A gait rehabilitation training device and apparatus

Abstract

This application discloses a walking rehabilitation training device and apparatus, relating to the field of medical rehabilitation equipment technology. A sensor module is used to collect gait data during the user's training phase. A controller reads the gait data collected by the sensor module according to a set cycle time and constructs a feature vector based on the gait data at the current cycle time. Based on the similarity between the feature vector and the weights of all categories, the category to which the feature vector belongs is determined, and the status label of the feature vector is output. A status action value table is queried to determine the intervention action matching the status label, and the auxiliary torque corresponding to the intervention action is output to the skeletal mechanical body. Differentiated torque intervention can be implemented according to the user's real-time status, improving training comfort and training effect. After completing the intervention action, the action value of the intervention action in the status action value table is adjusted based on the latest gait data fed back by the sensor module, so that the action value can adaptively optimize according to the user's physical condition and rehabilitation progress.

Description

Technical Field

[0001] This application relates to the field of medical rehabilitation equipment technology, and in particular to a walking rehabilitation training device and apparatus. Background Technology

[0002] Gait rehabilitation trainers are important devices used to assist users with lower limb motor dysfunction caused by spinal cord injury, stroke, or other reasons in their rehabilitation training. Current gait rehabilitation trainers typically employ fixed gait trajectories or simple impedance control modes.

[0003] Traditional gait rehabilitation methods typically use standardized gait data as control targets, ignoring the significant differences in plantar pressure among different users (e.g., varying height, weight, and spasticity levels), leading to poor training comfort and effectiveness. Furthermore, gait phase recognition (e.g., standing phase, swing phase) is usually based on fixed thresholds or temporal logic, resulting in rigid state recognition and an inability to provide adaptive rehabilitation training based on the user's evolving state. When insufficient muscle strength or gait abnormalities are detected, traditional gait rehabilitation trainers often only provide a preset, fixed assist torque (e.g., a uniform 10% increase in assistance). This intervention method cannot optimize for the user's needs at different times within the same state, resulting in poor training outcomes.

[0004] It is evident that improving training comfort and effectiveness is a problem that needs to be addressed by those skilled in the art. Summary of the Invention

[0005] The purpose of this application is to provide a walking rehabilitation training device and apparatus that can improve training comfort and training effectiveness.

[0006] This application provides a walking rehabilitation training device, including an exoskeleton mechanical body, a sensor module, and a controller; wherein the controller is connected to both the exoskeleton mechanical body and the sensor module; Sensor modules are used to collect gait data during the user's training phase; The controller reads gait data collected by the sensor module at set intervals and constructs a feature vector based on the gait data at the current interval. It then determines the category of the feature vector based on its similarity to the weights of all categories and outputs the feature vector's state label. Next, it queries the state-action value table to determine the intervention action matching the state label and outputs the corresponding auxiliary torque to the skeletal mechanical body. After completing the intervention action, it acquires the latest gait data from the sensor module and adjusts the action value of the intervention action in the state-action value table based on the latest gait data.

[0007] On one hand, the controller is used to compare the similarity of the feature vector with the weights of all categories in the category library when the feature vector does not belong to the first feature vector; and when the similarity between the feature vector and the target category weights meets the set similarity conditions, the feature vector is assigned to the target category and the category weights of the target category are updated. If the similarity between the feature vector and the weights of all categories in the category library does not meet the set similarity conditions, or if the feature vector belongs to the first feature vector, create the first category of the feature vector in the category library and use the feature vector as the category weight of the first category.

[0008] On one hand, the controller compares the feature vector with any class weight in the class library element by element, selects the minimum value of each element to form a similarity vector; and divides the sum of the elements of the similarity vector with the sum of the elements of the feature vector to obtain the similarity between the feature vector and any class weight in the class library.

[0009] On one hand, the controller is used to query all auxiliary actions that match the state label from the state action value table, select the auxiliary action with the highest action value from all auxiliary actions as the intervention action, and use the product of the percentage corresponding to the intervention action and the maximum torque of the joint corresponding to the intervention action as the auxiliary torque.

