Upper extremity exoskeleton control method and upper extremity exoskeleton

Abstract

The application discloses a kind of upper limb exoskeleton control method and upper limb exoskeleton, the method includes: by myoelectric sensor, pressure sensor and bending sensor respectively collect forearm surface muscle electric signal, hand pressure signal and finger bending angle signal;Extract muscle electric time domain feature from muscle electric signal, extract hand pressure feature from pressure signal, extract finger bending feature from bending signal;The forearm muscle movement intention is obtained by comparing muscle electric time domain feature with preset myoelectric activation threshold;Hand pressure feature and finger bending feature are compared with preset pressure activation threshold and preset bending activation threshold respectively, and hand movement intention is obtained;Based on forearm muscle movement intention and hand movement intention, the user movement intention is identified;The action of user movement intention is classified, and the upper limb action mode is obtained, and corresponding drive instruction is sent to the actuator of upper limb exoskeleton, and the hand-wrist coordination movement in line with human natural physiological habit.

Description

Technical Field

[0001] This invention relates to the field of upper limb exoskeleton technology, and more particularly to an upper limb exoskeleton control method and an upper limb exoskeleton. Background Technology

[0002] Neurological disorders, such as stroke, spinal cord injury, cerebral palsy, and muscular dystrophy, can significantly impair hand and wrist motor function, reducing a patient's ability to perform ADL (Activities of Daily Living), such as eating, dressing, and personal hygiene. In particular, wrist drop combined with weakened hand grip strength is a common and debilitating outcome. Existing rehabilitation tools often focus on either the hand or wrist separately and typically require supervised use in a clinical setting. Furthermore, age-related sarcopenia and neuromuscular degeneration in older adults can also lead to a decline in ADL abilities, making independence and task performance difficult.

[0003] Most current exoskeletons rely on manual input or fixed movement paths and lack intent awareness. Due to the lack of integrated hand-wrist coordination, intuitive control, and wearable design, existing upper limb exoskeletons are ineffective or difficult to use in the long term, especially in unsupervised rehabilitation at home. Summary of the Invention

[0004] The main objective of this invention is to provide an upper limb exoskeleton control method and an upper limb exoskeleton, aiming to solve the technical problem that existing upper limb exoskeletons are ineffective due to a lack of integrated hand-wrist coordination.

[0005] In a first aspect, the present invention provides a method for controlling an upper limb exoskeleton, comprising the following steps: Signal acquisition: Electromyography (EMG) signals from the forearm surface, pressure signals from the hand, and finger flexion angle signals are acquired using an EMG sensor, a pressure sensor, and a flexion sensor, respectively. Signal processing: Extracting electromyographic time-domain features from the electromyographic signals, extracting hand pressure features from the pressure signals, and extracting finger bending features from the bending signals; Intent recognition: The electromyographic temporal features are compared with preset electromyographic activation thresholds to obtain the forearm muscle movement intent; the hand pressure features and the finger bending features are compared with preset pressure activation thresholds and preset bending activation thresholds, respectively, to obtain the hand movement intent; based on the forearm muscle movement intent and the hand movement intent, the user's movement intent is identified. Action pattern: The user's movement intention is classified into action patterns to obtain upper limb action patterns, which include at least one of hand opening, hand grasping, forearm pronation, and forearm supination; Drive command: Based on the upper limb movement pattern, send corresponding drive commands to the actuators of the upper limb exoskeleton.

[0006] Optionally, the electromyographic sensor includes a first EMG sensor located 2-3 cm below the ulnar side of the elbow crease (pronator teres) and a second EMG sensor located below the radial head (supinator).

[0007] In a second aspect, the present invention also provides an upper limb exoskeleton, the upper limb exoskeleton comprising: a memory, a processor, and an exoskeleton control program stored in the memory and executable on the processor, the exoskeleton control program being configured to implement the steps of the upper limb exoskeleton control method described in any of the preceding claims.

[0008] This invention proposes an upper limb exoskeleton control method. In the signal acquisition step, the method independently acquires forearm electromyographic signals using an electromyography (EMG) sensor, and independently acquires hand pressure and flexion signals using pressure and flexion sensors, improving the independence, specificity, and accuracy of forearm and hand signal acquisition. In the signal processing step, EMG time-domain features are extracted from the EMG signals, hand pressure features from the pressure signals, and finger flexion features from the flexion signals. In the intent recognition step, forearm muscle movement intent and hand movement intent are obtained separately, and based on these, the true user movement intent is identified, effectively filtering out false signals caused by EMG noise, accidental physical contact, or unconscious muscle twitching, avoiding erroneous exoskeleton movements and improving safety. In the action pattern step, the user's movement intent is classified into action patterns, resulting in upper limb action patterns, such as at least one of hand opening, hand grasping, forearm pronation, and forearm supination, which conform to the natural physiological habits of hand-wrist coordination, thus providing driving commands for wrist-hand coordination. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating one aspect of the upper limb exoskeleton control method of the present invention; Figure 2 This is another flowchart illustrating the upper limb exoskeleton control method of the present invention; Figure 3 This is a schematic diagram of the upper limb exoskeleton of the present invention; Figure 4 This is an overall schematic diagram of the upper limb exoskeleton hidden hardware box and rehabilitation gloves of the present invention. Figure 5 This is an overall schematic diagram of the wrist joint motion mechanism of the present invention; Figure 6 This is a schematic diagram showing the arrangement of the first cable inside the C-shaped slide rail and slide block; Figure 7This is an overall schematic diagram of the wrist joint driving device of the present invention; Figure 8 This is a schematic diagram of the tension adjustment mechanism of the wrist joint drive device of the present invention; Figure 9 This is an exploded view of the adjusting rod and adjusting block of the wrist joint driving device of the present invention; Figure 10 This is a schematic diagram of the adjusting block and guide rod of the wrist joint driving device of the present invention; Figure 11 This is a front view of the roller of the wrist joint driving device of the present invention; Figure 12 This is a side view of the roller of the wrist joint driving device of the present invention; Figure 13 This is a schematic diagram of the internal structure of the roller of the wrist joint driving device of the present invention; Figure 14 This is a schematic diagram of the first cable inside the C-shaped slide rail and slide block. Detailed Implementation

[0010] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0011] Figure 1 This is a schematic flowchart of the first embodiment of the upper limb exoskeleton control method of the present invention. (Refer to...) Figure 1 The method includes the following steps: Step S3, Signal Acquisition: Electromyography (EMG) sensors, pressure sensors, and flexion sensors are used to acquire electromyographic signals from the forearm surface, pressure signals from the hand, and flexion angle signals from the fingers, respectively. The forearm muscle and hand signals are acquired separately using different sensors, improving the independence, specificity, and accuracy of the signal acquisition.

[0012] In one example, combining Table 1 and Figure 4 The electromyography sensor 44 includes a first EMG sensor 44a located 2-3 cm below the ulnar side of the elbow crease (pronator teres muscle) and a second EMG sensor 44b located below and behind the radial head (suppressor muscle); when the first EMG sensor 44a is activated and the second EMG sensor 44b is at rest, pronation movement (palm moving inward or downward) is determined; when the first EMG sensor 44a is at rest and the second EMG sensor 44b is activated, supination movement (palm moving outward or upward) is determined.

[0013]

[0014] Table 1. Forearm movement patterns, main controlling muscles and their locations, ECG sensor locations and algorithms In this study, both pronation and supination of the forearm involve the combined action of two antagonistic muscles (see Table 1). Therefore, we can conclude that: first, both pronation and supination require the simultaneous activity of two muscles; and second, these two muscles are antagonistic. Based on this, simultaneously detecting the electromyographic signals of the two antagonistic muscles to identify and determine forearm movement intentions can significantly reduce the probability of misjudgment and increase the accuracy of intention recognition.

[0015] In one example, since forearm movements are relatively simple and coarse, consisting only of forearm pronation or supination (internal and external rotation of the wrist joint), and forearm pronation or supination movements are not easy to detect forearm pressure and bending changes, electromyography (EMG) sensors can be used for forearm movements.

