A rehabilitation training system based on multi-modal perception and adaptive assistance

Abstract

This invention provides a rehabilitation training system based on multimodal perception and adaptive assistance, relating to the field of rehabilitation medicine technology. The rehabilitation training system includes: a wearable sensing module, a main control module, and an auxiliary execution device. When a user performs preset rehabilitation movements, the wearable sensing module collects active motion signals from the target area. These active motion signals include physiological electrical signals and motion state signals. Based on the active motion signals, the main control module assesses the degree of completion of the preset rehabilitation movements performed by the user and generates auxiliary motion signals based on this information. The auxiliary execution device applies auxiliary movements to the target area based on the auxiliary motion signals to assist the target area in completing the preset rehabilitation movements. This invention combines multimodal signal perception of the user's target area to obtain the user's movement intention and execution state, and adaptively assists the user in completing rehabilitation movements as accurately as possible.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation medicine technology, specifically to a rehabilitation training system based on multimodal perception and adaptive assistance. Background Technology

[0002] Rehabilitation medicine is an emerging discipline that aims to eliminate and alleviate functional impairments, compensate for and rebuild functional deficiencies, and improve and enhance various aspects of human function. In other words, it encompasses the prevention, diagnosis, assessment, treatment, training, and management of functional disorders. Exercise therapy, occupational therapy, and speech therapy are important components and methods of modern rehabilitation medicine.

[0003] In the fields of neurorehabilitation and orthopedic rehabilitation, although stroke, spinal cord injury or postoperative patients can recover at least some motor function through high-intensity repetitive training, existing rehabilitation methods have at least the following obvious limitations: large rehabilitation robots are expensive, require fixed locations and have poor flexibility; traditional home devices have limited functions, cannot provide personalized assistance based on the patient's real-time status and lack quantitative feedback; manual training relying on therapists suffers from low efficiency, difficulty in standardization and resource scarcity; current assistive devices usually use fixed assistive movements, resulting in a lack of precision and adaptability in assistance. Summary of the Invention

[0004] The purpose of this invention is to provide a rehabilitation training system based on multimodal perception and adaptive assistance. By combining multimodal signal perception of the user's target body parts, the system can obtain the user's movement intention and execution status of the target body parts, and adaptively assist the user to complete rehabilitation movements as accurately as possible. This system can provide personalized adaptive rehabilitation assistance for different users, and can meet the user's rehabilitation training needs more efficiently, safely and with high compliance. It is suitable for rehabilitation training in non-professional environments such as home and community.

[0005] To achieve the above objectives, the present invention provides a rehabilitation training system based on multimodal perception and adaptive assistance, comprising: Wearable sensing module, main control module, and auxiliary execution device; the wearable sensing module is worn by the user on the target part of the body; The wearable sensing module is used to collect active motion signals of the target area when the user performs preset rehabilitation actions. The active motion signals include: physiological electrical signals and motion state signals. The main control module is used to evaluate the degree of completion of the user's preset rehabilitation actions based on the active action signals; and to generate auxiliary action signals to assist the user in completing the preset rehabilitation actions based on the degree of completion information. The auxiliary execution device is used to apply auxiliary actions to the target area based on the auxiliary action signal, so as to assist the target area in completing the preset rehabilitation action.

[0006] In one embodiment, the wearable sensing module includes: an electromyography (EMG) sensor and an inertial measurement unit, wherein the EMG sensor is disposed at a designated muscle group in the target area and the inertial measurement unit is fixed to the target area; The electromyography sensor is used to collect electromyographic signals of a specified muscle group at the target site, and the physiological electrical signals include the electromyographic signals. The inertial measurement unit collects the acceleration and angular velocity of the target part, and the motion state signal includes: acceleration and angular velocity; The main control module is used for: Based on the electromyographic signals, the degree of active force exertion by the user when performing preset rehabilitation movements is assessed. Based on the motion state signal, assess the quality of the user's performance of preset rehabilitation movements; The completion level information includes: the active force exertion level information and the action completion quality information.

[0007] In one embodiment, the main control module is used for: Based on the root mean square value and median frequency of the electromyographic signal amplitude, the local state information of the muscle group corresponding to the target site is evaluated. Based on the frequency characteristic change rate of the median frequency of the electromyographic signal and the timing of activation of all designated muscle groups corresponding to the target site, the overall state information of the muscle groups when the user performs the preset rehabilitation movement is evaluated. The information on the degree of active force exertion includes: the local state information of the muscle group and the overall state information of the muscle group.

[0008] In one embodiment, the main control module is used to evaluate the quality of the user's execution of the preset rehabilitation movement based on a comparison between the motion state signal and the theoretical motion parameters corresponding to the preset rehabilitation movement.

