Exoskeleton cooperative control method and system based on brain-computer interface

Abstract

The present application relates to the technical field of exoskeleton cooperative control, and particularly relates to an exoskeleton cooperative control method and system based on a brain-computer interface, which combines electroencephalogram data and environmental crosstalk characteristic data to analyze crosstalk coupling characteristics of the brain-computer interface signal; according to the crosstalk coupling characteristic analysis result, the difficulty of man-machine matching between a user and an exoskeleton is analyzed in combination with the current execution action difficulty of the user; in combination with exoskeleton motion state data and target motion instructions, an exoskeleton execution action deviation risk caused by the difficulty of man-machine matching is predicted and analyzed; the cooperative safety of the exoskeleton man-machine system is comprehensively evaluated by fusing the action deviation risk prediction analysis result and the exoskeleton structure state data; and based on the cooperative safety evaluation result of the exoskeleton man-machine system, a differentiated exoskeleton cooperative control scheme is executed, so as to ensure the running safety of the exoskeleton man-machine system and the personal safety of the user in a complex crosstalk environment.

Description

Technical Field

[0001] This invention relates to the field of exoskeleton collaborative control technology, and in particular to a brain-computer interface-based exoskeleton collaborative control method and system. Background Technology

[0002] Exoskeleton robots, as intelligent electromechanical systems worn directly on the human body, decode the user's movement intentions through brain-computer interfaces and drive coordinated joint movements, playing a vital role in medical rehabilitation and industrial assistance. However, in practical applications, brain-computer interface signals are highly susceptible to the coupling effects of multiple sources of crosstalk, such as power frequency electromagnetic interference, electrooculography and electromyography artifacts, and motion artifacts introduced by electrode displacement. These factors severely degrade signal quality and consequently interfere with the decoding accuracy of movement intentions. When the exoskeleton performs movements of varying difficulty in a crosstalk environment, the coupling effect of crosstalk and movement difficulty further exacerbates the difficulty of human-machine matching: under high-difficulty movements, the damage to the signal caused by crosstalk leads to a significant deviation between the decoded instructions and the true intention, resulting in angle overshoot, response lag, or even erroneous movements. This causes strong cognitive conflict for the user, and the human-machine matching state deteriorates sharply.

[0003] Existing methods generally treat signal denoising, intent recognition, and motion control as independent modules, lacking real-time quantitative assessment of the dynamic changes in crosstalk characteristics and failing to output comprehensive indicators reflecting the degree of coupling between signals and crosstalk. Motion command correction also relies solely on physical feedback such as angle deviations, failing to incorporate human-machine matching status as a pre-assessment dimension into the control chain. More importantly, existing technologies lack a quantitative correlation model between crosstalk coupling degree, motion difficulty, and human-machine matching difficulty. This makes it difficult to objectively assess coordination quality based on neurophysiological signals and to predict risks and adjust strategies in advance before actual deviations occur, leading to problems such as perception gaps, prediction lags, and a single assessment dimension in complex working conditions. Summary of the Invention

[0004] To overcome the defects and shortcomings of existing technologies, this invention provides a brain-computer interface-based exoskeleton collaborative control method and system. It simultaneously collects EEG, motion, and environmental crosstalk data, calculates the crosstalk coupling degree and the difficulty coefficient of the executed action, and combines error-related potential characteristics with a human-machine matching difficulty analysis model to obtain the degree of human-machine matching difficulty. It then constructs a propagation feature vector, inputs it into a long short-term memory network to predict the risk of action deviation, and finally integrates the action deviation safety margin with the structural moment and stress safety margin to calculate the collaborative safety situation using a weighted average, and executes differentiated collaborative control schemes.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, embodiments of the present invention provide an exoskeleton collaborative control method based on a brain-computer interface, comprising the following steps: S1. Acquire exoskeleton structural state data, and simultaneously collect user's EEG signal data, exoskeleton motion state data, and environmental crosstalk characteristic data through multimodal brain-computer interface sensors, motion state sensors, and environmental perception sensors deployed on the exoskeleton robot. S2. Combining EEG signal data and environmental crosstalk characteristic data, analyze the crosstalk coupling characteristics of brain-computer interface signals; and based on the crosstalk coupling characteristic analysis results, combined with the difficulty of the user's current performance, analyze the degree of human-machine matching difficulty between the user and the exoskeleton. S3. Based on the analysis results of the difficulty of human-machine matching, and combined with the exoskeleton motion state data and target motion commands, predict and analyze the risk of exoskeleton execution deviation caused by human-machine matching difficulties. S4. Integrate the results of motion deviation risk prediction and analysis with exoskeleton structural status data to comprehensively assess the collaborative safety of the exoskeleton human-machine system. S5. Based on the assessment results of the collaborative safety of the exoskeleton human-machine system, implement differentiated exoskeleton collaborative control schemes.

