Exoskeleton safety detection method and device based on error determination domain dynamic reconstruction

Abstract

The application relates to the technical field of exoskeleton safety control, and particularly discloses an exoskeleton safety detection method and device based on error determination domain dynamic reconstruction, which synchronously collects human-machine coupling sensing data containing bandage tension, buckle pressure, joint angle, joint angular velocity and trunk posture angle at a preset sampling frequency, obtains connection constraint signals and motion state signals through signal classification and analog-digital conversion, obtains a normalized wearing coupling degree index through multi-path fusion calculation, drives a basic hyperellipsoid determination domain to perform dynamic reconstruction and apply stability constraint verification, extracts a multi-dimensional error vector from the motion state signals, and finally outputs a safety detection result through the attribution determination of the error vector and the stability constraint dynamic determination domain, so that effective differentiation between wearing abnormalities and motion abnormalities is realized.

Description

Technical Field

[0001] This application relates to the field of exoskeleton safety control technology, and more specifically, to an exoskeleton safety detection method and device based on dynamic reconstruction of error judgment domain. Background Technology

[0002] As a wearable human-machine collaborative system, the core challenge in the safety detection of exoskeleton robots lies in distinguishing between abnormal human-machine coupling and abnormal human movement. The exoskeleton and the human body achieve elastic coupling through flexible constraints such as straps and buckles. The coupling stiffness between the human and the machine dynamically fluctuates depending on the tightness of the clothing. When the straps loosen or the leg poles shift, the motion signals collected by the sensors will naturally drift. This signal drift caused by changes in the wearing state is highly similar to real motion anomalies at the data feature level, posing a fundamental challenge to the accurate determination of safety detection.

[0003] Current exoskeleton safety detection technologies generally employ a judgment structure combining a fixed threshold and a single error dimension. This structure assumes a constant human-machine coupling state and fails to incorporate the tightness of the garment into the judgment logic. This leads to normal signal drift when the garment becomes loose being misjudged as abnormal movement, frequently triggering unnecessary abrupt stops and degradation actions. Some existing technologies attempt to linearly amplify and adjust the threshold using a simple scaling factor, but this single scaling method lacks a stability constraint mechanism during changes in the judgment structure. Rapid changes in the garment's state can cause drastic fluctuations in the threshold, easily triggering closed-loop oscillations in the system, causing the judgment logic to repeatedly jump between misjudgments and missed judgments. Furthermore, existing solutions typically focus only on single angular or velocity deviations in error judgment, lacking a comprehensive consideration of multi-dimensional motion characteristics, making it difficult to fully characterize the true deviation state under complex motion scenarios. For example, patent application CN118744441A uses a single encoder with a fixed threshold judgment structure, completely disregarding the influence of the human-machine coupling state, failing to distinguish between abnormal garment wear and abnormal movement, thus hindering the commercialization of exoskeleton products. Summary of the Invention

[0004] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method and apparatus for exoskeleton safety detection based on dynamic reconstruction of the error determination domain.

[0005] According to one aspect of this application, a method for exoskeleton safety detection based on dynamic reconstruction of error determination domain is provided, comprising: S1, synchronously collect human-machine coupling sensing data at a preset sampling frequency. The human-machine coupling sensing data includes strap tension signal, buckle pressure signal, joint angle signal, joint angular velocity signal, and torso posture angle signal. S2, perform signal classification and analog-to-digital conversion on human-machine coupled sensing data to obtain exoskeleton connection constraint signals and joint motion state signals; S3, based on the pre-calibrated loosening reference value and fitting reference value, performs multi-path wearing state index quantification on the exoskeleton connection constraint signal and joint motion state signal to obtain a normalized wearing coupling index. S4. Dynamic reconstruction of error decision domain and constraint verification of the basic hyperellipsoid decision domain driven by normalized wear coupling degree to obtain stable constraint dynamic decision domain. S5, extract the multidimensional motion error features of the measured and expected values ​​of the joint motion state signal to obtain the joint motion error vector; S6 performs a domain-inside / outside attribution determination on the joint motion error vector and the dynamic decision domain of stability constraints to obtain a safety determination result.

[0006] According to another aspect of this application, an exoskeleton safety detection device based on dynamic reconstruction of error decision domain is provided, used in the aforementioned exoskeleton safety detection method based on dynamic reconstruction of error decision domain, comprising: The synchronous acquisition module is used to synchronously acquire human-machine coupling sensing data at a preset sampling frequency. The human-machine coupling sensing data includes strap tension signal, buckle pressure signal, joint angle signal, joint angular velocity signal, and torso posture angle signal. The signal processing module is used to classify and convert human-machine coupled sensing data into analog-to-digital signals to obtain exoskeleton connection constraint signals and joint motion state signals. The wearable state calculation module is used to quantify the wearable state index of the exoskeleton connection constraint signal and joint motion state signal based on the pre-calibrated loosening reference value and fitting reference value to obtain the normalized wearable coupling degree index. The decision domain reconstruction module is used to dynamically reconstruct the error decision domain and verify the constraints of the basic hyperellipsoid decision domain driven by the normalized wear coupling degree to obtain a stable constraint dynamic decision domain. The error extraction module is used to extract multi-dimensional motion error features from the measured and expected values ​​of the joint motion state signal to obtain the joint motion error vector. The safety determination module is used to determine the internal and external attribution of the joint motion error vector to the dynamic determination domain of stability constraints in order to obtain the safety determination result.

[0007] Compared with existing technologies, the advantages of this application are as follows: First, the fault tolerance boundary of the decision domain can be adaptively adjusted according to the change of the wearing coupling degree. When the wearer becomes loose, the error space where the signal drift is located is reasonably contained by the dynamic decision domain and no longer triggers false alarms, which greatly reduces the false judgment rate of the system compared with the fixed threshold scheme. Second, a dual truncation clamping logic of single-frame change rate constraint and absolute amplitude upper limit constraint is introduced. The decision domain parameters will not change drastically when the wearer state changes rapidly, and the closed-loop stability of the system is effectively guaranteed, avoiding the problem of decision logic oscillation in simple proportional adjustment schemes. Finally, a hyperellipsoidal decision structure containing three orthogonal dimensions of velocity, acceleration and attitude is adopted. Compared with single-dimensional scalar threshold decision, it can more comprehensively capture the real abnormal deviation patterns in the multi-dimensional motion feature space, and improve the detection sensitivity and accuracy of security threats in complex dynamic scenarios. Attached Figure Description

[0008] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 This is a schematic diagram of the structure of an exoskeleton system according to an embodiment of this application; Figure 2 This is a flowchart of an exoskeleton safety detection method based on dynamic reconstruction of error determination domain according to an embodiment of this application; Figure 3 This is a schematic diagram of the data flow of the exoskeleton safety detection method based on dynamic reconstruction of the error determination domain according to an embodiment of this application; Figure 4 This is a flowchart of step S6 of the exoskeleton safety detection method based on dynamic reconstruction of error determination domain according to an embodiment of this application; Figure 5 This is a block diagram of an exoskeleton safety detection device based on dynamic reconstruction of the error determination domain according to an embodiment of this application; Figure 6 This is a schematic diagram of the composition structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0010] To further illustrate the technical means and effects adopted by this application in order to achieve the intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of this application is provided in conjunction with the accompanying drawings and preferred embodiments.

[0011] In the technical solution of this application, a method for exoskeleton safety detection based on dynamic reconstruction of error determination domain is proposed. Figure 1 This is a schematic diagram of the structure of an exoskeleton system according to an embodiment of this application. Figure 1 As shown, the exoskeleton system achieves elastic coupling with the human limbs through flexible restraint components such as straps and buckles. The system includes the exoskeleton mechanical body, joint drive motors, joint encoders, inertial measurement units, strap tension sensors, and wearable coupling connection structures. The strap tension sensors are installed on the straps to detect the human-machine connection constraint force, the joint encoders are installed at the joints to collect joint angle and angular velocity signals, and the inertial measurement units are installed at the waist to collect torso posture angle signals.

[0012] like Figure 2 and Figure 3 As shown, the exoskeleton safety detection method based on dynamic reconstruction of error decision domain according to an embodiment of this application includes the following steps: S1 synchronously collects human-machine coupling sensing data at a preset sampling frequency. This data includes strap tension signals, buckle pressure signals, joint angle signals, joint angular velocity signals, and torso posture angle signals. The preset sampling frequency is 50 to 200 Hz. It should be understood that the accuracy of safety detection for exoskeletons, as wearable human-machine collaborative systems, fundamentally depends on a comprehensive perception of the human-machine coupling state. This perception requires simultaneously acquiring two fundamentally different types of physical information: one is mechanical constraint information directly reflecting the strength of the physical connection between the exoskeleton and the human body; the other is kinematic state information indirectly reflecting the human-machine motion following characteristics. Collecting only one type of signal will not provide complete data support for the accurate quantification of the wearable state and the multi-dimensional extraction of motion errors in subsequent steps.

