Medical rehabilitation training robot and control method, medium and equipment thereof

Abstract

The invention discloses a medical rehabilitation training robot, a control method thereof, a medium and equipment. The method comprises the following steps: acquiring user information, initializing a robot pose and generating an initial training strategy; executing training after the wearing state is qualified, collecting user feedback information and inputting the information into a neural network to generate a user training portrait; according to a deviation between a training level and a prediction target in a user training portrait calculation stage, generating lifting information and performing reverse mapping to obtain training defect parameters so as to adjust execution unit parameters; when the deviation exceeds a preset lifting threshold value, the stage prediction target is corrected, loop optimization is carried out until the standard is reached, and a dynamic adjustment strategy is generated; and judging a user state in real time to determine stage switching. Accurate dynamic adjustment of training parameters is realized through neural network modeling, adaptive transition of each training stage is ensured by using a closed-loop optimization mechanism, cooperative training of upper and lower limbs and dynamic auxiliary force adjustment can be effectively realized, and rehabilitation training adaptability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical rehabilitation training, and in particular to a medical rehabilitation training robot and a control method, medium and equipment thereof. Background Art

[0002] Existing medical rehabilitation training equipment is mostly designed for a single limb or movement type, lacking the integration of coordinated training of the upper and lower limbs and multimodal feedback, making it difficult to meet complex rehabilitation needs. Furthermore, it adopts a fixed structure and cannot adjust the auxiliary force in real time according to the user's movements, resulting in insufficient adaptability. Summary of the Invention

[0003] In view of the above problems, the present invention provides a medical rehabilitation training robot and its control method, medium and equipment to solve the problem that existing equipment cannot simultaneously support coordinated training of upper and lower limbs and dynamically adjust auxiliary force.

[0004] To achieve the above objectives, in a first aspect, the present application provides a control method for a medical rehabilitation training robot, comprising:

[0005] Obtain user information, initialize the robot's initial posture based on the user information, and generate an initial training strategy. The initial training strategy includes training categories, training content, and multiple stages of training levels.

[0006] Get the robot's wearing status. After the wearing status meets the wearing conditions, execute the initial training strategy for training and perform the following steps:

[0007] Generate feedback information collection mode based on training category, training content and preset target requirements to obtain user feedback information;

[0008] The training parameters of the initial training strategy are updated according to user feedback information to obtain a dynamic adjustment strategy, including:

[0009] Input user feedback information into the neural network model to obtain user training profiles;

[0010] According to the training level of the user training profile generation stage;

[0011] Generate a stage prediction target based on the stage training level and the preset training target, calculate the deviation between the stage training level and the stage prediction target, and generate improvement information;

[0012] Reverse-map the improvement information to the user training profile to obtain the training defect parameters;

[0013] Convert the training defect parameters into the training parameters of the current training content, and adjust the robot's execution unit according to the training parameters;

[0014] Determine whether the improvement information exceeds a preset improvement threshold. If so, generate a stage correction weight for the stage prediction target based on the stage training level and update the stage prediction target.

[0015] Repeat the above steps until the stage training level and the stage prediction target are within the preset target range, generate a dynamic adjustment strategy for the current stage training level and record it;

[0016] Obtain the current user status in real time and determine whether the current user status meets the preset user status for the next stage of training level;

[0017] If yes, proceed to the next training level in the initial training strategy;

[0018] If not, suspend training.

[0019] Furthermore, a feedback information collection mode and preset target requirements are generated based on the training category and training content, and user feedback information is obtained including:

[0020] Generate basic parameters based on user information, including age, medical history, and recovery status;

[0021] Generate preset target requirements based on user information and training content, including:

[0022] Construct a mapping relationship table between training content and training targets, and generate target mapping weights for multiple training targets based on basic parameters;

[0023] Adjust the lower target threshold of the training target according to the target mapping weight;

[0024] and, matching similar user groups in the database according to the basic parameters, obtaining the average of the upper target thresholds of the similar user groups, and updating it as the upper target threshold of the current training target;

[0025] Generate training target requirements by using the lower target threshold and the upper target threshold of each training target;

[0026] Arrange multiple training target requirements into preset target requirements according to the training sequence of the training content.

[0027] Furthermore, a feedback information collection mode and preset target requirements are generated based on the training category and training content, and user feedback information is obtained including:

[0028] Match the preset sensor collection group according to the training category and training content, and generate the preset collection frequency according to the user information. The preset sensor collection group includes the main sensor and the auxiliary sensor;

[0029] Acquire original feedback information of a preset sensor collection group according to a preset collection frequency, where the original feedback information includes dominant feedback information and auxiliary feedback information;

[0030] The original feedback information is time-aligned according to the temporal and spatial order of the dominant feedback information, and the original feedback information after time alignment is subjected to feature fusion to obtain user feedback information, including:

[0031] The original feedback information after time alignment is segmented into sliding windows, and the window length of the sliding window is configured to be determined according to the training content;

[0032] Extract the time-frequency domain joint features within each sliding window. The time-frequency domain joint features include mean, variance, zero-crossing rate, wavelet packet energy ratio, and coupling coefficient between the dominant feedback information and the auxiliary feedback information.

[0033] Calculate the spatiotemporal consistency index of multiple time-frequency domain joint features according to the training type, and generate the error correction weight based on the spatiotemporal consistency index:

[0034] And, inputting the time-frequency domain joint features into the feature importance weighting model to obtain the importance weight value;

[0035] According to the error correction weight and the importance weight value, the time-frequency domain joint features are converted into a user feedback information vector with unified dimension, which is the user feedback information.

[0036] Furthermore, the user feedback information is input into the neural network model to obtain the user training profile including:

[0037] Construct a cascade neural network model, including a feature encoding module, a spatiotemporal association module, and a portrait generation module;

[0038] The user feedback information vector is input into the feature encoding module for feature extraction to obtain encoding features, including:

[0039] A one-dimensional convolution layer is used to extract local features in the time domain, and the width of the convolution kernel is proportional to the length of the sliding window;

[0040] The temporal dependency features are captured through the long short-term memory network layer, and the number of hidden layer nodes is set to a multiple of the dimension of the user feedback information vector;

[0041] The encoded features are input into the spatiotemporal correlation module for cross-modal fusion to obtain fused features, including:

[0042] Calculate the attention weights of the dominant feedback features and the auxiliary feedback features:

[0043] Calibrate attention weights based on error correction weights;

[0044] The encoded features are fused according to the calibrated attention weights to obtain fused features;

[0045] Input the fused features into the portrait generation module and output the user training portrait, including:

[0046] Mapped to a high-dimensional feature space through a fully connected layer, the dimension of the high-dimensional feature space is a multiple of the dimension of the user feedback information vector;

[0047] The softmax function is used to generate a user training profile, which includes the ability values of multiple training parts of the user.

[0048] Furthermore, generating the current user training level based on the user training profile includes:

[0049] Input the ability value in the user training profile into the piecewise linear function and map it to the stage training level;

[0050] Calculate the stage training level based on the deviation between the level label and the ability value of the stage training level;

[0051] Reverse mapping the improvement information to the user training profile, the training defect parameters obtained include:

[0052] Obtaining level coefficients of multiple training parts in the stage training level;

[0053] Determine one by one whether the horizontal coefficient is within the range of the preset training threshold. If not, the training part associated with the horizontal coefficient is recorded as a defective part.

[0054] Extract the coding features and fusion features corresponding to the defect parts and construct the defect feature vector;

[0055] Calculate the cosine similarity between the defect feature vector and the standard feature vector of the same training part in a similar user group to generate the defect degree parameter;

[0056] The training defect parameters are generated according to the defect location, defect feature vector and defect degree parameters.