[0006] In one implementation of the present invention, step S2 involves analyzing the crosstalk coupling characteristics of the brain-computer interface signals by combining EEG signal data and environmental crosstalk feature data; specifically, this includes the following: S21. Perform bandpass filtering on the EEG signal data to separate out the motion-related signals. Rhythm band and Rhythm band and resting baseline band; S22. Calculate the energy of the signal in each frequency band window by window using short-time Fourier transform; perform windowing and fast Fourier transform on the EEG signal in the corresponding time window to obtain the power spectrum in the corresponding time window. S23. On the power spectrum within the corresponding time window, integrate and sum the power values ​​within the frequency band range to obtain the task-state frequency band power of the specified frequency band within the current time window; S24. Divide the difference between the task-state frequency band power and the resting-state frequency band average power within the corresponding time window by the resting-state frequency band average power to obtain the event-related desynchronization degree within the corresponding time window; wherein, the resting-state frequency band average power is the average value of the specified frequency band power within the last preset resting-state duration; S25. Obtain the average value of the signal power of the task frequency band and the signal power of the high-frequency noise frequency band within the corresponding time window; S26. Divide the signal power of the task frequency band within the corresponding time window by the mean of the signal power of the high-frequency noise frequency band to obtain the brain-to-electrical signal-to-noise ratio within the corresponding time window. S27. From the environmental crosstalk feature data, extract the preset narrowband frequency range containing the power frequency within the corresponding time window as the power frequency band power and the number of electrooculogram artifact events detected within the corresponding time window; simultaneously, calculate the interaction between each electromyography channel and the electroencephalogram signal. Coherence coefficient within the frequency band; S28. Divide the power frequency band power near the power frequency in the corresponding time window by the signal power of the task frequency band to obtain the power frequency interference intensity in the corresponding time window. S29. Divide the number of electrooculogram artifact events detected in the corresponding time window by the product of the sampling frequency and the duration of the time window to obtain the electrooculogram artifact density in the corresponding time window. S210, Combine all electromyographic channels and electroencephalogram signals within the corresponding time window. The absolute values ​​of the coherence coefficients within the frequency band are arithmetic averaged to obtain the electromyographic crosstalk index within the corresponding time window; S211. Obtain the standard deviation of EEG signal amplitude and the mean of the standard deviation of resting EEG signal amplitude within the corresponding time window; subtract the mean of the standard deviation of EEG signal amplitude within the corresponding time window and the mean of the standard deviation of resting EEG signal amplitude; divide the result of the subtraction by the mean of the standard deviation of resting EEG signal amplitude to obtain the intensity of motion artifacts within the corresponding time window. S212. Combine the event-related desynchronization degree, brain-to-electrical-to-noise ratio, power frequency interference intensity, electrooculogram artifact density, electromyography crosstalk index, and motion artifact intensity within the corresponding time window into a comprehensive crosstalk coupling feature vector, input it into the pre-trained crosstalk coupling degree calculation model, and obtain the crosstalk coupling degree within the corresponding time window.