[0016] In some embodiments, the primary function of the hand is grasping (other functions such as finger adduction and abduction are not very useful for basic daily activities). Finger flexion and extension movements are accompanied by changes in physical parameters of pressure and bending. Pressure sensors and bending sensors collect changes in hand pressure signals and finger bending angle signals, respectively, to infer and identify the user's intention. Since the changes in pressure and bending are significant during hand movements—that is, the changes in pressure signals and bending angle signals are obvious—a fusion method using pressure and bending sensors, and further combined at multiple points (using multiple pressure sensors and multiple bending sensors, such as pressure sensors and bending sensors on at least two fingers), can accurately identify the hand's movement intention and guide subsequent exoskeleton movements.

[0017] In this application, the motor center of the cerebral cortex sends motor nerve impulses, which are transmitted along the central nervous pathway to the forearm nerves. The neuromuscular signal transmission is then completed through the neuromuscular junction, causing contraction of the hand's motor muscles and generating electromyographic (EMG) signals. This ultimately manifests as forearm and / or hand movement (which may occur before or after the hand). Changes in pressure and flexion electrical signals are detected based on the hand movement. Theoretically, for recognizing user intent, EMG signals, as direct signals, are more accurate than pressure and flexion signals, which are indirect signals. However, EMG signals are very complex, especially non-invasive surface muscle EMG signals of the hand, which are subject to many interference factors (such as individual differences, EMG sensor location, conduction decay, etc.). It is difficult to accurately quantify hand opening and grasping movements, such as determining pressure magnitude, flexion angle, number of fingers moving, and hand coordination. Therefore, EMG signals alone cannot recognize and control fine hand movements. Thus, pressure sensors and flexion sensors are used to collect hand pressure signals and finger flexion angle signals, respectively.

[0018] It is understood that in other embodiments, a hand electromyography (EMG) unit is used to collect surface EMG signals of the hand. Surface EMG signals can provide qualitative motor intentions, improving the accuracy of intention detection. Specifically, the hand EMG unit includes a first EMG unit, a second EMG unit, and a third EMG unit. The first EMG unit collects surface EMG signals of the flexor digitorum superficialis muscles, the second EMG unit collects surface EMG signals of the flexor digitorum profundus muscles, and the third EMG unit collects surface EMG signals of the extensor digitorum muscles. The process of generating surface EMG signals for hand flexion and extension is as follows: the muscles controlling hand flexion and extension are actually located in the forearm (not in the hand). Finger flexion is caused by the contraction of the flexor digitorum superficialis and flexor digitorum profundus muscles. These surface EMG signals are collected in the upper-middle third of the palmar side of the forearm. The first and second EMG units are used to sense hand flexion activities. Finger extension is caused by the contraction of the extensor digitorum muscles. These surface EMG signals are collected in the upper-middle section of the dorsal side of the forearm. These muscle contractions cause the tendons of the fingers to stretch and contract, thereby causing hand flexion and extension activities. The third EMG unit is used to sense hand flexion and extension activities.

[0019] Specifically, the pressure sensor is used to detect the contact force between the fingertips and the palm. The pressure sensor can be a force-sensitive resistor (FSR).

[0020] Specifically, a flexible bending sensor can be selected as the bending sensor.

[0021] Step S4, Signal Processing: Extract electromyographic time-domain features from electromyographic signals, extract hand pressure features from pressure signals, and extract finger bending features from bending signals.

[0022] Specifically, the signal processing steps described above filter, amplify, and extract features from the electromyography (EMG), pressure, and bending signals, respectively, converting the raw sensor data into feature values ​​usable for intent recognition. The purpose of signal processing is to remove noise interference, retain effective features, and provide a reliable data foundation for subsequent threshold judgment and intent recognition.

[0023] Step S5, Intent Recognition: Compare the electromyographic temporal features with the preset electromyographic activation threshold to obtain the forearm muscle movement intent; compare the hand pressure features and finger bending features with the preset pressure activation threshold and preset bending activation threshold respectively to obtain the hand movement intent; based on the combination of forearm muscle movement intent and hand movement intent, identify the user's movement intent.

[0024] Specifically, the above intention recognition step compares the electromyographic temporal features, hand pressure features, and finger bending features with the corresponding preset activation thresholds to determine whether the user has an intention to rotate the forearm and an intention to grasp the hand.

[0025] Step S6, Action Pattern: Classify the user's movement intention to obtain the upper limb action pattern. The upper limb action pattern includes at least one of the following: hand opening, hand grasping, forearm pronation, and forearm supination, which are hand-wrist coordinated movements that conform to the natural physiological habits of the human body.

[0026] In real-world practice, there are four possible combinations of hand and forearm movements: (i) purely hand movements; (ii) purely forearm muscle movements; (iii) forearm movements followed by hand movements. For example, when reaching for a cup, the forearm rotates first before grasping the cup; (iv) hand movements occur first, followed by forearm muscle movements. For example, when drinking water, the hand grasps the cup first, then the forearm rotates to bring the cup to the mouth. All four of these movement combinations can occur in a user's daily activities.

[0027] Step S7, Drive Command: Based on the upper limb movement pattern, send corresponding drive commands to the actuators of the upper limb exoskeleton to drive hand-wrist coordinated movement.

[0028] In this embodiment, through the sequential execution of steps S3 to S7, the system can complete a full control closed loop from signal acquisition to drive output. After the user wears the exoskeleton, the system continuously monitors muscle electrical signals, pressure signals, and flexion signals, identifies the user's movement intentions in real time, and drives the actuators to provide auxiliary force, thereby assisting the user in completing hand grasping and forearm pronation and supination movements.

[0029] In some embodiments, combined with Figure 2 Before step S3, the method further includes: S1: Device initialization: Activate the electromyography sensor, pressure sensor, and flexion sensor, and establish a communication connection with the controller.

[0030] Specifically, the controller sends wake-up and configuration commands to each sensor, sets the sampling frequency of the electromyography sensor (e.g., 1000Hz) and the sampling frequencies of the pressure sensor and the bending sensor (e.g., both 10Hz); performs a hardware reset to eliminate power-on transients and baseline drift; and performs multimodal clock synchronization to assign a unified timestamp to the data of each channel.

[0031] It should be noted that step S1 can be omitted in some scenarios. For example, if an upper limb exoskeleton is used continuously to assist the user in continuous movements, the method can start directly from S3, without needing to initialize the device for each assisted movement.

[0032] In some embodiments, based on the first embodiment, a second embodiment of the upper limb exoskeleton control method of the present invention is proposed. Before step S3, the method further includes S2: threshold setting.

[0033] Step S2 specifically includes: S21: Collects the resting baselines of the electromyography sensor, pressure sensor, and bending sensor in the user's resting state to automatically eliminate environmental electromagnetic interference, electrode contact noise, and the muscle resting spasticity tension unique to hemiplegic patients, responding only to incremental signals of active force exertion, significantly improving the signal-to-noise ratio of weak signal extraction.

[0034] S22: Collect the maximum voluntary contraction values ​​(MVC / MVB) of each sensor (electromyography sensor, pressure sensor, and flexion sensor) when the user has maximum voluntary contraction.

[0035] S23: Based on the resting baseline and maximum voluntary contraction value of each sensor, determine the preset electromyographic activation threshold, preset pressure activation threshold, and preset flexion activation threshold respectively.

[0036] In this implementation, the aforementioned threshold setting establishes a personalized activation threshold for each user by collecting their resting baseline and maximum voluntary contraction value, overcoming physiological differences among patients in muscle volume, subcutaneous fat thickness, and residual muscle strength. The resting baseline is used to eliminate the influence of environmental noise and muscle resting activity, while the maximum voluntary contraction value is used to determine the user's individual ability range, ensuring that even early rehabilitation patients with weak muscle strength can effectively trigger the system. The combination of these two factors allows the activation threshold to adapt to the physiological differences among different users, improving the accuracy of intent recognition.