[0010] On one hand, the controller is used to determine the gait symmetry score based on the contact time of the plantar pressure of both feet after the intervention action is completed; to determine the reward value based on the current human-machine interaction torque, the historical maximum torque and the gait symmetry score; and to adjust the action value of the intervention action in the state action value table according to the set learning rate and reward value.

[0011] On the one hand, the controller is used to randomly select an auxiliary action as an intervention action from all auxiliary actions that match the state label in the state action value table, provided that the random selection probability is satisfied, after outputting the state label of the feature vector.

[0012] On the one hand, the sensor module is used to collect static data of the user when the user is stationary; and to use the static data as reference data; and to collect the user's gait data when the exoskeleton mechanical body moves the user to flexion and extension at a set speed; wherein, the gait data includes joint angles and plantar pressure values.

[0013] On one hand, the controller is used to determine the limit value of each type of data based on the baseline data and gait data. If each type of data exceeds its corresponding limit range in the gait data during the user training phase, then the controller constructs a feature vector based on the current gait data, and performs the operation steps of determining the category to which the feature vector belongs based on the similarity between the feature vector and the weights of all categories, and outputting the status label of the feature vector.

[0014] On one hand, the controller analyzes the baseline data and gait data according to data categories to determine the mean and standard deviation for each data category; based on the mean and standard deviation for each data category, it sets the warning range and limit range for each data category; where the warning range is less than the limit range; when each data category in the gait data during the user training phase exceeds its corresponding warning range, an alarm is issued; when each data category in the gait data during the user training phase exceeds its corresponding limit range, a feature vector is constructed based on the current gait data.

[0015] This application also provides a walking rehabilitation training device, including a construction unit, a category determination unit, a movement determination unit, an acquisition unit, and an adjustment unit; The construction unit is used to read the gait data collected by the sensor module according to the set cycle time, and construct the feature vector based on the gait data at the current cycle time. The category determination unit is used to determine the category to which the feature vector belongs based on the similarity between the feature vector and the weights of all categories, and outputs the status label of the feature vector. The action determination unit is used to query the state action value table, determine the intervention action that matches the state label, and output the auxiliary torque corresponding to the intervention action to the skeletal mechanical body. The acquisition unit is used to acquire the latest gait data fed back by the sensor module after the intervention action is completed; The adjustment unit is used to adjust the action value of intervention actions in the state action value table based on the latest gait data.

[0016] As can be seen from the above technical solution, the walking rehabilitation training device includes an exoskeleton mechanical body, a sensor module, and a controller; the controller is connected to both the exoskeleton mechanical body and the sensor module. The sensor module is used to collect gait data during the user's training phase. The controller reads the gait data collected by the sensor module according to a set cycle time and constructs a feature vector based on the gait data at the current cycle time. This abandons the rigid mode of using uniform standard gait data as the control target in traditional solutions, and can fully adapt to individual differences in plantar pressure among different users, providing reliable data support for improving the fit and comfort of rehabilitation training. The controller determines the category to which the feature vector belongs based on the similarity between the feature vector and the weights of all categories, and outputs the status label of the feature vector; this replaces the traditional gait phase recognition method based on fixed thresholds and temporal logic, breaking the limitations of rigid status recognition, and can perceive the dynamic state changes of the user during training in real time, achieving accurate recognition of gait status. The controller queries the status action value table to determine the intervention action matching the status label and outputs the auxiliary torque corresponding to the intervention action to the exoskeleton mechanical body. This application addresses the limitations of traditional devices that can only provide fixed auxiliary torques in situations of insufficient muscle strength or abnormal gait by selectively retrieving appropriate intervention movements and outputting corresponding auxiliary torques. It enables differentiated torque intervention based on the user's real-time state, effectively adapting to the user's rehabilitation assistance needs at different times within the same state, thus improving training comfort and effectiveness. After completing the intervention movement, the application acquires the latest gait data from the sensor module. Based on this data, it adjusts the movement value of the intervention movement in the state movement value table, allowing the rehabilitation intervention strategy to continuously and adaptively optimize according to the user's physical condition and rehabilitation progress. This avoids the training limitations caused by fixed intervention intensity, improving training quality and the adaptability of rehabilitation training. Attached Figure Description

[0017] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the structure of a walking rehabilitation training device provided in an embodiment of this application; Figure 2 A flowchart illustrating a method for identifying feature vector categories provided in this application embodiment; Figure 3 A flowchart illustrating a method for adjusting a state action value table provided in this application embodiment; Figure 4 This is a schematic diagram of the structure of a walking rehabilitation training device provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0020] The terms "comprising" and "having," and any variations thereof, in the specification and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may include steps or units not listed.

