Auxiliary system and method for handicapped with intelligent mechanical arm

Abstract

The invention relates to the technical field of body-equipped robots, in particular to a body-equipped intelligent mechanical arm handicapped assisting system and method.The method comprises the steps that electromyographic signals are collected in real time, electromyographic features are extracted, and the fatigue degree score and the future trend of a user are predicted through a layered deep learning model; based on the prediction result, establishing an adaptive compensation strategy, and dynamically amplifying the original control signal attenuated due to fatigue; continuously evaluating the signal-to-noise ratio and the stability of the compensated signal; if the effect is not good, triggering parameter fine tuning; and after the user state is recovered and the control performance reaches the standard, starting a progressive exit program, and smoothly cancelling compensation. According to the method, the complete closed-loop design of perception-prediction-compensation-evaluation is constructed, the accuracy and stability during control over the mechanical arm with the body are effectively improved, and the handicapped user can remarkably adapt to control over the mechanical arm with the body.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of embodied robots, in particular to an embodied intelligent mechanical arm disability assistance system and method. BACKGROUND

[0002] Muscle fatigue accumulation problems commonly exist in long-time control of mechanical arms by disabled persons. With the extension of use time, the control muscle group will gradually produce fatigue, resulting in a decrease in the amplitude of the electromyographic signal, a shift in the frequency component, and a decrease in signal stability. The fatigue-induced attenuation of the electromyographic signal directly affects the control accuracy and response speed of the mechanical arm, and in severe cases, can even cause control failure, forcing the user to interrupt the operation and rest.

[0003] Existing electromyographic control systems mostly use fixed signal processing parameters and control strategies, lacking real-time perception of user fatigue state and dynamic adaptation capability. Although some systems introduce simple signal gain adjustment, such adjustment is usually based on pre-set fixed rules and cannot accurately reflect individual differences and dynamic characteristics of fatigue development. At the same time, the existing technology lacks an effective feedback evaluation mechanism in the fatigue compensation process, making it difficult to determine whether the compensation effect meets the expectations, and also unable to exit the compensation mode in time after the user recovers from fatigue.

[0004] In particular, for severely disabled persons who need to rely on mechanical arms for daily activities for a long time, the problem of control performance decline caused by muscle fatigue is more prominent. This not only limits the continuous use time of the mechanical arm, but also affects the life quality and autonomy of the disabled user.

[0005] In view of the above, the present application proposes an embodied intelligent mechanical arm disability assistance system and method. SUMMARY

[0006] To achieve the above-mentioned purpose, the present application provides an embodied intelligent mechanical arm disability assistance system and method, and the specific technical solutions are as follows:

[0007] An embodied intelligent mechanical arm disability assistance method, comprising:

[0008] Collecting electromyographic signals when a user controls an embodied mechanical arm and extracting electromyographic features, the electromyographic features including time domain and frequency domain features of the electromyographic signals, analyzing the electromyographic signals through short-time Fourier transform and wavelet decomposition, establishing a feature vector sequence containing fatigue indicators including average power frequency, median frequency, and electromyographic amplitude decline rate, and recording the duration of controlling the mechanical arm;

[0009] A hierarchical fatigue state prediction model is constructed, the local muscle fatigue pattern is extracted by a bottom convolutional network, and the long-term cumulative effect is captured by a top gated recurrent unit, the input of the hierarchical fatigue state prediction model is the electromyographic feature, the historical fatigue state and the task intensity, and the output is the current fatigue degree score and the future fatigue trend;

[0010] An electromyographic signal fatigue compensation strategy is created, a compensation gain curve is generated according to the current fatigue degree score and the future fatigue trend, the original electromyographic signal is adaptively amplified, and the electromyographic signal attenuation caused by fatigue is compensated;

[0011] The signal-to-noise ratio and stability index of the compensated electromyographic signal are calculated, the electromyographic signal power spectral density change is analyzed, the current compensation strategy is maintained when the signal-to-noise ratio and stability index of the compensated electromyographic signal are improved, otherwise the compensation parameter fine tuning is triggered;

[0012] When the signal-to-noise ratio and stability index of the electromyographic signal are continuously detected to be improved, and the body-attached robot control performance index meets the preset recovery threshold, the compensation strength is gradually reduced according to the negative exponential law until the compensation function is exited.

[0013] Preferably, when the user controls the body-attached robot, the original electromyographic signal is collected through the arranged surface electromyographic electrode array, and the power frequency interference and motion trajectory are filtered through a band-pass filter;

[0014] The filtered electromyographic signal is processed in segments, the root mean square value, the average absolute value and the waveform length are calculated and extracted as time domain features in each time window; the power spectral density of each time period is calculated by using short-time Fourier transform, and the average power frequency and the median frequency are obtained as frequency domain features based on this;

[0015] The electromyographic amplitude decline rate is calculated by continuously monitoring the change trend of the electromyographic root mean square value, and the time domain features, frequency domain features and fatigue indicators are organized into a feature vector sequence in chronological order.

[0016] Preferably, the hierarchical fatigue state prediction model includes a bottom architecture and a top architecture;

[0017] The bottom architecture of the hierarchical fatigue state prediction model is a convolutional neural network, the convolutional neural network extracts local patterns from the feature vector sequence by stacking multiple convolutional structures and using multiple convolutional kernels of different sizes, and obtains fatigue time series pattern features;

[0018] A batch normalization layer is introduced in the convolutional neural network to standardize the output of each convolutional structure, and linear transformation is performed through scaling parameters and translation parameters.