[0057] In a second aspect, the present invention also provides a medical rehabilitation training robot, which is suitable for the method described in the first aspect. The robot includes a support component, a first training component, a second training component, a third training component and a control component. The support component includes a support body, a first support rod group, a second support rod group and a third support rod group. The support body is arranged in a vertical direction, and the first support rod group, the second support rod group and the third support rod group are arranged on the support body in sequence from top to bottom; the first training component is arranged on the first support rod group, and the first training component includes a laser projector, a projector and a camera. The first training component is arranged toward a preset training area, and the laser projector and the projector are both used to project instruction information to the preset training area, and the camera is used to collect the user's first action information; the second training component is arranged on the second support rod group, and the second training component includes a gripping armrest, a first adjustment group and a first collection group. The gripping armrest is used for the user to grip The first adjustment group is used to adjust the analog value of grasping the armrest, and the first acquisition group is used to collect the user's grasping feedback information; the third training component includes a pressure-sensitive floor mat, a multi-axis auxiliary rod group, a second adjustment group and a second acquisition group. The pressure-sensitive floor mat is arranged at the bottom of the support body, the multi-axis auxiliary rod group, the second adjustment group and the second acquisition group are all arranged on the third support rod group. The multi-axis auxiliary rod group is used to be tied to the user's lower limbs and assist the user's lower limb training. The second adjustment group is arranged on the multi-axis auxiliary rod group, and the second acquisition group is arranged on the multi-axis auxiliary rod group; the control component includes a display unit, a control unit, a voice broadcast unit and a visual recognition unit. The control component is arranged on the support body. The display unit is used to display training information. The control unit is electrically connected to the display unit, the voice broadcast unit, the visual recognition unit, the first training component, the second training component and the third training component respectively. The visual recognition unit is used to collect the user's second motion information.

[0058] Furthermore, the second support rod group includes a first upper limb support rod and a second upper limb support rod, and the first upper limb support rod is arranged on the support body; the second upper limb support rod is arranged on the support body, and the second upper limb support rod and the first upper limb support rod are arranged opposite to each other from bottom to top; the first adjustment group includes two first adjustable chains, two first adjustment drive units, two second adjustable chains, and two second adjustment drive units, the two first adjustable chains are arranged at both ends of the first upper limb support rod, and the first adjustable chain is suspended downward; the two first adjustment drive units are arranged at both ends of the first upper limb support rod, each first adjustment drive unit is connected to a first adjustable chain for transmission, and the first adjustment drive unit Used to adjust the damping of the first adjustable chain; two second adjustable chains are arranged at both ends of the second upper limb support rod, and the grasping armrest is arranged between the first adjustable chain and the second adjustable chain; two second adjustment drive units are arranged at both ends of the second upper limb support rod, each second adjustment drive unit is connected to a second adjustable chain in a transmission manner, and the second adjustment drive unit is used to adjust the damping of the second adjustable chain; the first acquisition group includes two pressure sensors, two tactile sensors, and two temperature sensors, the two pressure sensors are respectively arranged on the two grasping armrests; the two tactile sensors are respectively arranged on the two grasping armrests; the two temperature sensors are respectively arranged on the two grasping armrests.

[0059] Furthermore, the multi-axis auxiliary rod group includes two first auxiliary rods, two second auxiliary rods, and two third auxiliary rods. The two first auxiliary rods are respectively arranged at both ends of the third support rod group. A first binding belt is provided between the two first auxiliary rods, and the first binding belt is used to bind the user's hips; the two second auxiliary rods are transmission-connected to the first auxiliary rod, and a second binding belt is provided on the two second auxiliary rods, and the second binding belt is used to bind the user's thighs; the two third auxiliary rods are transmission-connected to the second auxiliary rod, and a third binding belt is provided on the two third auxiliary rods, and the third binding belt is used to bind the user's calves; the second adjustment group includes two third adjustment drive units, two fourth adjustment drive units, and two fifth adjustment drive units. The two third adjustment drive units are arranged at both ends of the third support rod group, and each third adjustment drive unit is transmission-connected to a first auxiliary rod, and the third adjustment drive unit is used to adjust the telescopic length and telescopic damping of the first auxiliary rod; the two fourth adjustment drive units are arranged at one end of the second auxiliary rod, and each fourth adjustment The driving unit is transmission-connected to the other end of a second auxiliary rod, and the fourth adjustment driving unit is used to adjust the telescopic length and telescopic damping of the second auxiliary rod; two fifth adjustment driving units are arranged at one end of the third auxiliary rod, and each fifth adjustment driving unit is transmission-connected to the other end of a third auxiliary rod, and the fifth adjustment driving unit is used to adjust the telescopic length and telescopic damping of the third auxiliary rod; the second acquisition group includes two first pressure-sensitive sensors, two second pressure-sensitive sensors, two third pressure-sensitive sensors, two first inertial measurement units, two second inertial measurement units, and two third inertial measurement units. The two first pressure-sensitive sensors are respectively arranged on the two first straps; the two second pressure-sensitive sensors are respectively arranged on the two second straps; the two third pressure-sensitive sensors are respectively arranged on the two third straps; the two first inertial measurement units are respectively arranged on the two first straps; the two second inertial measurement units are respectively arranged on the two second straps; and the two third inertial measurement units are respectively arranged on the two third straps.

[0060] In a third aspect, the present invention further provides a computer-readable storage medium storing computer program instructions, which implement the method described in the first aspect when executed by a processor.

[0061] In a fourth aspect, the present invention further provides an electronic device comprising a memory and a processor, wherein the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method described in the first aspect.

[0062] Different from the existing technology, the above technical solution provides a medical rehabilitation training robot and its control method, medium and equipment. The method includes: obtaining user information to initialize the robot's initial posture and generate an initial training strategy; executing training after confirming that the wearing status meets the conditions, generating a user training profile by collecting user feedback information and inputting it into the neural network model; calculating the deviation value between the stage training level and the stage prediction target based on the user training profile, generating improvement information and reverse mapping to obtain training defect parameters, and then dynamically adjusting the training parameters of the robot execution unit; when the improvement information exceeds the preset improvement threshold, updating the stage prediction target through the stage correction weight, cyclically optimizing until the training level meets the standard, and generating a dynamic adjustment strategy for the current stage; judging the user status in real time to decide whether to enter the next stage training level. The above technical solution realizes precise dynamic adjustment of training parameters through neural network modeling, and uses a closed-loop optimization mechanism to ensure adaptive transition of each training stage. It can effectively realize coordinated training of upper and lower limbs and dynamic auxiliary force adjustment, and improve the adaptability of rehabilitation training.

[0063] The above-mentioned records related to the content of the invention are only an overview of the technical solution of this application. In order to enable ordinary technicians in this field to understand the technical solution of this application more clearly, and then implement it according to the text of the specification and the contents recorded in the drawings, and to make the above-mentioned purposes and other purposes, features and advantages of this application easier to understand, the following is an explanation in combination with the specific implementation methods and drawings of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The drawings are only used to illustrate the principles, implementation methods, applications, features and effects of the specific embodiments of the present invention and other related contents, and are not to be considered as limiting the present application.

[0065] In the drawings of the specification:

[0066] Figure 1 A method step diagram of steps S101 to S113 of the control method described in the specific embodiment;

[0067] Figure 2 This is a schematic diagram of the specific structure of the medical rehabilitation training robot described in the specific implementation method;

[0068] Figure 3 This is a schematic diagram of the front view of the medical rehabilitation training robot described in the specific embodiment;

[0069] Figure 4 This is a side view schematic diagram of the medical rehabilitation training robot described in the specific embodiment;

[0070] Figure 5 Schematic diagram of the structure of the electronic device described in the specific implementation.

[0071] The reference numerals in the above drawings are described as follows:

[0072] 1. Support components;

[0073] 11. Support the main body;

[0074] 12. The first pole group;

[0075] 13. The second support rod group;

[0076] 131. First upper limb support bar;

[0077] 132. Second upper limb support bar;

[0078] 14. The third support rod group;

[0079] 2. First training component;

[0080] 21. Laser projector;

[0081] 22. Camera;

[0082] 3. Second training component;

[0083] 31. Grab the handrails;

[0084] 32. First adjustment group;

[0085] 321, first adjustable chain;

[0086] 322. First adjustment drive unit;

[0087] 323, second adjustable chain;

[0088] 324. Second adjustment drive unit;

[0089] 4. The third training component;

[0090] 41. Pressure-sensitive floor mats;

[0091] 42. Multi-axis auxiliary rod group;

[0092] 421, first auxiliary rod;

[0093] 422, second auxiliary rod;

[0094] 423, third auxiliary rod;

[0095] 424, first binding belt;

[0096] 5. Control components;

[0097] 51. Display unit.