[0009] In one embodiment, the main control module is used for: The root mean square value of the amplitude of the electromyographic signal is calculated using a sliding window method, and the root mean square value is used as the instantaneous activation intensity value of the specified muscle group. The median frequency of the frequency domain signal is obtained from the frequency domain signal of the electromyographic signal, and the local fatigue state of the specified muscle group is obtained based on tracking the decreasing trend of the median frequency within a continuous time window. The local state information of the muscle group includes: the instantaneous activation intensity value and local fatigue state of the specified muscle group.

[0010] In one embodiment, the main control module is used for: The median frequency of the frequency domain signal is obtained from the frequency domain signal of the electromyography signal, and the linear regression slope of the median frequency over a unit time is obtained as the overall muscle fatigue state when the user performs the preset rehabilitation action. Based on the activation sequence and lag time between active and antagonistic muscle groups in all designated muscle groups corresponding to the target site, the overall muscle group coordination status of all designated muscle groups is evaluated when the user performs a preset rehabilitation movement. The overall muscle group status information includes the overall muscle fatigue state and the overall muscle group cooperation state.

[0011] In one embodiment, the main control module is used for: Based on the motion state signal, the real-time motion trajectory, key joint angles, and motion speed curve of the target body when performing the preset rehabilitation action are obtained; The real-time motion trajectory, key joint angles, and motion speed curves are compared with the theoretical motion trajectory, theoretical joint angles, and theoretical speed curves in the theoretical motion parameters to obtain the trajectory position deviation, key joint angle error, and motion speed deviation. The quality information of the action completion includes: the trajectory position deviation, the key joint angle error, and the movement speed deviation.

[0012] In one embodiment, the main control module includes: a state space definition unit, an action space configuration unit, a reward function design unit, and a policy network unit; The state space definition unit is used to construct a multi-dimensional state vector representing the completion state of the user's execution of the preset rehabilitation action based on the completion degree information. The action space configuration unit is used to define an executable set of discrete auxiliary actions. Each auxiliary action in the set of auxiliary actions is defined with: auxiliary torque, timing of action application, direction of action, and auxiliary mode. The reward function design unit is used to define a multi-objective reward function that includes action completion accuracy, active action participation, energy consumption efficiency, and training safety. The policy network unit is used to generate an optimal auxiliary action policy based on the current multidimensional state vector and the set of auxiliary actions. The auxiliary action signal includes: the auxiliary torque of the optimal auxiliary action policy, the timing of the action application, the direction of the action, and the auxiliary mode.

[0013] In one embodiment, the auxiliary execution device includes: a drive unit and a control unit connected in communication; the drive unit is disposed in contact with the target part; The control unit is used to generate a drive signal based on the auxiliary action signal and send it to the drive unit; The drive unit is used to apply auxiliary actions to the target part under the control of the drive signal.

[0014] In one embodiment, the rehabilitation training system further includes: a human-computer interaction module; The human-computer interaction module includes: a gamified interaction unit and a data reporting unit; The gamified interaction unit is used to map the preset rehabilitation actions to operations within a preset game, and to update the progress of the preset game based on the degree of completion information of the user's execution of the preset rehabilitation actions. The data reporting unit is used to generate and display rehabilitation reports that include training duration, active force statistics, and accuracy of movement execution. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a rehabilitation training system based on multimodal perception and adaptive assistance in the first embodiment of the present invention. Detailed Implementation

[0016] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings to provide a clearer understanding of the purpose, features, and advantages of the present invention. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of the present invention, but are merely illustrative of the essential spirit of the technical solution of the present invention.

[0017] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known apparatuses, structures, and techniques associated with this application may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0018] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0019] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0020] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to include the meaning of “or / and” unless otherwise expressly stated herein.

[0021] In the following description, in order to clearly demonstrate the structure and working method of the present invention, a number of directional terms will be used. However, terms such as "front", "back", "left", "right", "outer", "inner", "outer", "inner", "up", and "down" should be understood as convenient terms and not as limiting terms.

[0022] The first embodiment of the present invention relates to a rehabilitation training system based on multimodal perception and adaptive assistance, which is applied to rehabilitation training equipment to assist users in completing rehabilitation movements during rehabilitation training.

[0023] Please refer to Figure 1 The rehabilitation training system includes: a wearable sensing module 1, a main control module 2, and an auxiliary execution device 3. In some embodiments, the rehabilitation training system further includes: a human-computer interaction module 4; the wearable sensing module 1 is worn by the user on the target area, and the main control module 2 is communicatively connected to the wearable sensing module 1, the auxiliary execution device 3, and the human-computer interaction module 4.