[0021] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] Next, we will describe in detail a walking rehabilitation training device provided in the embodiments of this application. Figure 1 The present application provides a schematic diagram of the structure of a walking rehabilitation training device, which includes an exoskeleton mechanical body 11, a sensor module 12 and a controller 13; wherein the controller 13 is connected to the exoskeleton mechanical body and the sensor module 12 respectively.

[0023] There are various types of walking rehabilitation training equipment, such as basic walking aids, weight-reducing gait training equipment, robot-assisted equipment, end-effector trajectory-based equipment, and simple functional training equipment.

[0024] The exoskeleton mechanical body 11 may include a hip joint drive module, a knee joint drive module, an ankle joint drive module, and a lower limb linkage.

[0025] Sensor module 12 is used to collect gait data during the user's training phase. Gait data may include joint angles and plantar pressure values.

[0026] The sensor module 12 may include an angle encoder integrated at the joint, a thin-film pressure sensor integrated at the strap, and an array of plantar pressure sensors integrated at the sole of the shoe. The thin-film pressure sensor can be used to detect human-machine interaction torque.

[0027] The controller 13 is used to read the gait data collected by the sensor module 12 according to the set cycle time, and construct a feature vector based on the gait data at the current cycle time.

[0028] The specific value of the cycle time can be flexibly set according to actual needs. When the control accuracy requirement is relatively high, the cycle time can be set shorter; when the control accuracy requirement is not high, the cycle time can be set longer. In practical applications, the cycle time can be set to 10ms, and the controller 13 reads gait data once every 10ms, constructing a multi-dimensional feature vector from the read gait data.

[0029] For example, the multidimensional feature vector is: [left hip angle, right hip angle, left ankle angle, right ankle angle, left knee angle, right knee angle, left foot pressure peak, right foot plantar pressure peak].

[0030] Since the angle and pressure have different dimensions, they need to be normalized to the range [0, 1] through preprocessing to obtain the normalized feature vector, denoted as X(t).

[0031] Controller 13 can embed an adaptive resonance theory network chip or algorithm to detect new user states in real time.

[0032] In a specific implementation, the controller 13 can determine the category to which the feature vector belongs based on the similarity between the feature vector and the weights of all categories, and output the status label of the feature vector.

[0033] For each new category detected, its feature vector can be used as its category weight. To facilitate the management of existing categories, a category library can be built, and the category weights of existing categories can be recorded in the category library.

[0034] After identifying the user's current state, controller 13 needs to determine the appropriate force. Controller 13 can maintain a dynamic two-dimensional table in memory, namely the State-Action Value Table (Q-table). The rows of the State-Action Value Table are the state labels (State_IDs) discovered by the ART network, and the columns are all possible auxiliary actions.

[0035] Each auxiliary action has its corresponding action value. The higher the action value, the more it meets the needs of the current state.

[0036] The controller 13 queries the status action value table, determines the intervention action that matches the status label, and outputs the auxiliary torque corresponding to the intervention action to the skeletal mechanical body.

[0037] An intervention action can be the auxiliary action with the highest action value among all auxiliary actions corresponding to the current state.

[0038] Different auxiliary torques represent different auxiliary actions. In the state-action value table, a specific auxiliary torque can be set as an auxiliary action for each state.

[0039] Considering that the maximum torque varies among different users and at different joints, to ensure that the torque of the auxiliary action better reflects individual user differences, a percentage can be set for each state in the state-action value table instead of a specific auxiliary torque. One percentage represents one auxiliary action. For example, action set A = {0%, 10%, 20%, 30%}.

[0040] Taking the auxiliary action as a percentage as an example, in the specific implementation, the controller 13 can query all auxiliary actions that match the state label from the state action value table, select the auxiliary action with the highest action value from all auxiliary actions as the intervention action, and use the product of the percentage corresponding to the intervention action and the maximum torque of the joint corresponding to the intervention action as the auxiliary torque.