[0037] It should be noted that step S2 can be omitted in some scenarios. For example, if an upper limb exoskeleton is used continuously to assist the user in continuous movements, the method can directly start from S3 without determining the preset activation threshold of each sensor for each assisted movement. Alternatively, the user can directly input the activation threshold of each sensor without data collection. Optionally, the system performs threshold setting once before the user's first use, when the user changes, or when the usage time reaches a preset duration.

[0038] In one embodiment, for an electromyography (EMG) sensor, steps S21 and S22 specifically involve: S211: During the period of 5 to 10 seconds when the user is in a completely relaxed state, the electromyography sensor continuously records the electromyographic signals of the muscle during this time period and calculates the root mean square (RMS) value as the resting baseline.

[0039] For example, the amplitude of the raw resting electromyography signal (EMG) is approximately ±5 µV to 10 µV, and the root mean square value of the resting baseline (RMS) is approximately 0.008 mV to 0.012 mV.

[0040] S221: The electromyography (EMG) sensor acquires the root mean square (RMS) value of the electromyographic signal during the user's maximal voluntary contraction (MVC, such as pronation or supination of the forearm when the user clenches their fist with maximum force), and calculates the normalized resting baseline of the EMG. For example, the RMS value of MVC. ≈0.10mV~0.15 mV, calculate The normalized resting baseline for electromyography (EMG) is approximately 5% to 10% of the maximum voluntary contraction value. This baseline value will be stored in non-volatile memory and used to define preset EMG activation thresholds. Different users have different skin impedances and degrees of muscle atrophy, resulting in different resting baselines for EMG. Each user's baseline is individually calibrated, with a unified judgment standard. Signals within 5% to 10% of the maximum voluntary contraction value (MVC) are considered pure noise to avoid interference and false triggering. Using MVC as a reference, even weak EMG signals from patients with mild hemiplegia and muscular dystrophy can be uniformly quantified and identified.

[0041] For the pressure sensor, steps S21 and S22 are as follows: S212: Record the baseline with the user's hand completely relaxed and fingers in a naturally extended position (i.e., no physical contact, no bending force). For example, the pressure baseline value (i.e., sensor noise floor) in the no-contact state is approximately 0.05N to 0.2N. That is, FSR baseline (no contact): 0.05N–0.2 N (noise floor).

[0042] S222: The user's maximum voluntary contraction fingertip force, calculated as the ratio of the normalized resting pressure baseline to the maximum voluntary contraction force. For example, the maximum voluntary contraction fingertip force. 30 N, the normalized resting pressure baseline is approximately 1% to 3% of the maximum voluntary contractile force, that is, .

[0043] For the bending sensor, steps S21 and S22 are as follows: S213: With the user's finger fully extended in the default position, set its angle to the bending baseline via calibration mapping. .

[0044] S223: The maximum voluntary bending angle of the user's finger, calculated as the ratio of the normalized resting bending baseline to the maximum voluntary bending angle. For example, 0° The normalized bending resting baseline is 90°. .

[0045] By calibrating and normalizing the specific values ​​mentioned above, the noise floor of each sensor and the physiological limits of the user can be accurately quantified. This allows for the calculation of scientific and reasonable personalized activation thresholds for users in different physiological states and at different stages of disease, effectively avoiding false triggering or failure to trigger due to fixed thresholds.

[0046] Furthermore, in the third embodiment of the upper limb exoskeleton control method of the present invention, S23 includes: S231: Preset electromyographic activation threshold (EMG) thr =EMG baseline +k1 (EMG MVC - EMG baseline ), of which EMG baseline EMG MVC These are the resting baseline and maximum voluntary contraction value of the electromyography sensor, respectively, and k1 is the first safety factor, where 5% < k1 < 20%.

[0047] Based on this, firstly, the (EMG) in the formula MVC - EMG baseline The system precisely defines the user's effective dynamic force range, using the resting baseline as the calculation starting point. Mathematically, it automatically eliminates environmental electromagnetic noise and the unique resting muscle spasm tension of hemiplegic patients, ensuring a response only to active incremental force exertion. Secondly, considering the high-frequency, low-amplitude, and low signal-to-noise ratio physical characteristics of surface muscle electrical signals, the first safety factor k1 is strictly limited to between 5% and 20%. If it is below 5%, minute motion artifacts or involuntary muscle twitches can easily exceed the threshold, leading to erroneous exoskeleton movements; if it is above 20%, patients with residual muscle weakness and nerve damage will be unable to trigger the device.

[0048] Specifically, due to The estimate is obtained .

[0049] S232: Preset pressure activation threshold F thr =F baseline +k2 (F MVC -F baseline ), where F baseline F MVC These represent the resting baseline and maximum spontaneous contraction value of the pressure sensor, respectively, with k2 being the second safety factor, where 2% < k2 < 12%. For example, F baseline It is 0.2 N. 30 N, k2 is 4.4%, then F thr =0.05F MVC .

[0050] Based on this, firstly, since pressure sensors (such as force sensors) measure direct physical mechanical contact, their signal stability is high and low-frequency noise is low. Therefore, the lower limit of the second safety factor k2 is set to 2%. This makes the threshold closer to the resting baseline, enabling extremely sensitive capture of the initial mechanical signal when the fingertip just makes slight contact with the object, significantly improving the sensitivity to weak grasping intentions. Secondly, the lower threshold range of 2% to 12% ensures that grasping intentions can be identified as early as possible and smoothly switched to the 'proportional grasping assistance' mode, so that the assisting force is linearly output with the slight increase of fingertip force, meeting the rehabilitation needs of fine motor skills such as pinching small objects, and avoiding abrupt changes in assisting force. Thirdly, the formula takes the resting baseline when the user is wearing gloves as the starting point, effectively eliminating the false contact force caused by glove pre-tightening force and sensor zero-point drift, preventing false triggering when there is no contact with the object.

[0051] It should be noted that surface muscle electromyography (SEMG) signals are weak bioelectrical signals, highly susceptible to interference from changes in skin impedance and motion artifacts, exhibiting large noise variance. Therefore, a relatively high lower limit (5%) is required to ensure anti-interference safety. In contrast, pressure sensors measure macroscopic physical and mechanical signals, which are stable and have a high signal-to-noise ratio. Therefore, a lower lower limit (2%) can be used to pursue ultimate trigger sensitivity. This differentiated threshold configuration based on the heterogeneous characteristics of multimodal sensors demonstrates the system's deep adaptation to complex physiological and physical signals, achieving synergistic control effects that cannot be achieved with a single fixed threshold or a threshold of the same proportion.

[0052] S233: Preset bending activation threshold θ thr =θ baseline +k3 (θ MVC -θ baseline ), where θ baseline θ MVC These represent the resting baseline and maximum spontaneous contraction value of the bending sensor, respectively, with k3 serving as the third safety factor: 4% < k3 < 8%. Based on this, firstly, to address the inherent mechanical hysteresis effect of the bending sensor and the flexor muscle spasm common in hemiplegic patients, the lower limit of k3 is set at 4%, effectively filtering out sensor hysteresis rebound noise and unconscious micro-tremors of the fingers; secondly, the upper limit is strictly controlled at 8%, ensuring that the system can respond instantly when the fingers produce slight but effective active flexion, avoiding the dragging sensation caused by delayed intervention of auxiliary force, and achieving smooth interaction with extremely low latency.

[0053] In this embodiment, the values ​​of the three safety factors k1, k2, and k3 are determined based on the noise characteristics and signal stability of each sensor. The electromyography (EMG) sensor signal has relatively high noise, therefore k1 ranges from 5% to 20%; the pressure sensor signal is relatively stable, so k2 ranges from 2% to 12%; the bending sensor signal falls between the two, so k3 ranges from 4% to 8%. By appropriately selecting the safety factors, noise-induced false triggering can be avoided while ensuring sensitivity, thus improving the reliability of the system.

[0054] In some embodiments, based on the first embodiment, a fourth embodiment of the upper limb exoskeleton control method of the present invention is proposed. Step S4, the step of extracting electromyographic temporal features from electromyographic signals, specifically includes: S411: The time-domain characteristics of electromyography, including root mean square and mean absolute value, are calculated using a sliding window. S412: Normalize the calculated electromyographic temporal features based on the maximum voluntary contraction value.