[0019] Preferably, the top layer architecture of the hierarchical fatigue state prediction model is a gated recurrent unit network, which is used to model the time-dependent relationship of fatigue state and selectively remember and forget historical information through update gate and reset gate mechanisms;

[0020] The input of the hierarchical fatigue state prediction model is in the form of a triple, including an electromyographic feature vector, a historical fatigue state vector obtained by weighted averaging of fatigue degree scores in a sliding time window, and a task intensity scalar obtained by comprehensive evaluation and normalization of robot arm load, movement speed, and operation complexity;

[0021] The output layer of the hierarchical fatigue state prediction model is designed as a double-branch structure, with the first branch outputting the current fatigue degree score and the second branch predicting a sequence of future fatigue degree scores.

[0022] Preferably, the creation of the electromyographic signal fatigue compensation strategy includes designing a dynamic compensation gain function, the base value of which is obtained from the current fatigue degree score through a nonlinear mapping relationship;

[0023] A pre-compensation factor is calculated in combination with the rate of change of the future fatigue trend to adjust the comprehensive compensation gain, and the base compensation value is multiplied by the pre-compensation factor to obtain the comprehensive compensation gain;

[0024] An exponential weighted moving average method is used to perform time domain smoothing on the comprehensive compensation gain to generate a smoothed compensation gain curve.

[0025] Preferably, the electromyographic signal fatigue compensation strategy further includes decomposing the original electromyographic signal into multiple frequency band components through a band-pass filter, and setting differentiated compensation gains for each frequency band component according to the degree of influence of fatigue on different frequency components for selective compensation of frequency;

[0026] The compensated electromyographic signal is obtained by weighted summation of the frequency band components, and an upper limit constraint and a rate limit of the compensation gain are set;

[0027] A real-time evaluation mechanism for the compensation effect is established, the compensation effect is quantified by calculating the energy ratio of the signals before and after compensation, and the compensation parameters are adjusted adaptively according to the evaluation results.

[0028] Preferably, the signal-to-noise ratio and stability index of the compensated electromyography signal are calculated, including: in the sliding time window, the electromyography signal is decomposed into effective signal component and noise component, the signal-to-noise ratio is calculated by estimating the signal energy during active contraction and the signal variance during rest; the stability index is obtained by analyzing the time domain fluctuation characteristics of the electromyography signal, including calculating the coefficient of variation of the signal envelope, the zero-crossing rate of the signal and the spectral stability index, the signal envelope is extracted by Hilbert transform; the Welch method is used to estimate the power spectral density in each window, and the spectral stability index is calculated by comparing the power spectral density distribution of adjacent windows;

[0029] The signal-to-noise ratio and stability index are weighted and fused to construct a unified quality score to quantitatively evaluate the compensation effect.

[0030] Preferably, when the compensation parameter fine-tuning is triggered, the adaptive fine-tuning mechanism of the compensation parameter is triggered, and the parameter adjustment amount is calculated by the gradient descent method;

[0031] When the quality score does not improve after continuous adjustment, the parameter reset strategy is started, the compensation parameter is restored to the historical optimal configuration, and the historical information of each parameter adjustment is recorded to construct a parameter optimization trajectory database to provide data support for subsequent optimization.

[0032] Preferably, a multi-dimensional exit condition judgment mechanism is established, and when the signal-to-noise ratio and stability index sequence continuously meet the preset threshold value for a specified number of times, it is determined that the signal quality has reached stable improvement;

[0033] The control performance index of the embodied robotic arm is evaluated, including trajectory tracking accuracy, response delay and operation smoothness, and the index is weighted and fused into a comprehensive control performance evaluation function, and when the value of the comprehensive control performance evaluation function meets the preset recovery threshold, it is determined that the control performance meets the standard;

[0034] When the conditions of signal quality improvement and control performance recovery are met at the same time, the gradual exit process of compensation strength is started, and the compensation gain decreases according to the negative exponential law; during the exit process, safety monitoring is implemented by monitoring the performance fluctuation index and the rebound of the fatigue degree score, and if an abnormality occurs, the exit process is suspended or terminated; when the compensation gain is reduced to the threshold value of no compensation state and the control performance remains stable, the exit is completed.

[0035] A embodied intelligent robotic arm disability assistance system for implementing the embodied intelligent robotic arm disability assistance method, comprising: a physiological signal sensing module, a fatigue state prediction module, a control signal compensation module, a compensation effect evaluation module and a compensation gradual exit module;

[0036] The physiological signal sensing module is configured to collect electromyographic signals when a user controls the embodied robotic arm and extract electromyographic features, the electromyographic features including time domain and frequency domain features of the electromyographic signals, analyze the electromyographic signals through short-time Fourier transform and wavelet decomposition, and establish a feature vector sequence including fatigue indicators of average power frequency, median frequency and electromyographic amplitude drop rate, while recording the duration of controlling the robotic arm;

[0037] The fatigue state prediction module is configured to construct a hierarchical fatigue state prediction model, extract local muscle fatigue patterns through a bottom convolutional network, capture long-term cumulative effects through a top gated recurrent unit, and input electromyographic features, historical fatigue states and task intensities into the hierarchical fatigue state prediction model to output a current fatigue degree score and a future fatigue trend.

[0038] The control signal compensation module is configured to create an electromyographic signal fatigue compensation strategy, generate a compensation gain curve according to the current fatigue degree score and the future fatigue trend, and adaptively amplify the original electromyographic signals to compensate for electromyographic signal attenuation caused by fatigue.

[0039] The compensation effect evaluation module is configured to calculate the signal-to-noise ratio and stability indicators of the compensated electromyographic signals, analyze the power spectral density changes of the electromyographic signals, maintain the current compensation strategy when the signal-to-noise ratio and stability indicators of the compensated electromyographic signals are detected to be improved, and otherwise trigger compensation parameter fine-tuning.

[0040] The compensation gradual exit module is configured to gradually reduce the compensation strength according to a negative exponential law when the signal-to-noise ratio and stability indicators of the electromyographic signals are continuously detected to be improved, and the control performance indicators of the embodied robotic arm meet a preset recovery threshold, until the compensation function is exited.