[0013] Specifically, at the arrival of each sampling clock cycle, the hardware timer of the main control processing unit generates a unified sampling trigger signal, which is simultaneously sent to the data acquisition front-end circuits of all sensor channels. Upon receiving the trigger signal, each sensor channel completes a physical quantity to electrical signal conversion and latching operation at the same time. Specifically, the strap tension sensor converts the mechanical deformation of the strap into a proportional analog voltage output, the buckle pressure sensor converts the normal pressure of the buckle contact surface into a corresponding analog voltage output, the joint encoder converts the rotational angular displacement of the joint axis into a corresponding digital pulse count value or analog voltage value, and the IMU inertial measurement unit calculates the spatial attitude of the human torso using its internal accelerometer and gyroscope and outputs digital values ​​of pitch and roll angles. After all sensor channels complete data latching at the same sampling time, the data acquisition circuit reads the signal values ​​of each channel sequentially according to the preset channel order, adds a unified timestamp, and packages them into a human-machine coupled sensing data frame for the current sampling cycle. This data frame is then written to the system buffer, awaiting signal classification and analog-to-digital conversion processing in the subsequent step S2.

[0014] In the first embodiment of this application, a high-precision solution with tension sensors is adopted for rehabilitation-grade and industrial-grade exoskeletons with high detection accuracy requirements. The hardware configuration includes one TSB-100 tension sensor (range 0-100N) mounted on the lumbar strap to collect tension signals; one identical tension sensor mounted on the thigh strap to collect tension signals; one MPU6050 inertial measurement unit (IMU) mounted on the lumbar region (sampling frequency 100Hz) to collect pitch and roll angle signals of the human torso; and an E6B2-CWZ6C encoder (1000 lines resolution) mounted on the hip joint to collect hip joint angle and angular velocity signals. The data acquisition circuit uses an STM32F407 processor with a preset sampling frequency of 100Hz and a corresponding sampling period of 10 milliseconds. In this configuration, the system synchronously collects a complete human-machine coupled sensing data frame every 10 milliseconds, including waist strap tension, thigh strap tension, hip joint angle, hip joint angular velocity, and trunk pitch and roll angles.

[0015] In the second embodiment of this application, a low-cost solution without tension sensors is adopted for cost-sensitive consumer-grade exoskeletons. The hardware configuration only includes a lumbar IMU (Inertial Measurement Unit) model MPU6050 with a sampling frequency of 100Hz and a hip joint encoder model E6B2-CWZ6C with a resolution of 1000 lines; no tension or pressure sensors are configured. Data acquisition and processing utilizes an STM32F103 processor, with the preset sampling frequency also set to 100Hz. Under this configuration, the strap tension and buckle pressure signals in the human-machine coupling sensing data are empty. The system only collects joint angle signals, joint angular velocity signals, and torso posture angle signals. Subsequent calculations of wearability indicators will rely entirely on indirect methods, i.e., indirectly estimating the degree of human-machine coupling through motion state signals, achieving basic safety detection functions without additional hardware costs.

[0016] S2 performs signal classification and analog-to-digital conversion on the human-machine coupled sensing data to obtain exoskeleton connection constraint signals and joint motion state signals. It should be understood that different types of sensors have different analog output characteristics, sampling timestamp biases, and signal-to-noise ratio levels at the physical level. If the unprocessed mixed data stream is directly used for subsequent wearable state quantization and motion error extraction, signals with different physical properties will interfere with each other, and hardware noise and timing deviations will affect the accuracy of subsequent calculations. Therefore, it is necessary to perform signal classification and analog-to-digital conversion on the human-machine coupled sensing data, logically decouple the mixed data stream according to its physical properties, and apply targeted signal conditioning operations to obtain exoskeleton connection constraint signals and joint motion state signals that meet the input requirements of subsequent algorithms.

[0017] Specifically, in this embodiment, step S2 includes: sub-step S21, performing channel separation processing on human-machine coupled sensing data based on a fixed step-size sampling clock to obtain discrete constraint state sequence and discrete motion state sequence; sub-step S22, performing time-domain smoothing reconstruction on the discrete constraint state sequence to obtain exoskeleton connection constraint signal; and sub-step S23, performing multi-source hardware phase alignment processing on the discrete motion state sequence to obtain joint motion state signal.

[0018] First, sub-step S21 is executed. The system uses a high-frequency hardware timer to provide a sampling clock reference with a fixed step size. The frequency of this clock reference is consistent with the sampling frequency preset in step S1, and the sampling frequency range is 50 to 200 Hz. At each sampling clock trigger moment, the data acquisition circuit performs synchronous analog-to-digital conversion on the analog output signals of all sensor channels, truncating the time-continuous analog voltage signal into a discrete digital sequence with a finite step size.

[0019] After completing the analog-to-digital conversion, the system uses a preset sensor physical pin mapping matrix to perform channel separation on the mixed-acquired digital sequences. This channel separation process extracts and categorizes the channel data belonging to the strap tension sensor and buckle pressure sensor into discrete constraint-state sequences, and extracts and categorizes the channel data belonging to the joint encoder and inertial measurement unit into discrete motion-state sequences. Its mathematical expression is as follows:

[0020]

[0021] in, This refers to the multidimensional original analog signal vector of human-machine coupled sensing data in the continuous time domain. The system's preset discrete sampling period corresponds to a sampling frequency range of 50 to 200 Hz. This is the discrete time step index value. Extract the diagonal matrix for constraint sensor channels to separate tension and pressure channels. Extract the diagonal matrix for motion sensor channels to separate angle, angular velocity, and attitude channels. This is the quantized feature vector of the discrete constrained state sequence at time k. This is the quantized feature vector of the discrete motion sequence at time k. The channel extraction diagonal matrix is ​​constructed as follows: the matrix dimension is equal to the total number of sensor channels, the elements on the diagonal corresponding to the target sensor type are set to 1, and the elements at other positions are set to 0. Selective extraction of target channels can be achieved through matrix multiplication.

[0022] Then, sub-step S22 is executed. Since the tension and pressure sensors are easily affected by the mechanical vibrations of the exoskeleton in actual working environments, the discrete constraint state sequence inevitably contains high-frequency noise spikes. If these are not suppressed, they will affect the accuracy of subsequent wearable state index calculations. This sub-step introduces a first-order infinite impulse response low-pass filter model, performing differential smoothing operations on the noisy samples collected at the current moment and the smoothed reference values ​​accumulated over historical time steps. The filtering result is used to attenuate high-frequency hardware spikes, assembling the reconstructed signal sequence into a continuous standard signal stream characterizing the physical coupling strength. The calculation formula for this time-domain smoothed reconstruction is:

[0023] in, This represents the vector sample value of the discrete constrained state sequence at the current k-th step. This is the damping coefficient of a first-order low-pass filter, and its value range is... This determines the proportion of trust in the current discrete new sample. This is the filtered output result that was completed and temporarily stored in the previous time step. This represents the time-domain reconstruction result of the exoskeleton connection constraint signal at step k. (Filter damping coefficient) A smaller value results in a stronger filtering effect but a greater signal response delay; a larger value results in a faster response but a weaker noise suppression capability. In practical applications, The value can be 0.3. This value can effectively suppress the high-frequency noise of mechanical vibration generated during the walking process of the exoskeleton at a sampling frequency of 100Hz, while ensuring that the response delay of the trend of the strap tension change is within an acceptable range.

[0024] Finally, sub-step S23 is executed. Since the joint encoder and the IMU (Inertial Measurement Unit) are different types of hardware devices, each with its own independent internal clock source, there is a slight phase deviation between the sampling times of different hardware during actual operation. Without alignment, spurious deviations due to timing misalignment will occur in subsequent motion error calculations. This sub-step first identifies and reads the asynchronous hardware timestamps independently generated by each distributed hardware in the discrete motion sequence. Then, it establishes the system clock of the main control processing unit as the global standard time axis, calculates the phase deviation difference between each heterogeneous sensor relative to this standard time axis, and finally introduces a linear interpolation calculation model to perform mapping projection between the sampling values ​​of two adjacent offset moments. This forces the asynchronously acquired motion data to be resampled to the synchronization node of the global standard time axis, eliminating the timing shearing effect of multi-source data. The formula for this phase alignment calculation is:

[0025] in, and Let the two adjacent heterogeneous hardware real sampling timestamps in the discrete motion sequence satisfy the following conditions: , and For discrete motion state sequences in and Discrete motion bias parameters recorded at time points. This refers to the target alignment and synchronization node time on the global standard timeline. For joint motion state signals in The output result after the synchronization node completes phase alignment.

[0026] In the first embodiment of this application, for a high-precision solution with tension sensors, one tension sensor is installed in the waist strap, one tension sensor is installed in the thigh strap, one IMU (Inertial Measurement Unit) is installed in the waist, and an encoder is installed in the hip joint. The total number of sensor channels is the sum of the output channels of the aforementioned sensors. A diagonal matrix is ​​extracted from the channels. In the diagram, the diagonal positions corresponding to the two tension sensors are set to 1, and the remaining positions are set to 0; the channel extracts the diagonal matrix. In the diagram, the diagonal positions corresponding to the inertial measurement unit and the joint encoder are set to 1, and the remaining positions are set to 0. First-order low-pass filter damping coefficient. The value is set to 0.3. The typical phase deviation between the inertial measurement unit and the joint encoder is in the range of 0.5 to 2 milliseconds. After the above linear interpolation alignment process, the timing synchronization accuracy of multi-source motion data can be controlled within 0.1 milliseconds.

[0027] In the second embodiment of this application, for a low-cost solution without tension sensors, an STM32F103 processor is used for data acquisition and processing. Only a waist IMU inertial measurement unit and a hip joint encoder are installed, without configuring a tension sensor. At this time, the discrete constraint state sequence is an empty set during channel separation. The system skips the time-domain smooth reconstruction sub-step and only performs multi-source hardware phase alignment processing on the discrete motion state sequence to obtain the joint motion state signal. The calculation of wearable state indicators in subsequent steps will rely entirely on indirect paths.