[0041] Different joints have their own corresponding maximum joint torque. In practical applications, the maximum joint torque corresponding to different joints can be obtained in advance through testing before the user uses the walking rehabilitation training equipment for formal training, and recorded in the controller 13.

[0042] After the intervention action is completed, in order to understand the effect of the intervention action, the controller 13 can obtain the latest gait data fed back by the sensor module 12; based on the latest gait data, the action value of the intervention action in the state action value table is adjusted.

[0043] For example, when controller 13 determines that the current state is State_ID=k, it looks at the state k row in the Q table, selects the auxiliary action with the highest current action value, and executes the auxiliary torque corresponding to the auxiliary action.

[0044] In this embodiment of the application, in order to better understand the user's physical condition, the user's baseline data can be obtained before the user performs formal training.

[0045] In a practical implementation, the sensor module 12 can collect static data of the user when the user is stationary, and use the static data as reference data.

[0046] The static state can be when the user is fully relaxed after wearing the device, maintaining a standing or sitting posture. At this time, the data collected by the sensor module 12 can define the "zero torque point". For example, the value collected by the plantar pressure sensor at this moment can be used as the user's "weight baseline", and the subsequent real-time pressure can be compared with it to calculate the "center of gravity shift".

[0047] In addition to collecting static data, sensor module 12 can also collect dynamic data.

[0048] In a specific implementation, when the exoskeleton mechanical body 11 drives the user to perform flexion and extension movements at a set speed, the sensor module 12 can collect gait data of the user during the testing phase; the gait data during the testing phase may include joint angles and plantar pressure values.

[0049] The speed can be set to an extremely low speed, such as 0.05 m / s.

[0050] By guiding the user's lower limbs through a complete flexion and extension movement, the values ​​of the angle encoder at the joint are recorded. This differs from standard gait; it records the mechanical resistance and range of motion of the joint under passive range of motion.

[0051] The controller 13 can determine the limit range for each type of data based on the baseline data and the gait data during the testing phase. In addition to periodically identifying the user's state, the controller 13 can also construct a feature vector based on the current gait data when each type of data in the user's training phase exceeds its corresponding limit range. It then performs the following steps: based on the similarity between the feature vector and the weights of all categories, it determines the category to which the feature vector belongs and outputs the status label of the feature vector. This allows the controller to adjust the applied auxiliary torque in a timely manner, helping the user to adjust to a comfortable training mode.

[0052] The determination of the range of limits for each type of data can rely on the mean and standard deviation of each type of data.

[0053] In a specific implementation, the controller 13 can analyze the baseline data and gait data during the testing phase according to data categories, determine the mean and standard deviation corresponding to each data category, and set the warning range and limit range for each data category based on the mean and standard deviation corresponding to each data category; wherein, the warning range is less than the limit range.

[0054] An alert is issued when each type of gait data exceeds its corresponding warning range during the user training phase.

[0055] If any type of gait data exceeds its corresponding limit during the user training phase, a feature vector is constructed based on the current gait data to re-identify the user's state and adjust the applied auxiliary torque in a timely manner.

[0056] For each type of data (left hip angle, right hip angle, left ankle angle, right ankle angle, left knee angle, right knee angle, left foot pressure peak, right foot plantar pressure peak), its mean (μ) and standard deviation (σ) can be calculated.

[0057] The user-specific warning range can be set to μ±2σ; the limit range can be set to μ±3σ. Subsequently, whenever the real-time data from the sensor exceeds the individualized limit range, the system knows that the user's actions have deviated from their personal baseline, indicating a possible need or an anomaly.

[0058] The above description uses the example of selecting the auxiliary action with the highest action value for intervention. In this embodiment, to prevent the algorithm from getting stuck in local optima, such as always only applying 5 Nm (15 Nm would be better, but this has not been tested), the controller 13 can randomly select an auxiliary action to execute with a very small probability, such as ε=0.1, when applying the auxiliary torque.

[0059] In a specific implementation, after outputting the state label of the feature vector, the controller 13 can randomly select one auxiliary action as the intervention action from all auxiliary actions that match the state label in the state action value table, provided that the random selection probability is satisfied.