[0098] 6. Electronic equipment;

[0099] 61. Memory;

[0100] 62. Processor. DETAILED DESCRIPTION

[0101] In order to explain in detail the possible application scenarios, technical principles, specific solutions that can be implemented, and the purpose and effects of this application, the following is a detailed description of the specific embodiments listed in conjunction with the accompanying drawings. The embodiments described herein are only used to more clearly illustrate the technical solutions of this application and are therefore only examples and are not intended to limit the scope of protection of this application.

[0102] References to "embodiments" herein mean that the specific features, structures, or characteristics described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the word "embodiment" in various places in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or relevance to other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the various technical features mentioned in the embodiments can be combined in any manner to form a corresponding implementable technical solution.

[0103] Unless otherwise defined, the technical terms used herein have the same meanings as those generally understood by those skilled in the art to which this application belongs; the use of relevant terms herein is only for describing specific embodiments and is not intended to limit this application.

[0104] In the description of this application, the term "and / or" is used to describe a logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and both A and B exist. In addition, the character " / " in this document generally indicates that the objects before and after are in a logical "or" relationship.

[0105] In this application, terms such as "first" and "second" are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual quantity, priority or sequence relationship between these entities or operations.

[0106] Without further limitations, in this application, the words "include", "comprise", "have" or other similar open-ended expressions used in sentences are intended to cover non-exclusive inclusion. These expressions do not exclude the presence of additional elements in the process, method or product that includes the elements, so that the process, method or product that includes a series of elements may include not only those defined elements, but also other elements that are not explicitly listed, or also include elements inherent to such process, method or product.

[0107] Consistent with the understanding in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceed" are understood to exclude the number itself; expressions such as "above," "below," and "within" are understood to include the number itself. Furthermore, in the description of the embodiments of this application, "multiple" means two or more (including two), and similar expressions related to "multiple," such as "multiple groups" and "multiple times," are also understood in this manner, unless otherwise specifically defined.

[0108] In the description of the embodiments of the present application, the space-related expressions used, such as "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "vertical", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicate the orientation or position relationship based on the orientation or position relationship shown in the specific embodiments or drawings, and are only for the convenience of describing the specific embodiments of the present application or facilitating the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, it should not be understood as a limitation on the embodiments of the present application.

[0109] See also Figure 1 In a first aspect, this embodiment provides a control method for a medical rehabilitation training robot, comprising:

[0110] S101, obtaining user information, initializing the robot's initial posture according to the user information, and generating an initial training strategy, the initial training strategy including training category, training content, and multiple stages of training levels;

[0111] S102: Obtain the wearing state of the robot. After the wearing state meets the wearing conditions, execute the initial training strategy for training and perform the following steps:

[0112] S103: Generate a feedback information collection mode based on the training category, training content, and preset target requirements to obtain user feedback information;

[0113] The training parameters of the initial training strategy are updated according to user feedback information to obtain a dynamic adjustment strategy, including:

[0114] S104: Input the user feedback information into the neural network model to obtain a user training profile;

[0115] S105, generating a training level according to the user training profile;

[0116] S106: Generate a stage prediction target based on the stage training level and the preset training target, calculate the deviation between the stage training level and the stage prediction target, and generate improvement information;

[0117] S107, reverse mapping the improvement information to the user training profile to obtain training defect parameters;

[0118] S108, converting the training defect parameters into training parameters of the current training content, and adjusting the robot's execution unit according to the training parameters;

[0119] S109: Determine whether the improvement information exceeds a preset improvement threshold. If so, generate a stage correction weight for the stage prediction target based on the stage training level, and update the stage prediction target.

[0120] S110, repeat the above steps until the stage training level and the stage prediction target are within the preset target range, generate a dynamic adjustment strategy for the current stage training level and record it;

[0121] S111, obtaining the current user status in real time, and determining whether the current user status meets the preset user status for the next stage of training level;

[0122] S112: If yes, enter the next training level in the initial training strategy;

[0123] S113. If not, suspend training.

[0124] In step S101, user information may include the range of limb movement, muscle strength level, etc. The robot posture is initialized by adjusting the drive unit of the robot's support component, and a phased initial strategy is generated based on the training category and training content. The training categories include perception training, gross movement training, fine movement training and muscle training, and the training content includes head training, upper limb training, lower limb training, body training and mixed training.

[0125] In step S102, the wearing state is verified by the sensor information acquired by the first acquisition group and the second acquisition group and the posture information captured by the camera.

[0126] In step S103, the generation process of the feedback information collection mode can be understood as: according to the training category and training category, the corresponding sensor activation strategy is called from the preset rule library, such as the visual recognition unit tracking eye movements and the camera capturing the head turning angle; based on the user's rehabilitation stage, the data collection mode is set, such as the sampling rate and accuracy threshold.

[0127] For an example of preset target requirements, please refer to "gaze retention time ≥ 2 seconds", which is dynamically generated by the standard value matching the user's age and ability level in the rule library. The rule library integrates medical guidelines and historical rehabilitation data, and dynamically generates adaptive target parameters based on training categories (such as gross motor training) and content (such as lower limb training). User feedback information is obtained through sensor data obtained in real time by multiple sensors, such as pressure sensor data for grasping the handrail and motion trajectory of the lower limb inertial measurement unit.

[0128] In step S104, the neural network model extracts features through multi-source data fusion, including grip force fluctuation frequency, lower limb joint range of motion, etc., and further generates a user training portrait to quantify parameters such as gross movement coordination score and fine movement stability index.

[0129] In step S105 , the stage training level is obtained by comparing the portrait parameters with the preset rehabilitation standards (such as age-matched movement benchmarks), and the stage training level can be divided into low, intermediate and high levels.

[0130] In step S106, the stage prediction target is based on the preset target corresponding to the current stage training level (such as "intermediate" requires a grip force of 30N±5%). By calculating the deviation between the actual level and the target, improvement information is generated (such as "upper limb strength needs to be improved by 20%") to provide a quantitative basis for the dynamic adjustment strategy.

[0131] In steps S107 and S108, training defect parameters are converted into training parameter adjustments (e.g., a 20% increase in auxiliary rod damping) using a pre-set mapping rule library (e.g., a "strength defect-resistance coefficient comparison table"). Based on the correlation between defect type and training content, the rule library defines parameter correction factors. For example, coordination deviation is linearly converted into an increment of movement repetitions or a resistance gradient value, ensuring a quantitative correspondence between defect parameters and execution instructions.

[0132] In step S109, the preset improvement threshold is defined by historical data. If the deviation improvement rate exceeds the limit, the correction weight (such as target value × 0.8) is calculated according to the training level, and the stage target value is reduced or increased according to the weight to prevent sudden changes in training intensity and ensure gradual intensity.

[0133] In step S110, a preset target range is set based on the user's ability level in the profile and medical rehabilitation standards. For example, the target range for lower limb coordination is defined as ±15% of the target value. Through cyclical evaluation and parameter adjustment, the training level gradually approaches the target. Once the target is achieved, the dynamic strategy is locked and stored as a benchmark for the next stage.

[0134] In step S111 to step S113, the process of the steps can be understood as: the current user status is monitored in real time through the sensor network, and if the conditions for the next stage are met, the training level is upgraded; otherwise, a pause is prompted through the voice broadcast unit to avoid excessive load.

[0135] This embodiment achieves full-cycle personalized adaptation of rehabilitation training through dynamic strategy generation and closed-loop adjustment mechanism: initializing training strategy based on user information, generating feedback collection mode and target requirements in combination with training categories and training content, building user training portraits through multi-source data fusion, quantifying stage training levels and calculating deviation values with preset targets, generating improvement information and reverse mapping it to training defect parameters, and then converting it into dynamic adjustment instructions for the execution unit. The preset improvement threshold and target range are set based on medical standards and historical data, and the training level is gradually approached to the target through cyclic evaluation and parameter adjustment. During the process, the user status is judged in real time to control stage upgrades or pauses. This method integrates defect location, parameter correction and intensity control into a closed-loop system, and realizes step-by-step training from gross motor coordination to muscle strengthening through rule base mapping and neural network analysis, taking into account individual ability differences and training safety, avoiding the risk of intensity mutation, and improving the scientificity and efficiency of the rehabilitation process.