[0007] In one implementation of the present invention, step S2 analyzes the difficulty of human-machine matching between the user and the exoskeleton based on the crosstalk coupling characteristic analysis results and the difficulty of the user's current action; specifically, it includes the following steps: S213. Calculate the execution difficulty coefficient of the main body and the execution difficulty coefficient of the exoskeleton; perform a weighted sum of the execution difficulty coefficient of the main body and the execution difficulty coefficient of the exoskeleton to obtain the execution difficulty coefficient of the corresponding execution action within the corresponding time window; S214. Extract error-related potential features from the frontal central region electrodes of the EEG signal. The error-related potential is the negative deflection component of the brainwave generated in a preset time window after the action is performed when the exoskeleton's action does not match the user's movement intention. The extracted error-related potential features include the peak amplitude ratio of error-related potentials, the average power ratio within the error-related potential window, and the area under the waveform curve of the error-related potential. S215. Obtain the crosstalk coupling degree within the current time window, the execution difficulty coefficient of the corresponding action within the current time window, the electromyographic crosstalk index and the brain electrical signal-to-noise ratio, as well as the error-related potential characteristics within the current time window, combine them into a human-machine matching feature vector, input it into a pre-trained human-machine matching difficulty analysis model, and obtain the human-machine matching difficulty between the user and the exoskeleton within the current time window; the human-machine matching difficulty analysis model adopts the Gaussian process regression algorithm.

[0008] In one implementation of the present invention, the construction process of the human-machine matching difficulty analysis model includes: Acquire historical exoskeleton operation experiments under different crosstalk environments and different movement difficulty conditions; In each group of exoskeleton operation experiments, the brain-to-noise ratio, electromyographic crosstalk index, crosstalk coupling degree, difficulty coefficient of the action to be performed, and error-related potential characteristics were calculated. The brain-to-noise ratio, electromyographic crosstalk index, crosstalk coupling degree, difficulty coefficient of the action to be performed, and error-related potential characteristics of each exoskeleton operation experiment were combined into a historical human-machine matching feature vector. Obtain training labels for the difficulty of human-machine matching in each group of exoskeleton operation experiments; combine the historical human-machine matching feature vectors of all exoskeleton operation experiments and the corresponding training labels for the difficulty of human-machine matching into a human-machine matching dataset, and divide the human-machine matching dataset into a human-machine matching training set and a human-machine matching validation set. A Gaussian process regression model is constructed, with the covariance function chosen as a combination of radial basis function kernel and white noise kernel. Based on the human-machine matching training set, the human-machine matching feature vector is used as the input feature of the Gaussian process regression model, and the training label of human-machine matching difficulty is used as the output target of the Gaussian process regression model. The Gaussian process regression model is trained, and the kernel function hyperparameter is determined by maximizing the log marginal likelihood, resulting in an initial human-machine matching difficulty analysis model. The initial human-machine matching difficulty analysis model is validated using a human-machine matching validation set. The root mean square error on the validation set is calculated, and the initial human-machine matching difficulty analysis model whose root mean square error on the validation set is less than or equal to a preset error threshold is output as the final human-machine matching difficulty analysis model.

[0009] In one implementation of the present invention, step S3, based on the analysis results of the difficulty of human-machine matching, combines exoskeleton motion state data and target motion commands to predict and analyze the risk of exoskeleton execution deviation caused by human-machine matching difficulties; including the following specific steps: S31. Compare the exoskeleton motion state data within the corresponding time window with the target motion command joint by joint, and calculate the degree of angular deviation, angular velocity deviation and motion synchronization delay of each joint within the corresponding time window. S32. Take the root mean square values ​​of the degree of angular deviation, the degree of angular velocity deviation, and the motion synchronization delay of all joints within the corresponding time window to obtain the risk of angular deviation, angular velocity deviation, and motion synchronization delay caused by human-machine matching difficulties within the corresponding time window. S33. The degree of difficulty in human-machine matching between the user and the exoskeleton in the current time window, the peak amplitude ratio of error-related potentials in the error-related potential characteristics, the electromyographic crosstalk index, and the risk of angle deviation, angular velocity deviation, and motion synchronization delay caused by human-machine matching difficulties in the current time window, as well as the first-order difference of the risk of angle deviation caused by human-machine matching difficulties in the current time window, are time-aligned to construct a propagation feature vector. S34. Input the propagation feature vector into the pre-trained motion deviation risk prediction model to obtain the risk of exoskeleton motion deviation caused by human-machine matching difficulties within the current time window.