[0055] In this embodiment, a sliding window technique is used to extract temporal features from the electromuscular signal, effectively capturing the short-term activation characteristics of the muscle. The window duration is set to 100-250ms, which reflects the physiological response time of the muscle while meeting the requirements of real-time control. Normalization is performed based on the maximum voluntary contraction value, ensuring the comparability of feature values ​​from different users or different usage sessions.

[0056] Furthermore, based on the first embodiment, a fifth embodiment of the upper limb exoskeleton control method of the present invention is proposed, which further includes the following steps before the step of extracting electromyographic time-domain features from muscle electromotive force signals: S410: The electromyography signal is passed sequentially through a differential instrumentation amplifier with a gain of 500–1000, an analog bandpass filter of 20 Hz–450 Hz, a notch filter of 50 / 60 Hz, and an analog-to-digital converter with a sampling frequency of at least 1000 Hz.

[0057] Specifically, for electromyography (EMG) signal processing, the raw EMG signals acquired from the forearm muscles undergo multi-stage signal conditioning and feature extraction. The raw EMG signals are first amplified by a differential instrumentation amplifier with a gain of 500-1000 ohms to enhance the low-amplitude bioelectrical signals. The amplified signal then passes through an analog bandpass filter with a passband range of 20Hz-450Hz to filter out motion artifacts and high-frequency noise. A 50Hz notch filter is subsequently applied to eliminate power frequency interference.

[0058] The filtered signal is digitized by an analog-to-digital converter with a sampling frequency of at least 1000Hz, and the digital electromyography (EMG) signal is processed in real time by an embedded controller. In the feature extraction stage, time-domain features are calculated within a sliding window, typically 100-250ms, including root mean square (RMS) value, mean absolute value (MAV), waveform length, and zero-crossing rate. RMS and MAV are primarily used in real-time applications due to their robustness and low computational cost. The extracted features are normalized based on the maximum spontaneous contraction value obtained during the calibration stage. Finally, a smoothing operation, such as moving average filtering, is applied to reduce signal fluctuations before the processed features are sent to the pattern recognition stage.

[0059] In one optional implementation, the original electromyographic signal amplitude is approximately ±1mV. After passing through a fourth-order Butterworth bandpass filter with a passband of 20Hz-450Hz and a 50Hz notch filter, it undergoes full-wave rectification and moving average smoothing with a window duration of 100ms. The processed resting-state RMS value is approximately 0.01mV, and the activated-state RMS value is approximately 0.08mV.

[0060] Based on the first embodiment, a sixth embodiment of the upper limb exoskeleton control method of the present invention is proposed, which further includes the following steps before the step of extracting hand pressure features from pressure signals: S421: Passes the pressure signal through a low-pass filter with a cutoff frequency of 5–10 Hz, calibrates it, converts it to standard force units, and normalizes it.

[0061] In this embodiment, optionally, the force normalization formula is: the normalized force value is equal to the fingertip force divided by the maximum voluntary contraction force.

[0062] For pressure signal processing, force-sensitive sensors measure the contact force in the fingertip and palm areas. Each force-sensitive sensor operates under a constant voltage supply, and current variations are converted into voltage signals by a signal conditioning circuit. The raw force-sensitive sensor signal is passed through a low-pass filter with a cutoff frequency of 5-10 Hz to remove high-frequency noise and transient spikes. The filtered signal is calibrated and mapped to an estimated fingertip force value according to a calibration curve. The force signal is normalized based on the maximum voluntary contraction force.

[0063] For the bending sensor processing, a bending sensor mounted on the back of the finger measures the resistance change corresponding to finger bending. This resistance change is converted into a voltage signal via a voltage divider circuit. After noise reduction by a low-pass filter, the signal is calibrated and mapped to the finger bending angle. The bending angle is normalized based on the maximum voluntary bending angle.

[0064] Optionally, the normalization formula for the bending angle is: the normalized value of the bending angle is equal to the finger bending angle divided by the maximum voluntary bending angle.

[0065] All sensor signals—electrical muscle signals, force-sensitive resistor signals, and bending signals—are time-synchronized and processed in real time within the embedded controller. Each modality is independently filtered and normalized before being transmitted to the threshold judgment and pattern recognition stages. This modular preprocessing ensures noise robustness and allows for independent and joint interpretation of user intent.

[0066] In one optional embodiment, the original voltage signal of the force-sensitive resistor ranges from 0.2 to 2.5V. After being low-pass filtered with a cutoff frequency of 10Hz, it is mapped to a force value according to a calibration curve; for example, 1.2V corresponds to 5N. The original voltage signal of the bending sensor ranges from 1.0 to 3.0V. After being low-pass filtered with a cutoff frequency of 5Hz, it is calibrated and mapped to an angle value; for example, 2.0V corresponds to 45 degrees.

[0067] Based on the first embodiment, a seventh embodiment of the upper limb exoskeleton control method of the present invention is proposed. In this embodiment, a hand electromyography (EMG) unit for acquiring hand muscle intentions is further included. The hand EMG unit includes a first EMG unit, a second EMG unit, and a third EMG unit. The first EMG unit collects surface EMG signals of the superficial flexor digitorum muscles, the second EMG unit collects surface EMG signals of the deep flexor digitorum muscles, and the third EMG unit collects surface EMG signals of the extensor digitorum muscles. The first, second, and third EMG units are cross-validated with pressure sensors and bending sensors to improve the accuracy of hand movement recognition. Step S5 includes: S51: When the intention of the hand muscles is consistent with the intention of the hand movement Figure 1 Upon verification of the user's intention to move, proceed to the action mode step.

[0068] S52: When the intention of the hand muscles is inconsistent with the intention of the hand movement, it is judged as a false trigger.

[0069] S53: When the comparison results of hand pressure characteristics and preset pressure activation threshold are inconsistent with the comparison results of finger bending characteristics and preset bending activation threshold, it is verified that there is no user movement intention.

[0070] In this embodiment, the first case is the consistency case, i.e. the strong intent case.

[0071] The surface electromyography (EMG) signal was 0.05mV, exceeding the preset hand muscle activation threshold, indicating a finger flexion intention. The flexion angle was 15 degrees, exceeding the 9-degree activation threshold. The force-sensitive resistor pressure was 2N, exceeding the 1.5N activation threshold. With the EMG signal, force-sensitive resistor, and flexion sensor all confirming a movement intention, the system determined it to be a strong intention, and the actuator was activated in proportional-assisted mode.

[0072] The second scenario is an inconsistency situation, i.e. a conflict case.

[0073] Example 1: The surface electromyography (EMG) signal is 0.11mV, which is greater than the preset threshold, indicating an intention to flex the finger; however, the flexion angle is 2 degrees, which is less than the activation threshold; the force-sensitive resistor pressure is 0.1N, which is also less than the activation threshold; meaning no physical movement is detected. In this case, because the force-sensitive resistor and flexion sensor are not activated, the system determines that there is no physical movement matching, rejects the false trigger, and the actuator does not move. Therefore, when the hand EMG unit is used alone, it may be affected by noise, muscle crosstalk, or unexpected contractions; hand pressure characteristics and finger flexion characteristics serve as confirmation signals to improve the reliability and safety of the system.

[0074] Example 2: Unintentional Contact. The surface electromyography (EMG) signal is below a preset threshold, the bending angle is 5 degrees (less than the activation threshold), and the force-sensitive resistor pressure is 2N, thus contact is detected. This is considered accidental contact. Because the bending angle does not meet the activation condition, although contact force is detected, the system still determines there is no intention to move and does not trigger the auxiliary action. The above weighted decision-making mechanism effectively avoids accidental activation of the EMG signal and false triggering of external contact, improving the system's safety and reliability.