[0028] S3, based on pre-calibrated loosening and fitting reference values, performs multi-path wearing state index quantification on the exoskeleton connection constraint signal and joint motion state signal to obtain a normalized wearing coupling index. The pre-calibrated loosening and fitting reference values ​​are calibration parameters determined by averaging corresponding signal values ​​collected under laboratory conditions, simulating different wearing states. These include signal calibration values ​​when the exoskeleton is completely loose from the human body and signal calibration values ​​when the exoskeleton is in normal fitting with the human body. It should be understood that the exoskeleton connection constraint signal and joint motion state signal exist in the form of mechanical and kinematic dimensions, respectively, and cannot be directly used to drive the dynamic reconstruction of the error judgment domain in subsequent steps. The subsequent step S4 requires a single scalar that can uniformly characterize the current tightness of human-machine coupling as a driving variable for adjusting the judgment domain. This scalar must have a clear physical meaning and a normalized value range to establish a deterministic mapping relationship between the wearing state and the boundary of the judgment domain. Therefore, based on the pre-calibrated loosening and fitting reference values, it is necessary to quantify the multi-path wearing state index of the exoskeleton connection constraint signal and joint motion state signal, and fuse the dispersed multi-source heterogeneous signals into a normalized wearing coupling index to provide a unified driving input for the dynamic reconstruction of the decision domain.

[0029] Specifically, in this embodiment, step S3 includes: sub-step S31, based on pre-calibrated loosening reference value and fitting reference value, performing direct coupling quantization on the exoskeleton connection constraint signal to obtain a direct coupling degree sub-index; sub-step S32, performing indirect posture following feature quantization on the joint motion state signal to obtain an indirect coupling degree sub-index; sub-step S33, based on preset path confidence weight allocation coefficients, performing weighted summation and boundary truncation normalization on the direct coupling degree sub-index and the indirect coupling degree sub-index to obtain a normalized wear coupling degree index.

[0030] First, in sub-step S31, the system receives the exoskeleton connection constraint signal and extracts the measured amplitude of real-time physical tension or pressure. Then, it retrieves the pre-stored zero-point calibration parameters and full-scale calibration parameters from the system memory. The zero-point calibration parameters are the signal calibration values ​​when the exoskeleton is completely loose from the human body, and the full-scale calibration parameters are the signal calibration values ​​when the exoskeleton is normally in contact with the human body. These two calibration parameters were obtained through prior experimental calibration. The calibration method involved simulating different wearing states in a laboratory environment, collecting corresponding signal values, and averaging them. Using linear proportional mapping logic, the relative proportion of real-time physical tension within the dynamic range from loosening to contact is calculated to obtain the direct coupling degree sub-index. The calculation formula is as follows:

[0031] in, As a sub-index of direct coupling degree, This refers to the real-time measured environmental tension or pressure scalar value extracted from the exoskeleton connection constraint signal. This is the signal calibration value when the exoskeleton is completely detached from the human body. These are the signal calibration values ​​under the condition that the exoskeleton fits perfectly and completely within the human body. and It can be obtained through preliminary experimental calibration. The calibration method is as follows: in a laboratory environment, simulate different wearing states, collect the corresponding signal values, and take the average value to determine the result.

[0032] Then, sub-step S32 is executed. The system receives joint motion state signals, performs time-domain differentiation and extremum search on the human torso posture angle data in the signals, and calculates the maximum relative sway amplitude of the human posture within a unit time window. Simultaneously, the exoskeleton robot joint angle sequence in the signals is compared with the expected human motion trajectory sequence, and the absolute value of the motion phase deviation between the two is calculated using a cross-correlation algorithm. The posture sway amplitude and phase deviation value are introduced into a penalty deduction function to calculate the indirect coupling degree sub-index characterizing the human-machine motion following characteristics. The calculation formula is:

[0033] in, The amplitude of the current actual posture sway is extracted from the joint motion state signal, specifically the maximum fluctuation value of the posture angle per unit time, calculated from the trunk pitch and roll angles collected by the IMU. The system presets the maximum allowable human body sway amplitude threshold. The actual absolute value of the human-machine motion phase deviation, calculated from the joint motion state signal, is obtained by using a phase difference calculation algorithm between the exoskeleton joint motion signal acquired by the joint encoder and the human motion signal acquired by the IMU. The system's preset maximum allowable human-machine motion phase deviation threshold. and For the sway weight parameter and phase difference weight parameter in the indirect penalty determination, satisfying The constraints and weighting coefficients can be determined through experimental adjustments based on the type of exoskeleton product (rehabilitation, industrial, or consumer grade). This is an indirect coupling degree sub-index.

[0034] Finally, sub-step S33 is executed. The system receives the direct coupling degree sub-index and indirect coupling degree sub-index generated in the above two sub-steps respectively. Based on the sensor type and reliability model of the exoskeleton, it retrieves the preset fusion weight allocation coefficients, prioritizing the weight of the direct path, i.e., the sensor measured signal, to ensure calculation accuracy. The two sets of sub-indexes are weighted and summed to obtain the preliminary fusion comprehensive state quantity. Then, upper and lower boundary truncation operators are constructed to perform a forced numerical delimitation operation on the comprehensive state quantity, ensuring that the output single index is strictly clamped and mapped within the closed interval [0,1]. The calculation formula is:

[0035] in, For the scalar value of the direct coupling degree sub-index, For the scalar value of the indirect coupling degree sub-index, The fusion confidence weights are assigned to the direct sensor decision path. To assign fusion confidence weights to the indirect motion estimation decision path, and require , This is the final quantification result of the normalized wearable coupling degree index. When both direct and indirect signals exist, a weighted fusion method is used to calculate the final wearable state index. The fusion weight coefficient is set to prioritize the weight of the direct path to ensure the calculation accuracy of the normalized wearable coupling degree index.

[0036] In the first embodiment of this application, for a high-precision solution with a tension sensor, the loosening reference value is determined through preliminary experimental calibration. The value is 5N, which is the tension signal value when the exoskeleton is completely loosened, and it conforms to the reference value. The value is 45N, which is the tension signal value when the exoskeleton is properly fitted. The weighting coefficient in the indirect path. The value is 0.6. The value is 0.4. (Fusing weight coefficient) The value is 0.7. The value is set to 0.3 to prioritize the weight of the measured signal from the direct path sensor. Under this configuration, when the measured tension of the strap is 45N, the direct coupling degree sub-index... A calculation result of 1.0 indicates a perfect fit; when the measured tension of the strap is 25N, The calculation result is 0.5, indicating moderate loosening.

[0037] In the second embodiment of this application, for a low-cost solution using a tension-free sensor, the system skips the direct coupling metric sub-step, and the wearable state indicators rely entirely on indirect path calculations. The threshold for the maximum allowable human body posture sway in the indirect path... Calibration is set at 15 degrees, maximum phase difference threshold. The calibration is 0.5 radians, with a weighting coefficient. The value is 0.6. The value is 0.4. (Fusing weight coefficient) Set to 0, Setting it to 1 means that the normalized wearable coupling degree index is exactly equal to the calculated result of the indirect coupling degree sub-index. With this configuration, no additional hardware cost is required, and quantitative evaluation of the wearable state can be achieved through the indirect path, meeting the safety detection requirements of consumer-grade exoskeletons.

[0038] S4 involves dynamically reconstructing the error decision domain and verifying constraints on the basic hyperellipsoidal decision domain driven by the normalized wear coupling degree to obtain a stable constrained dynamic decision domain. It should be understood that the normalized wear coupling degree index itself is only a state description quantity and has not yet been established as a correlation with the decision logic of security detection. The subsequent step S6 requires a multidimensional decision domain with clear mathematical boundaries as a spatial benchmark for security adjudication. The fault-tolerant boundary of this decision domain must be able to adaptively adjust with changes in the wear state to avoid misjudging normal signal drift as abnormal movement when the wearer becomes loose. Simultaneously, the dynamic adjustment process of the decision domain must be subject to stability constraints to prevent drastic changes in the decision structure from causing oscillations in the system's closed-loop control. Therefore, it is necessary to use the normalized wear coupling degree index as the driving variable to dynamically reconstruct the error decision domain of the preset basic hyperellipsoidal decision domain and apply stability constraints to obtain a stable constrained dynamic decision domain that can both adapt to changes in the wear state and ensure stable system operation.

[0039] Specifically, in this embodiment, step S4 includes: sub-step S41, calculating the dimensional dynamic amplification coefficient of the normalized wear coupling index based on the preset maximum amplification gain base and nonlinear curvature index of each dimension to obtain the dimensional threshold amplification coefficient matrix; sub-step S42, reconstructing the basic boundary hyperellipsoid parameters of the pre-stored basic hyperellipsoid boundary constant based on the dimensional threshold amplification coefficient matrix to obtain the preliminary reconstruction decision domain parameter matrix; sub-step S43, performing oscillation suppression and numerical convergence verification on the preliminary reconstruction decision domain parameter matrix based on the dual truncation clamping logic of single-frame change rate constraint and absolute amplitude upper limit constraint to obtain the stable constraint dynamic decision domain.