[0010] In one implementation of the present invention, step S4 integrates the motion deviation risk prediction and analysis results with the exoskeleton structural status data to comprehensively assess the collaborative safety of the exoskeleton human-machine system; this includes the following specific steps: S41. Extract the absolute value of the actual joint torque of each joint, the actual stress of each link, the joint design torque safety threshold, and the yield strength of the link material from the exoskeleton structural state data within the current time window. S42. Divide the absolute value of the actual joint torque of each joint by the safety threshold of the joint design torque, and subtract 1 from the result to obtain the safety margin of the joint torque of each joint; divide the actual stress of each link by the yield strength of the link material, and subtract 1 from the result to obtain the safety redundancy of the link stress of each link. S43. Extract the minimum value of the joint torque safety margin of all joints as the torque safety margin of the weakest joint; extract the minimum value of the link stress safety redundancy of all links as the stress safety redundancy of the weakest link. S44. Extract the risk of exoskeleton execution deviation caused by human-machine matching difficulties within the current time window; subtract the risk of exoskeleton execution deviation caused by human-machine matching difficulties within the current time window from 1 to obtain the safety margin of exoskeleton execution deviation within and outside the current time window. S45. The safety margin of the skeleton's execution action deviation, the safety margin of the weakest joint torque, and the safety redundancy of the weakest link stress are weighted and summed to obtain the collaborative safety status of the skeleton human-machine system inside and outside the current time window.

[0011] In one implementation of the present invention, step S5 involves executing a differentiated exoskeleton collaborative control scheme based on the assessment results of the collaborative safety of the exoskeleton human-machine system; this includes the following specific steps: S51. Extract the collaborative safety status of the human-machine skeletal system inside and outside the current time window; S52. Preset a collaborative safety threshold. When the collaborative safety status of the exoskeleton human-machine system inside and outside the current time window is greater than the collaborative safety threshold, the exoskeleton human-machine system is judged to be in a safe state and the conventional collaborative control scheme is executed. When the collaborative safety status of the exoskeleton human-machine system inside and outside the current time window is less than or equal to the collaborative safety threshold, the exoskeleton human-machine system is judged to be in a dangerous state and the protective collaborative control scheme is executed.

[0012] Secondly, embodiments of the present invention also provide an exoskeleton collaborative control system based on a brain-computer interface, including: The data acquisition module is used to acquire exoskeleton structural state data and simultaneously acquire the user's EEG signal data, exoskeleton motion state data and environmental crosstalk characteristic data through multimodal brain-computer interface sensors, motion state sensors and environmental perception sensors deployed on the exoskeleton robot. The human-machine matching analysis module is used to analyze the crosstalk coupling characteristics of brain-computer interface signals by combining EEG signal data and environmental crosstalk feature data; and based on the crosstalk coupling feature analysis results, combined with the difficulty of the user's current performance action, it analyzes the degree of difficulty of human-machine matching between the user and the exoskeleton. The execution status prediction module is used to predict and analyze the risk of exoskeleton execution deviation caused by human-machine matching difficulties, based on the analysis results of the difficulty of human-machine matching, combined with exoskeleton motion status data and target motion commands. The collaborative safety monitoring module is used to integrate the results of motion deviation risk prediction and analysis with exoskeleton structural status data to comprehensively assess the collaborative safety of the exoskeleton human-machine system. The collaborative control module is used to execute differentiated exoskeleton collaborative control schemes based on the assessment results of the collaborative safety of the exoskeleton human-machine system.

[0013] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. This invention establishes an objective evaluation model for the difficulty of human-machine matching between users and exoskeletons under crosstalk environment by jointly quantifying the crosstalk coupling degree of brain-computer interface signals and the difficulty of performing actions, which can perceive the quality of human-machine collaboration in real time from the neurophysiological level. 2. This invention constructs a propagation feature vector by using the difficulty of human-machine matching as the core input, and uses a long short-term memory network to predict the risk of action deviation. This enables the early warning of risk to be issued before the actual deviation exceeds the limit, solving the problem of the lag in traditional methods that only passively correct errors after the deviation occurs. 3. This invention integrates motion deviation safety margin with structural torque and stress safety margin to construct a dual collaborative safety assessment system covering both the neural control level and the physical and mechanical level, ensuring the operational safety of the exoskeleton human-machine system and the personal safety of the user in complex crosstalk environments. Attached Figure Description