[0024] The wearable sensing module 1 is used to collect active motion signals from the target area when the user performs preset rehabilitation movements. The active motion signals include physiological electrical signals and motion state signals. The wearable sensing module 1 collects the active motion signals at a preset frequency, such as 100Hz. The target area is a joint connection point, such as the upper limb (e.g., forearm, upper arm), lower limb (e.g., thigh, calf), or head and neck area.

[0025] The human-computer interaction module 4 includes a display module, which is used to display a visual training task. The training task includes several preset rehabilitation actions that need to be performed. The user can view the preset rehabilitation actions that need to be performed through the display module.

[0026] The wearable sensing module 1 includes a sensor for acquiring physiological electrical signals and a sensor for acquiring motion state signals. For example, the sensor for acquiring physiological electrical signals is an electromyography (EMG) sensor, which acquires EMG signals. The EMG sensor is deployed on a designated muscle group at the target site. For instance, if the target site is the thigh, the EMG sensor is deployed on the user's quadriceps; if the target site is the upper arm, the EMG sensor is deployed on the user's biceps. The sensor for acquiring motion state signals is an inertial measurement unit (IMU), which is fixed to the target site. The motion state signals acquired by the IMU include acceleration and angular velocity. The IMU may include, for example, a triaxial accelerometer and a triaxial gyroscope. The EMG sensor is deployed on the designated muscle group at the target site, and the IMU is fixed to the target site.

[0027] The wearable sensing module 1 can be worn on at least one target part of the user's body, so that active motion signals can be collected for each target part during rehabilitation of a single target part or simultaneous rehabilitation of multiple target parts.

[0028] The main control module 2 is used to evaluate the degree of completion of the user's preset rehabilitation actions based on the active action signal; and to generate auxiliary action signals to assist the user in completing the preset rehabilitation actions based on the degree of completion information.

[0029] As can be seen, the main control module 2 is divided into two functional sub-modules: a state assessment sub-module and a reinforcement learning strategy sub-module. The state assessment sub-module is used to assess the degree of completion of the user's preset rehabilitation actions based on the active action signal. The reinforcement learning strategy sub-module is used to generate auxiliary action signals to assist the user in completing the preset rehabilitation actions based on the degree of completion information.

[0030] The specific functions of the status assessment submodule are as follows: When a user performs a preset rehabilitation exercise, the wearable sensing module 1 can collect active motion signals from the target area and send them to the main control module 2. The active motion signals indicate the specific situation of the user performing the preset rehabilitation exercise. Based on the active motion signals, the main control module 2 can evaluate the degree of completion of the user's performance of the preset rehabilitation exercise. The degree of completion information indicates the specific execution of the user's performance of the preset rehabilitation exercise, such as the muscle state of the target area and the completion status of the preset rehabilitation exercise.

[0031] Subsequently, the reinforcement learning strategy submodule compares the actual execution of the preset rehabilitation action indicated by the completion level information with the ideal completion level corresponding to the preset rehabilitation action to obtain the actual deviation information between the two; and based on the actual deviation information, generates an auxiliary action signal to assist the user in completing the preset rehabilitation action; the auxiliary action signal indicates the assistance required for the user to perform the preset rehabilitation action relative to the ideal completion level corresponding to the preset rehabilitation action; and sends the auxiliary action signal to the auxiliary execution device 3.

[0032] For example, the main control module 2 is used to: evaluate the degree of active force exertion when the user performs a preset rehabilitation movement based on the electromyographic signal; evaluate the quality of movement completion when the user performs the preset rehabilitation movement based on the motion state signal; the degree of completion information includes: the degree of active force exertion information and the quality of movement completion information.

[0033] Electromyographic (EMG) signals indicate the local state of a specified muscle group at a target location and the overall state of the entire muscle group. The information on the degree of active force exertion includes both the local and overall state of the muscle group. Therefore, the local and overall state of the muscle group at the target location can be evaluated based on EMG signals to obtain information on the degree of active force exertion, as detailed below: Based on the root mean square (RMS) value and median frequency of the electromyographic (EMG) signal amplitude, the local state information of the designated muscle group corresponding to the target site is evaluated. This local state information includes the activation and fatigue states of the designated muscle group. For example, the EMG signal acquired by the EMG sensor is a time-domain signal, which can be preprocessed. Preprocessing methods include, but are not limited to, bandpass filtering (10-500Hz) and power frequency notch filtering. Then, a sliding window method is used to calculate the RMS value of the EMG signal amplitude, with a sliding window size of 250ms and an overlap rate of 50%. This RMS value is used as the instantaneous activation intensity value of the designated muscle group, which characterizes the activation state of the designated muscle group. Thus, the activation states of all designated muscle groups corresponding to the target site can be obtained.