[0040] First, in sub-step S41, the system receives the normalized wear coupling index and, for the three independent physical error dimensions of velocity, acceleration, and attitude, retrieves the system's preset nonlinear adjustment function model. The normalized wear coupling index is substituted into the function model as an independent variable. Utilizing its monotonically decreasing characteristic (i.e., the lower the fit, the higher the amplification factor), the three independent tolerance boundary amplification ratios adapted to the current tightness are calculated. These three independent ratio coefficients are then vectorized and combined to generate a dimension threshold amplification coefficient matrix. This adjustment function satisfies the following constraints: when the normalized wear coupling index decreases from 1 to 0, the adjustment function value monotonically increases from 1, meaning the looser the fit, the larger the threshold for each dimension, adapting to signal drift; the adjustment function is a nonlinear function or a piecewise linear function, rather than a single proportional scaling, ensuring adjustment accuracy; different dimensions can have different adjustment functions, for example, the adjustment range for the velocity error dimension is greater than that for the attitude error dimension, adapting to the drift characteristics of different error types. The calculation formula is:

[0041] in, This is a quantitative characteristic value of the normalized wearable coupling degree index. The dimension indexes for the multidimensional error boundary represent the velocity dimension. acceleration dimension and posture dimension , The elements in the matrix representing the dynamic magnification factor of the i-th dimension, i.e., the dimension threshold magnification factor, are calculated. The maximum amplification gain base for the i-th dimension is preset by the system, which controls the maximum amplification magnitude of the threshold value for that dimension. The system uses a preset nonlinear curvature exponent of the i-th dimension to control the curve shape of the amplification coefficient as it changes with the wearing state. This function satisfies the following condition: hour That is, it does not enlarge when fully fitted, when hour That is, the boundary condition at which the maximum magnification is reached when completely separated. That is, when it is slightly loose, and For example, The threshold is magnified to twice its original value; For example, The threshold is magnified to 1.5 times, and the larger the nonlinear curvature index, the more conservative the magnification is when there is slight loosening.

[0042] Next, in sub-step S42, the system extracts pre-calibrated basic three-dimensional hyperellipsoidal boundary constants from the static storage area for the current exoskeleton product model under ideal, fully fitted conditions (i.e., when the normalized wear coupling index equals 1). These constants include basic velocity thresholds, basic acceleration thresholds, and basic attitude thresholds. Using element-wise matrix operations, each amplification factor in the dimension threshold amplification factor matrix is ​​scaled and multiplied with its corresponding basic boundary constant. The resulting new threshold parameters are then packaged to initially establish a hyperellipsoidal feature parameter matrix encompassing the current error tolerance upper limit. The mathematical expression for the basic error judgment domain under fully fitted conditions is the three-dimensional hyperellipsoidal equation:

[0043] in, This represents the deviation value in the velocity error dimension, specifically the difference between the measured value of the exoskeleton joint angular velocity and the estimated value of the human joint angular velocity. This refers to the deviation value in the acceleration error dimension, which is the difference between the measured value of the exoskeleton joint angular acceleration and the estimated value of the human joint angular acceleration. This represents the deviation value of the posture error dimension, which is the difference between the measured values ​​of the exoskeleton joint posture and the estimated values ​​of the human joint posture. Based on the base speed threshold, Based on the basic acceleration threshold, The base attitude threshold is used. The initial reconstruction decision domain parameter matrix after scaling by the dynamic amplification factor is:

[0044]

[0045]

[0046] in, , , The three coefficient components in the dimension threshold magnification coefficient matrix are... , , To initially reconstruct the three-dimensional threshold parameters in the decision domain parameter matrix, the mathematical expression for the adjusted dynamic error decision domain is as follows:

[0047] This expression indicates that when the wearer becomes loose, causing the normalized wear coupling index to decrease, the adjustment function values ​​of each dimension increase, and the threshold parameters of the corresponding dimensions increase accordingly. The fault tolerance boundary of the hyperellipsoidal decision domain is extended differentially in each dimension, thereby encompassing the normal signal drift caused by the wearer's looseness within the extended decision domain and avoiding misjudging it as motion abnormality.

[0048] Finally, sub-step S43 is executed. The dynamic adjustment of the decision domain must meet the preset stability constraints to prevent changes in the decision structure from causing instability in the closed loop of the system. The stability constraints are implemented in the following way.

[0049] The rate of change constraint requires that the change in the decision domain parameter in a single frame does not exceed 5% of the base value, and the change per second does not exceed 20% of the base value. Specifically, the change in a single frame is calculated by dividing the absolute value of the difference between the decision domain parameter in the current frame and the decision domain parameter in the previous frame by the base parameter value. The change per second is calculated by dividing the absolute value of the difference between the initial value and the final value of the decision domain parameter in the current second by the base parameter value. The magnitude of change constraint requires that the maximum adjustment range of the decision domain parameter does not exceed three times the base value, i.e., the adjusted threshold. , , All values ​​are no greater than 3 times the corresponding base threshold. The control influence constraint requires that the influence of the change in the decision domain on the control output does not exceed 30% of the maximum assist torque. The influence magnitude is calculated by dividing the absolute value of the difference between the control output torque after the decision domain adjustment and the control output torque before the decision domain adjustment by the maximum assist torque value.

[0050] The system receives the decision domain parameters after the current calculation period, i.e., the current frame, has been calculated by the adjustment function. , , Simultaneously, the system retrieves the historical final decision domain parameters from the previous calculation cycle (i.e., the previous frame output) from the microcontroller's cache. A difference operation is performed between the two to calculate the gradient of the current frame's parameter change, and this gradient is compared and verified against the system's preset maximum rate of change for a single frame (i.e., 5% of the base value). A numerical truncation clamping logic is introduced: if the gradient exceeds the safe rate limit, the current decision domain parameter is forcibly restricted to the boundary between the previous frame's parameter plus or minus the limit; if it does not exceed this limit, the original calculated value is retained. After completing the rate of change constraint, an absolute amplitude constraint is added: an absolute upper limit check is performed on the clamped parameters to ensure that the threshold parameter in any dimension does not exceed three times the corresponding base threshold. If the adjusted decision domain parameter does not meet the above stability constraints, it is forcibly restricted within the constraint range, and a warning signal is triggered and output to the exoskeleton control system to prompt the wearer to check the wearing status. The final three-dimensional threshold parameter, after the above double convergence constraint verification and correction, is solidified and encapsulated to form a stable constraint dynamic decision domain.

[0051] In the first embodiment of this application, for a high-precision scheme with a tension sensor, the basic velocity threshold is determined through experimental calibration. The baseline acceleration threshold is 0.4 rad / s. 2 rad / s 2 Basic attitude threshold The angle is 8°. The nonlinear adjustment functions for all dimensions adopt a unified form. The single-frame parameter change rate is constrained to no more than 5% of the base value, the change per second is constrained to no more than 20% of the base value, and the maximum adjustment range is constrained to no more than 3 times the base value.

[0052] The operation process under this configuration is as follows: when the normalized wearable coupling index is equal to 1.0, i.e., a perfect fit, the adjustment function values ​​of each dimension are all 1, and the decision domain keeps the basic threshold unchanged, i.e., the velocity threshold is 0.4 rad / s and the acceleration threshold is 2 rad / s. 2 The posture threshold is 8°, at which point the system's sensitivity is highest, accurately detecting real motion anomalies. When the normalized wear coupling index equals 0.5 (slight looseness), the adjustment function value is... The decision thresholds were adjusted to a velocity threshold of 0.6 rad / s and an acceleration threshold of 3 rad / s. 2 The posture threshold is 12° to accommodate signal drift caused by slight loosening and avoid misjudgment. When the normalized wear coupling index equals 0, indicating complete disengagement, the adjustment function value is... The decision thresholds were adjusted to a velocity threshold of 1.2 rad / s and an acceleration threshold of 6 rad / s. 2 The attitude threshold is 24° to accommodate signal drift caused by severe loosening. All adjustments meet the stability constraints of a change rate not exceeding 5% per frame and a change amplitude not exceeding 3 times, and there is no oscillation during system operation.

[0053] In the second embodiment of this application, for a low-cost solution using a tension-free sensor, the adjustment function adopts a piecewise linear form: when the normalized wear coupling index is in the range of 0.7 to 1.0, the adjustment function value is equal to 1, meaning that the judgment domain is not adjusted and maintains the basic threshold when the wearer fits snugly; when the normalized wear coupling index is in the range of 0.3 to 0.7, the adjustment function value increases linearly from 1 to 2, meaning that the judgment domain is linearly amplified when the wearer is slightly to moderately loose; when the normalized wear coupling index is in the range of 0 to 0.3, the adjustment function value is fixed at 2, meaning that the judgment domain is fixedly amplified by 2 times when the wearer is severely loose, avoiding excessive amplification that could lead to missed detections. The single-frame parameter change rate is also constrained to no more than 5% of the basic value, and the maximum adjustment amplitude is constrained to no more than twice the basic value. This piecewise linear adjustment function has lower computational complexity than nonlinear functions, adapting to the computing power limitations of consumer-grade exoskeleton processors. Furthermore, by limiting the maximum amplification factor to 2 times instead of 3 times, a balance is achieved between reducing the false positive rate and preventing missed detections suitable for consumer-grade application scenarios.

[0054] S5 involves extracting multidimensional motion error features from the measured and expected values ​​of the joint motion state signal to obtain a joint motion error vector. It should be understood that subsequent step S6 requires an error vector containing multiple motion feature dimensions as a judgment input. This vector must comprehensively characterize the degree of deviation between the actual motion state of the exoskeleton and the expected motion state of the human body in the three orthogonal dimensions of velocity, acceleration, and posture. Therefore, multidimensional motion error features are extracted from the joint motion state signal output in step S2, and the measured kinematic parameters of the exoskeleton and the expected kinematic parameters of the human body are cross-subtracted dimension by dimension to construct a joint motion error vector that can be substituted into the hyperellipsoidal decision domain for spatial attribution determination.