[0041] The application has the following beneficial effects: The application accurately depicts electromyographic fatigue evolution and spectral drift, constructs a robust fatigue indicator sequence using average power frequency, median frequency and amplitude drop rate, and correlates task duration, improves the distinguishability and time sequence traceability of fatigue states under different muscle groups and task intensities, and provides high signal-to-noise ratio input for subsequent modeling.

[0042] The hierarchical model constructed by the application combines local morphological detection of the convolutional network and long-range dependence modeling of the GRU, takes into account transient fatigue features and cumulative load effects, fuses historical fatigue and task intensity of synchronous input, outputs a calibratable fatigue degree score and trend prediction, enhances the generalization and early warning ability of non-stationary electromyography, and reduces overfitting and delay.

[0043] The application amplifies the electromyographic amplitude attenuated with fatigue deepening based on the adaptive gain curve driven by prediction, maintains the dynamic range and controllability of the control signal, reduces the overuse and repeated calibration frequency of the user; the compensation strategy takes into account the real-time score and trend, avoids overcompensation to introduce jitter, and improves the stability and comfort of interaction.

[0044] The application verifies the compensation effect and adaptively fine-tunes the parameters through the closed-loop evaluation of SNR and stability indicators, ensures that the spectral energy distribution and signal consistency evolve in the expected direction; when the effect is insufficient, it adjusts immediately, suppresses noise amplification and drift accumulation, and maintains the operation of the robot arm in a robust interval.

[0045] After meeting the stable improvement and performance recovery threshold, the application uses a negative exponential method to smoothly reduce the compensation strength, avoids sudden changes that cause control performance to fall back and users to be uncomfortable; the gradual exit mechanism maintains the sensitivity to residual fatigue, realizes seamless transition from compensation dependence to natural control, and improves long-term usability. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A body-integrated intelligent robot arm disability assistance method flowchart is provided for the application;

[0047] Figure 2 A hierarchical fatigue state prediction model construction flowchart is provided for the application;

[0048] Figure 3 An electromyographic signal fatigue compensation flowchart is provided for the application;

[0049] Figure 4 A compensation signal quality evaluation and fine-tuning flowchart is provided for the application;

[0050] Figure 5 A body-integrated intelligent robot arm disability assistance system structure diagram is provided for the application. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application will be described in detail below in conjunction with the drawings of the specification.

[0052] In the following description, many specific details are set forth in order to provide a thorough understanding of the application, but the application can also be implemented in other ways different from the description, and those skilled in the art can make similar extensions without departing from the connotation of the application, therefore the application is not limited by the specific embodiments disclosed below.

[0053] Second, the "one embodiment" or "an embodiment" referred to herein can include, but is not limited to, any particular feature, structure, or characteristic including those described in this specification. The various embodiments described throughout this specification are not necessarily mutually exclusive, but can be implemented in combination with each other.

[0054] Embodiment 1

[0055] Reference Figures 1 to 4 For the first embodiment of the present application, as shown in the accompanying drawings, a body-possessed intelligent mechanical arm disability assistance method is provided. Figure 1

[0056] Step 1: Collect the electromyographic signals of the user when controlling the body-possessed mechanical arm and extract the electromyographic features, including the time domain and frequency domain features of the electromyographic signals, analyze the electromyographic signals through short-time Fourier transform and wavelet decomposition, establish a feature vector sequence containing the fatigue index of the average power frequency, median frequency, and electromyographic amplitude decline rate, and record the duration of controlling the mechanical arm.

[0057] Specifically, in the process of the user operating the body-possessed mechanical arm, the electromyographic signals of the target muscle group are collected through a surface electromyographic electrode array, and the electrode array is arranged on the surface of the key control muscle groups such as the forearm flexor muscle group, the extensor muscle group, the biceps brachii, and the triceps brachii. The original electromyographic signals are digitized and collected at a preset sampling frequency, and the power frequency interference and motion trajectory are filtered through a band-pass filter. For example, the sampling frequency can be set to be within the range of 1000 Hz to 2000 Hz, and the passband range of the band-pass filter can be set to be within the range of 20 Hz to 500 Hz.

[0058] Extract the time domain features of the electromyographic signals, segment the filtered electromyographic signals, and calculate the root mean square value of the electromyographic signals in each time window , wherein represents the electromyographic signal amplitude of the i-th sampling point in the j-th time window, represents the total number of sampling points in the time window. At the same time, the average absolute value and waveform length ; the time domain features can directly reflect the change rule of muscle contraction intensity, providing basic data support for subsequent fatigue state evaluation.

[0059] Extract the frequency domain features of the electromyographic signals, perform time-frequency analysis on the electromyographic signals through short-time Fourier transform, and calculate the power spectral density of each time period through a sliding window method. For the j-th time window, the short-time Fourier transform expression is , wherein

[0060] , wherein​​​ Represents the window function. Indicates from the first The signal sequence starting from each sampling point Represents frequency variables. Indicates the sampling frequency. Indicates the window length. Calculates the average power frequency based on the power spectral density. ,in Represents frequency Power spectral density at that point and These represent the lower and upper limits of the analysis band, respectively. Median frequency. Defined as the frequency points that equally divide the area under the power spectral density curve, satisfying The decreasing trend of these two frequency domain indices can effectively characterize the development process of muscle fatigue.

[0061] The muscle fatigue index is calculated by continuously monitoring the trend of changes in electromyography (EMG) amplitude and determining the rate of decrease in EMG amplitude. As an indicator of muscle fatigue, among This represents the root mean square value of electromyography (EMG) at the start of the task. This represents the root mean square value of electromyography (EMG) at the current moment. The time-domain features, frequency-domain features, and fatigue indicators are organized into a feature vector sequence in chronological order. ,in Indicates the first The feature vector at each time step This represents the total number of time sampling points.