[0034] The median frequency of the electromyographic (EMG) signal is obtained from its frequency domain signal, and the local fatigue state of the specified muscle group is determined by tracking the decreasing trend of the median frequency within a continuous time window. Specifically, the EMG signal is a time domain signal, which is subjected to a Fast Fourier Transform (FFT) to obtain its frequency domain signal. The median frequency of this frequency domain signal is then obtained. Subsequently, by tracking the decreasing trend of the EMG signal of the specified muscle group within a continuous time window of this median frequency, the local fatigue state of the specified muscle group is obtained. The greater the decreasing trend, the higher the degree of fatigue in the specified muscle group. The length of the continuous time window is a preset duration.

[0035] The local state information of the muscle group includes the instantaneous activation intensity value and local fatigue state of the specified muscle group.

[0036] Based on the rate of change of the frequency characteristics of the median frequency of the electromyographic signal and the timing of activation of all designated muscle groups corresponding to the target site, the overall muscle state information when the user performs a preset rehabilitation exercise is evaluated. Specifically, the median frequency of the frequency domain signal is obtained from the frequency domain signal of the electromyographic signal, and the linear regression slope of the median frequency over a unit time is obtained. The linear regression slope quantifies the rate of change of the frequency domain characteristics of the electromyographic signal and can be used as a measure of the overall muscle fatigue state when the user performs the preset rehabilitation exercise; wherein, the larger the linear regression slope of the median frequency of the electromyographic signal over a unit time, the higher the overall muscle fatigue level of the target site.

[0037] Based on the activation sequence and lag time between agonist and antagonist muscle groups in all designated muscle groups corresponding to the target site, the overall muscle group coordination state of all designated muscle groups is evaluated when the user performs a preset rehabilitation movement. Specifically, the designated muscle groups in the target site include agonist and antagonist muscle groups. The agonist and antagonist muscle groups work together to complete the movement, and there is a certain temporal relationship between them. Therefore, the activation time of the agonist muscle group can be determined by the electromyographic signal of the antagonist muscle group, and the activation time of the antagonist muscle group can be determined by the electromyographic signal of the antagonist muscle group. Then, the peak value of the cross-correlation function on the activation sequence of the agonist and antagonist muscle groups in the target site is obtained to determine the activation sequence coordination between the agonist and antagonist muscle groups in the target site. Based on the theoretical activation time difference and the actual activation time difference between the agonist and antagonist muscle groups in the target site, the activation lag time between the agonist and antagonist muscle groups in the target site is determined. The activation sequence coordination and activation lag time characterize the overall muscle group coordination state of all designated muscle groups in the target location, representing the temporal coordination and synergistic contraction level of different muscle groups in the target site.

[0038] The overall muscle group status information includes the overall muscle fatigue state of the target area and the overall muscle group cooperation state.

[0039] Based on the motion state signal, the quality of the user's performance of the preset rehabilitation movement is evaluated; specifically, based on the comparison between the motion state signal and the theoretical movement parameters corresponding to the preset rehabilitation movement, the quality of the user's performance of the preset rehabilitation movement is evaluated.

[0040] For example, based on the motion state signal, the real-time motion trajectory, key joint angles, and motion velocity curve of the target part when performing the preset rehabilitation action are obtained; for example, the motion state signal collected by the inertial measurement unit includes acceleration and angular velocity; the real-time motion trajectory and motion velocity curve of the target part can be obtained by fitting the motion state signal; based on the fusion algorithm of complementary filtering or Kalman filtering, the key joint angles of the target part can be calculated from the motion state signal, including: the precise posture angle and joint angle of the target part in three-dimensional space, for example, for the upper limb, the precise posture angle between the forearm and the upper arm, and the joint angle between the upper arm and the body, and the joint angle between the upper arm and the forearm can be calculated.

[0041] The theoretical motion parameters corresponding to the preset rehabilitation movements define the theoretical conditions for performing the preset rehabilitation movements, including: theoretical motion trajectory, theoretical joint angle, and theoretical velocity curve.

[0042] Therefore, by comparing the real-time motion trajectory, key joint angles, and motion velocity curves with the theoretical motion trajectory, theoretical joint angles, and theoretical velocity curves in the theoretical motion parameters, the trajectory position deviation, key joint angle error, and motion velocity deviation are obtained; specifically: The real-time motion trajectory is compared with the theoretical motion trajectory to obtain the trajectory position deviation between the two; for example, the position deviation between corresponding points of the real-time motion trajectory and the theoretical motion trajectory is calculated, and then the average value is taken as the trajectory position deviation between the two.

[0043] The difference between the critical joint angle and the corresponding theoretical joint angle is calculated as the critical joint angle error. If the critical joint angle error is less than 0, it means that the target part has not moved to the theoretical joint angle desired by the preset rehabilitation movement. If the critical joint angle error is greater than 0, it means that the target part has moved to a higher than the theoretical joint angle desired by the preset rehabilitation movement.