[0055] Specifically, in this embodiment, step S5 includes: sub-step S51, performing machine-measured kinematic state differential solution on the exoskeleton joint angle sequence in the joint motion state signal to obtain the machine-measured motion state sequence, and separating the human reference motion state sequence; sub-step S52, mapping and projecting the human reference motion state sequence from the torso to the target joint space using a biomechanical kinematic projection matrix to obtain the human expected motion state sequence; sub-step S53, performing multi-dimensional error fusion reconstruction on the machine-measured motion state sequence and the human expected motion state sequence to obtain the joint motion error vector.

[0056] First, in sub-step S51, the system receives the joint motion state signal output from step S2 and separates the physical pose data of the exoskeleton robot body from the original sensor data of the human torso in the logical space. For the separated joint angle sequence of the exoskeleton robot body, the first derivative is obtained using a discrete-time first-order forward difference algorithm to generate measured angular velocity data of the exoskeleton. The generated measured angular velocity data is then substituted into the difference model to obtain the second derivative, generating measured angular acceleration data of the exoskeleton. The measured angles, angular velocities, and angular accelerations of the robot body are sequentially aligned and encapsulated to construct the measured motion state sequence of the machine. Simultaneously, the human torso data, which did not participate in the differential calculation, is directly packaged into a human reference motion state sequence for use in subsequent human intention estimation models. The calculation formula for this differential solution is:

[0057]

[0058]

[0059] in, This refers to the measured joint angle of the exoskeleton at the k-th discrete moment, extracted from the joint motion state signal. These are the measured joint angles of the exoskeleton at the (k-1)th discrete moment, temporarily stored from the previous control cycle. The preset discrete computation time step, i.e., the sampling period, is the system's preset time step. The measured angular velocity of the exoskeleton is obtained through first-order difference calculation. The measured angular acceleration of the exoskeleton is obtained through second-order difference calculation. This is the encapsulated column vector result of the machine's measured motion state sequence at time k.

[0060] Then, in sub-step S52, the system receives the human reference motion state sequence separated in the previous sub-step and extracts IMU feature data such as the human torso pitch and roll angles. A pre-established lower limb biomechanical kinematic projection matrix is ​​introduced, containing empirical parameter mapping relationships determined based on human anatomical structure and gait kinematics, mapping the posture input in the torso space to the expected activity space of the corresponding exoskeleton joints, such as the hip or knee joints. A target motion trajectory feature column vector containing expected angles, expected angular velocities, and expected angular accelerations is generated within the target joint space to characterize the ideal human body's expected homing response to external devices. The calculation formula for this mapping projection is:

[0061] in, This is the multi-dimensional posture feature matrix of the human torso at time k, extracted from the human reference motion state sequence, containing torso pitch and roll angle data acquired by the IMU. The system uses a pre-defined biomechanical kinematic projection matrix to map trunk pose to the target joint space of the lower limbs. The expected joint angles of the human body are calculated using projection. The expected joint angular velocities of the human body are calculated using projection. The expected joint angular acceleration of the human body is calculated using projection. This is the encapsulated column vector result of the expected human motion state sequence at time k. The parameters of the biomechanical kinematic projection matrix can be determined by fitting the mapping relationship between trunk posture and lower limb joint movement using the least squares method after collecting data from a standard gait experiment on healthy subjects.

[0062] Finally, in sub-step S53, the system synchronously loads the machine's measured motion state sequence and the human's expected motion state sequence at the same timestamp. It then performs a cross-subtraction operation on the corresponding element dimensions of the two column vectors to calculate the independent velocity deviation values ​​(the difference between the measured exoskeleton joint angular velocity and the estimated human joint angular velocity), acceleration deviation values ​​(the difference between the measured exoskeleton joint angular acceleration and the estimated human joint angular acceleration), and posture deviation values ​​(the difference between the measured exoskeleton joint posture and the estimated human joint posture). To adapt to the coordinate axis definition of the three-dimensional hyperellipsoidal spatial boundary determination model in subsequent step S6, the three independent physical scalars are spatially orthogonalized and combined according to a strict dimensional order: velocity as the first dimension, acceleration as the second dimension, and posture as the third dimension, outputting the final joint motion error vector. The calculation formula for this multi-dimensional error fusion reconstruction is as follows:

[0063] in, , and This refers to the scalar elements of velocity, acceleration, and attitude in the machine's measured motion state sequence. , and These are the expected velocity, acceleration, and attitude scalar elements in the expected motion state sequence of the human body. The velocity deviation value is calculated by taking the cross-subtraction result corresponding to the velocity dimension. The acceleration deviation value is the result of cross-subtraction calculation corresponding to the acceleration dimension. The cross-subtraction calculation result corresponding to the attitude dimension is the attitude angle deviation value. This is the result of reconstructing the column vector of the joint motion error vector in three-dimensional space at time k.

[0064] In the first embodiment of this application, for a high-precision scheme with a tension sensor, under this configuration, assuming the current time k exoskeleton hip joint encoder reading is 25.0 degrees and the previous time k-1 reading is 24.5 degrees, the measured angular velocity of the exoskeleton is calculated using first-order difference. The value is 50 degrees per second. Assuming the angular velocity at the previous moment was 48 degrees per second, the second-order difference calculation yields the measured angular acceleration of the exoskeleton. The value is 200 degrees per second squared. Simultaneously, the trunk posture data from the lumbar IMU is mapped using a biomechanical kinematic projection matrix to calculate the expected angular velocity of the human body. The expected angular acceleration of the human body is 45 degrees per second. The expected joint angle in the human body is 180 degrees per second squared. The angle is 24.8 degrees. Therefore, the three components of the joint motion error vector are: velocity deviation... 50 minus 45 equals 5 degrees per second, acceleration deviation. 200 minus 180 equals 20 degrees per second squared, attitude deviation. The difference between 25.0 and 24.8 equals 0.2 degrees. This error vector will be substituted into the dynamic decision domain of the stability constraint in step S6 to determine whether the error belongs to the domain or not.

[0065] In the second embodiment of this application, for a low-cost solution using a tension-free sensor, due to limited processor computing power, a simplified linear approximation model can be used for the biomechanical kinematic projection matrix. This establishes a single proportional mapping relationship between the trunk pitch angle and the expected hip joint angle, reducing the computational complexity of matrix operations. Under this simplified configuration, the computational logic for differential solving and error fusion reconstruction is completely consistent with the high-precision solution; only the dimension and number of parameters of the projection matrix are reduced, yet a three-dimensional joint motion error vector that meets the requirements for subsequent domain assignment can still be generated.

[0066] S6 performs an inside-outside classification determination on the joint motion error vector and the dynamic decision domain of stability constraints to obtain a safety judgment result. It should be understood that the ultimate goal of exoskeleton safety detection is to determine whether the current motion deviation exceeds the reasonable allowable range after correction for the wearing state. That is, to determine whether the joint motion error vector is inside or outside the hyperellipsoidal space defined by the dynamic decision domain of stability constraints, and based on this, output a clear safety or abnormal judgment result for the downstream exoskeleton's underlying motor control system to execute corresponding physical degradation, shutdown, or warning operations. Therefore, performing an inside-outside classification determination on the joint motion error vector and the dynamic decision domain of stability constraints transforms the geometric inclusion relationship in multi-dimensional space into executable safety control instructions, completing the closed loop of the entire safety detection method.

[0067] Specifically, in the embodiments of this application, such as Figure 4As shown, step S6 includes: sub-step S61, calculating the spatial distance between the components of each dimension of the joint motion error vector and the fault tolerance parameters of each dimension of the dynamic decision domain of the stability constraint to obtain a normalized spatial error distance scalar; sub-step S62, performing domain-inner-outer envelope crossing discrimination and logical assignment on the normalized spatial error distance scalar to obtain a Boolean feature value of the security state; sub-step S63, based on the frame encapsulation rules and security response strategy mapping table of the underlying communication protocol, performing instruction encoding and verification encapsulation on the Boolean feature value of the security state to obtain a security determination result.

[0068] First, in sub-step S61, the system parses the received dynamic decision domain of stability constraints and extracts the dynamic boundary tolerance parameters in three dimensions, namely, the allowable extreme values ​​of velocity, acceleration, and attitude determined after stability constraint verification in step S4. Simultaneously, the system parses the received joint motion error vector and extracts its three-dimensional real-time error components, namely, the real-time velocity deviation, acceleration deviation, and attitude deviation output in step S5. These real-time error components are then divided by their corresponding dynamic boundary tolerance parameters to perform dimensionless processing in the spatial dimensions, eliminating scale differences between different physical dimensions. Using the hyperellipsoidal quadratic equation of Euclidean spatial distance, the sum of squares of each dimension component after dimensionless processing is calculated to determine the comprehensive deviation scalar of the current machine motion error relative to the central reference point in the constraint space. The calculation formula is as follows:

[0069] in, , , These are the dynamic boundary tolerance parameters for velocity, acceleration, and attitude dimensions extracted at time k from the self-stabilizing constraint dynamic decision domain. , , These are the measured values ​​of velocity deviation, acceleration deviation, and attitude deviation extracted from the joint motion error vector at time k. This is the deviation value obtained by solving the normalized spatial error distance scalar at time k. When this scalar value is less than or equal to 1, it indicates that the joint motion error vector satisfies the hyperellipsoid equation constraint of the dynamic decision domain, that is, the error vector is located inside the envelope domain; when this scalar value is greater than 1, it indicates that the joint motion error vector does not satisfy the hyperellipsoid equation constraint, that is, the error vector penetrates the envelope boundary.