[0075] Optionally, the surface electromyography (EMG) signals from the first, second, and third EMG units typically appear slightly earlier than the finger flexion angle signal (Flex signal) and the hand pressure signal (FSR signal). The Flex signal appears slightly earlier than the FSR signal because muscle activation precedes actual finger or hand movement, and finger flexion then generates contact force. This method evaluates within a predefined short temporal correlation window (e.g., 0.5–2 seconds). This is because human neuromuscular and mechanical responses introduce a natural delay between muscle activation and observable movement. Time-window-based correlation improves system robustness while maintaining real-time performance. If a corresponding Flex and / or FSR activation is detected within this time window following EMG activation, the system considers the event a valid intent; otherwise, it is considered a false activation.

[0076] Specifically, the surface electromyography (EMG) signal processing methods for the first, second, and third EMG units are the same as those for the muscle EMG signals of the EMG sensor, and will not be elaborated further here. Optionally, in step S2, the resting baseline of the hand EMG unit in the user's resting state is acquired; the maximum voluntary contraction value of the hand EMG unit during the user's maximum voluntary contraction is acquired; and a preset hand muscle activation threshold is determined based on the resting baseline and the maximum voluntary contraction value of the hand EMG unit. Optionally, in step S4, the temporal features of the hand muscles are extracted from the surface EMG signal, such as filtering, amplifying, and extracting features from the surface EMG signal, converting the raw sensor data into feature values ​​that can be used for intention recognition. Optionally, in step S5, the temporal features of the hand muscles are compared with the preset hand muscle activation threshold to obtain the hand muscle intention.

[0077] Furthermore, based on the first embodiment, an eighth embodiment of the upper limb exoskeleton control method of the present invention is proposed. In this embodiment, step S6 includes: S61: A support vector machine classifier is used to receive the temporal features of muscles and identify the forearm movement categories of the upper limb movement patterns. The forearm movement categories include forearm pronation, forearm supination, and forearm relaxation. S62: A rule-based multimodal fusion strategy is adopted to identify the hand movement categories of upper limb movement patterns. The hand movement categories include hand opening, proportional grasping assistance, and maximum grasping assistance.

[0078] In this embodiment, the processed multimodal features are interpreted through a hybrid decision framework to identify user intent, which combines classification-based forearm motion recognition and rule-based grasping intent fusion.

[0079] Specifically, for proportional gripping assistance, this method configures multiple discrete levels of assistance force. For example, the gripping assistance force of the smart glove can be set to four levels as needed: 25%, 50%, 75%, and 100% of the maximum assistance force. In practical applications, the corresponding assistance force level is dynamically matched based on the physical properties of the object being gripped (such as weight, stiffness, and deformability). For instance, when gripping easily deformable objects like empty water bottles, 25% assistance force is used to prevent the bottle from being crushed; while when gripping heavier objects like full water bottles, 50% or higher assistance force is used to prevent the object from slipping.

[0080] It should be noted that this application does not strictly limit the number or specific values ​​of the assistive level classifications. The four fixed levels mentioned above are merely examples. In actual rehabilitation or assistive scenarios, the assistive level can be divided into any number of discrete levels based on the patient's functional status and specific task requirements. For example, the system can use assistive levels in 10% increments (10%, 20%, ..., 100%), or even more refined 5% increments (5%, 10%, 15%, ..., 100%).

[0081] The advantages of using quantified discrete levels of assistance are: it effectively improves the patient's ability to actively control grip force and reduces unintended force fluctuations. Smaller force increments (i.e., more discrete intervals) provide finer grip force adjustment, allowing users to more accurately match assistive force to the task at hand. The control algorithm maps the user's grip force intention to predefined assist intervals based on real-time measurements of fingertip force and finger curvature, with each interval corresponding to a specific percentage of assist output from the actuator. As the number of assist intervals increases, the resolution of grip force assistance and the accuracy of intention-based force control also improve.

[0082] Furthermore, in an alternative implementation, once the user has been sufficiently trained and adapted to the system, the control strategy can smoothly transition from a discrete assistance level to a continuous proportional assistance mode. In this continuous mode, the actuator's output force is proportional to the real-time grip force intent detected by the algorithm. This design, while maintaining the same underlying sensing and control hardware framework, provides a smoother, more natural, and biomechanically compliant grip assistance experience.

[0083] Specifically, for forearm pronation and supination classification based on electromyography (EMG) signals, the processed EMG features are used to identify the user's movement intention through supervised classification. A support vector machine (SVM) classifier is employed due to its high accuracy and good applicability in low-dimensional EMG feature spaces. During the calibration phase, labeled EMG data corresponding to different movement categories are collected, including forearm pronation, forearm supination, and relaxation. Extracted features, such as root mean square (RMS) and mean absolute value (MAV), form feature vectors, which are used to train the classifier offline or during the initial calibration phase. In real-time operation, the classifier receives the input feature vectors and assigns them to one of the predefined movement categories. To improve robustness, a decision smoothing strategy is employed, such as majority voting or confidence thresholding of continuous windows, to ensure that transient noise does not trigger unexpected movements. The movement category recognition results are mapped as follows: pronation triggers counter-clockwise motor drive, supination triggers clockwise motor drive, and relaxation does not trigger drive, i.e., remains idle. This classification output is then transmitted to the drive command phase for execution.

[0084] For grasping intent detection, a rule-based multimodal fusion strategy is employed, using fingertip force and finger bending angle. The controller applies dual thresholds and logic: activation thresholds include a pressure activation threshold and a bending activation threshold. The decision rules are as follows: if the fingertip force is less than the pressure activation threshold, or the finger bending angle is less than the bending activation threshold, it is determined that there is no grasping intent and the controller is in release mode; if the fingertip force is greater than or equal to the pressure activation threshold and less than the maximum autonomous pressure value, and the finger bending angle is greater than or equal to the bending activation threshold and less than the maximum autonomous bending value, it is determined that proportional grasping assistance is enabled; if the fingertip force is greater than or equal to the maximum autonomous pressure value, and the finger bending angle is greater than or equal to the maximum autonomous bending value, it is determined that maximum grasping assistance is enabled. This rule-based fusion strategy ensures that grasping assistance is triggered only when active finger movement and contact force are detected simultaneously, thereby reducing false triggers.

[0085] The ultimate intent is determined by combining two pathways: electromyography (EMG) classification results control forearm pronation and supination movements, while force-sensitive resistors and flexion sensors are fused to control hand grasping. These two control flows can operate independently or synchronously to achieve coordinated hand and wrist movements.

[0086] Furthermore, step S62 specifically includes: S621: When the hand pressure feature is less than the preset pressure activation threshold and the finger bending feature is less than the preset bending activation threshold, the hand action category is hand opening; S622: When the hand pressure characteristics are greater than or equal to the preset pressure activation threshold and less than the maximum autonomous pressure value, and the finger bending characteristics are greater than or equal to the preset bending activation threshold and less than the maximum autonomous bending value, the hand movement category is proportional grasping assistance. S623: When the hand pressure characteristics are greater than or equal to the maximum voluntary pressure value and the finger bending characteristics are greater than or equal to the maximum voluntary bending value, the hand movement category is maximum grasping assistance.

[0087] In this embodiment, to further verify the effectiveness of the above-mentioned intent recognition method, the accuracy of intent recognition for different sensing modalities is analyzed below.

[0088] To verify the effectiveness of the above control methods, human trials were conducted. Specifically, eight healthy adult subjects were recruited for the trial, and their basic information is shown in Table 2.

[0089]

[0090] Table 2 Basic Information of Subjects During the experiment, each subject wore a soft robotic glove and a forearm pronation / supination exoskeleton, performing hand grasping and forearm pronation / supination movements according to the control method described above. The intention-based grasping experiment demonstrated satisfactory performance among the subjects. Each subject, wearing the exoskeleton and using the multimodal intention control algorithm, completed multiple consecutive trials using everyday objects. The average accuracy for each object ranged from 82% to 91% among the subjects, with an overall average accuracy of approximately 87% ± 9.9%. The stacked bar chart indicates that most trials were successfully completed with a low number of failures, demonstrating that the system can reliably interpret motor intentions and provide timely assistance. These results highlight the responsiveness and reliability of the exoskeleton in recognizing motor intentions in simulated rehabilitation tasks.