[0136] In some embodiments, generating a feedback information collection mode and preset target requirements based on the training category and training content, obtaining user feedback information includes:

[0137] Generate basic parameters based on user information, including age, medical history, and recovery status;

[0138] Generate preset target requirements based on user information and training content, including:

[0139] Construct a mapping relationship table between training content and training targets, and generate target mapping weights for multiple training targets based on basic parameters;

[0140] Adjust the lower target threshold of the training target according to the target mapping weight;

[0141] and, matching similar user groups in the database according to the basic parameters, obtaining the average of the upper target thresholds of the similar user groups, and updating it as the upper target threshold of the current training target;

[0142] Generate training target requirements by using the lower target threshold and the upper target threshold of each training target;

[0143] Arrange multiple training target requirements into preset target requirements according to the training sequence of the training content.

[0144] In this embodiment, basic parameters are used to quantify the basic capabilities of users and constrain the target setting range. For example, the lower limb training target threshold of elderly users needs to be adapted to the degree of joint degeneration, while users with a history of cardiovascular disease need to limit the upper limit of intensity. The mapping relationship table of training content-training goals is a preset rule base, which defines the target types and initial threshold ranges corresponding to different training contents. For example, the upper limb training mapping grip force target range is 10N-50N. The target mapping weight is generated by calculating the basic parameters. For example, the age parameter uses a segmented weight coefficient, and the initial target threshold is scaled by the weight to generate the lower target threshold.

[0145] Similar user group matching is based on historical rehabilitation data in the database. A sample of users with consistent basic parameters (e.g., age ±5 years, common medical history classification) and training content are selected. The upper limit of their training targets is extracted and averaged. For example, if the average upper limit of upper limb grip strength for 10 similar users is 40N, the upper limit target threshold for the current user is set to 40N. Together, the lower and upper target thresholds constitute the training target requirement. Multiple training target requirements are organized into a pre-set target requirement sequence based on training order (e.g., strength first, then coordination), ensuring that the training process follows a logical progression from easy to difficult and intensified in stages.

[0146] This embodiment dynamically adjusts upper and lower thresholds to prevent individual differences from causing targets to deviate from actual capabilities, while also leveraging group data to enhance target rationality. For users in the early stages of rehabilitation, the lower threshold is lowered to ensure training feasibility, while the upper threshold is referenced to the group to avoid overly conservative targets. The synergy between the rule base and the database ensures that preset target requirements are both medically standardized and personalized, providing a precise benchmark for subsequent feedback collection and strategy adjustment.

[0147] In some embodiments, generating a feedback information collection mode and preset target requirements based on the training category and training content, obtaining user feedback information includes:

[0148] Match the preset sensor collection group according to the training category and training content, and generate the preset collection frequency according to the user information. The preset sensor collection group includes the main sensor and the auxiliary sensor;

[0149] Acquire original feedback information of a preset sensor collection group according to a preset collection frequency, where the original feedback information includes dominant feedback information and auxiliary feedback information;

[0150] The original feedback information is time-aligned according to the temporal and spatial order of the dominant feedback information, and the original feedback information after time alignment is subjected to feature fusion to obtain user feedback information, including:

[0151] The original feedback information after time alignment is segmented into sliding windows, and the window length of the sliding window is configured to be determined according to the training content;

[0152] Extract the time-frequency domain joint features within each sliding window. The time-frequency domain joint features include mean, variance, zero-crossing rate, wavelet packet energy ratio, and coupling coefficient between the dominant feedback information and the auxiliary feedback information.

[0153] Calculate the spatiotemporal consistency index of multiple time-frequency domain joint features according to the training type, and generate the error correction weight based on the spatiotemporal consistency index:

[0154] And, inputting the time-frequency domain joint features into the feature importance weighting model to obtain the importance weight value;

[0155] According to the error correction weight and the importance weight value, the time-frequency domain joint features are converted into a user feedback information vector with unified dimension, which is the user feedback information.

[0156] In this embodiment, the preset sensor collection group refers to a preconfigured sensor combination based on the training category and content. The primary sensor is responsible for capturing core motion features, while the auxiliary sensors provide additional environmental or contextual motion data. The preset collection frequency is dynamically set based on user information, for example, high-frequency sampling is used for fine motor training and low-frequency sampling is used for gross motor training.

[0157] The raw feedback information is aligned in time and space to eliminate timing deviations between sensors. Specifically, alignment is based on the timestamp of the dominant feedback information, and the auxiliary feedback information is interpolated and matched along the time axis to ensure data synchronization. Sliding window segmentation adjusts the window length based on the training content. For example, upper limb grasping training uses a short window to capture instantaneous force, while lower limb gait training uses a long window to analyze cyclical movements.

[0158] In the joint time-frequency domain features, the mean reflects the average level of the signal (such as the mean of grip force), the variance represents the volatility (such as the amplitude of foot pressure fluctuation), the zero-crossing rate indicates the frequency change of the signal (such as the rate of change of joint angle), the wavelet packet energy ratio analyzes the energy distribution of different frequency bands (such as high-frequency jitter energy), and the coupling coefficient quantifies the correlation strength between the dominant and auxiliary sensors (such as the synchronization of eye movement and head turning).

[0159] The spatiotemporal consistency metric generates error correction weights by comparing the spatiotemporal distribution differences in data from different sensors, reducing noise interference. The feature importance weighting model automatically assigns weights based on the training type; for example, wavelet packet energy is given a higher weight in coordination training. Ultimately, weighted fusion generates a uniformly dimensioned user feedback information vector, which is then standardized and used by the subsequent policy generation module.

[0160] This embodiment improves the comprehensiveness and reliability of feedback information through multi-sensor collaboration, feature fusion and dynamic weighting. The dominant and auxiliary sensor combinations are matched according to the training category and content, the acquisition frequency is dynamically set, the time deviation of multi-source data is eliminated through time alignment, and the sliding window segmentation strategy is configured based on the training content to extract joint features in the time and frequency domains. The spatiotemporal consistency index is combined with the feature importance weighted model to generate error correction weights and feature weight values, which are then weighted and fused to form a user feedback information vector with unified dimensions. This method enhances data comprehensiveness through the complementarity of the dominant and auxiliary sensors. The time and frequency domain feature extraction adapts to different action characteristics (such as instantaneous grasping and periodic gait). The error correction mechanism effectively suppresses noise interference. The dynamic weighting of features highlights the core training indicators, providing highly reliable input for strategy adjustment while taking into account the efficiency of data processing and the integrity of key information.

[0161] In some embodiments, inputting user feedback information into a neural network model to obtain a user training profile includes:

[0162] Construct a cascade neural network model, including a feature encoding module, a spatiotemporal association module, and a portrait generation module;

[0163] The user feedback information vector is input into the feature encoding module for feature extraction to obtain encoding features, including:

[0164] A one-dimensional convolution layer is used to extract local features in the time domain, and the width of the convolution kernel is proportional to the length of the sliding window;

[0165] The temporal dependency features are captured through the long short-term memory network layer, and the number of hidden layer nodes is set to a multiple of the dimension of the user feedback information vector;

[0166] The encoded features are input into the spatiotemporal correlation module for cross-modal fusion to obtain fused features, including:

[0167] Calculate the attention weights of the dominant feedback features and the auxiliary feedback features:

[0168] Calibrate attention weights based on error correction weights;

[0169] The encoded features are fused according to the calibrated attention weights to obtain fused features;

[0170] Input the fused features into the portrait generation module and output the user training portrait, including:

[0171] Mapped to a high-dimensional feature space through a fully connected layer, the dimension of the high-dimensional feature space is a multiple of the dimension of the user feedback information vector;

[0172] The softmax function is used to generate a user training profile, which includes the ability values of multiple training parts of the user.

[0173] In this embodiment, the cascade neural network model is a deep learning architecture composed of a feature encoding module, a spatiotemporal association module, and a portrait generation module connected in series, and is used to generate a user capability portrait from multi-dimensional feedback information.