[0044] By comparing the motion speed curve with the theoretical speed curve and calculating the correlation coefficient between the two, the motion speed deviation can be obtained. The higher the correlation coefficient, the smaller the motion speed deviation.

[0045] Based on the above, the quality of a user's execution of preset rehabilitation movements can be comprehensively quantified from three aspects: spatial trajectory accuracy (trajectory position deviation), joint angle accuracy (joint angle deviation), and movement smoothness (movement speed deviation).

[0046] The specific functions of the reinforcement learning strategy submodule are as follows: The reinforcement learning policy sub-module in the main control module 2 includes: a state space definition unit, an action space configuration unit, a reward function design unit, and a policy network unit; The state space definition unit is used to construct a multi-dimensional state vector representing the completion state of the user's execution of the preset rehabilitation action based on the completion level information. For example, the activation state, local fatigue state, overall muscle fatigue state of the target area, and overall muscle coordination state of the muscle groups corresponding to the target area are quantitatively integrated to obtain the multi-dimensional state vector. This multi-dimensional state vector quantitatively represents the completion quality of the user's execution of the preset rehabilitation action. Furthermore, the multi-dimensional state vector may also include: the duration of the current training phase and the quantitative representation value of the number of repetitions of the completed preset rehabilitation action, thereby introducing temporal context features and enabling a more comprehensive representation of the user's state in executing the preset rehabilitation action.

[0047] The action space configuration unit is used to define a set of executable, discretized auxiliary actions. Each auxiliary action in the set is defined with: auxiliary torque, timing of action application, direction of action, and auxiliary mode. In other words, the discretized auxiliary action set defines all executable auxiliary actions, including all implementable auxiliary actions. Each auxiliary action is a combination of auxiliary torque, timing of action application, direction of action, and auxiliary mode. For example: The auxiliary torque is divided into 5 discrete torque levels in the range of 0-15 N•m. Each torque level corresponds to a torque value, which are 3 N•m, 6 N•m, 9 N•m, 12 N•m and 15 N•m respectively.

[0048] The timing of applying the movement is defined as the three key phases of the rehabilitation movement cycle: the beginning, the middle, and the end.

[0049] The direction of the movement defines the two dimensions of flexion and extension, indicating the angle at which the auxiliary movement is applied.

[0050] Assist modes include full assist, half assist, and drag mode.

[0051] Based on the combination of the above parameters, the motion space configuration unit defines a total of 108 executable auxiliary movements, ensuring that the rehabilitation training system can output refined auxiliary movement strategies.

[0052] The reward function design unit defines a multi-objective reward function that includes action completion accuracy, active participation, energy consumption efficiency, and training safety. For example, the multi-objective reward function constructed by the reward function design unit includes variables such as trajectory position error, electromyographic characteristics, motor power consumption, and safety penalties, as well as positive reward weights, energy consumption penalty weights, and safety penalty weights. It is evident that this multi-objective reward function rewards action accuracy and user active participation, while penalizing detected abnormal movement patterns or situations exceeding safety boundaries. The positive reward weights, energy consumption penalty weights, and safety penalty weights can be dynamically adjusted according to the user's rehabilitation stage to achieve personalized optimization goals.

[0053] The policy network unit is used to generate an optimal auxiliary action policy based on the current multidimensional state vector and the set of auxiliary actions. The auxiliary action signal includes: the auxiliary torque of the optimal auxiliary action policy, the timing of the action application, the direction of the action, and the auxiliary mode.

[0054] For example, the policy network unit adopts a deep deterministic policy gradient algorithm based on a dual-network architecture of Actor (policy network) and Critic (value network). The Actor policy network is used to determine the optimal auxiliary action policy from the set of auxiliary actions based on the current multidimensional state vector, that is, to determine the optimal combination of auxiliary torque, timing of action application, direction of action, and auxiliary mode; for example, it is a fully connected neural network containing three hidden layers (the number of nodes in the three hidden layers are 256, 128, and 64, respectively), using the ReLU activation function.

[0055] The Critic value network is used to evaluate the long-term expected return of different combinations of multidimensional state vectors and auxiliary actions. Its output is used to guide the parameter updates of the Actor policy network. The Critic value network stores historical training data through an experience replay mechanism and uses target network technology to ensure training stability, thereby achieving continuous optimization of policy network parameters.

[0056] After obtaining the optimal auxiliary action strategy, the reinforcement learning strategy submodule sends an auxiliary action signal to the auxiliary execution device 3, which consists of the auxiliary torque, timing of action application, direction of action, and auxiliary mode of the optimal auxiliary action strategy.