[0091] It is important to note that the different sensing modalities in the system serve different functional purposes, rather than acting as redundancy. Specifically, surface electromyography (SEMG) signals are used to decode forearm pronation and supination intentions, while the fusion of force-sensitive resistors and flexion sensors is used to detect hand grasping intentions. Due to this functional separation, it is not possible to directly compare the cross-modal intention recognition accuracy of all sensor combinations, as each sensor is not designed to detect the same type of movement, but rather is evaluated within its own functional domain.

[0092] Furthermore, the employed logic-based sensor fusion strategy enhances robustness and safety, and reduces false triggering, by ensuring that assistance is triggered only when both the bending and force sensing channels detect a consistent intent. Therefore, this invention emphasizes task-specific optimization, where each sensing modality is selected and validated based on its intended motion detection function, thereby achieving a generally reliable and efficient human-computer interaction system.

[0093] This invention also proposes an upper limb exoskeleton. The upper limb exoskeleton includes a memory, a processor, and an exoskeleton control program stored in the memory and executable on the processor. The exoskeleton control program is configured to implement the steps of the upper limb exoskeleton control method described above.

[0094] For details, please refer to Figure 2 The upper limb rehabilitation exoskeleton also includes a hardware box 1, an upper arm fixation mechanism 2, an elbow joint support 3, a forearm fixation mechanism 4, a wearable rehabilitation glove 5, a first cable 6, a second cable 7, and a wrist joint movement mechanism 8. The elbow joint support 3 is hinged to the upper arm fixation mechanism 2, and the forearm fixation mechanism 4 is fixedly mounted on the elbow joint support 3. With the help of the elbow joint support 3, the forearm fixation mechanism 4 can rotate relative to the upper arm fixation mechanism 2 when force is applied. It is understood that the upper arm fixation mechanism 2 is used to bind to the upper arm, the forearm fixation mechanism 4 is used to bind to the forearm, and the elbow joint support 3 is positioned corresponding to the elbow joint. The wrist joint movement mechanism 8 is fixedly mounted on the forearm fixation mechanism 4, and the rehabilitation glove 5 is connected to the hardware box 1 via the second cable 7. One end of the first cable 6 is connected to the hardware box 1, and the other end is connected to the wrist joint movement mechanism 8. The hardware box 1 drives the first cable 6 to extend and retract to drive the C-shaped guide rail 82 to slide relative to the slide block 81. One end of the second cable 7 is connected to the hardware box 1, and the other end is connected to the rehabilitation glove 5. The hardware box 1 drives the second cable 7 to extend and retract, thereby driving the fingers of the rehabilitation glove 5 to perform extension and flexion movements. It is understood that the hardware box 1 integrates a wrist joint drive device 9 and a hand drive device (not shown in the figure). The wrist joint drive device is connected to the first cable 6 to drive the extension and retraction of the first cable 6. The hand drive device is connected to the second cable 7 to drive the extension and retraction of the second cable 7. It is also understood that the hardware box 1 integrates a control unit (not shown in the figure) and a power supply unit (not shown in the figure). The hardware box 1 is hung on the waist or back of the body. Integrating the drive device inside the hardware box 1 reduces the weight of other upper limb mechanisms.

[0095] Please see Figure 3The upper arm fixing mechanism 2 includes an upper arm support 21, an upper arm sleeve 22, and an upper arm strap 23. One end of the upper arm support 21 is hinged to the elbow joint support 3, and the other end of the upper arm support 21 is fixedly connected to the upper arm sleeve 22. Specifically, the upper arm sleeve 22 is an open arc plate structure, and the upper arm strap 23 is fixed to the upper arm sleeve 22. It is understood that the upper arm strap 23 is a Velcro structure, but it is not limited thereto. In this embodiment, there are two upper arm straps 23, but it is not limited thereto.

[0096] Please continue to refer to this. Figure 3 The forearm fixation mechanism 4 includes a forearm support 41, a forearm sleeve 42, and a forearm strap 43. The forearm support 41 is fixedly connected to the other end of the elbow joint support 3. The forearm sleeve 42 is fixedly connected to the forearm support 41; specifically, the forearm sleeve 42 is an open, arc-shaped plate structure. The forearm strap 43 is fixed to the forearm sleeve 42; it is understood that the forearm strap 43 is a Velcro structure, but not limited thereto. Preferably, the forearm fixation mechanism 4 also includes an electromyography (EMG) sensor 44, which is embedded in the forearm support 41 to detect forearm muscle activity, thereby estimating the user's intention during wrist joint movement.

[0097] Specifically, the rehabilitation glove 5 integrates a flexible bending sensor at the top of the finger and a piezoresistive force sensor at the fingertip and middle phalanx. Through the pressure sensor and the flexible bending sensor, it actively senses the intention to move, enabling the second drive device to respond immediately and drive the second cable 7 to assist the fingers in flexion and extension movements. The structure and principle of the rehabilitation glove 5 are well known to those skilled in the art and will not be described in detail here.

[0098] Please refer to the following: Figures 4 to 5 The wrist joint motion mechanism 8 is shown. The wrist joint motion mechanism 8 includes a slide 81, a C-shaped guide rail 82, and a cable guide assembly (not shown). Furthermore, the wrist joint motion mechanism 8 may also include any one of a cable fixing connector 84, a cable sheath 85, a limiter 86, a passive element 87, and a hand support sheath 88.

[0099] Please see Figure 5The slide 81 has a first wiring channel 811 inside for the first cable 6 to pass through. A C-shaped guide rail 82 is slidably mounted on the slide 81, and the C-shaped guide rail 82 has a second wiring channel 821 inside for the first cable 6 to pass through. The wrist joint drive device 9 drives the first cable 6 to extend and retract, thereby driving the C-shaped guide rail 82 to slide relative to the slide 81. A cable guide assembly is located inside the slide 81. The first cable 6 passes through the slide 81 after being wound around the cable guide assembly, and is parallel to the length direction of the slide 81. It can be understood that the length direction of the slide 81 is parallel to the length direction of the forearm. With the help of the cable guide assembly, the exit direction of the first cable 6 is forcibly changed. The first cable 6 no longer intersects the forearm at an angle, but passes through parallel to the length direction of the slide 81 (i.e., simulating the axis direction of the human forearm). This design completely solves the problem of physical friction and snagging between the first cable 6 and the arm or exoskeleton support, ensuring the smoothness of movement. Because the first cable 6 is constrained and guided by the cable guide assembly before exiting the slide 81, unnecessary bending and lateral tension at the exit point are avoided. This makes the power transmission from the drive controller to the wrist joint more direct and less lossy, reduces the frictional resistance of the first cable 6, and thus improves the upper limb rehabilitation exoskeleton's ability to capture subtle movement intentions and the accuracy of overall control, making rehabilitation training movements more delicate and precise. Traditional wrist joint motion mechanisms without a guide structure are prone to frequent bending or wear of the first cable 6 at the exit point, posing a risk of breakage with long-term use. The wrist joint motion structure of this invention uses a cable guide assembly to provide a smooth transition path for the first cable 6, effectively preventing fatigue damage caused by excessive bending of the first cable 6, and significantly improving the durability and safety of the first cable 6 and the entire wrist joint motion structure.

[0100] Please see Figure 5 There are two limiters 86, which are respectively located at both ends of the C-shaped guide rail 82. With the help of the limiters 86, the sliding stroke of the C-shaped guide rail 82 is limited to prevent the C-shaped guide rail 82 from slipping off the slide block 81.

[0101] Specifically, the slide 81 has two first wiring channels 811 inside, which are spaced apart from each other in the vertical direction. For easy distinction, they will be referred to as the upper wiring channel 811a and the lower wiring channel 811b. It is understood that both first wiring channels 811 are equipped with cable guide components.