[0062] At the same time, the cumulative time the user controls the robotic arm is continuously recorded. This information is used as an important reference indicator for the cumulative effect of fatigue. By establishing a correspondence between the timestamp sequence and the feature vector sequence, the time process of fatigue development can be accurately tracked. For example, the feature vector can be collected and updated every 30 seconds to ensure the real-time monitoring of fatigue status.

[0063] This step, through multi-dimensional and multi-scale signal processing and feature extraction methods, can comprehensively capture fatigue-related information contained in electromyographic signals, providing a reliable data foundation for subsequent fatigue state prediction and compensation control, and effectively improving the ability of the assistive system to perceive the user's muscle state and the accuracy of its response.

[0064] Step 2: Construct a hierarchical fatigue state prediction model. The bottom-level convolutional network extracts local muscle fatigue patterns, while the top-level gated recurrent units capture long-term cumulative effects. The inputs to the hierarchical fatigue state prediction model are electromyographic features, historical fatigue states, and task intensity. The outputs are the current fatigue level score and future fatigue trend. (See also...) Figure 2A flow chart of the hierarchical fatigue state prediction model construction process for this step is shown.

[0065] Specifically, the hierarchical fatigue state prediction model is constructed using a deep learning architecture, which includes a bottom layer architecture and a top layer architecture. A one-dimensional convolutional neural network is deployed in the bottom layer architecture to process the feature vector sequence extracted in step 1. The convolutional layer of the one-dimensional convolutional neural network extracts local patterns from the input features through multiple convolution kernels of different sizes. The convolution operation of the convolutional layer includes gradually extracting fatigue time series pattern features from simple to complex by stacking multiple convolutional structures.

[0066] On the basis of the convolutional network, a batch normalization layer is introduced to standardize the output of each layer, followed by a linear transformation through learnable scaling and shifting parameters, ensuring that the one-dimensional convolutional neural network has sufficient expressive power.

[0067] A gated recurrent unit network is deployed in the top layer architecture to model the time-dependent relationship of fatigue state. The update gate calculation formula of the gated recurrent unit is The reset gate calculation formula is wherein represents the update gate output at the moment, represents the reset gate output at the moment, and represent the weight matrix of the update gate and the reset gate, respectively, represents the hidden state at the previous moment, represents the input feature at the current moment, and represent the corresponding bias vectors. The candidate hidden state is calculated by , wherein represents the candidate state weight matrix, represents the bias vector, represents the element-wise multiplication. The final hidden state is updated as The gating mechanism enables the hierarchical fatigue state prediction model to selectively remember and forget historical information, accurately capturing the long-term effects of fatigue accumulation.

[0068] The input of the hierarchical fatigue state prediction model is designed in a triple form, including the real-time electromyographic feature vector , the historical fatigue state vector , and the task intensity scalar . The historical fatigue state vector is obtained by weighted averaging the fatigue degree scores within the sliding time window, with the weights assigned according to the time decay function , wherein represents the decay coefficient, denotes the current time, denotes the historical time. The task intensity is evaluated according to the robot arm load, motion speed and operation complexity, and is normalized to a unified dimension range. For example, the task intensity can be divided into ten levels, corresponding to scalar values of 0.1 to 1.0 respectively.

[0069] The output layer of the hierarchical fatigue state prediction model is designed as a double-branch structure, and the output layer includes a first branch and a second branch. The first branch of the output layer outputs the current fatigue degree score , wherein denotes the score weight vector, denotes the final hidden layer output, denotes the score bias term, and the sigmoid function maps the score to the interval of 0 to 1. The second branch of the output layer predicts the future fatigue trend, and outputs a sequence of fatigue degree scores at multiple future time steps through a linear layer , wherein denotes the prediction weight matrix, denotes the prediction bias vector. For example, the fatigue degree score change every 30 seconds within the next 5 minutes can be predicted.

[0070] The hierarchical fatigue state prediction model is trained in an end-to-end supervised learning manner, and the loss function of the training adopts mean square error; the network parameters are iteratively updated by an optimization algorithm, so that the hierarchical fatigue state prediction model can accurately predict the fatigue state and its development trend.

[0071] This step fully utilizes the local feature extraction capability of the convolutional network and the time series modeling advantage of the recurrent network by constructing a hierarchical deep learning model, realizes accurate evaluation and trend prediction of the muscle fatigue state, provides reliable decision basis for subsequent adaptive compensation control, and significantly improves the intelligent level and user experience of the disabled assistance system.

[0072] Step 3: Create an electromyographic signal fatigue compensation strategy, generate a compensation gain curve according to the current fatigue degree score and future fatigue trend, and adaptively amplify the original electromyographic signal to compensate for the electromyographic signal attenuation caused by fatigue. Referring to Figure 3 , the electromyographic signal fatigue compensation flowchart for this step.

[0073] Specifically, based on the current fatigue degree score and future fatigue trend prediction results output by step 2, a dynamic compensation gain function is designed. The base value of the dynamic compensation gain function is obtained by mapping the fatigue degree score, and the mapping relationship is , wherein denotes the base compensation value, denotes the current fatigue degree score, denotes the fatigue compensation coefficient, denotes the non-linear adjustment exponent. This non-linear mapping ensures a moderate compensation at mild fatigue, while enhancing the compensation at severe fatigue to avoid control instability caused by over-compensation.

[0074] The compensation value is adjusted prospectively in combination with the future fatigue trend. The rate of change of the future fatigue trend is calculated wherein denotes the predicted fatigue value at the future time instant, denotes the prediction time step. When an accelerated rise in fatigue trend is detected, a pre-compensation factor is introduced, wherein denotes the pre-compensation weight coefficient, is the max function. The overall compensation gain is calculated as The constructed pre-compensation strategy can cope with the upcoming fatigue peak in advance, maintaining the continuity of control performance.