[0060] As can be seen from the above technical solution, the walking rehabilitation training device includes an exoskeleton mechanical body, a sensor module, and a controller; the controller is connected to both the exoskeleton mechanical body and the sensor module. The sensor module is used to collect gait data during the user's training phase. The controller reads the gait data collected by the sensor module according to a set cycle time and constructs a feature vector based on the gait data at the current cycle time. This abandons the rigid mode of using uniform standard gait data as the control target in traditional solutions, and can fully adapt to individual differences in plantar pressure among different users, providing reliable data support for improving the fit and comfort of rehabilitation training. The controller determines the category to which the feature vector belongs based on the similarity between the feature vector and the weights of all categories, and outputs the status label of the feature vector; this replaces the traditional gait phase recognition method based on fixed thresholds and temporal logic, breaking the limitations of rigid status recognition, and can perceive the dynamic state changes of the user during training in real time, achieving accurate recognition of gait status. The controller queries the status action value table to determine the intervention action matching the status label and outputs the auxiliary torque corresponding to the intervention action to the exoskeleton mechanical body. This application addresses the limitations of traditional devices that can only provide fixed auxiliary torques in situations of insufficient muscle strength or abnormal gait by selectively retrieving appropriate intervention movements and outputting corresponding auxiliary torques. It enables differentiated torque intervention based on the user's real-time state, effectively adapting to the user's rehabilitation assistance needs at different times within the same state, thus improving training comfort and effectiveness. After completing the intervention movement, the application acquires the latest gait data from the sensor module. Based on this data, it adjusts the movement value of the intervention movement in the state movement value table, allowing the rehabilitation intervention strategy to continuously and adaptively optimize according to the user's physical condition and rehabilitation progress. This avoids the training limitations caused by fixed intervention intensity, improving training quality and the adaptability of rehabilitation training.

[0061] Figure 2 A flowchart of a method for identifying feature vector categories provided in this application embodiment, the method including: S201: Determine whether the feature vector belongs to the first feature vector.

[0062] In the embodiments of this application, an Adaptive Resonance Theory (ART) network can be used to identify category matching and clustering.

[0063] When the system is first started, the ART network is empty. At this time, the feature vector belongs to the first feature vector, and the operation of S205 can be executed.

[0064] If a feature vector does not belong to the first feature vector, it means that there is already a category in the category library, and S202 can be executed at this time.

[0065] When the first feature vector X1 arrives, since there is no existing category, the controller can automatically create category 1 and use X1 as the category weight W1 for that category.

[0066] When the second feature vector X2 arrives, calculate the similarity between X2 and all existing categories (currently only category 1). A fuzzy matching function can be used to determine the similarity.

[0067] S202: Compare the similarity between the feature vector and the weights of all categories in the category library.

[0068] In its implementation, the controller compares the feature vector element-by-element with any class weight in the class library, selecting the minimum value among the elements to form a similarity vector. The sum of the elements in the similarity vector is then divided by the sum of the elements in the feature vector to obtain the similarity between the feature vector and any class weight in the class library.

[0069] For example, suppose the feature vector X = [0.8, 0.3, 0.6] and the class weight W1 = [0.7, 0.5, 0.4]. By selecting the minimum value of each element, a similarity vector can be formed, that is, similarity vector = [min(0.8, 0.7), min(0.3, 0.5), min(0.6, 0.4)] = [0.7, 0.3, 0.4].

[0070] Similarity (Match) = |W1| / |X| = (0.7+0.3+0.4) / (0.8+0.3+0.6) = 0.82.

[0071] S203: Determine whether the similarity between the feature vector and the weights of all categories in the category library meets the set similarity conditions.

[0072] In this embodiment of the application, a similarity threshold can be set, and ρ can be used to represent the similarity threshold. If Match≥ρ, it is assigned to an existing category, and if Match≥ρ, it is assigned to a new category.

[0073] In practical applications, ρ can be set to 0.95.

[0074] If the similarity between the feature vector and the target category weights meets the set similarity conditions, it means that the feature vector can be assigned to an existing category, and S204 can be executed. If the similarity between the feature vector and the weights of all categories in the category library does not meet the set similarity conditions, it means that a new category needs to be set for the feature vector, and S205 can be executed.