[0102] Please see Figure 5 and Figure 14For example, the pronation cable 6a passes sequentially through the second wiring channel 821 and the lower wiring channel 811b, and under the guidance of the cable guide assembly 83, it finally exits the slide 81 in a direction parallel to its length, and finally connects to the wrist joint drive device 9. Similarly, the supination cable 6b passes sequentially through the second wiring channel 821 and the upper wiring channel 811a, and under the guidance of the cable guide assembly, it finally exits the slide 81 in a direction parallel to its length, and finally connects to the wrist joint drive device 9. Therefore, when the user's movement intention is detected, the wrist joint drive device 9 drives the pronation cable 6a and the supination cable 6b to move in opposite directions, which allows the C-shaped guide rail 82 to slide relative to the slide 81.

[0103] Please see Figures 6 to 13 The present invention demonstrates the wrist joint drive device 9 provided by the present invention.

[0104] The wrist joint drive device 9 includes a support member 91, a driver 92, a drive shaft 93, a roller 94, and a tension adjustment mechanism 95.

[0105] The actuator 92 is mounted on the support member 91; the drive shaft 93 is connected to the actuator 92 and rotatably supported on the support member 91. The actuator 92 is preferably a Dynamixel servo motor, capable of precisely controlling the forward and reverse rotation and rotation angle of the drive shaft 93, thereby precisely controlling the pronation and supination amplitude of the wrist joint. Specifically, the actuator 92 and the drive shaft 93 are connected by a coupling 96. Specifically, the end of the drive shaft 93 is supported in the support member 91 by a bearing 99.

[0106] The reel 94 is fitted onto the drive shaft 93 and rotates synchronously with it. Specifically, the reel 94 has an anti-rotation hole 941 in the axial direction for the drive shaft 93 to pass through. The cross-section of the drive shaft 93 passing through the anti-rotation hole 941 is adapted to the anti-rotation hole 941. For example, the anti-rotation hole 941 is D-shaped. The anti-rotation hole 941 and the D-shaped cross-section of the drive shaft 93 are matched to achieve circumferential fixation of the reel 94 and the drive shaft 93, ensuring that the reel 94 rotates synchronously without relative slippage when the drive shaft 93 rotates, thus ensuring the stability of power transmission.

[0107] The reel 94 is provided with a winding section 942 for winding the first cable 6, realizing the unwinding and rewinding of the first cable 6. Specifically, the reel 94 is provided with two winding sections 942 for winding the first cable 6 in opposite directions, and the winding directions of the first cable 6 on the two winding sections 942 are opposite. Optionally, the middle part of the first cable 6 is anchored on the reel 94, and the two cable sections separated by the midpoint are wound on the winding sections 942 in opposite winding directions to avoid the first cables 6 located on the two winding sections 942 from affecting each other. Specifically, the first cable 6 is divided into a pronation cable 6a and a supination cable 6b with the midpoint as the boundary. When the reel 94 rotates in the forward direction, the pronation cable 6a is wound up and the supination cable 6b is released, driving the wrist joint to perform a pronation movement. When the reel 94 rotates in the reverse direction, the supination cable 6b is wound up and the pronation cable 6a is released, driving the wrist joint to perform a supination movement. The wrist joint is driven bidirectionally by a single reel 94 and a single first cable 6, resulting in a more compact structure and lighter weight.

[0108] A better approach is to combine Figure 10 , Figure 11 and Figure 12 The winding wheel 94 has a built-in transition channel 943 in its middle section for the middle portion of the first cable 6 to pass through. Both ends of the transition channel 943 lead to two winding sections 942, each with a through hole 944 communicating with the transition channel 943. The middle portion of the first cable 6 passes through the transition channel 943, and both ends emerge from their corresponding through holes 944 and are then wound in opposite directions around the winding section 942, achieving centered anchoring and orderly winding of the cable. Specifically, the winding wheel 94 also has threaded fixing holes 945 for fastener installation. The winding section 942 communicates with the outside through the threaded fixing holes 945. Fasteners are used to anchor the first cable 6 located within the inner layer of the winding section 942. The middle portion of the first cable 6 is locked in place by fasteners (such as screws) engaging with the threaded fixing holes 945, preventing the first cable 6 from slipping during winding and ensuring reliable transmission. For example, the transition channel 943 is a spiral channel. The spiral channel has a continuous and smooth transition without sharp corners or right-angle bends, making it easier for the first cable 6 to run. The first cable 6 passes through the spiral channel from the center of the reel 94, and the force is symmetrical. The reel 94 does not become eccentric or vibrate when it rotates.

[0109] The tension adjustment mechanism 95 is mounted on the support member 91 and presses against the first cable 6 that has left the winding part 942 to adjust the tension of the first cable 6. This compensates for slack caused by repeated stretching and contraction of the first cable 6, long-term use, or temperature changes, ensuring that the first cable 6 always maintains optimal tension. This eliminates transmission lag, inaccurate response, and mechanism play caused by slack, thereby significantly improving the accuracy, immediacy, and reliability of motion control. The tension adjustment mechanism 95 is linked to the first cable 6 to adjust its tension. In this embodiment, there are two tension adjustment mechanisms 95, one for the pre-spinning cable 6a and the other for the post-spinning cable 6b, independently adjusting the tension of both cables to meet the tension requirements of bidirectional motion.

[0110] Specifically, the tension adjustment mechanism 95 includes an adjustment block 951 and an adjustment rod 952. The adjustment block 951 is located in the output direction X of the first cable 6, and its position is adjustable on the support member 91 in the output direction X. The adjustment rod 952 is used to push the adjustment block 951 to move in the output direction X. Therefore, when the first cable 6 leaves the reel 94, it needs to pass around the adjustment block 951. The adjustment block 951 and the adjustment rod 952 push the adjustment block 951 closer to the reel 94 in the output direction X, causing the first cable 6 to tighten, lengthen, and increase its bending amplitude, thus increasing the tension force on the first cable 6. Conversely, when the adjustment block 951 moves away from the reel 94 in the output direction X, the tension force on the first cable 6 decreases. Specifically, the adjusting rod 952 has external threads, and the adjusting block 951 has a threaded hole 9512 for the adjusting rod 952 to pass through. The head of the adjusting rod 952 abuts against the side of the adjusting block 951 away from the reel 94, preventing the adjusting block 951 from disengaging from the adjusting rod 952. The end of the adjusting rod 952 is threadedly installed on the support member 91. When the adjusting rod 952 is rotated, the rotational motion of the adjusting rod 952 is converted into the linear motion of the adjusting block 951. The linear motion of the adjusting block 951 causes a change in the tension of the first cable 6. For example, rotating the adjusting rod 952 clockwise moves the adjusting block 951 closer to the reel 94, increasing the tension on the first cable 6; rotating the adjusting rod 952 counterclockwise moves the adjusting block 951 away from the reel 94, reducing the tension on the first cable 6. The operation is simple and convenient, and the tension can be precisely adjusted.

[0111] The wrist joint drive device 9 of the present invention further includes an elastic element 97. Optionally, the elastic element 97 is sleeved on the adjusting rod 952 to achieve circumferential limiting of the elastic element 97. One end of the elastic element 97 abuts against the adjusting block 951 and the other end abuts against the supporting member 91. The elastic element 97 is preferably a compression spring. When the wrist joint motion mechanism is subjected to a sudden impact load, the impact force is transmitted to the adjusting block 951 through the first cable 6. The adjusting block 951 compresses the elastic element 97, and the elastic element 97 undergoes elastic deformation to absorb the impact load, preventing the impact force from acting directly on the first cable 6, the reel 94, and the driver 92, effectively protecting the transmission components, extending the service life of the device, and at the same time buffering the vibration caused by the impact, improving the patient's comfort.

[0112] Specifically, the extension and retraction direction of the elastic element 97 is consistent with the output direction X, which allows the impact load to be directly transmitted along the output direction X and absorbed by the elastic element 97, without lateral force component or friction loss.

[0113] Specifically, the adjusting block 951 has a countersunk hole 9513 on the side near the elastic member 97, and the end of the elastic member 97 is embedded in the countersunk hole 9513, which can achieve precise centering of the elastic member 97, prevent deviation, prevent dislodgement, and prevent lateral pressure bending.