[0075] The time-domain smoothing of the compensation gain curve adopts the exponential weighted moving average method, and the smoothed gain value is wherein denotes the smoothed gain value at the time instant, denotes the smoothing factor, which ranges from 0 to 1, denotes the overall compensation gain at the time instant.

[0076] A frequency-dependent compensation strategy is designed for the myoelectric signals of different frequency components. The original myoelectric signal is decomposed into multiple frequency band components by a band-pass filter, and the center frequency of the th frequency band is with a bandwidth of . The compensation gain of each frequency band is set differently according to the degree of fatigue influence on different frequency components, and the calculation formula is:

[0077] ;

[0078] wherein, denotes the compensation gain of the th frequency band, denotes the compensation adjustment coefficient of the th frequency band, denotes the peak frequency with the most significant fatigue influence, denotes the standard deviation of frequency selectivity, is the natural exponential function. The frequency-selective compensation method can more accurately restore the specific frequency components attenuated due to fatigue.

[0079] The compensated myoelectric signal is obtained by weighted sum of each frequency band component, expressed as wherein denotes the compensated myoelectric signal, denotes the signal component of the i-th frequency band, denotes the signal component of the i-th frequency band, denotes the total number of frequency bands. For example, the frequency range from 20 Hz to 500 Hz can be divided into 8 frequency bands, and each frequency band is extracted by a Butterworth filter.

[0080] To prevent overcompensation, an upper limit constraint and a rate limit of the compensation gain are set. The maximum value of the compensation gain is limited to wherein denotes the maximum compensation factor, denotes the cumulative control time, denotes the time decay constant. The gain rate limit is wherein denotes the maximum allowed rate. When the calculated gain rate exceeds the limit, it is constrained within the safe range by a gradient clipping technique.

[0081] In addition, a real-time evaluation mechanism of the compensation effect is established, and the compensation effect is quantified by calculating the energy ratio of the signal before and after compensation wherein denotes the evaluation time window length. When the energy ratio deviates from the target range, the compensation parameters are adjusted adaptively to ensure that the compensated signal can effectively resist the influence of fatigue and will not cause signal distortion.

[0082] This step can dynamically adjust the compensation strategy according to the real-time fatigue state and development trend of the user by constructing a multi-level and adaptive fatigue compensation strategy, can effectively restore the control signal attenuation caused by muscle fatigue, maintain the precision and stability of the robot arm control, can significantly prolong the effective operation time of the user on the equipment such as the embodied robot arm, and improve the practicability and user satisfaction of the assistive device.

[0083] Step 4: Calculate the signal-to-noise ratio and stability index of the compensated myoelectric signal, analyze the power spectral density change of the myoelectric signal by a sliding window, and maintain the current compensation strategy when the signal-to-noise ratio and stability index of the compensated myoelectric signal are improved, otherwise trigger the compensation parameter fine tuning. Referring to Figure 4 , the compensation signal quality evaluation and fine tuning flowchart of this step.

[0084] Specifically, the quality of the compensated myoelectric signal output by step 3 is evaluated, including calculating the signal-to-noise ratio index and calculating the stability index. The calculation of the signal-to-noise ratio index includes: in a sliding time window, the myoelectric signal is decomposed into effective signal component and noise component, and the signal-to-noise ratio calculation formula is wherein denotes the effective signal power, denotes the noise power. The effective signal power is estimated by the signal energy during active contraction: where denotes the signal amplitude at the th sampling point, denotes the signal mean value, denotes the number of sampling points during contraction period. The noise power is estimated by the signal variance during rest period, ensuring accurate distinction between the effective control signal and background noise.

[0085] The calculation of the stability index includes: obtaining the stability index by analyzing the time-domain fluctuation characteristics of the electromyographic signal, calculating the coefficient of variation of the signal envelope in a continuous time period where denotes the standard deviation of the signal envelope, denotes the mean value of the signal envelope. The signal envelope is extracted by Hilbert transform, where denotes the original signal, denotes the signal after Hilbert transform. The coefficient of variation is smaller, indicating that the signal is more stable and the control performance is more reliable. At the same time, the zero-crossing rate of the signal is calculated where denotes the number of sampling points within the analysis window, denotes the sign function, and the stability of the zero-crossing rate reflects the consistency of the frequency components of the signal.

[0086] The electromyographic signal is continuously monitored using a sliding window method, with a window length of sampling points and a window step length of sampling points. In each window, the power spectral density is calculated, estimated using the Welch method, where denotes the number of segments, denotes the normalization factor of the window function, denotes the Fourier transform of the th segment of the signal. By comparing the power spectral density distribution of adjacent windows, the spectral stability index is calculated:

[0087] ;

[0088] where denotes the total number of frequency points, and denote the power spectral density of the current window and the previous window, respectively. For example, the window length can be set to 512 sampling points, and the step length is 128 sampling points, achieving a 75% window overlap.

[0089] An integrated evaluation index system is established, and the signal-to-noise ratio and stability index are integrated into a unified quality score , wherein , , denote the weight coefficient of each index and satisfy , denotes the reference signal-to-noise ratio level. By comparing the quality score changes before and after compensation , denotes the quality score after compensation, denotes the quality score before compensation, and the compensation effect is quantitatively evaluated. When , it is determined that the compensation strategy is effective, wherein denotes the improvement threshold.

[0090] When the quality score of the myoelectric signal is calculated and does not reach the expected improvement or worsens, an adaptive fine-tuning mechanism of the compensation parameters is triggered. The parameter adjustment amount is calculated by the gradient descent method, , wherein denotes the newly calculated compensation parameter vector, denotes the old compensation parameter vector, denotes the compensation parameter vector, denotes the learning rate, denotes the quality loss function, denotes the target quality level; the gradient estimation adopts the finite difference method:

[0091]

[0092] , wherein denotes a small perturbation amount, and an exemplary learning rate can be initially set to 0.01 and dynamically adjusted according to the convergence condition.