[0057] The assistive device 3 is also a wearable device that can be worn on the target area, such as an exoskeleton or an integrated training device.

[0058] Based on the auxiliary action signal, the auxiliary execution device 3 can determine the auxiliary action that needs to be applied to the target area and apply it to the target area to assist the target area in completing the preset rehabilitation action.

[0059] Specifically, the auxiliary execution device 3 includes a drive unit, a control unit, and a safety monitoring unit.

[0060] The drive unit, acting as the actuator, directly contacts the target area (either fixed or non-fixed) and outputs physical assistance force. Specifically, for drive units used in upper limb joint training, a frameless servo motor paired with a planetary reducer serves as the actuator to achieve high-precision torque control, providing continuously adjustable torque from 0-20 N•m with a response latency of less than 10 milliseconds. For drive units used in lower limb rehabilitation training, a fiber-reinforced pneumatic artificial muscle serves as the actuator, adjusting the working air pressure from 0-600 kPa to achieve compliant, bio-muscle-like assistance.

[0061] The control unit controls the drive unit to output physical assistance force to the target area. It is equipped with a real-time operating system and runs a high-frequency servo drive algorithm. Based on the received auxiliary action signal for the target area, the control unit converts the auxiliary action signal into a drive signal. Based on the real-time received auxiliary action signal, it analyzes the target torque curve and impedance parameters to obtain drive commands, and converts the drive commands into drive signals for the drive unit of the target area through a drive model. The drive signal represents the magnitude, direction, and duration of the auxiliary torque required by the drive unit to output the auxiliary action. Thus, the drive signal can control the drive unit to output physical assistance actions, assisting the target area to accurately complete the preset rehabilitation action; for example, converting the drive command into a motor current command or a pneumatic valve PWM signal.

[0062] Based on the above, the control unit can generate a drive signal based on the auxiliary action signal and send it to the drive unit. The drive unit then applies an auxiliary action to the target part under the control of the drive signal.

[0063] In some embodiments, for preset rehabilitation actions that are executed periodically, the control unit can also perform iterative learning, and output the corrected and precise assistance in the current cycle based on the execution deviation of the previous cycle, so as to achieve continuous optimization of the assistance effect.

[0064] The safety monitoring unit includes a torque sensor, which is installed at the location of the output physical auxiliary force of the drive unit to detect the actual auxiliary torque value output by the drive unit in real time. The safety monitoring unit determines the instantaneous rate of change of the auxiliary torque based on the actual auxiliary torque value detected in real time by the torque sensor. If the instantaneous rate of change of the auxiliary torque exceeds the preset torque change threshold (e.g., 50 N•m / s), the braking and stopping drive unit is triggered in time, the drive unit is controlled to stop outputting physical auxiliary force, and an early warning is activated.

[0065] The safety monitoring unit also includes a position detection sensor (e.g., an absolute encoder), which is fixed to the target part to detect the position of the target part. Thus, the safety monitoring unit can continuously track the joint position through the position detection sensor. When the displacement of the target part exceeds the preset displacement threshold, the braking and stopping drive unit is triggered in time, the drive unit is controlled to stop outputting physical assistance force, and an early warning is activated.

[0066] The safety monitoring unit also establishes a user electromyography-output torque correlation model. When the electromyography signal indicates that the user's sudden voluntary contraction conflicts with the direction of the auxiliary movement, the auxiliary mode is switched to zero impedance mode within 50 milliseconds to ensure the user's autonomous movement of the target part.

[0067] In some embodiments, the human-computer interaction module includes a gamified interaction unit and a data reporting unit; the gamified interaction unit is used to map the preset rehabilitation actions to operations in a preset game, and update the progress of the preset game based on the completion information of the user's performance of the preset rehabilitation actions; the data reporting unit is used to generate and display a rehabilitation report that includes training duration, active exertion statistics, and action execution accuracy.

[0068] The gamified interactive interface mapping actions mapped by the gamified interactive unit include: upper limb action mapping group, lower limb action mapping group and compound action mapping group. The upper limb motion mapping group maps the user's shoulder joint abduction / adduction and elbow joint flexion / extension movements in real time to the up-and-down movements of the arm of the virtual character in the preset game, as well as the grasping and releasing operations of objects.

[0069] The lower limb motion mapping group maps the user's hip flexion / extension and knee flexion / extension movements in real time to the standing, sitting, walking, and stepping movements of the virtual character in the preset game.

[0070] The composite motion mapping group maps the coordinated movements of the user's upper and lower limbs to virtual tasks within the preset game that require the cooperation of both hands, such as rowing, drumming, and catching and tossing balls.