[0114] The wrist joint drive device 9 also includes a guide rod 98 that provides guidance for the movement of the adjustment block 951. The length direction of the guide rod 98 is consistent with the output direction X. The guide rod 98 is fixed on the support member 91. The adjustment block 951 is provided with a guide hole 9511 for the guide rod 98 to pass through. The guide rod 98 is arranged parallel to the adjustment rod 952 to ensure that the adjustment block 951 does not deflect during linear movement, thus ensuring the accuracy of tension adjustment and the smoothness of the movement of the adjustment block 951.

[0115] Electromyography (EMG) signals are highly dependent on individual differences and are easily affected by electrode placement, skin impedance, and noise. They are also often unreliable in patients with muscle weakness or paralysis.

[0116] Multimodal sensing technology combines two types of sensors to improve response speed and reliability. For example, by embedding a pressure sensor in the fingertip and placing a bending sensor along the finger segment, it is possible to detect both contact force and joint movement

[69] . This hybrid sensing system can achieve context awareness and adaptive control, which is especially beneficial for users with some autonomous motor abilities.

[0117] A related study indicated:

[75] This suggests that the estimated force exceeds 5% of the maximum voluntary contraction (MVC) fingertip force, and the user can withstand this force for at least one hour without significant fatigue. It is important to understand that the specific degree of finger flexion varies depending on factors such as hand anatomy, dexterity, and joint health. However, the fingers tend to bend considerably when people clench their fists. In this case, the maximum voluntary contraction (MVB) may reach 90° or higher for some individuals. There are differences in maximum muscle strength, maximum voluntary contraction, and estimated force among older adults and those with weaker muscles. Therefore, we can modify formula (16) by considering appropriate variable values ​​selected based on individual capabilities.

[0118] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0119] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0120] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0121] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for controlling an upper limb exoskeleton, characterized in that, Includes the following steps: Signal acquisition: Electromyography (EMG) signals from the forearm surface, pressure signals from the hand, and finger flexion angle signals are acquired using an EMG sensor, a pressure sensor, and a flexion sensor, respectively. Signal processing: Extracting electromyographic time-domain features from the electromyographic signals, extracting hand pressure features from the pressure signals, and extracting finger bending features from the bending signals; Intent recognition: The electromyographic temporal features are compared with a preset electromyographic activation threshold to obtain the forearm muscle movement intent; The hand pressure characteristics and finger bending characteristics are compared with preset pressure activation thresholds and preset bending activation thresholds, respectively, to obtain the hand movement intention; Based on the forearm muscle movement intention and the hand movement intention, the user's movement intention is identified; Action pattern: The user's movement intention is classified into action patterns to obtain upper limb action patterns, which include at least one of hand opening, hand grasping, forearm pronation, and forearm supination; Drive command: Based on the upper limb movement pattern, send corresponding drive commands to the actuators of the upper limb exoskeleton.

2. The upper limb exoskeleton control method as described in claim 1, characterized in that, Prior to the signal acquisition step, the method further includes threshold setting, which includes: The resting baselines of the electromyography sensor, the pressure sensor, and the flexion sensor were collected respectively under the user's resting state; Collect the maximum spontaneous contraction value of each sensor when the user experiences maximum spontaneous contraction; Based on the resting baseline and maximum voluntary contraction value of each sensor, the preset electromyographic activation threshold, preset pressure activation threshold, and preset flexion activation threshold are determined respectively.

3. The upper limb exoskeleton control method as described in claim 2, characterized in that, The step of determining the preset electromyographic activation threshold, preset pressure activation threshold, and preset flexion activation threshold based on the resting baseline and maximum voluntary contraction value of each sensor includes: Preset electromyography activation threshold (EMG) thr =EMG baseline +k1 (EMG MVC - EMG baseline ), of which EMG baseline EMG MVC These are the resting baseline and maximum voluntary contraction value of the electromyography sensor, respectively, with k1 being the first safety factor, where 5% < k1 < 20%. Preset pressure activation threshold F thr =F baseline +k2 (F MVC -F baseline ), where F baseline F MVC Here, k1 represents the resting baseline and maximum spontaneous contraction value of the pressure sensor, respectively, and k2 is the second safety factor, where 2% < k2 < 12%. Preset bending activation threshold θ thr =θ baseline +k3 (θ MVC -θ baseline ), where θ baseline θ MVC These are the resting baseline and maximum spontaneous contraction value of the bending sensor, respectively, and k3 is the third safety factor, where 4% < k3 < 8%.

4. The upper limb exoskeleton control method as described in claim 1, characterized in that, In the signal processing step, the step of extracting electromyographic time-domain features from the electromyographic signal specifically includes: The time-domain characteristics of electromyography, including root mean square and mean absolute value, were calculated using a sliding window method. The calculated electromyographic temporal features were normalized based on the maximum voluntary contraction value.

5. The upper limb exoskeleton control method as described in claim 1, characterized in that, Prior to the step of extracting electromyographic time-domain features from the electromyographic signal, the following steps are also included: The electromyography signal is passed sequentially through a differential instrumentation amplifier with a gain of 500–1000, an analog bandpass filter of 20 Hz–450 Hz, a notch filter of 50 / 60 Hz, and an analog-to-digital converter with a sampling frequency of at least 1000 Hz.

6. The upper limb exoskeleton control method as described in claim 1, characterized in that, Prior to the step of extracting hand pressure features from the pressure signal, the following steps are also included: The pressure signal is passed through a low-pass filter with a cutoff frequency of 5–10 Hz, calibrated, converted to standard force units, and normalized. And / or, prior to the step of extracting finger bending features from the bending signal, the steps include: passing the bending signal through a low-pass filter, calibrating, converting it to standard angle units, and normalizing it.

7. The upper limb exoskeleton control method as described in claim 1, characterized in that, It also includes a hand electromyography (EMG) unit for acquiring hand muscle intentions. The hand EMG unit includes a first EMG unit, a second EMG unit, and a third EMG unit. The first EMG unit acquires surface EMG signals of the flexor digitorum superficialis, the second EMG unit acquires surface EMG signals of the flexor digitorum profundus, and the third EMG unit acquires surface EMG signals of the extensor digitorum. In the intent recognition step, when the hand muscle intent is consistent with the hand movement intent, there is a user movement intent, and the process proceeds to the action mode step. When the intention of the hand muscles is inconsistent with the intention of the hand movement, it is determined to be a false trigger; When the comparison result of the hand pressure feature with the preset pressure activation threshold and the comparison result of the finger bending feature with the preset bending activation threshold are inconsistent, it is concluded that there is no user movement intention.

8. The upper limb exoskeleton control method as described in claim 1, characterized in that, In the action classification step, a support vector machine classifier is used to receive the temporal features of the muscles and identify the forearm action category of the upper limb action pattern. The forearm action category includes forearm pronation, forearm supination and forearm relaxation. A rule-based multimodal fusion strategy is adopted to identify the hand movement categories of upper limb movement patterns, wherein the hand movement categories include hand opening, proportional grasping assistance, and maximum grasping assistance.

9. The upper limb exoskeleton control method as described in claim 8, characterized in that, The rule-based multimodal fusion strategy is specifically as follows: When the hand pressure feature is less than a preset pressure activation threshold and the finger bending feature is less than a preset bending activation threshold, the hand action category is hand opening; When the hand pressure feature is greater than or equal to a preset pressure activation threshold and less than the maximum autonomous pressure value, and the finger bending feature is greater than or equal to a preset bending activation threshold and less than the maximum autonomous bending value, the hand movement category is proportional gripping assistance. When the hand pressure characteristic is greater than or equal to the maximum voluntary pressure value and the finger bending characteristic is greater than or equal to the maximum voluntary bending value, the hand movement category is maximum gripping assistance.

10. An upper limb exoskeleton, characterized in that, The upper limb exoskeleton includes: a memory, a processor, and an exoskeleton control program stored in the memory and executable on the processor, the exoskeleton control program being configured to implement the steps of the upper limb exoskeleton control method as described in any one of claims 1 to 9.

Images