[0093] Further, a constraint mechanism for parameter adjustment is established to prevent over-adjustment from causing control instability of the embodied robotic arm. The upper and lower limit of parameter change is set , and the maximum amplitude of single adjustment is When the quality score still does not improve after continuous multiple adjustments, a parameter reset strategy is started, which can restore the compensation parameters to the historical optimal configuration to avoid falling into local optimum. The historical information of each parameter adjustment is recorded, and a parameter optimization trajectory database is constructed to store parameter optimization records and provide experience data for subsequent intelligent optimization.

[0094] This step, by monitoring the quality score of the compensated signal in real time and establishing a closed-loop feedback adjustment mechanism, ensures that the fatigue compensation strategy always remains in its optimal working state, adaptively responding to the compensation needs of different users and different levels of fatigue. This effectively improves the reliability of electromyographic signals and the accuracy of robotic arm control, providing a more stable and efficient assistive control experience for users with disabilities.

[0095] Step 5: When the signal-to-noise ratio and stability of the electromyographic signal are continuously improved, and the control performance of the robotic arm meets the preset recovery threshold, the compensation intensity is gradually reduced according to the negative exponential law until the compensation function is exited.

[0096] Specifically, a multi-dimensional exit condition judgment mechanism is established to continuously monitor the signal-to-noise ratio calculated in step 4. and stability Indicators. Set up a continuous detection counter. When the signal-to-noise ratio satisfies And the stability index meets As time progresses, the counter increments, where... Indicates the first Signal-to-noise ratio at any given moment. Indicates the first Stability metrics at any given time. and These represent the corresponding thresholds. The counter is reset to zero when any indicator falls below the threshold. Only when... Only when the signal quality has reached a stable and improved state is it considered that the signal quality has been stabilized. This indicates the required number of consecutive satisfactions. For example, the number of consecutive satisfactions can be set to 20, corresponding to a stable monitoring time of approximately 3 seconds.

[0097] Simultaneously, the control performance indicators of the embodied robotic arm are evaluated, including trajectory tracking accuracy, response latency, and operational smoothness. Trajectory tracking accuracy is obtained by calculating the root mean square error between the actual trajectory and the desired trajectory.

[0098] ;

[0099] in Indicates the first The actual location at that moment Indicates the desired position. This indicates the number of sampling points within the evaluation period. Response delay is calculated using cross-correlation analysis to determine the time difference between the control command and the executed action. ,in Indicates control signal With execution signal The cross-correlation function. Operational smoothness is quantized by the spectral entropy of the acceleration signal. ,in This represents the normalized power spectral density of the acceleration signal.

[0100] Construct a comprehensive control performance evaluation function:

[0101] ;

[0102] in , , Represents the weighting coefficients of each performance index and satisfies , , These represent the maximum acceptable trajectory tracking accuracy error and response delay, respectively. Indicates the reference smoothness level. When When the control performance is deemed to have reached the recovery standard, among which... This indicates the preset recovery threshold. This multi-index comprehensive evaluation method ensures the full recovery of the robotic arm's control quality.

[0103] When both signal quality improvement and control performance recovery conditions are met simultaneously, a gradual exit process for the compensation intensity is initiated. The compensation gain decreases according to a negative exponential law. ,in Indicates the first Compensation gain for each exit step, This indicates the current compensation gain value. This represents the decay rate parameter. This represents the minimum hold-up gain. The negative exponential decay law ensures a rapid reduction in compensation intensity initially, followed by a slow approach to a state without compensation, avoiding control fluctuations that might result from a sudden withdrawal of compensation. For example, the decay rate parameter can be set to 0.1, corresponding to approximately 10 steps to complete the main decay process.

[0104] During the compensation exit process, a safety monitoring mechanism will be implemented; performance fluctuation indicators will be defined. ,when At that time, the exit process will be paused and the current compensation level will be maintained, whereby... Indicates the maximum permissible performance volatility. For the first The comprehensive control performance evaluation function value for each exit step. Simultaneously, monitor the rebound of fatigue status; if the fatigue score is detected to rise again and the change exceeds a set threshold... ,Right now Immediately terminate the exit process and restore to the previous stable compensation state to prevent performance degradation caused by premature exit from compensation.

[0105] Establish a completion criterion for compensation exit, when the compensation gain decreases to And continuous When the control performance remains stable in a monitoring period, the fatigue compensation is exited, wherein represents a small threshold value close to the no-compensation state, represents the number of monitoring periods required for stability confirmation.

[0106] After exiting the fatigue compensation, the fatigue monitoring function is kept active, ready to reactivate the compensation strategy at any time when new signs of fatigue are detected. Data of the entire compensation period, including the fatigue development curve, the compensation parameter change trajectory and the control performance evolution process, are recorded to provide data support for optimizing the compensation strategy.

[0107] By implementing such a gradual and safe and controllable compensation exit strategy, it is ensured that the user can smoothly transition to the normal control state after fatigue recovery, avoiding the case of over-compensation, while ensuring the continuity and stability of the entire auxiliary control process, maximizing the effective operation time of the disabled user, and improving the adaptability and practical value of the embodied intelligent robotic arm auxiliary system.

[0108] Embodiment 2

[0109] With reference to Figure 5 , a second embodiment of the present application provides an embodied intelligent robotic arm disability assistance system.

[0110] The system comprises a physiological signal sensing module, a fatigue state prediction module, a control signal compensation module, a compensation effect evaluation module, and a compensation gradual exit module.