[0174] The user feedback information vector is input into the feature encoding module for feature extraction. A one-dimensional convolutional layer uses a sliding convolution kernel to scan the user feedback information vector (such as grip force time series data) along the time axis. The convolution kernel width is proportional to the sliding window length to ensure coverage of the entire action segment. Different convolution kernels capture features of different granularities and output local feature maps (such as peak values and fluctuation patterns) to characterize the instantaneous characteristics of the action.

[0175] The long short-term memory (LSTM) network layer processes sequential data through memory cells and a gating mechanism consisting of input, forget, and output gates. The number of hidden nodes is set as a multiple of the vector dimension to ensure that the network capacity matches the temporal complexity. A higher multiple improves the network's ability to express complex temporal patterns, while a lower multiple avoids overfitting. This multiple can be adapted to different training content (e.g., fine motor skills require a higher multiple), ensuring the generalization and adaptability of feature encoding. Memory cells retain historical states, and the gating mechanism dynamically regulates information flow to model long-term dependencies in action sequences (e.g., force decay during continuous grasping). The output hidden state vector represents the global temporal pattern.

[0176] The one-dimensional convolutional layer extracts local patterns (such as the peak interval of grip force) from the time domain signal, and the long short-term memory network layer models the long-term temporal dependency of the action sequence (such as the trend of force fluctuation). The combination of the two forms an encoding feature that combines local details and global temporal regularities, providing structured input for subsequent modules.

[0177] Attention weights are calculated based on the correlation between features and are adjusted using error correction weights, such as auxiliary feature weights to suppress noise interference. Calibrated attention weights are used to weight the fused encoding features to generate fused features, strengthening valid correlated features and weakening anomalous data.

[0178] The fully connected layer of the portrait generation module maps the fused features to a high-dimensional space to enhance the feature expression capability. The softmax function converts the high-dimensional features into a probability distribution of the ability values of multiple training parts of the user to form a user training portrait. The ability value reflects the quantitative level of each part in the training target and is used for subsequent strategy adjustment.

[0179] This embodiment uses a hierarchical process of cascading modules to gradually abstract raw feedback information into an interpretable user capability profile. The feature encoding module balances local details with global temporal sequences, the spatiotemporal correlation module uses an attention mechanism to achieve cross-modal data complementarity, and the profile generation module integrates medical evaluation logic to output a multi-dimensional capability assessment, providing a precise basis for personalized rehabilitation strategies.

[0180] In some embodiments, generating the current user training level according to the user training profile includes:

[0181] Input the ability value in the user training profile into the piecewise linear function and map it to the stage training level;

[0182] Calculate the stage training level based on the deviation between the level label and the ability value of the stage training level;

[0183] Reverse mapping the improvement information to the user training profile, the training defect parameters obtained include:

[0184] Obtaining level coefficients of multiple training parts in the stage training level;

[0185] Determine one by one whether the horizontal coefficient is within the range of the preset training threshold. If not, the training part associated with the horizontal coefficient is recorded as a defective part.

[0186] Extract the coding features and fusion features corresponding to the defect parts and construct the defect feature vector;

[0187] Calculate the cosine similarity between the defect feature vector and the standard feature vector of the same training part in a similar user group to generate the defect degree parameter;

[0188] The training defect parameters are generated according to the defect location, defect feature vector and defect degree parameters.

[0189] In this embodiment, a piecewise linear function is a preset discretization mapping tool that converts the continuous ability values in the user's training profile into staged training levels. The segmented intervals are set based on medical rehabilitation standards. Level labels are defined for each stage, and the deviation is calculated by calculating the absolute difference between the ability value and the corresponding level median. For example, if the intermediate median is 0.7 and the actual value is 0.8, the deviation is 0.1. The smaller the deviation, the closer the stage training level is to the core competency of the level.

[0190] The level coefficient refers to the proportional coefficient of the capability value of each training part relative to the preset training threshold within the stage training level. The preset training threshold is set by the rule base based on the training part type. If the level coefficient exceeds the threshold, the part is identified as defective. The defect feature vector is constructed by extracting the encoding features of the defect part in the feature encoding module and the fusion features of the spatiotemporal correlation module to represent the motion characteristics of the part.

[0191] Calculate the cosine similarity between the defect feature vector and the standard feature vector of the same training part in the similar user group, and quantify the difference between the defect feature vector and the standard feature vector of the similar user group. Preferably, the standard feature vector is the feature mean of the qualified samples of similar users in the database. For example, if the similarity between the upper limb coordination defect feature vector and the standard vector is 0.6, then the defect degree parameter is 0.4. The defect degree parameter is combined with the defect part type to generate a training defect parameter to guide subsequent strategy adjustment. The above steps convert the abstract ability value into an operational defect parameter through segmented mapping and feature comparison, which not only adapts to individual differences but also ensures the objectivity of defect positioning.

[0192] This embodiment achieves accurate defect location through structured mapping and feature comparison. The ability value in the user training portrait is mapped to the stage training level through a piecewise linear function. The stage level is quantified based on the deviation metric. The defect location is determined in combination with the preset training threshold. The coding features and fusion features are extracted to construct the defect feature vector. The defect degree parameters are generated by comparing the standard features of similar user groups through cosine similarity, and finally the training defect parameters are synthesized. This method adapts to individual ability differences through discretized mapping and threshold judgment, uses feature vector similarity analysis to ensure the objectivity of defect assessment, converts abstract ability values into operational defect parameters, provides an accurate basis for dynamic strategy adjustment, and combines group data to enhance the rationality and explainability of defect location, thereby improving the pertinence of rehabilitation training.

[0193] See also Figures 2 to 4In the second aspect, the present embodiment further provides a medical rehabilitation training robot, which is applicable to the method described in the first aspect. The robot includes a support component 1, a first training component 2, a second training component 3, a third training component 4 and a control component 5. The support component 1 includes a support body 11, a first support rod group 12, a second support rod group 13 and a third support rod group 14. The support body 11 is arranged in a vertical direction, and the first support rod group 12, the second support rod group 13 and the third support rod group 14 are arranged on the support body 11 from top to bottom in sequence; the first training component 2 is arranged on the first support rod group 12, the first training component 2 includes a laser projector 21, a projector and a camera 22, the first training component 2 is arranged toward a preset training area, the laser projector 21 and the projector are both used to project instruction information to the preset training area, and the camera 22 is used to collect the user's first action information; the second training component 3 is arranged on the second support rod group 13, the second training component 3 includes a gripping armrest 31, a first adjustment group 32 and a first collection group, and the gripping armrest 31 is used to For the user to grasp, the first adjustment group 32 is used to adjust the analog value of the grasping armrest 31, and the first acquisition group is used to collect the user's grasping feedback information; the third training component 4 includes a pressure-sensitive floor mat 41, a multi-axis auxiliary rod group 42, a second adjustment group and a second acquisition group. The pressure-sensitive floor mat 41 is arranged at the bottom of the support body 11, and the multi-axis auxiliary rod group 42, the second adjustment group and the second acquisition group are all arranged on the third support rod group 14. The multi-axis auxiliary rod group 42 is used to be tied to the user's lower limbs and assist the user's lower limb training. The second adjustment group is arranged on the multi-axis auxiliary rod group 42, and the second acquisition group is arranged on the multi-axis auxiliary rod group 42; the control component 5 includes a display unit 51, a control unit, a voice broadcast unit and a visual recognition unit. The control component 5 is arranged on the support body 11. The display unit 51 is used to display training information. The control unit is electrically connected to the display unit 51, the voice broadcast unit, the visual recognition unit, the first training component 2, the second training component 3, and the third training component 4 respectively. The visual recognition unit is used to collect the user's second motion information.

[0194] In this embodiment, the support body 11 is a vertically arranged metal frame that provides rigid support and secures the rod groups. The first rod group 12, the second rod group 13, and the third rod group 14 are arranged from top to bottom, carrying the upper limb vision training, grip training, and lower limb assistance modules, respectively.

[0195] The laser projector 21 and the projector of the first training component 2 project dynamic visual markers, such as moving light spots or graphics, to a preset area, and the camera 22 synchronously captures the movement trajectory of the user's upper limbs following the markers (i.e., the first movement information), forming a closed loop of visual guidance and movement correction.