[0014] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the overall process of the brain-computer interface-based exoskeleton collaborative control method of the present invention; Figure 2 This is a flowchart analyzing the crosstalk coupling characteristics in step S2 of the brain-computer interface-based exoskeleton collaborative control method of the present invention. Figure 3 This is a flowchart analyzing the difficulty of human-machine matching in step S2 of the brain-computer interface-based exoskeleton collaborative control method of the present invention. Figure 4 This is a flowchart of step S3 in the brain-computer interface-based exoskeleton collaborative control method of the present invention; Figure 5 This is a schematic diagram of the exoskeleton collaborative control system based on a brain-computer interface of the present invention. Detailed Implementation

[0015] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0016] Example 1 Brain-computer interface-based exoskeleton collaborative control technology mainly achieves human-machine collaborative movement by decoding the user's motor intentions in their brain signals and driving the exoskeleton to perform corresponding actions. Existing technologies mainly revolve around three core aspects: brain signal preprocessing and feature extraction, motor intention decoding and recognition, and feedback-based control command adjustment.

[0017] In terms of signal preprocessing, existing methods mostly use adaptive filtering and wavelet transform with fixed parameters to remove power frequency interference and electrooculogram artifacts. The technical idea is to suppress noise through preset filtering parameters. However, due to the lack of a real-time response mechanism to the dynamic changes in noise characteristics, the signal-to-noise ratio fluctuates greatly in complex crosstalk environments, which significantly affects the accuracy of subsequent feature extraction.

[0018] In terms of motion intention recognition, existing technologies are gradually evolving from single classifiers to ensemble learning models. Convolutional neural networks are used to extract local spatiotemporal features of EEG signals, support vector machines are used to construct classification boundaries in a high-dimensional feature space, and random forests are combined to capture global statistical patterns of signals. Finally, the intention recognition results are output using a weighted voting method, which improves the accuracy of intention recognition compared to single models.

[0019] In terms of dynamic adjustment of control commands, existing technologies have begun to introduce reinforcement learning algorithms for online optimization of motion state parameters. By defining a state space that includes joint angle deviation, angular velocity deviation and joint torque, and combining it with a reward function designed based on deviation and safety, the adaptive correction of control commands is achieved by iteratively updating the state value function, so that motion deviation can gradually converge to a preset range.

[0020] However, the aforementioned existing technologies generally share a common problem: they all fail to fully consider the impact of the coupling effect between crosstalk and movement difficulty on the human-machine matching state when brain-computer interface signals perform movements of varying difficulty in a crosstalk environment. Specifically, existing technologies treat signal denoising, intent recognition, and motion control as relatively independent functional modules, lacking a unified evaluation framework to quantify the matching difficulty between the user and the exoskeleton in a crosstalk environment. They also fail to incorporate this matching difficulty as a precondition into subsequent movement deviation risk prediction and safety assessment systems. This results in existing systems struggling to detect the deterioration trend of human-machine collaboration under complex crosstalk conditions, often triggering passive error correction only after movement deviations have significantly exceeded limits, leading to delayed responses and impacting collaborative safety. Therefore, there is an urgent need for a collaborative control method that can quantify crosstalk coupling in real time, objectively assess the difficulty of human-machine matching, and use this as a core input throughout the entire process of movement deviation risk prediction and collaborative safety assessment.

[0021] like Figure 1 As shown, this embodiment provides a brain-computer interface-based exoskeleton collaborative control method, which specifically includes the following steps: S1. Acquire exoskeleton structural state data, and simultaneously collect user's EEG signal data, exoskeleton motion state data, and environmental crosstalk characteristic data through multimodal brain-computer interface sensors, motion state sensors, and environmental perception sensors deployed on the exoskeleton robot. In this embodiment, the exoskeleton structural status data includes torque sensor readings of each joint, strain gauge readings of each load-bearing link, and static structural parameters such as the joint design torque safety threshold and the yield strength of the link material. All of these can be collected by torque sensors and stress sensors pre-deployed on the output end of the exoskeleton joints and the surface of the links at a sampling frequency of not less than 100Hz, and transmitted in real time to the data processing unit through the internal bus of the exoskeleton control system.

[0022] The acquisition of EEG signal data used a dry electrode EEG acquisition device with electrodes laid out according to the international 10-20 system, focusing on covering the C3 and C4 regions of the motor cortex and the Fz, FCz, and Cz regions of the prefrontal cortex. The sampling frequency was set to 250Hz, and the signal bandwidth covered the motor-related μ rhythm in the 8 to 13Hz band and the β rhythm in the 13 to 30Hz band, ensuring that key neurophysiological features such as event-related desynchronization and error-related potentials during motor imagination and motor execution could be fully captured.