[0111] The physiological signal sensing module is configured to collect the electromyographic signals of the user when controlling the embodied robotic arm and extract electromyographic features, the electromyographic features including time domain and frequency domain features of the electromyographic signals, analyze the electromyographic signals through short-time Fourier transform and wavelet decomposition, establish a feature vector sequence containing fatigue indicators such as average power frequency, median frequency, and electromyographic amplitude decline rate, and record the duration of controlling the robotic arm.

[0112] The fatigue state prediction module is configured to construct a hierarchical fatigue state prediction model, extract local muscle fatigue patterns through a bottom convolutional network, and capture long-term cumulative effects through a top gated recurrent unit, the input of the hierarchical fatigue state prediction model being the electromyographic features, historical fatigue states, and task intensity, and the output being a current fatigue degree score and a future fatigue trend.

[0113] The control signal compensation module is configured to create an electromyographic signal fatigue compensation strategy, generate a compensation gain curve according to the current fatigue degree score and the future fatigue trend, and adaptively amplify the original electromyographic signals to compensate for the attenuation of the electromyographic signals caused by fatigue.

[0114] The compensation effect evaluation module is configured to calculate a signal-to-noise ratio and a stability index of the compensated electromyography signal, analyze a power spectral density change of the electromyography signal, maintain a current compensation strategy when detecting that the signal-to-noise ratio and the stability index of the compensated electromyography signal are improved, and trigger a compensation parameter fine adjustment otherwise.

[0115] The compensation gradual exit module is configured to gradually reduce the compensation strength according to a negative exponential law until the compensation function is exited when the signal-to-noise ratio and the stability index of the electromyography signal are continuously detected to be improved and a control performance index of the body-attached robot arm satisfies a preset recovery threshold.

[0116] In the embodiments provided by the present application, it should be understood that the disclosed device and method can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There can be another division during actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units can be indirect couplings or communication connections through some communication interfaces, devices or units, and can be electrical, mechanical or other forms.

[0117] The embodiments of the present application are described above with reference to the drawings, but the present application is not limited to the above specific embodiments. The above specific embodiments are only illustrative, but not restrictive. A person of ordinary skill in the art can make changes, modifications, replacements and variations to the above embodiments without departing from the purpose and protection scope of the present application under the inspiration of the present application. These are all within the protection scope of the present application.

Claims

1. A method for assisting people with disabilities using a unibody intelligent robotic arm, characterized in that, include: Electromyography (EMG) signals are collected when the user controls the robotic arm and EMG features are extracted. The EMG features include the time domain and frequency domain features of the EMG signals. The EMG signals are analyzed by short-time Fourier transform and wavelet decomposition to establish a feature vector sequence of fatigue indicators including average power frequency, median frequency, and rate of decrease of EMG amplitude. At the same time, the duration of controlling the robotic arm is recorded. A hierarchical fatigue state prediction model is constructed. The bottom layer convolutional network extracts local muscle fatigue patterns, and the top layer gated recurrent unit captures long-term cumulative effects. The input of the hierarchical fatigue state prediction model is electromyographic features, historical fatigue state and task intensity, and the output is the current fatigue level score and future fatigue trend. A fatigue compensation strategy for electromyography (EMG) signals is created. Based on the current fatigue level score and future fatigue trend, a compensation gain curve is generated to adaptively amplify the original EMG signal and compensate for the attenuation of EMG signals caused by fatigue. Calculate the signal-to-noise ratio and stability index of the compensated electromyographic signal, analyze the change in the power spectral density of the electromyographic signal, and maintain the current compensation strategy when an improvement in the signal-to-noise ratio and stability index of the compensated electromyographic signal is detected; otherwise, trigger fine-tuning of the compensation parameters. When the signal-to-noise ratio and stability of the electromyographic signal are continuously improved, and the control performance of the robotic arm meets the preset recovery threshold, the compensation intensity is gradually reduced according to the negative exponential law until the compensation function is exited.

2. The method for assisting the disabled with a body-worn intelligent robotic arm according to claim 1, characterized in that, When a user operates the robotic arm, raw electromyography signals are collected through an array of surface electromyography electrodes and filtered out power frequency interference and motion trajectory through a bandpass filter. The filtered electromyographic signal is segmented, and the root mean square value, mean absolute value and waveform length are calculated and extracted as time domain features within each time window. The power spectral density of each time period is calculated using short-time Fourier transform, and the average power frequency and median frequency are obtained as frequency domain features based on this. The rate of decrease in electromyographic amplitude is calculated by continuously monitoring the trend of the root mean square value of electromyography (EMG), and the time-domain features, frequency-domain features, and fatigue indicators are organized into a feature vector sequence in chronological order.

3. The method for assisting the disabled with a body-worn intelligent robotic arm according to claim 2, characterized in that, The hierarchical fatigue state prediction model includes a bottom-level architecture and a top-level architecture; The underlying architecture of the hierarchical fatigue state prediction model is a convolutional neural network. The convolutional neural network extracts local patterns from the feature vector sequence by stacking multiple convolutional structures and using multiple convolutional kernels of different sizes to obtain fatigue temporal pattern features. Introducing batch normalization layers into convolutional neural networks standardizes the output of each convolutional structure by performing linear transformations using scaling and translation parameters.

4. The method for assisting the disabled with a body-worn intelligent robotic arm according to claim 3, characterized in that, The top-level architecture of the hierarchical fatigue state prediction model is a gated recurrent unit network, which is used to model the time dependency of fatigue state and selectively remember and forget historical information through update gate and reset gate mechanism. The input of the hierarchical fatigue state prediction model is in the form of a triplet, including an electromyographic feature vector, a historical fatigue state vector, and a task intensity scalar. The historical fatigue state vector is obtained by weighted averaging of fatigue level scores within a sliding time window, and the task intensity is obtained by comprehensive evaluation and normalization based on the robotic arm load, movement speed, and operational complexity. The output layer of the hierarchical fatigue state prediction model is designed with a dual-branch structure. The first branch outputs the current fatigue level score, while the second branch predicts the fatigue level score sequence for multiple future time steps.