[0075] S204: Assign the feature vector to the target category and update the category weights of the target category.

[0076] Based on the distance mentioned above, if the similarity (X2, W1) ≥ ρ, it means that X2 is very similar to category 1. In this case, X2 can be classified into category 1, and the category weight W1 of category 1 can be updated.

[0077] S205: Create a first category for the feature vector in the category library and use the feature vector as the category weight of the first category.

[0078] If the similarity (X2, W1) < ρ, it means that X2 is a novel pattern, such as a user suddenly having a spasm or experiencing foot drop. The ART network does not force it to be classified into category 1, but instead creates a new neuron in the network layer, that is, category 2, and uses X2 as the category weight W2 of that category.

[0079] Whenever data X(t) at time t is processed, the ART network outputs a state label (State_ID). For example, State_ID=1 represents "normal support phase", State_ID=2 represents "normal oscillation phase", and State_ID=3 represents "the first appearance of the pre-spasticity state".

[0080] In this embodiment, the state classification is completed by matching the similarity between the feature vector and the weight of each category and the corresponding state label is output. The dynamic comparison mechanism of data similarity replaces the fixed judgment rule to realize the intelligent recognition of gait state. This breaks the traditional rigid state judgment logic and can accurately identify the dynamically changing gait state during the user's training process. It adapts to the fluctuation of the user's body state at different training stages and ensures the real-time performance and accuracy of gait state recognition.

[0081] Figure 3 A flowchart of a method for adjusting a state action value table provided in this application embodiment, the method including: S301: After the intervention action is completed, the gait symmetry score is determined based on the plantar pressure contact time of both feet.

[0082] The more symmetrical the support time of the left and right feet after the intervention is completed, the higher the gait symmetry score.

[0083] According to the plantar pressure contact time (1-(left T-right T) / (left T+right T)), if it is 0, it is completely asymmetrical, and if it is 1, it is completely symmetrical.

[0084] S302: Determine the reward value based on the current human-computer interaction torque, the historical maximum torque, and the gait symmetry score.

[0085] If the force applied is just right, the user feels that it is effortless, and the reading of the human-computer interaction torque is 0 or very small, then the reward is high; if the force applied is too large or too small, it leads to human-computer conflict, the interaction torque is large, and the reward is low.

[0086] After an intervention is performed, the data returned by the sensor module determines whether the action was good or bad.

[0087] In practice, the reward value can be calculated using the following formula: Reward=w1 (1 - |current human-computer interaction torque| / historical maximum torque) + w2 (Gait symmetry score); Here, w1 and w2 are both weights of the reward function, and w1 + w2 = 1.

[0088] In practical applications, we can set w1=0.6 (adversarial priority) and w2=0.4 (gait priority).

[0089] Two weighting coefficients are used to adjust which of the two objectives is more important.

[0090] The term (1-|current human-computer interaction torque| / historical maximum torque) is objective 1, which is to reduce human-computer interaction.

[0091] The historical maximum torque can be selected from the maximum torque of human-computer interaction within the last 30 seconds. The smaller the pulling force between the user and the walking rehabilitation training equipment (current human-computer interaction torque), the higher the score for this item; When the torque is close to 0, it means that the machine assistance is just right, the user exerts no effort, there is no resistance, and the reward is high; the greater the torque, the more unsuitable the assistance is, there is mutual competition, strong human-machine resistance, and the reward becomes lower.

[0092] The gait symmetry score is objective 2. The more symmetrical the support / swing of the left and right legs and feet, the higher the gait score and the greater the reward; limping, large differences in force exertion / time between the left and right sides, result in a lower score and a lower reward.

[0093] S303: Adjust the action value of the intervention action in the state action value table based on the set learning rate and reward value.

[0094] After the device completes an action and receives the reward value R, it will review and learn from the experience, modify the Q table, and make it easier to select better actions next time.

[0095] Update formula: Q(State_k, Action_a) = Q(State_k, Action_a) + α [RQ(State_k, Action_a)]; Where Q(State_k, Action_a) represents the action value corresponding to action a in state k, and α is the learning rate. This means that for state k, if performing action a yields a reward value R, the action value of the corresponding action in the Q table will be updated, and the system will be more inclined to choose action a the next time it encounters state k.