[0196] Preferably, a spherical structure is provided at the end of the gripping armrest 31 of the second training component 3, which can simultaneously train the palm strength of the user when gripping; the first adjustment group 32 can drive the gripping armrest 31 to lift or lower through a motor or hydraulic device to simulate gripping scenes at different heights, and the first acquisition group records the gripping force changes through a pressure sensor.

[0197] The pressure-sensitive floor mat 41 of the third training component 4 is preferably a carpet-type structure with a built-in pressure-sensitive resistor, which is electrically connected to the support body 11 to monitor the pressure distribution and center of gravity offset of the foot in real time; the multi-axis auxiliary rod group 42 adopts a multi-joint soft connection sheet design, which fits the hip, knee and other joints through flexible binding straps when tied to the lower limbs. The second adjustment group dynamically adjusts the auxiliary force based on the motion data, and the second acquisition group can collect lower limb motion parameters through strain gauges and angle sensors.

[0198] The visual recognition unit of the control component 5 analyzes the data from the camera 22 to track the user's facial and body movements (i.e., the second action information); the voice broadcast unit guides the user to perform facial reaction training through instructions, such as turning the head according to the voice prompt; the display unit 51 synchronously displays the training feedback; the control unit integrates multi-source data and coordinates each component to dynamically adjust the training mode.

[0199] The second training component 3 of this embodiment provides a retractable gripping armrest 31. Combined with the spherical end structure of the gripping armrest 31, it not only trains upper limb lifting and lowering movements but also strengthens hand grip. The flexible connecting sheets of the multi-axis assistive rod assembly 42 conform to the lower limb joints, providing movement assistance while avoiding the discomfort of a rigid structure. The pressure-sensitive mat 41 accurately monitors foot pressure through changes in resistance, and combined with lower limb motion parameter analysis, improves the accuracy of gait stability assessment. The control component 5 uses visual recognition and voice commands to guide the user through five sensory reflexes such as head rotation and eye tracking, expanding the scope of rehabilitation. The control unit integrates data from each component in real time, dynamically adjusting the visual guidance difficulty, grip resistance, and lower limb assistance strength to achieve progressive training from gross to fine motor skills. The multimodal feedback mechanism enhances interactivity and instant error correction capabilities, addressing the limited functionality and poor adaptability of traditional equipment, significantly improving the comprehensiveness and personalization of rehabilitation training.

[0200] In some embodiments, the second support rod group 13 includes a first upper limb support rod 131 and a second upper limb support rod 132, and the first upper limb support rod 131 is arranged on the support body 11; the second upper limb support rod 132 is arranged on the support body 11, and the second upper limb support rod 132 is arranged opposite to the first upper limb support rod 131 from bottom to top; the first adjustment group 32 includes two first adjustable chains 321, two first adjustment drive units 322, two second adjustable chains 323, and two second adjustment drive units 324, the two first adjustable chains 321 are arranged at both ends of the first upper limb support rod 131, and the first adjustable chain 321 is suspended downward; the two first adjustment drive units 322 are arranged at both ends of the first upper limb support rod 131, and each first adjustment drive unit 322 is transmission connected to a first adjustable chain 321 The first adjustment drive unit 322 is used to adjust the damping of the first adjustable chain 321; the two second adjustable chains 323 are arranged at both ends of the second upper limb support rod 132, and the grasping armrest 31 is arranged between the first adjustable chain 321 and the second adjustable chain 323; the two second adjustment drive units 324 are arranged at both ends of the second upper limb support rod 132, and each second adjustment drive unit 324 is transmission-connected to a second adjustable chain 323, and the second adjustment drive unit 324 is used to adjust the damping of the second adjustable chain 323; the first acquisition group includes two pressure sensors, two tactile sensors, and two temperature sensors. The two pressure sensors are respectively arranged on the two grasping armrests 31; the two tactile sensors are respectively arranged on the two grasping armrests 31; and the two temperature sensors are respectively arranged on the two grasping armrests 31.

[0201] In this embodiment, the first upper limb support rod 131 and the second upper limb support rod 132 are both horizontally extending metal rods, which are respectively fixed in the middle area of the support body 11, and the two are relatively distributed from bottom to top to form an upper and lower clamping structure.

[0202] The first adjustable chain 321 and the second adjustable chain 323 are flexible connecting parts. The first adjustable chain 321 hangs downward from both ends of the first upper limb support rod 131, and the second adjustable chain 323 extends upward from both ends of the second upper limb support rod 132. The gripping armrest 31 is suspended in the middle through two groups of first adjustable chains 321 and two groups of second adjustable chains 323, forming an adjustable gripping interface.

[0203] The first adjustment drive unit 322 and the second adjustment drive unit 324 are damping adjustment devices with built-in motors. The first adjustment drive unit 322 is installed at both ends of the first upper limb support rod 131, and the second adjustment drive unit 324 is installed at both ends of the second upper limb support rod 132. The lifting and lowering resistance of the gripping armrest 31 is controlled by changing the damping coefficient of the first adjustable chain 321 and the second adjustable chain 323.

[0204] Preferably, a pressure sensor is embedded in the surface of the gripping armrest 31 to quantify the user's gripping force; a tactile sensor is distributed on the contact surface of the armrest to detect the gripping contact state and duration; a temperature sensor is integrated inside the armrest to monitor the changes in palm temperature during training and provide comprehensive feedback on the physiological state.

[0205] This embodiment realizes multi-dimensional adjustment of the height and resistance of the gripping armrest 31 through the coordinated design of the first upper limb support rod 131, the second upper limb support rod 132 and the first adjustable chain 321, the second adjustable chain 323, and adapts to different training modes such as lifting and pulling down. The first acquisition group is configured to include multiple types of sensors including pressure sensors, tactile sensors and temperature sensors, which can comprehensively collect grip strength, contact stability and physiological indicators, providing a data basis for dynamically adjusting training intensity. The independent adjustment function of the damping of the first adjustable chain 321 and the damping of the second adjustable chain 323 supports differentiated training of the bilateral upper limbs, which helps to correct the problem of strength imbalance. This embodiment not only ensures the flexibility of training movements, but also improves the level of refinement of rehabilitation assessment, meeting the personalized needs from basic grip to advanced complex movements.

[0206] In some embodiments, the multi-axis auxiliary rod group 42 includes two first auxiliary rods 421, two second auxiliary rods 422, and two third auxiliary rods 423. The two first auxiliary rods 421 are respectively arranged at both ends of the third support rod group 14. A first binding belt 424 is provided between the two first auxiliary rods 421. The first binding belt 424 is used to bind the user's hips; the two second auxiliary rods 422 are transmission-connected to the first auxiliary rods 421. A second binding belt is provided on each of the two second auxiliary rods 422. The second binding belt is used to bind the user's thighs; the two third auxiliary rods 42 3 is connected to the second auxiliary rod 422 by transmission, and the two third auxiliary rods 423 are respectively provided with a third binding belt, which is used to bind the user's calf; the second adjustment group includes two third adjustment drive units, two fourth adjustment drive units, and two fifth adjustment drive units. The two third adjustment drive units are arranged at both ends of the third support rod group 14, and each third adjustment drive unit is connected to a first auxiliary rod 421 by transmission, and the third adjustment drive unit is used to adjust the telescopic length and telescopic damping of the first auxiliary rod 421; the two fourth adjustment drive units are arranged At one end of the second auxiliary rod 422, each fourth adjustment drive unit is transmission-connected to the other end of a second auxiliary rod 422, and the fourth adjustment drive unit is used to adjust the telescopic length and telescopic damping of the second auxiliary rod 422; two fifth adjustment drive units are arranged at one end of the third auxiliary rod 423, and each fifth adjustment drive unit is transmission-connected to the other end of a third auxiliary rod 423, and the fifth adjustment drive unit is used to adjust the telescopic length and telescopic damping of the third auxiliary rod 423; the second acquisition group includes two first pressure-sensitive sensors, two second pressure-sensitive sensors, two third pressure-sensitive sensors, two first inertial measurement units, two second inertial measurement units, and two third inertial measurement units. The two first pressure-sensitive sensors are respectively arranged on the two first straps 424; the two second pressure-sensitive sensors are respectively arranged on the two second straps; the two third pressure-sensitive sensors are respectively arranged on the two third straps; the two first inertial measurement units are respectively arranged on the two first straps 424; the two second inertial measurement units are respectively arranged on the two second straps; and the two third inertial measurement units are respectively arranged on the two third straps.