5. A method for assisting the disabled with a body-worn intelligent robotic arm according to claim 4, characterized in that, The fatigue compensation strategy for creating electromyographic signals includes: designing a dynamic compensation gain function, the base value of which is obtained from the current fatigue level score through a nonlinear mapping relationship; The pre-compensation factor is calculated based on the rate of change of the future fatigue trend. The comprehensive compensation gain is then adjusted by multiplying the base compensation value by the pre-compensation factor to obtain the comprehensive compensation gain. An exponentially weighted moving average method is used to smooth the overall compensation gain in the time domain, generating a smooth compensation gain curve.

6. A method for assisting the disabled with a body-worn intelligent robotic arm according to claim 5, characterized in that, The electromyographic signal fatigue compensation strategy further includes: decomposing the original electromyographic signal into multiple frequency band components through a bandpass filter, and setting differentiated compensation gain for each frequency band component according to the degree of influence of fatigue on different frequency components, so as to selectively compensate for the frequency. The compensated electromyographic signal is obtained by weighted summation of the components in each frequency band, and upper limit constraints and rate of change limits are set for the compensation gain. Establish a real-time evaluation mechanism for the compensation effect, quantify the compensation effect by calculating the energy ratio of the signal before and after compensation, and adaptively adjust the compensation parameters based on the evaluation results.

7. A method for assisting the disabled with a body-worn intelligent robotic arm according to claim 6, characterized in that, The signal-to-noise ratio (SNR) and stability indices of the compensated electromyographic (EMG) signal were calculated, including: within a sliding time window, the EMG signal was decomposed into effective signal components and noise components, and the SNR was calculated by estimating the signal energy during active contraction and the signal variance during rest; stability indices were obtained by analyzing the temporal fluctuation characteristics of the EMG signal, including calculating the coefficient of variation of the signal envelope, the zero-crossing rate of the signal, and the spectral stability index, with the signal envelope extracted using Hilbert transform; the power spectral density within each window was estimated using the Welch method, and the spectral stability index was calculated by comparing the power spectral density distributions of adjacent windows; The signal-to-noise ratio and stability indicators are weighted and fused to construct a unified quality score, which quantifies the compensation effect.

8. A method for assisting the disabled with a body-worn intelligent robotic arm according to claim 7, characterized in that, When the compensation parameter fine-tuning is triggered, the parameter adjustment amount is calculated using the gradient descent method. When the quality score does not improve after multiple consecutive adjustments, a parameter reset strategy is initiated to restore the compensation parameters to their historical optimal configuration. At the same time, historical information of each parameter adjustment is recorded to build a parameter optimization trajectory database, providing data support for subsequent optimizations.

9. A method for assisting the disabled with a unibody intelligent robotic arm according to claim 8, characterized in that, A multi-dimensional exit condition judgment mechanism is established. By setting a continuous detection counter, when the signal-to-noise ratio and stability index sequence continuously meet the preset threshold for a specified number of times, it is determined that the signal quality has reached a stable improvement. The control performance indicators of the robotic arm are evaluated, including trajectory tracking accuracy, response delay and operation smoothness. The indicators are weighted and integrated into a comprehensive control performance evaluation function. When the value of the comprehensive control performance evaluation function meets the preset recovery threshold, the control performance is deemed to be up to standard. When the conditions for signal quality improvement and control performance recovery are met simultaneously, a gradual exit process for compensation intensity is initiated, and the compensation gain decreases according to a negative exponential law. During the exit process, safety monitoring is implemented by monitoring performance fluctuation indicators and fatigue level score rebound. If any abnormality occurs, the exit process is paused or terminated. When the compensation gain decreases to the threshold of the uncompensated state and the control performance remains stable, the exit is completed.

10. A unibody intelligent robotic arm disability assistance system, used to implement the unibody intelligent robotic arm disability assistance method according to any one of claims 1 to 9, characterized in that, include: The module includes a physiological signal sensing module, a fatigue state prediction module, a control signal compensation module, a compensation effect evaluation module, and a compensation gradual exit module. The physiological signal sensing module is used to collect electromyographic signals and extract electromyographic features when the user controls the robotic arm. The electromyographic features include the time domain and frequency domain features of the electromyographic signals. The electromyographic signals are analyzed by short-time Fourier transform and wavelet decomposition to establish a feature vector sequence of fatigue indicators including average power frequency, median frequency, and electromyographic amplitude decline rate. At the same time, the duration of controlling the robotic arm is recorded. The fatigue state prediction module is used to construct a hierarchical fatigue state prediction model. It extracts local muscle fatigue patterns through a bottom-level convolutional network and captures long-term cumulative effects through a top-level gated recurrent unit. The inputs of the hierarchical fatigue state prediction model are electromyographic features, historical fatigue state, and task intensity, and the outputs are the current fatigue level score and future fatigue trend. The control signal compensation module is used to create an electromyographic signal fatigue compensation strategy, generate a compensation gain curve based on the current fatigue level score and future fatigue trend, adaptively amplify the original electromyographic signal, and compensate for the attenuation of the electromyographic signal caused by fatigue. The compensation effect evaluation module is used to calculate the signal-to-noise ratio and stability index of the electromyographic signal after compensation, analyze the change in the power spectral density of the electromyographic signal, and maintain the current compensation strategy when the signal-to-noise ratio and stability index of the electromyographic signal after compensation are detected; otherwise, it triggers fine-tuning of the compensation parameters. The progressive compensation exit module gradually reduces the compensation intensity according to a negative exponential law when the signal-to-noise ratio and stability of the electromyographic signal are continuously detected to improve, and the control performance of the robotic arm meets the preset recovery threshold, until the compensation function is exited.

Images