[0023] The exoskeleton's motion data is collected synchronously through encoders and inertial measurement units in each joint, including the actual angle, actual angular velocity, and actual angular acceleration of each joint. The sampling frequency is aligned with the EEG acquisition frequency, and the time synchronization accuracy is ensured to be within 1ms through hardware trigger signals.

[0024] The acquisition of environmental crosstalk characteristic data includes real-time monitoring of power frequency electromagnetic interference, estimation of motion artifact intensity caused by environmental vibration, and surface electromyography (EMG) signals from each channel of the EMG sensor. The EMG signals cover key muscle groups assisted by the exoskeleton (such as the quadriceps femoris and biceps brachii), and the sampling frequency is set to 1000Hz to ensure the integrity of high-frequency components in the EMG signals.

[0025] Based on the above, this embodiment synchronously collects neural signals, mechanical motion signals, environmental interference signals, and structural mechanics signals under a unified timestamp. This provides a multi-dimensional cross-validation basis for calculating crosstalk coupling degree and human-machine matching difficulty in subsequent steps, avoiding causal analysis bias caused by asynchronous data time. At the same time, the redundancy and complementarity of multimodal signals also improves the system's fault tolerance when a single sensor fails.

[0026] S2. Combining EEG signal data and environmental crosstalk characteristic data, analyze the crosstalk coupling characteristics of brain-computer interface signals; and based on the crosstalk coupling characteristic analysis results, combined with the difficulty of the user's current performance, analyze the degree of human-machine matching difficulty between the user and the exoskeleton. S3. Based on the analysis results of the difficulty of human-machine matching, and combined with the exoskeleton motion state data and target motion commands, predict and analyze the risk of exoskeleton execution deviation caused by human-machine matching difficulties. S4. Integrate the results of motion deviation risk prediction and analysis with exoskeleton structural status data to comprehensively assess the collaborative safety of the exoskeleton human-machine system. S5. Based on the assessment results of the collaborative safety of the exoskeleton human-machine system, implement differentiated exoskeleton collaborative control schemes.

[0027] In a specific embodiment, such as Figure 2As shown, step S2 combines EEG signal data and environmental crosstalk characteristic data to analyze the crosstalk coupling characteristics of brain-computer interface signals; specifically, it includes the following: S21. Perform bandpass filtering on the EEG signal data to separate out the motion-related signals. Rhythm band and Rhythm band and resting baseline band; S22. Calculate the energy of the signal in each frequency band window by window using short-time Fourier transform; perform windowing and fast Fourier transform on the EEG signal in the corresponding time window to obtain the power spectrum in the corresponding time window; in the windowing calculation process, the window width is set to 1 second, i.e., 250 sampling points, and the overlap rate of adjacent windows is set to 50%. The EEG signal in each time window is first multiplied by the Hamming window function to reduce spectral leakage, and then the 250 data points after windowing are subjected to fast Fourier transform to obtain the complex spectrum of the signal in the window at each frequency point. Take the square of the modulus of the complex spectrum to obtain the power spectrum of the corresponding window. S23. On the power spectrum within the corresponding time window, integrate and sum the power values ​​within the frequency band range to obtain the task-state frequency band power of the specified frequency band within the current time window; S24. Divide the difference between the task-state frequency band power and the resting-state frequency band average power within the corresponding time window by the resting-state frequency band average power to obtain the event-related desynchronization degree reflecting the intensity of motion intention within the corresponding time window; wherein, the resting-state frequency band average power is the average value of the specified frequency band power within the last preset duration of the resting state; in this embodiment, the preset duration can be the last 30 seconds of the resting state; the resting-state frequency band average power can be obtained by taking the data from the last 30 seconds of the resting state, performing short-time Fourier transform window-by-window calculation with the same window width and overlap rate, and then taking the arithmetic mean of the frequency band power of all windows; S25. Obtain the average value of the task frequency band signal power and the high-frequency noise frequency band signal power within the corresponding time window, where the task frequency band is... Rhythm band and The union of the rhythm frequency bands and the signal power of the task frequency band are obtained by integrating and summing the power values ​​within the task frequency band range on the power spectrum of the current time window obtained by the short-time Fourier transform; the high-frequency noise frequency band is the frequency band with a frequency higher than the upper limit of the task frequency band on the same power spectrum; the average signal power of the high-frequency noise frequency band is obtained by summing the power values ​​of each frequency point within the high-frequency noise frequency band range and dividing by the number of frequency points. S26. Divide the signal power of the task frequency band within the corresponding time window by the average signal power of the high-frequency noise frequency band to obtain the brain-to-noise ratio within the corresponding time window, so that the signal quality within different time windows can be directly compared without being affected by absolute amplitude fluctuations. S27. From the environmental crosstalk feature data, extract a preset narrowband frequency range containing the power frequency within the corresponding time window as the power band power near the power frequency, and the number of electrooculography artifact events detected within the corresponding time window; for example, in this embodiment, the average power within the preset narrowband frequency range containing the power frequency of 50Hz, i.e., the 49 to 51Hz frequency band, is extracted as the power band power near the power frequency; and the amplitude threshold is used. The number of blinking or saccade artifacts was identified by combining slope detection, and the number of electrooculography artifacts detected within the corresponding time window was recorded; simultaneously, the correlation between each electromyographic channel and the electroencephalogram (EEG) signal was calculated. Coherence coefficient within the frequency band.