[0071] As described above, the gamified interaction unit achieves deep integration of physiological signals and virtual tasks through a motion mapping engine.

[0072] For example, the interface includes three functional units: upper limb motion mapping group, lower limb motion mapping group, and compound motion mapping group. The upper limb motion mapping group precisely maps shoulder abduction / adduction movements to arm extension and fruit picking operations in a virtual orchard harvesting scenario, and transforms elbow flexion / extension movements into finer operations such as stirring and pouring in a simulated cooking game. The lower limb motion mapping group maps hip flexion / extension movements to pedaling movements in a virtual cycling game, and knee flexion / extension movements to stair climbing movements in a simulated mountain climbing scenario. The compound motion mapping group integrates coordinated upper and lower limb movements into paddling movements in a virtual rowing game, requiring users to simultaneously complete a compound movement pattern of trunk flexion, upper limb paddling, and lower limb extension.

[0073] In this embodiment, the gamified interaction unit integrates a multimodal feedback system. When the user's actions meet the standards of the preset rehabilitation actions, the system simultaneously triggers three positive stimuli: visual effects, auditory feedback, and tactile feedback. The visual effects are manifested as particle flashes and dynamic progress bar filling in the interface; the auditory feedback provides real-time operation guidance through spatial sound effects; and the tactile feedback enhances the realism of the training by driving the subtle vibrations of the unit.

[0074] The data reporting unit is built on a visualization and analysis platform using WebGL technology, providing real-time data display and in-depth analysis capabilities. The data reporting unit comprises three functional modules: The real-time data display module is used to dynamically display key indicators such as the effective training time of this training, the completion rate of preset rehabilitation movements, and the curve of the user's active exertion ratio. The progress analysis module is used to generate trend charts of improved movement accuracy and radar charts of improved muscle activation patterns by comparing historical data. The intelligent report generation module is used to automatically generate rehabilitation reports that include training intensity analysis, fatigue assessment, and personalized suggestions.

[0075] In this embodiment, when a user uses the rehabilitation training system to perform rehabilitation training on a target area, the wearable sensing module and the auxiliary execution device are worn at the target location. The user actively performs preset rehabilitation movements. During this process, the wearable sensing module collects active movement signals from the target area. These active movement signals include two modes of signals: physiological electrical signals and motion state signals. The active movement signals characterize the user's completion status of the preset rehabilitation movements. Based on these active movement signals, the degree of completion of the preset rehabilitation movements can be evaluated. The degree of completion of the preset rehabilitation movements is then compared with the theoretical movements of the preset rehabilitation movements to obtain the deviation between the two. This deviation generates an auxiliary movement signal to assist the user in completing the preset rehabilitation movements. The auxiliary execution device then applies the auxiliary movement to the target area based on this auxiliary movement signal to assist the target area in completing the preset rehabilitation movements.

[0076] Based on the above, by combining multimodal signal perception of the user's target body parts, the system can obtain the user's motor intention and execution status of the target body parts, and adaptively assist the user to complete rehabilitation movements as accurately as possible. This can provide personalized and adaptive rehabilitation assistance for different users, which can meet the user's rehabilitation training needs more efficiently, safely and with high compliance. It is suitable for rehabilitation training in non-professional environments such as home and community.

[0077] The preferred embodiments of the present invention have been described in detail above, but it should be understood that, if necessary, aspects of the embodiments can be modified to utilize aspects, features, and concepts from various patents, applications, and publications to provide other embodiments.

[0078] In light of the detailed description above, these and other changes can be made to the embodiments. Generally, the terminology used in the claims should not be considered limited to the specific embodiments disclosed in the specification and claims, but should be understood to include all possible embodiments together with the full scope of equivalents enjoyed by these claims.

Claims

1. A rehabilitation training system based on multimodal perception and adaptive assistance, characterized in that, include: Wearable sensing module, main control module, and auxiliary execution device; The wearable sensing module is worn by the user on the target area; The wearable sensing module is used to collect active motion signals of the target area when the user performs preset rehabilitation actions. The active motion signals include: physiological electrical signals and motion state signals. The main control module is used to evaluate the degree of completion of the user's preset rehabilitation actions based on the active action signals; and to generate auxiliary action signals to assist the user in completing the preset rehabilitation actions based on the degree of completion information. The auxiliary execution device is used to apply auxiliary actions to the target area based on the auxiliary action signal, so as to assist the target area in completing the preset rehabilitation action.