[0207] In this embodiment, the first auxiliary rod 421, the second auxiliary rod 422 and the third auxiliary rod 423 are all retractable metal rods. The first auxiliary rod 421 is fixed to the hip by a first binding belt 424, the second auxiliary rod 422 is fixed to the thigh by a second binding belt, and the third auxiliary rod 423 is fixed to the calf by a third binding belt.

[0208] The length and damping of the first auxiliary rod 421 are adjusted by the third adjustment drive unit to adapt to the hip motion range of different users; the second auxiliary rod 422 is controlled by the fourth adjustment drive unit, and the third auxiliary rod 423 is controlled by the fifth adjustment drive unit to achieve step-by-step adaptation of lower limb joint motion assistance.

[0209] The first, second, and third straps 424 have built-in pressure-sensitive sensors to monitor pressure distribution within the strapping area to prevent excessive localized compression. A first inertial measurement unit (IMU) is located on the first strap 424, a second IMU on the second strap, and a third IMU on the third strap. These units collect posture, velocity, and angle data of lower limb movement using accelerometers and gyroscopes. The third, fourth, and fifth adjustment drive units dynamically adjust the extension and retraction resistance of the auxiliary rod based on sensor feedback. For example, they reduce damping during knee flexion to reduce the load, or increase resistance during extension to enhance muscle strength training.

[0210] This embodiment realizes segmented and precise assistance of the hip, knee and ankle joints of the lower limbs through the combined design of a multi-axis auxiliary rod and a first binding belt 424, a second binding belt and a third binding belt. The pressure-sensitive sensor and the inertial measurement unit work together to synchronously monitor the contact pressure and movement trajectory of the limbs, providing multi-dimensional data support for the dynamic adjustment of the auxiliary force. The third adjustment drive unit, the fourth adjustment drive unit and the fifth adjustment drive unit independently control the telescopic parameters of the corresponding auxiliary rod according to real-time data, which can not only adapt to the strength requirements of different rehabilitation stages, but also avoid sports injuries. Through segmented binding and flexible adjustment, this embodiment improves the coordination and safety of lower limb training while ensuring the freedom of joint movement, and provides reliable support for the progressive rehabilitation of complex movements.

[0211] In a third aspect, this embodiment further provides a computer-readable storage medium storing computer program instructions, which implement the method described in the first aspect when executed by a processor.

[0212] The computer program involved in this embodiment can be stored in a computer device readable storage medium, which includes but is not limited to a disk, a magnetic tape, a magnetic card, a floppy disk, a flash memory, an optical disc, an optical card, a read-only memory (ROM), a random access memory (RAM), an erasable programmable ROM (EPROM) and an electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical or chemical structures that can achieve similar or equivalent functions to the storage media listed above, such as DNA, RNA, proteins and other units with information storage capabilities. In a specific embodiment, the storage medium involved can be one of the above-mentioned media types or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiment can be stored in a single medium in a centralized manner or in a distributed manner in multiple media. The memory containing the computer device readable storage medium can be a non-volatile memory or a random access memory. These computer device readable storage media can be built into the device or can be connected to the device involved in the embodiment as an external device or part of an external device. In some embodiments, a memory having a computer-readable storage medium is deployed locally; in other embodiments, a solution of deploying the memory away from the processor may also be adopted, such as a network-attached memory accessed via an RF circuit or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), etc., or a suitable combination thereof, as long as the computer device can access the memory. In addition, the computer program involved in the embodiment can be stored in plaintext / ciphertext form, or can be designed as training data, which can be integrated and reorganized through model training and implicitly stored in the parameter state of a deep neural network or other machine learning model.

[0213] See also Figure 5 In a fourth aspect, this embodiment further provides an electronic device 6, comprising a memory 61 and a processor 62, wherein the memory 61 is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor 62 to implement the method described in the first aspect.

[0214] The processor described in this embodiment can be implemented by hardware, firmware, software or a combination thereof, and can use a circuit, a single or multiple application-specific integrated circuits (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, or at least one of a microprocessor. It also includes other physical, biological or chemical structures that can achieve similar or equivalent functions to the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in each embodiment of the present application, or any combination of the steps mentioned therein.

[0215] Different from the existing technology, the above technical solution has the following beneficial effects:

[0216] The present invention provides a medical rehabilitation training robot and its control method, medium and equipment. The control method is based on a dynamic strategy generation mechanism, constructs a user training profile through a neural network model, combines piecewise linear functions with thresholds to determine the training level of the quantified stage, uses feature fusion and cosine similarity analysis to locate training defect parameters, and dynamically adjusts the execution unit parameters to form an "evaluation-adjustment-advancement" closed loop. By adapting to individual differences through a preset rule library and group data, the gradual and safe training intensity is guaranteed, and the training progresses step by step from gross movement coordination to fine control and muscle strengthening, thereby improving the targeting of training. The first training component of the medical rehabilitation training robot implements visual guidance, the second training component implements adjustable grip resistance, and the third training component implements multi-axis lower limb assistance, supporting upper and lower limb coordinated training and multimodal feedback. The pressure-sensitive floor mat, inertial measurement unit and visual recognition unit form a multi-source perception network to collect movement, strength and physiological data in real time. The above technical solution achieves precise adaptation of the entire rehabilitation training cycle through the collaboration of modular hardware architecture and intelligent control methods.

[0217] Finally, it should be noted that although the above embodiments have been described in the specification and drawings of this application, this does not limit the scope of patent protection of this application. All technical solutions generated by replacing or modifying equivalent structures or equivalent processes based on the essential concepts of this application using the contents recorded in the specification and drawings of this application, as well as directly or indirectly implementing the technical solutions of the above embodiments in other related technical fields, are included in the scope of patent protection of this application.

Claims

1. A control method for a medical rehabilitation training robot, characterized in that: include: Obtain user information, initialize the robot's initial posture based on the user information, and generate an initial training strategy. The initial training strategy includes training categories, training content, and multiple stages of training levels. Get the robot's wearing status. After the wearing status meets the wearing conditions, execute the initial training strategy for training and perform the following steps: Generate feedback information collection mode based on training category, training content and preset target requirements to obtain user feedback information; The training parameters of the initial training strategy are updated according to user feedback information to obtain a dynamic adjustment strategy, including: Input user feedback information into the neural network model to obtain user training profiles; According to the training level of the user training profile generation stage; Generate a stage prediction target based on the stage training level and the preset training target, calculate the deviation between the stage training level and the stage prediction target, and generate improvement information; Reverse-map the improvement information to the user training profile to obtain the training defect parameters; Convert the training defect parameters into the training parameters of the current training content, and adjust the robot's execution unit according to the training parameters; Determine whether the improvement information exceeds a preset improvement threshold. If so, generate a stage correction weight for the stage prediction target based on the stage training level and update the stage prediction target. Repeat the above steps until the stage training level and the stage prediction target are within the preset target range, generate a dynamic adjustment strategy for the current stage training level and record it; Obtain the current user status in real time and determine whether the current user status meets the preset user status for the next stage of training level; If yes, proceed to the next training level in the initial training strategy; If not, suspend training.

2. The control method of the medical rehabilitation training robot according to claim 1, characterized in that: Generate feedback information collection mode based on training category and content and preset target requirements, and obtain user feedback information including: Generate basic parameters based on user information, including age, medical history, and recovery status; Generating the preset target requirement according to the user information and the training content includes: Constructing a training content-training target mapping relationship table, and generating target mapping weights of multiple training targets based on the basic parameters; adjusting a lower target threshold of a training target according to the target mapping weight; and, matching similar user groups in a database according to the basic parameters, obtaining an average of upper target thresholds of the similar user groups, and updating the average as the upper target threshold of the current training target; Generating a training target requirement by using a lower target threshold and an upper target threshold of each training target; The plurality of training target requirements are arranged into the preset target requirements according to the training sequence of the training contents.