[0096] The above update formula can be simplified to Q new =Q old +α (RQ old ); For example, Q old =0.5, R=0.9, Q new =0.5 + 0.1 × (0.9 - 0.5) = 0.5 + 0.04 = 0.54. The change from 0.5 to 0.54 increases the action value. If R > Q old This indicates that the effect was better than expected, so the value of the action is increased; if R old This indicates that the effect was worse than expected, so the value of the action should be reduced.

[0097] In this embodiment, after each round of intervention is completed, the latest gait data fed back by the sensor is collected in a timely manner. Based on the actual gait performance after intervention, the action value of the corresponding intervention action in the state action value table is adjusted in reverse. This realizes the dynamic iterative optimization of intervention actions and auxiliary torque strategies, allowing the rehabilitation intervention method to continuously and adaptively adjust with the user's rehabilitation progress and improvement of physical function, gradually optimizing the intervention adaptability, strengthening the effects of gait correction and muscle strength training, and realizing personalized, progressive and efficient rehabilitation training.

[0098] Figure 4 ​A schematic diagram of the structure of a walking rehabilitation training device provided in this application embodiment includes a construction unit 41, a category determination unit 42, an action determination unit 43, an acquisition unit 44, and an adjustment unit 45; The construction unit 41 is used to read the gait data collected by the sensor module according to the set cycle time, and construct a feature vector based on the gait data at the current cycle time. The category determination unit 42 is used to determine the category to which the feature vector belongs based on the similarity between the feature vector and all category weights, and output the status label of the feature vector. The action determination unit 43 is used to query the state action value table, determine the intervention action that matches the state label, and output the auxiliary torque corresponding to the intervention action to the skeletal mechanical body. The acquisition unit 44 is used to acquire the latest gait data fed back by the sensor module after the intervention action is completed; Adjustment unit 45 is used to adjust the action value of intervention actions in the state action value table based on the latest gait data.

[0099] Figure 4 For a description of the features in the corresponding embodiments, please refer to Figure 1 The relevant descriptions of the corresponding embodiments will not be repeated here.

[0100] As can be seen from the above technical solution, gait data collected by the sensor module is read according to a set periodic time, and a feature vector is constructed based on the gait data at the current periodic time. This abandons the rigid mode of traditional solutions that use uniform standard gait data as the control target, and can fully adapt to individual differences in plantar pressure among different users, providing reliable data support for improving the fit and comfort of rehabilitation training. Based on the similarity between the feature vector and the weights of all categories, the category to which the feature vector belongs is determined, and the status label of the feature vector is output; this replaces the traditional gait phase recognition method based on fixed thresholds and temporal logic, breaking through the rigidity of status recognition, and can perceive the dynamic state changes of the user during training in real time, achieving accurate recognition of gait status. The status action value table is queried to determine the intervention action matching the status label, and the auxiliary torque corresponding to the intervention action is output to the skeletal mechanical body. This application addresses the limitations of traditional devices that can only provide fixed auxiliary torques in situations of insufficient muscle strength or abnormal gait by selectively retrieving appropriate intervention movements and outputting corresponding auxiliary torques. It enables differentiated torque intervention based on the user's real-time state, effectively adapting to the user's rehabilitation assistance needs at different times within the same state, thus improving training comfort and effectiveness. After completing the intervention movement, the application acquires the latest gait data from the sensor module. Based on this data, it adjusts the movement value of the intervention movement in the state movement value table, allowing the rehabilitation intervention strategy to continuously and adaptively optimize according to the user's physical condition and rehabilitation progress. This avoids the training limitations caused by fixed intervention intensity, improving training quality and the adaptability of rehabilitation training.

[0101] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0102] The foregoing has provided a detailed description of a walking rehabilitation training device and apparatus provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only intended to help understand the method and core ideas of this application. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A walking rehabilitation training device, characterized in that, It includes an exoskeleton mechanical body, a sensor module, and a controller; wherein the controller is connected to both the exoskeleton mechanical body and the sensor module. The sensor module is used to collect gait data during the user's training phase; The controller is configured to read gait data collected by the sensor module at set intervals, construct a feature vector based on the gait data at the current interval, determine the category to which the feature vector belongs based on the similarity between the feature vector and the weights of all categories, and output the status label of the feature vector; query the status action value table, determine the intervention action matching the status label, and output the auxiliary torque corresponding to the intervention action to the skeletal mechanical body; after completing the intervention action, obtain the latest gait data fed back by the sensor module; and adjust the action value of the intervention action in the status action value table based on the latest gait data.