[0070] Next, sub-step S62 is executed. The core task of this sub-step is to transform the continuous distance scalar into a discrete security or anomaly binary determination result.

[0071] In the first implementation, a single-frame instantaneous decision paradigm is used to complete the identification of envelope crossings inside and outside the domain. The system receives the normalized spatial error distance scalar at the current k-th moment and directly compares this scalar value with the standard critical threshold of the hyperellipsoidal surface in the normalized mathematical model, i.e., the constant 1.0. If the distance scalar is less than or equal to the critical threshold, it means that the error vector falls within the three-dimensional fault-tolerant envelope domain, and it is determined to be a normal drift caused by reasonable physical factors such as loose clothing, although there is a deviation, and is assigned a Boolean logic truth value of 1 to indicate safety; if the distance scalar is greater than the critical threshold, it means that the error has crossed the envelope boundary, and is determined to be a real human-machine confrontation or instability or other motion abnormality, and is assigned a Boolean logic false value of 0 to indicate danger. The decision formula is:

[0072] in, Let be the value of the normalized spatial error distance scalar at time k, and 1.0 be the physical envelope boundary constant after the hyperellipsoidal spatial normalization mapping. The value is a binary logic level representing the Boolean characteristic of the safety state, with a value of 1 indicating safety and a value of 0 indicating anomaly. This implementation treats each frame as an independent, memoryless, isolated event, and outputs a judgment based solely on the instantaneous comparison result of the current frame. It has low computational complexity and is suitable for application scenarios with high real-time requirements and relatively stable operating environments.

[0073] Then, in the actual physical scenario of exoskeleton wear and operation, the out-of-bounds behavior of the error distance scalar presents two fundamentally different modes in the time dimension. The first is the transient pulse disturbance mode, whose typical triggering scenarios include single-frame mechanical impact caused by the wearer stepping on a protrusion on the ground during walking, isolated signal spikes generated by interference from surrounding electromagnetic devices to the exoskeleton joint encoder, and brief overshoot caused by inertial abrupt changes during the moment of normal gait switching. The common feature of these out-of-bounds events is that the duration is extremely short, usually only 1 to 3 control frames, and the error distance scalar will quickly fall back into the envelope domain after the disturbance disappears. In essence, it does not constitute a real human-machine safety threat. The second type is the persistent trend deviation mode. Typical triggering scenarios include active resistance between the human body and the exoskeleton (i.e., the wearer attempts to resist the exoskeleton's assist direction), jamming or stalling of the exoskeleton's mechanical transmission chain, and instability and oscillation in the control system due to parameter drift. A common characteristic of these boundary-crossing events is that the error distance scalar exhibits a persistent or progressively worsening trend across multiple frames (usually more than four frames), constituting a real and imminent motion safety hazard in physical terms. The single-frame hard threshold determination mechanism of the first implementation lacks the ability to model the aforementioned temporal persistence relationship, and therefore cannot distinguish between these two physically completely different boundary-crossing modes at the algorithmic level. It treats all boundary-crossing frames equally and immediately triggers anomaly detection, leading to frequent misjudgments of transient disturbances as real motion anomalies.

[0074] To address the aforementioned issues, the second implementation method establishes an intermediate decision layer with temporal memory characteristics between the normalized spatial error distance scalar and the final safe state Boolean feature value to effectively separate the two boundary crossing modes. Specifically, in the second implementation method of this application, sub-step S62 includes: based on the instantaneous comparison of the hyperellipsoidal standard critical threshold, marking the normalized spatial error distance scalar for boundary crossing events and encoding the gradient direction to obtain the boundary crossing state and gradient composite feature value; based on the positive gradient weighted amplification coefficient, performing a dynamic balance calculation of the energy charging and discharging of the boundary crossing state and gradient composite feature value to obtain the boundary crossing persistence accumulation; and through an inequality comparison logic with a preset persistence judgment threshold, performing persistence boundary crossing discrimination and binarization logic assignment processing on the boundary crossing persistence accumulation to obtain the safe state Boolean feature value.

[0075] First, based on instantaneous comparison using the hyperellipsoidal standard critical threshold, the normalized spatial error distance scalar is marked for out-of-bounds events and encoded with gradient directions to obtain a composite feature quantity of out-of-bounds state and gradient. The system receives the normalized spatial error distance scalar at the current k-th moment and simultaneously retrieves the historical value of the scalar temporarily stored in the system cache from the previous moment. The current scalar value is compared with the hyperellipsoidal standard critical threshold to generate an instantaneous out-of-bounds Boolean flag for the current frame. Simultaneously, the first-order temporal difference between the current and previous scalar values ​​is calculated to obtain the inter-frame gradient. The reason for introducing the inter-frame gradient in addition to the out-of-bounds flag is that in exoskeleton operation scenarios, two frames in the same out-of-bounds state may have completely different physical meanings. A positive gradient means the error is accelerating away from the safety center, corresponding to a dangerous trend of mechanical jamming or worsening human-machine interaction; a negative gradient means the error is returning to the safety center, corresponding to the natural recovery process after a transient impact. Finally, the instantaneous out-of-bounds Boolean flag and the inter-frame gradient are combined and encapsulated into a composite feature quantity of out-of-bounds state and gradient. The formula for calculating the instantaneous out-of-bounds Boolean flag is:

[0076] in, Let be the normalized spatial error distance scalar at time k. Let be the standard envelope boundary constant in the normalized space of the hyperellipsoid. This is a Boolean flag indicating the instantaneous boundary breach in the current frame. A value of 1 indicates that an envelope penetration event has occurred in the current frame, and a value of 0 indicates that the current frame is inside the envelope. The formula for calculating the inter-frame gradient is:

[0077] in, Let be the normalized spatial error distance scalar at time k. This refers to the normalized spatial error distance scalar historical values ​​temporarily stored in the system cache at the previous time step. The gradient values ​​represent the inter-frame variation. Positive values ​​indicate that the error deviation is worsening, while negative values ​​indicate that the error deviation is recovering.

[0078] Then, based on the positive gradient weighted amplification coefficient, a dynamic balance calculation of charging and discharging is performed on the composite feature of the boundary crossing state and the gradient to obtain the cumulative amount of boundary crossing persistence. This step constructs a virtual energy pool with charging-discharging dynamic balance characteristics: when the instantaneous boundary crossing is marked as 1, the accumulator performs a charging operation, and the charging amplitude is weighted and amplified by the positive component of the inter-frame changing gradient, so that the boundary crossing event with the faster deterioration rate injects more energy; when the instantaneous boundary crossing is marked as 0, the accumulator does not charge, but only performs natural energy dissipation according to the preset exponential decay factor. In the exoskeleton operation scenario, the direct effect of this charging-discharging dynamic balance mechanism is: the small amount of energy injected by the transient isolated pulse (boundary crossing lasting only 1 to 3 frames) is rapidly dissipated by exponential decay in subsequent frames without boundary crossing, and the accumulator always remains at a low level and cannot trigger an alarm; while the energy injected by the persistent real anomaly (continuous boundary crossing in multiple frames and the gradient is continuously positive) continues to accumulate and rise due to the lack of decay window, and the accumulator quickly rises to a high level. This achieves effective separation of two physically different boundary-crossing modes in the time domain. The formula for calculating the exponentially decaying time-series accumulation is:

[0079] in, This refers to the historical values ​​of the accumulator temporarily stored in the system cache from the previous moment. It is an exponential decay factor, with a value range of 1. The energy dissipation rate of the accumulator during non-boundary frames determines the system's forgetting rate of transient disturbances. This refers to the instantaneous out-of-bounds Boolean label component extracted from the composite feature of the out-of-bounds state and gradient. The inter-frame variation gradient component is extracted from the composite feature of the out-of-bounds state and the gradient. This is a gradient-weighted amplification factor that controls the increase in energy intensity due to the deterioration trend, ensuring that rapidly deteriorating anomalies trigger alarms faster than slowly deteriorating anomalies. As a positive truncation operator, it retains only the gradient contribution in the deterioration direction while excluding the gradient in the regression direction from the energy calculation. This is a constant representing the saturation limit of the accumulator (e.g., a value of 10.0), used to prevent numerical overflow in extreme and persistent abnormal scenarios. This is the result of solving for the cumulative amount of boundary crossing at time k.

[0080] It should be noted that when The energy-charging term is zero overall, and the accumulator retains only the decay term. It shows an exponential decreasing trend; when and At that time, the charging item is The accumulator receives accelerated charging beyond the baseline value. This mathematical characteristic ensures that continuous out-of-bounds behavior during deterioration can break through the judgment threshold at the fastest rate.