3. The control method of the medical rehabilitation training robot according to claim 1, characterized in that: Generate feedback information collection mode based on training category and content and preset target requirements, and obtain user feedback information including: Matching a preset sensor collection group according to the training category and training content, and generating a preset collection frequency according to the user information, the preset sensor collection group including a leading sensor and an auxiliary sensor; Acquire original feedback information of the preset sensor acquisition group according to a preset acquisition frequency, wherein the original feedback information includes dominant feedback information and auxiliary feedback information; The original feedback information is time-aligned according to the spatiotemporal order of the dominant feedback information, and feature-fused on the time-aligned original feedback information to obtain the user feedback information, including: Performing sliding window segmentation on the original feedback information after time sequence alignment, where the window length of the sliding window is configured to be determined according to the training content; Extracting time-frequency domain joint features within each sliding window, wherein the time-frequency domain joint features include mean, variance, zero-crossing rate, wavelet packet energy ratio, and coupling coefficient between dominant feedback information and auxiliary feedback information; Calculate the spatiotemporal consistency indexes of the multiple time-frequency domain joint features according to the training type, and generate error correction weights according to the spatiotemporal consistency indexes: and inputting the time-frequency domain joint features into a feature importance weighting model to obtain an importance weight value; The time-frequency domain joint feature is converted into a user feedback information vector with unified dimension according to the error correction weight and the importance weight value, which is the user feedback information.

4. The control method of the medical rehabilitation training robot according to claim 3, characterized in that: Input user feedback information into the neural network model to obtain the user training profile including: Construct a cascade neural network model, including a feature encoding module, a spatiotemporal association module, and a portrait generation module; The user feedback information vector is input into the feature encoding module for feature extraction to obtain encoding features, including: A one-dimensional convolution layer is used to extract local features in the time domain, and the width of the convolution kernel is proportional to the length of the sliding window; Capturing temporal dependency features through a long short-term memory network layer, where the number of hidden layer nodes is set to a multiple of the dimension of the user feedback information vector; The encoded features are input into the spatiotemporal correlation module for cross-modal fusion to obtain fused features, including: Calculate the attention weights of the dominant feedback features and the auxiliary feedback features: calibrating the attention weights according to the error correction weights; Fusing the encoded features according to the calibrated attention weights to obtain fused features; Input the fused features into the portrait generation module and output the user training portrait, including: Mapping to a high-dimensional feature space through a fully connected layer, where the dimension of the high-dimensional feature space is a multiple of the dimension of the user feedback information vector; A softmax function is used to generate a user training profile, which includes the ability values of multiple training parts of the user.

5. The control method of the medical rehabilitation training robot according to claim 4, characterized in that: Generating the current user training level based on the user training profile includes: Input the ability value in the user training profile into a piecewise linear function and map it to a stage training level; Calculate the stage training level based on the deviation between the level label and the ability value of the stage training level; Reverse mapping the improvement information to the user training profile, the training defect parameters obtained include: Obtaining level coefficients of multiple training parts in the training level of the stage; Determining one by one whether the horizontal coefficients are within a preset training threshold range, and if not, recording the training parts associated with the horizontal coefficients as defective parts; Extract the coding features and fusion features corresponding to the defect parts and construct the defect feature vector; Calculate the cosine similarity between the defect feature vector and the standard feature vector of the same training part in a similar user group to generate the defect degree parameter; The training defect parameters are generated according to the defect location, defect feature vector and defect degree parameter.

6. A medical rehabilitation training robot, characterized in that: The method according to any one of claims 1 to 5, wherein the robot comprises: The support assembly includes a support body, a first support rod group, a second support rod group, and a third support rod group, wherein the support body is arranged in a vertical direction, and the first support rod group, the second support rod group, and the third support rod group are arranged on the support body in order from top to bottom; a first training assembly, disposed on the first support rod assembly, comprising a laser projector, a projector, and a camera, the first training assembly being disposed toward a preset training area, the laser projector and the projector being configured to project instruction information toward the preset training area, and the camera being configured to capture first movement information of the user; a second training assembly, disposed on the second support rod assembly, comprising a gripping armrest, a first adjustment group, and a first acquisition group, wherein the gripping armrest is for a user to grip, the first adjustment group is for adjusting an analog value of the gripping armrest, and the first acquisition group is for acquiring gripping feedback information of the user; A third training component includes a pressure-sensitive floor mat, a multi-axis auxiliary rod group, a second adjustment group, and a second acquisition group, wherein the pressure-sensitive floor mat is arranged at the bottom of the support body, the multi-axis auxiliary rod group, the second adjustment group, and the second acquisition group are all arranged on the third support rod group, the multi-axis auxiliary rod group is used to be tied to the user's lower limbs and assist the user in lower limb training, the second adjustment group is arranged on the multi-axis auxiliary rod group, and the second acquisition group is arranged on the multi-axis auxiliary rod group; The control component includes a display unit, a control unit, a voice broadcast unit and a visual recognition unit. The control component is arranged on the supporting body. The display unit is used to display training information. The control unit is electrically connected to the display unit, the voice broadcast unit, the visual recognition unit, the first training component, the second training component and the third training component respectively. The visual recognition unit is used to collect the user's second action information.

7. The medical rehabilitation training robot according to claim 6, characterized in that: The second support rod group includes: a first upper limb support rod, provided on the support body; A second upper limb support rod is provided on the support body, and the second upper limb support rod is arranged opposite to the first upper limb support rod from bottom to top; The first adjustment group includes: Two first adjustable chains are provided at both ends of the first upper limb support rod, and the first adjustable chains are suspended downward; Two first adjustment drive units are provided at both ends of the first upper limb support rod, each of the first adjustment drive units is drivingly connected to one of the first adjustable chains, and the first adjustment drive unit is used to adjust the damping of the first adjustable chain; Two second adjustable chains are provided at both ends of the second upper limb support rod, and the gripping armrest is provided between the first adjustable chain and the second adjustable chain; Two second adjustment drive units are provided at both ends of the second upper limb support rod, each second adjustment drive unit is drivingly connected to one second adjustable chain, and the second adjustment drive unit is used to adjust the damping of the second adjustable chain; The first acquisition group includes: Two pressure sensors are respectively arranged on two gripping armrests; Two tactile sensors are respectively provided on the two gripping armrests; Two temperature sensors are respectively arranged on the two gripping armrests.

8. The medical rehabilitation training robot according to claim 6, characterized in that: The multi-axis auxiliary rod group includes: Two first auxiliary rods are respectively provided at two ends of the third support rod group, and a first binding belt is provided between the two first auxiliary rods, wherein the first binding belt is used to bind the user's hips; Two second auxiliary rods are transmission-connected to the first auxiliary rod, and each of the two second auxiliary rods is provided with a second binding belt, and the second binding belt is used to bind the user's thigh; Two third auxiliary rods are transmission-connected to the second auxiliary rod, and each of the two third auxiliary rods is provided with a third binding belt, and the third binding belt is used to bind the user's calf; The second adjustment group includes: Two third adjustment drive units are provided at both ends of the third support rod group, each of the third adjustment drive units is in transmission connection with one of the first auxiliary rods, and the third adjustment drive units are used to adjust the telescopic length and telescopic damping of the first auxiliary rod; Two fourth adjustment drive units are provided at one end of the second auxiliary rod, each of the fourth adjustment drive units is in transmission connection with the other end of one of the second auxiliary rods, and the fourth adjustment drive units are used to adjust the telescopic length and telescopic damping of the second auxiliary rod; Two fifth adjustment drive units are provided at one end of the third auxiliary rod, each of the fifth adjustment drive units is in transmission connection with the other end of one of the third auxiliary rods, and the fifth adjustment drive units are used to adjust the telescopic length and telescopic damping of the third auxiliary rod; The second acquisition group includes: Two first pressure-sensitive sensors are respectively arranged on the two first binding belts; Two second pressure-sensitive sensors are respectively provided on the two second binding belts; Two third pressure-sensitive sensors are respectively provided on the two third binding belts; Two first inertial measurement units, respectively provided on the two first straps; two second inertial measurement units, respectively provided on the two second straps; The two third inertial measurement units are respectively arranged on the two third straps.

9. A computer-readable storage medium storing computer program instructions, characterized in that: The computer program instructions implement the method according to any one of claims 1 to 5 when executed by a processor.

10. An electronic device comprising a memory and a processor, characterized in that: The memory is configured to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method according to any one of claims 1 to 5.

Images