2. The gait rehabilitation training device according to claim 1, characterized in that, The controller is configured to, when the feature vector does not belong to the first feature vector, compare the similarity of the feature vector with all category weights in the category library; and when the similarity between the feature vector and the target category weights meets the set similarity conditions, assign the feature vector to the target category and update the category weights of the target category. If the similarity between the feature vector and the weights of all categories in the category library does not meet the set similarity conditions, or if the feature vector belongs to the first feature vector, a first category for the feature vector is created in the category library, and the feature vector is used as the category weight of the first category.

3. The gait rehabilitation training device according to claim 2, characterized in that, The controller is configured to compare the feature vector with any category weight in the category library element by element, select the minimum value of each element to form a similarity vector; and perform a division operation between the sum of the elements of the similarity vector and the sum of the elements of the feature vector to obtain the similarity between the feature vector and any category weight in the category library.

4. The gait rehabilitation training device according to claim 1, characterized in that, The controller is configured to query all auxiliary actions that match the state label from the state action value table, select the auxiliary action with the highest action value from all auxiliary actions as the intervention action, and use the product of the percentage corresponding to the intervention action and the maximum torque of the joint corresponding to the intervention action as the auxiliary torque.

5. The gait rehabilitation training device according to claim 1, characterized in that, The controller is configured to, after completing the intervention action, determine a gait symmetry score based on the time of contact with the ground by the plantar pressure of both feet; determine a reward value based on the current human-machine interaction torque, the historical maximum torque, and the gait symmetry score; and adjust the action value of the intervention action in the state action value table according to the set learning rate and the reward value.

6. The walking rehabilitation training device according to claim 1, characterized in that, The controller is configured to, after outputting the state label of the feature vector, randomly select one auxiliary action as the intervention action from all auxiliary actions that match the state label in the state action value table, provided that the random selection probability is satisfied.

7. The gait rehabilitation training device according to claim 1, characterized in that, The sensor module is used to collect static data of the user when the user is stationary. The static data is used as the baseline data; while the exoskeleton mechanical body drives the user to perform flexion and extension movements at a set speed, the user's gait data is collected; wherein, the gait data includes joint angles and plantar pressure values.

8. The gait rehabilitation training device according to claim 7, characterized in that, The controller is configured to determine the limit value for each type of data based on the baseline data and the gait data; if each type of data exceeds its corresponding limit range in the gait data during the user training phase, then a feature vector is constructed based on the current gait data, and the operation steps of determining the category to which the feature vector belongs based on the similarity between the feature vector and the weights of all categories, and outputting the status label of the feature vector are performed.

9. The gait rehabilitation training device according to claim 8, characterized in that, The controller is configured to analyze the baseline data and the gait data according to data categories, determine the mean and standard deviation corresponding to each data category, and set a warning range and a limit range for each data category based on the mean and standard deviation of each data category, wherein the warning range is smaller than the limit range; when each data category in the gait data during the user training phase exceeds its corresponding warning range, an alarm is issued; when each data category in the gait data during the user training phase exceeds its corresponding limit range, a feature vector is constructed based on the current gait data.

10. A walking rehabilitation training device, characterized in that, It includes a construction unit, a category determination unit, an action determination unit, an acquisition unit, and an adjustment unit; The construction unit is used to read gait data collected by the sensor module according to a set cycle time, and construct a feature vector based on the gait data at the current cycle time. The category determination unit is used to determine the category to which the feature vector belongs based on the similarity between the feature vector and all category weights, and output the status label of the feature vector; The action determination unit is used to query the state action value table, determine the intervention action that matches the state label, and output the auxiliary torque corresponding to the intervention action to the skeletal mechanical body. The acquisition unit is used to acquire the latest gait data fed back by the sensor module after the intervention action is completed; The adjustment unit is used to adjust the action value of the intervention action in the state action value table based on the latest gait data.

Images