[0081] Finally, by comparing the cumulative amount of boundary crossing with the preset persistence threshold using inequality logic, the system performs persistence limit screening and binarization logic assignment to obtain the Boolean feature value of the safe state. The physical meaning of the persistence threshold in the exoskeleton safety detection scenario is to characterize the energy threshold representing how many consecutive frames of boundary crossing energy are needed to confirm a true persistent danger. The value of this threshold must satisfy the following scenario constraints: under an exponential decay factor of 0.85, the energy accumulated by isolated 1 to 3 frames of transient pulses will never reach this threshold after natural decay in subsequent frames, while the energy accumulated by more than 4 consecutive frames of continuous boundary crossing will inevitably exceed this threshold. If the cumulative amount is less than the persistence threshold, the current system is determined to be in a safe operating state, encompassing both normal operation and transient disturbance sub-cases, and a Boolean logic truth value is assigned; if the cumulative amount is greater than or equal to the persistence threshold, the current system is confirmed to have experienced a persistent real motion anomaly, and a Boolean logic false value is assigned. The determination formula is:

[0082] in, Let be the value of the cumulative amount of boundary crossing persistence at time k. The preset persistent threshold for the system represents the minimum cumulative energy threshold required to confirm a real hazard. This is the binary logic level of the Boolean characteristic value of the safety state. A value of 1 indicates safety, and a value of 0 indicates abnormality.

[0083] Through the aforementioned temporal persistence crossing discrimination mechanism based on exponential decay accumulation, an intermediate decision layer with temporal memory characteristics is established between the normalized spatial error distance scalar and the final safe state Boolean characteristic value. This makes the safety determination no longer rely on the isolated results of instantaneous comparison of a single frame, but comprehensively considers the persistence and evolution direction of the boundary crossing event on the time axis.

[0084] Transient isolated pulses, due to insufficient duration and rapid dissipation of injected energy by exponential decay in subsequent frames, prevent the accumulator from reaching the persistence threshold, thus being automatically filtered out at the algorithm level without triggering false alarms. Continuous real anomalies, on the other hand, are caused by the superposition of out-of-bounds charging across multiple consecutive frames without a decay window, causing the accumulator to quickly exceed the persistence threshold, ensuring that real dangers are identified and responded to in a timely and accurate manner.

[0085] The second implementation significantly reduces the false trigger rate of the exoskeleton safety detection system caused by transient disturbances without increasing the cost of additional hardware sensors, while maintaining the ability to respond quickly to real continuous motion abnormalities, thus improving the operational continuity and user experience reliability of exoskeleton products in complex dynamic scenarios such as rehabilitation training and industrial operations.

[0086] Finally, in sub-step S63, the system receives the safety state Boolean feature value generated by the above screening. Based on the exoskeleton's underlying communication protocol, such as CAN bus or EtherCAT protocol, it retrieves the mapping table corresponding to the system's degradation response strategy: if the Boolean feature value is in a safe state (value 1), the control code that maintains the current assist torque is retrieved; if the Boolean feature value is in an abnormal state (value 0), the control code that triggers torque attenuation, emergency stop, and audible and visual warnings is forcibly retrieved. The corresponding control code is then spliced ​​with frame headers and tails, and a cyclic redundancy check code is added to encapsulate it into a standardized machine instruction stream that meets the requirements of the communication physical layer. This stream serves as the final input for the safety control system, completing the overall safety detection closed loop. The mathematical expression of this encapsulation process is:

[0087] in, The polarized state parameters are the Boolean eigenvalues ​​of the safe state. The system's preset security response policy mapping function is based on the input. The returned corresponding hardware degradation control action word includes actions such as normal operation or shutdown and load relief. This is a cyclic redundancy check code calculated based on the payload data. For data frame packing and encapsulation operators based on communication protocols, A standardized message sequence presented as a result of a security assessment.

[0088] In a specific implementation example of the first embodiment, a rehabilitation-grade exoskeleton suitable for a relatively stable operating environment is used. The hardware configuration includes an IMU (Inertial Measurement Unit) mounted in the lumbar region (model MPU6050, sampling frequency 100Hz) and an encoder mounted in the hip joint (model E6B2-CWZ6C, resolution 1000 lines). Assume that the stability constraint dynamic decision domain parameters output in step S4 at the current moment are a permissible extreme value of 45 degrees per second for velocity, a permissible extreme value of 300 degrees per second squared for acceleration, and a permissible extreme value of 15 degrees for posture, corresponding to a normalized wear coupling index of 0.5, i.e., a slightly loose state. Assume that the joint motion error vector output in step S5 is a velocity deviation of 5 degrees per second, an acceleration deviation of 20 degrees per second squared, and a posture deviation of 0.2 degrees. Then the normalized spatial error distance scalar is calculated as follows: This value is much smaller than the critical threshold of 1.0, so it is considered safe. The Boolean characteristic value of the safe state is assigned to 1, and the system maintains normal power-assisted operation. If at another moment the joint motion error vector is a velocity deviation of 50 degrees per second, an acceleration deviation of 350 degrees per second squared, and an attitude deviation of 18 degrees, then the normalized spatial error distance scalar is calculated as follows: If the value is greater than the critical threshold of 1.0, it is judged as abnormal, the Boolean characteristic value of the safety state is assigned to 0, and the system executes torque decay and shutdown warning.

[0089] In a specific implementation example of the second method, it is suitable for industrial-grade exoskeletons operating in complex environments with high requirements for false triggering rates. (Exponential decay factor) Setting it to 0.85 means that the accumulator decays by 15% per frame when there are no out-of-bounds frames. After approximately 15 frames (150 milliseconds), the accumulated amount decays to less than 10% of the initial value. Gradient-weighted amplification factor. The value is set to 0.5 to control the impact of the deteriorating trend on the charge intensity gain. The accumulator saturation upper limit constant is set to 10.0 to prevent numerical overflow. The persistence threshold is also set. Setting it to 3.0 corresponds to approximately four consecutive frames of out-of-bounds movement before triggering, under a decay factor of 0.85. Assuming that within a certain timeframe, the first frame experiences a single-frame out-of-bounds movement with a positive gradient of 0.2, the charging energy would be... The accumulated amount increased from 0 to 1.1; in the second frame, it did not cross the boundary, and the accumulated amount decreased to [amount missing]. The third frame did not exceed the limit, and the accumulated amount decayed to [value missing]. Subsequent frames showed no boundary violations, and the accumulated amount continued to decay exponentially, never reaching the threshold of 3.0. The system did not trigger anomaly detection, successfully filtering out the transient pulse disturbance. Suppose that in another time period, four consecutive frames all exceeded the boundary with a positive gradient of 0.3. The accumulated amount in the first frame would be 1.15, in the second frame 0.85 × 1.15 + 1.15 = 2.1275, in the third frame 0.85 × 2.1275 + 1.15 = 2.9584, and in the fourth frame 0.85 × 2.9584 + 1.15 = 3.6646. At this point, the accumulated amount exceeded the threshold of 3.0, and the system confirmed a persistent real motion anomaly. The Boolean characteristic value for the safety state was assigned 0, triggering safety degradation measures. Through the above-mentioned time-series continuous crossover discrimination mechanism based on exponential decay accumulation, without increasing the cost of additional hardware sensors, transient isolated pulses are automatically filtered out and do not trigger false alarms because the number of continuous frames is insufficient and the exponential decay of subsequent frames rapidly dissipates their injected energy. Continuous real anomalies are promptly and accurately identified and responded to because of the superposition of out-of-bounds charging in multiple consecutive frames.

[0090] In summary, the exoskeleton safety detection method based on dynamic reconstruction of the error decision domain according to the embodiments of this application is explained. It fundamentally solves the industry pain point of confusion between wearing abnormalities and movement abnormalities in the prior art through the dynamic reconstruction mechanism of the error decision domain. While significantly reducing the misjudgment rate caused by loose wearing, it ensures the system reliability of the decision structure during the dynamic adjustment process through the stability constraint mechanism, and improves the comprehensiveness of detecting real abnormalities in complex movement scenarios by constructing a multidimensional hyperellipsoidal decision domain.

[0091] Furthermore, an exoskeleton safety detection device based on dynamic reconstruction of the error determination domain is also provided.

[0092] Figure 5 This is a block diagram of an exoskeleton safety detection device based on dynamic reconstruction of the error determination domain, according to an embodiment of this application. Figure 5 As shown, the exoskeleton safety detection device 400 based on error judgment domain dynamic reconstruction according to an embodiment of this application includes: a synchronous acquisition module 410, used to synchronously acquire human-machine coupling sensing data at a preset sampling frequency, the human-machine coupling sensing data including strap tension signal, buckle pressure signal, joint angle signal, joint angular velocity signal, and torso posture angle signal; a signal processing module 420, used to perform signal classification and analog-to-digital conversion processing on the human-machine coupling sensing data to obtain exoskeleton connection constraint signal and joint motion state signal; and a wear state calculation module 430, used to calculate the exoskeleton connection constraint signal and joint motion state signal based on pre-calibrated loosening reference value and fitting reference value. The skeletal connection constraint signal and joint motion state signal are quantized using multi-path wear state indices to obtain a normalized wear coupling degree index; the decision domain reconstruction module 440 is used to dynamically reconstruct the error decision domain and verify the constraints of the basic hyperellipsoidal decision domain driven by the normalized wear coupling degree to obtain a stable constraint dynamic decision domain; the error extraction module 450 is used to extract multi-dimensional motion error features from the measured and expected values ​​of the joint motion state signal to obtain a joint motion error vector; the safety determination module 460 is used to determine the domain-inside / outside attribution of the joint motion error vector and the stable constraint dynamic decision domain to obtain a safety determination result.

[0093] The implementation method of the device involved in this embodiment is the same as that of the method described above, and will not be repeated here.