[0028] In one specific embodiment, a single electromyographic channel is correlated with electroencephalographic signals in... The calculation process for the coherence coefficient within the frequency band is as follows: Extract the EEG signal spectrum of the corresponding electrode (e.g., C3 or C4 electrode) in the motor cortex within the corresponding time window and the EMG signal spectrum of a single EMG channel within the corresponding time window; The EEG signal spectrum and EMG signal spectrum were respectively processed by removing the mean, and the mean-removed EEG signal spectrum and EMG signal spectrum were obtained respectively. The corresponding time window is divided into several sub-segments according to a fixed overlap rate (the overlap rate is set to 50% in this embodiment). Each sub-segment has a length of 128 points. The EEG signal spectrum and EMG signal spectrum after mean removal processing of several sub-segments are extracted. The EEG signal spectrum and EMG signal spectrum after mean removal processing of several sub-segments are windowed using a Hamming window to reduce spectral leakage, resulting in the windowed EEG signal spectrum and EMG signal spectrum of several sub-segments. The EEG and EMG signal spectra after windowing of each sub-segment are subjected to Fast Fourier Transform to obtain the complex spectra of the EEG and EMG signal spectra after windowing of the corresponding sub-segment. The complex value at frequency f in the complex spectrum of the windowed EEG signal spectrum of each segment is squared modulowise. Then, the complex value at frequency f in the complex spectrum of all windowed EEG signal spectra after squared modulowise processing is... The complex values ​​at each point are averaged to obtain the EEG signal at the corresponding time window in terms of frequency. The self-power spectral density at the location; The frequency of the complex spectrum of the electromyographic signal after windowing of each segment is obtained. The complex conjugate of the complex value at the location; the frequency in the complex spectrum of the electromyographic signal after windowing each segment. The complex conjugate of the complex value at a given frequency is multiplied by the complex value at frequency f in the complex spectrum of the EEG signal spectrum after windowing of the corresponding sub-segment to obtain the EEG and EMG signals of each sub-segment at frequency f. The cross-power spectral density at frequency f is calculated by averaging the cross-power spectral densities of EEG and EMG at frequency f for all sub-segments, thus obtaining the cross-power spectral density of EEG and EMG at frequency f for the corresponding time window. EEG and EMG at corresponding time windows in terms of frequency The square of the modulus of the cross-power spectral density at a given point divided by the corresponding time window of the EEG signal at a given frequency. The self-power spectral density at a given frequency is similar to that of electromyography signals. The product of the self-power spectral density at a given point yields the corresponding time window. Inner frequency of the frequency band The coherence coefficient at the corresponding time window; The arithmetic mean of the coherence coefficients at all frequency points within the frequency band is used to obtain the correlation between a single electromyographic channel and the electroencephalogram (EEG) signal within the corresponding time window. Coherence coefficient within the frequency band; S28. Divide the power frequency band power near the power frequency in the corresponding time window by the signal power of the task frequency band to obtain the power frequency interference intensity in the corresponding time window, so that the degree of power frequency pollution under different signal intensities can be uniformly measured. S29. Divide the number of EEG artifact events detected within the corresponding time window by the product of the sampling frequency and the duration of the time window to obtain the EEG artifact density within the corresponding time window. This converts the absolute count value into the probability of artifact occurrence per unit sampling point, eliminating the influence of sampling duration differences. S210, Combine all electromyographic channels and electroencephalogram signals within the corresponding time window. The absolute values ​​of the coherence coefficients within the frequency band