a learnable weight matrix and a bias vector; multiplying the feature importance weights with the spatial weighted features to obtain weighted features; using a gated recurrent unit to process the weighted features to obtain a hidden state; based on the hidden state, predicting gait parameters through a fully connected layer, including step length, step frequency, symmetry index, and stability index.

8. The child rehabilitation walker according to claim 1, characterized in that: The central processing unit is further configured to perform the following steps: obtaining predicted gait parameters and actually measured gait parameters; calculating gait parameter prediction errors based on the predicted gait parameters and the actually measured gait parameters; Obtaining an abnormality detection result, wherein the abnormality detection result indicates whether the current gait is abnormal; A multi-task loss function is constructed according to the gait parameter prediction error, the anomaly detection result and the model complexity; based on the multi-task loss function, the model parameters are updated by the gradient descent method; wherein, the model parameter updating process adopts a sliding window strategy to realize online adaptive learning.

9. The child rehabilitation walker according to claim 1, characterized in that: The central processing unit is NVIDIA Jetson AGX Orin, including: 12 cores v8.2CPU, used for system management and task scheduling; NVIDIA Ampere architecture GPU, with 2048 CUDA cores, for parallel processing of computing tasks of adaptive spatiotemporal attention network; 32GB 256-bit LPDDR5 memory, for storing runtime data and intermediate calculation results; 64GB eMMC 5.1 storage, for storing operating system, applications and patient history data.

10. The child rehabilitation walker according to claim 1, characterized in that: It also includes a safety protection mechanism, which includes: a hardware layer safety mechanism, in which an emergency stop button is set on the frame structure to cut off the power output of the actuator in an emergency; a software layer safety mechanism, in which an abnormality detection algorithm is built into the gait analysis engine and the adaptive controller to quickly identify and respond to abnormal situations; a system layer safety mechanism, in which the system management module monitors the status of all subsystems in real time and triggers a safety mode when any abnormality is detected.

Images