[0094] According to embodiments of this application, this application also provides an electronic device and a readable storage medium.

[0095] Figure 6A schematic block diagram of an example electronic device that can be used to implement embodiments of this application is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0096] like Figure 6 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0097] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0098] The computing unit 801 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as the exoskeleton safety detection method based on dynamic reconstruction of the error decision domain. For example, in some embodiments, the exoskeleton safety detection method based on dynamic reconstruction of the error decision domain can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the exoskeleton safety detection method based on dynamic reconstruction of the error decision domain described above can be performed. Alternatively, in other embodiments, the computing unit 801 may be configured by any other suitable means (e.g., by means of firmware) to perform an exoskeleton safety detection method based on error decision domain dynamic reconstruction.

[0099] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0100] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0101] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0102] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0103] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0104] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0105] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this application can be achieved, and this is not limited herein.

[0106] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of those features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0107] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for safety detection of exoskeletons based on dynamic reconstruction of error decision domain, characterized in that, include: S1, synchronously collect human-machine coupling sensing data at a preset sampling frequency. The human-machine coupling sensing data includes strap tension signal, buckle pressure signal, joint angle signal, joint angular velocity signal, and torso posture angle signal. S2, perform signal classification and analog-to-digital conversion on human-machine coupled sensing data to obtain exoskeleton connection constraint signals and joint motion state signals; S3, based on the pre-calibrated loosening reference value and fitting reference value, performs multi-path wearing state index quantification on the exoskeleton connection constraint signal and joint motion state signal to obtain a normalized wearing coupling index. S4. Dynamic reconstruction of error decision domain and constraint verification of the basic hyperellipsoid decision domain driven by normalized wear coupling degree to obtain stable constraint dynamic decision domain. S5, extract the multidimensional motion error features of the measured and expected values ​​of the joint motion state signal to obtain the joint motion error vector; S6 performs a domain-inside / outside attribution determination on the joint motion error vector and the dynamic decision domain of stability constraints to obtain a safety determination result.

2. The exoskeleton safety detection method based on dynamic reconstruction of the error determination domain according to claim 1, characterized in that, Step S2 includes: Based on a fixed step-size sampling clock, channel separation processing is performed on human-machine coupled sensing data to obtain discrete constraint state sequences and discrete motion state sequences; The discrete constraint state sequence is smoothed and reconstructed in the time domain to obtain the exoskeleton connection constraint signal; Multi-source hardware phase alignment processing is performed on discrete motion state sequences to obtain joint motion state signals.

3. The exoskeleton safety detection method based on dynamic reconstruction of the error determination domain according to claim 1, characterized in that, Step S3 includes: Based on the pre-calibrated loosening reference value and fitting reference value, the direct coupling metric of the exoskeleton connection constraint signal is quantified to obtain the direct coupling degree sub-index. Indirect attitude following feature quantization is performed on the joint motion state signal to obtain the indirect coupling degree sub-index; Based on the preset path confidence weight allocation coefficient, the direct coupling degree sub-indices and indirect coupling degree sub-indices are weighted and summed and normalized by boundary truncation to obtain the normalized wear coupling degree index.

4. The exoskeleton safety detection method based on dynamic reconstruction of error determination domain according to claim 1, characterized in that, Step S4 includes: Based on the preset maximum amplification gain base and nonlinear curvature index of each dimension, the normalized wear coupling index is dynamically amplified by dimension to obtain the dimension threshold amplification coefficient matrix. Based on the dimension threshold magnification coefficient matrix, the pre-stored basic hyperellipsoid boundary constants are used to reconstruct the basic boundary hyperellipsoid parameters to obtain the preliminary reconstruction decision domain parameter matrix. Based on the dual truncation clamping logic of single-frame change rate constraint and absolute amplitude upper limit constraint, the parameter matrix of the preliminary reconstructed decision domain is subjected to oscillation suppression and numerical convergence verification to obtain a stable constraint dynamic decision domain.

5. The exoskeleton safety detection method based on dynamic reconstruction of the error determination domain according to claim 1, characterized in that, Step S5 includes: The machine-measured kinematic state differential solution is performed on the exoskeleton joint angle sequence in the joint motion state signal to obtain the machine-measured motion state sequence, and the human reference motion state sequence is separated out. The human body reference motion state sequence is mapped and projected from the trunk to the target joint space using a biomechanical kinematic projection matrix to obtain the human body expected motion state sequence. Multidimensional error fusion reconstruction is performed on the machine's measured motion state sequence and the human body's expected motion state sequence to obtain the joint motion error vector.

6. The exoskeleton safety detection method based on dynamic reconstruction of the error determination domain according to claim 1, characterized in that, Step S6 includes: The spatial distance between the components of each dimension of the joint motion error vector and the fault tolerance parameters of each dimension of the dynamic decision domain of the stability constraint is calculated to obtain the normalized spatial error distance scalar. The normalized spatial error distance scalar is subjected to domain-inside / outside envelope traversal discrimination and logical assignment to obtain the safe state Boolean characteristic value; Based on the underlying communication protocol's frame encapsulation rules and security response strategy mapping table, the security state Boolean feature values ​​are encoded and verified to obtain the security determination result.

7. The exoskeleton safety detection method based on dynamic reconstruction of the error decision domain according to claim 4, characterized in that, Based on a dual-truncation clamping logic of single-frame change rate constraint and absolute amplitude upper limit constraint, oscillation suppression and numerical convergence verification are performed on the parameter matrix of the preliminary reconstructed decision domain to obtain a stable constraint dynamic decision domain, including: The inter-frame change gradient is calculated for the initial reconstructed decision domain parameter matrix and the historical final decision domain parameters output in the previous calculation cycle. The change gradient is compared and verified with the preset maximum change rate extreme value of a single frame. If the change gradient exceeds the maximum change rate extreme value of a single frame, the current decision domain parameter is forcibly truncated and clamped to the boundary of the parameter of the previous frame plus or minus the extreme value. Perform an absolute amplitude upper limit check on the decision domain parameters after the rate of change truncation process to ensure that the threshold parameters of each dimension do not exceed the preset multiple upper limit of the corresponding basic threshold. If the adjusted decision domain parameters do not meet the above stability constraints, they will be forcibly limited to the constraint range and an early warning signal will be triggered and output to the exoskeleton control system. The final three-dimensional threshold parameters, which have undergone double convergence constraint verification and correction, will be solidified and encapsulated to form a stable constraint dynamic decision domain.

8. The exoskeleton safety detection method based on dynamic reconstruction of the error determination domain according to claim 6, characterized in that, The normalized spatial error distance scalar is subjected to in-domain and out-of-domain envelope traversal discrimination and logical assignment to obtain the safe state Boolean characteristic value, including: Based on instantaneous comparison of the standard critical threshold of the hyperellipsoid, the normalized spatial error distance scalar is marked with out-of-bounds events and encoded with gradient direction to obtain the composite feature quantity of out-of-bounds state and gradient. Based on the positive gradient weighted amplification coefficient, the dynamic balance solution of the charging and discharging of the cross-boundary state and the gradient composite feature is performed to obtain the cumulative amount of boundary crossing persistence. By comparing the cumulative amount of boundary crossing persistence with the preset persistence determination threshold using inequality logic, the persistence limit overrun detection and binarization logic assignment are performed on the cumulative amount of boundary crossing to obtain the Boolean feature value of the safe state.

9. The exoskeleton safety detection method based on dynamic reconstruction of the error determination domain according to claim 6, characterized in that, Based on the frame encapsulation rules and security response policy mapping table of the underlying communication protocol, the Boolean characteristic values ​​of the security state are encoded and verified to obtain the security determination result, including: Based on the underlying communication protocol, the safety response strategy mapping table is retrieved. If the safety state Boolean characteristic value is a safe state, the control code that maintains the current assist torque is retrieved. If the safety state Boolean characteristic value is an abnormal state, the control code that executes torque decay, emergency stop and triggers audible and visual warnings is retrieved. The retrieved control code is spliced ​​with frame headers and trailers and a cyclic redundancy check code is added. It is then encapsulated into a standardized machine instruction stream that meets the requirements of the communication physical layer and output as the security determination result.

10. An exoskeleton safety detection device based on dynamic reconstruction of error decision domain, used to execute the exoskeleton safety detection method based on dynamic reconstruction of error decision domain as described in any one of claims 1-9, characterized in that, include: The synchronous acquisition module is used to synchronously acquire human-machine coupling sensing data at a preset sampling frequency. The human-machine coupling sensing data includes strap tension signal, buckle pressure signal, joint angle signal, joint angular velocity signal, and torso posture angle signal. The signal processing module is used to classify and convert human-machine coupled sensing data into analog-to-digital signals to obtain exoskeleton connection constraint signals and joint motion state signals. The wearable state calculation module is used to quantify the wearable state index of the exoskeleton connection constraint signal and joint motion state signal based on the pre-calibrated loosening reference value and fitting reference value to obtain the normalized wearable coupling degree index. The decision domain reconstruction module is used to dynamically reconstruct the error decision domain and verify the constraints of the basic hyperellipsoid decision domain driven by the normalized wear coupling degree to obtain a stable constraint dynamic decision domain. The error extraction module is used to extract multi-dimensional motion error features from the measured and expected values ​​of the joint motion state signal to obtain the joint motion error vector. The safety determination module is used to determine the internal and external attribution of the joint motion error vector to the dynamic determination domain of stability constraints in order to obtain the safety determination result.

Images