2. The rehabilitation training system according to claim 1, characterized in that, The wearable sensing module includes an electromyography (EMG) sensor and an inertial measurement unit, wherein the EMG sensor is deployed at a designated muscle group in the target area and the inertial measurement unit is fixed to the target area. The electromyography sensor is used to collect electromyographic signals of a specified muscle group at the target site, and the physiological electrical signals include the electromyographic signals. The inertial measurement unit collects the acceleration and angular velocity of the target part, and the motion state signal includes: acceleration and angular velocity; The main control module is used for: Based on the electromyographic signals, the degree of active force exertion by the user when performing preset rehabilitation movements is assessed. Based on the motion state signal, assess the quality of the user's performance of preset rehabilitation movements; The completion level information includes: the active force exertion level information and the action completion quality information.

3. The rehabilitation training system according to claim 2, characterized in that, The main control module is used for: Based on the root mean square value and median frequency of the electromyographic signal amplitude, the local state information of the muscle group corresponding to the target site is evaluated. Based on the frequency characteristic change rate of the median frequency of the electromyographic signal and the timing of activation of all designated muscle groups corresponding to the target site, the overall state information of the muscle groups when the user performs the preset rehabilitation movement is evaluated. The information on the degree of active force exertion includes: the local state information of the muscle group and the overall state information of the muscle group.

4. The rehabilitation training system according to claim 2, characterized in that, The main control module is used to evaluate the quality of the user's execution of the preset rehabilitation movements by comparing the motion state signal with the theoretical motion parameters corresponding to the preset rehabilitation movements.

5. The rehabilitation training system according to claim 3, characterized in that, The main control module is used for: The root mean square value of the amplitude of the electromyographic signal is calculated using a sliding window method, and the root mean square value is used as the instantaneous activation intensity value of the specified muscle group. The median frequency of the frequency domain signal is obtained from the frequency domain signal of the electromyographic signal, and the local fatigue state of the specified muscle group is obtained based on tracking the decreasing trend of the median frequency within a continuous time window. The local state information of the muscle group includes: the instantaneous activation intensity value and local fatigue state of the specified muscle group.

6. The rehabilitation training system according to claim 3, characterized in that, The main control module is used for: The median frequency of the frequency domain signal is obtained from the frequency domain signal of the electromyography signal, and the linear regression slope of the median frequency over a unit time is obtained as the overall muscle fatigue state when the user performs the preset rehabilitation action. Based on the activation sequence and lag time between active and antagonistic muscle groups in all designated muscle groups corresponding to the target site, the overall muscle group coordination status of all designated muscle groups is evaluated when the user performs a preset rehabilitation movement. The overall muscle group status information includes the overall muscle fatigue state and the overall muscle group cooperation state.

7. The rehabilitation training system according to claim 4, characterized in that, The main control module is used for: Based on the motion state signal, the real-time motion trajectory, key joint angles, and motion speed curve of the target body when performing the preset rehabilitation action are obtained; The real-time motion trajectory, key joint angles, and motion speed curves are compared with the theoretical motion trajectory, theoretical joint angles, and theoretical speed curves in the theoretical motion parameters to obtain the trajectory position deviation, key joint angle error, and motion speed deviation. The quality information of the action completion includes: the trajectory position deviation, the key joint angle error, and the movement speed deviation.

8. The rehabilitation training system according to claim 1, characterized in that, The main control module includes: a state space definition unit, an action space configuration unit, a reward function design unit, and a policy network unit; The state space definition unit is used to construct a multi-dimensional state vector representing the completion state of the user's execution of the preset rehabilitation action based on the completion degree information. The action space configuration unit is used to define an executable set of discrete auxiliary actions. Each auxiliary action in the set of auxiliary actions is defined with: auxiliary torque, timing of action application, direction of action, and auxiliary mode. The reward function design unit is used to define a multi-objective reward function that includes action completion accuracy, active action participation, energy consumption efficiency, and training safety. The policy network unit is used to generate an optimal auxiliary action policy based on the current multidimensional state vector and the set of auxiliary actions. The auxiliary action signal includes: the auxiliary torque of the optimal auxiliary action policy, the timing of the action application, the direction of the action, and the auxiliary mode.

9. The rehabilitation training system according to claim 1, characterized in that, The auxiliary execution device includes: a drive unit and a control unit connected in communication; the drive unit is configured to contact the target part. The control unit is used to generate a drive signal based on the auxiliary action signal and send it to the drive unit; The drive unit is used to apply auxiliary actions to the target part under the control of the drive signal.

10. The rehabilitation training system according to claim 1, characterized in that, The rehabilitation training system also includes: a human-computer interaction module; The human-computer interaction module includes: a gamified interaction unit and a data reporting unit; The gamified interaction unit is used to map the preset rehabilitation actions to operations within a preset game, and to update the progress of the preset game based on the degree of completion information of the user's execution of the preset rehabilitation actions. The data reporting unit is used to generate and display rehabilitation reports that include training duration, active force statistics, and accuracy of movement execution.

Images