are arithmetic averaged to obtain the electromyographic crosstalk index within the corresponding time window; S211. Obtain the standard deviation of EEG signal amplitude and the mean of the standard deviation of resting EEG signal amplitude within the corresponding time window; subtract the mean of the standard deviation of EEG signal amplitude within the corresponding time window and the mean of the standard deviation of resting EEG signal amplitude; divide the result of the subtraction by the mean of the standard deviation of resting EEG signal amplitude to obtain the intensity of motion artifacts within the corresponding time window, which can reflect the relative intensity of artifacts introduced by electrode displacement caused by user or exoskeleton movement; S212. Combine the event-related desynchronization degree, brain-to-electrical-to-noise ratio, power frequency interference intensity, electrooculogram artifact density, electromyography crosstalk index and motion artifact intensity within the corresponding time window into a comprehensive crosstalk coupling feature vector, input it into the pre-trained crosstalk coupling degree calculation model, and obtain the crosstalk coupling degree within the corresponding time window. The construction process of the crosstalk coupling degree calculation model includes: Historical brain-computer interface (BCI) experimental data was acquired, including EEG signal recordings from multiple experimental trials. Each trial simultaneously recorded both pure EEG signals and mixed EEG signals after the injection of crosstalk signals of known intensity. For each trial, the event-related desynchronization degree, EEG-to-noise ratio, power line interference intensity, electrooculogram (EOG) artifact density, electromyography (EMG) crosstalk index, and motion artifact intensity were calculated. These seven indicators were combined into a comprehensive crosstalk coupling feature vector. The crosstalk coupling degree label for each trial was calculated. The comprehensive crosstalk coupling feature vectors and corresponding crosstalk coupling degree labels from all trials were combined to form a crosstalk coupling dataset, which was then divided into a crosstalk coupling training set and a crosstalk coupling validation set. The formula for calculating the crosstalk coupling degree label is as follows: ; In the formula, For crosstalk coupling degree label, The power of the mixed EEG signal after injecting a crosstalk signal of known strength. For pure EEG signal power; A kernel ridge regression model is constructed, with radial basis functions chosen as the kernel function. Based on the crosstalk coupling training set, the comprehensive crosstalk coupling feature vector is used as the input feature of the kernel ridge regression model, and the crosstalk coupling degree label is used as the output target. The kernel ridge regression model is trained, and the weight parameters of the model are determined by minimizing the sum of the mean square error loss and the regularization term, resulting in an initial crosstalk coupling degree calculation model. The initial crosstalk coupling degree calculation model is validated using a crosstalk coupling validation set. The root mean square error on the validation set is calculated, and the initial crosstalk coupling degree calculation model whose root mean square error on the validation set is less than or equal to a preset error threshold is output as the crosstalk coupling degree measure. The model is calculated as follows: In a specific embodiment, the combined crosstalk coupling feature vectors and labels of all experimental trials are used to form a dataset, which is then divided into a training set and a validation set in a 7:3 ratio. During training, the regularization strength parameter of the kernel ridge regression is selected from the candidate values ​​of 0.01, 0.1, 1, 10, and 100 through five-fold cross-validation to minimize the root mean square error of the validation set. Similarly, the kernel width parameter of the radial basis function is determined from the candidate values ​​of 0.001, 0.01, 0.1, 1, and 10 through grid search. The model training convergence criterion is terminated when the root mean square error of the validation set decreases by less than 1‰ over ten consecutive training rounds. Based on the above, this embodiment achieves efficient nonlinear synthesis of multidimensional crosstalk information by inputting seven crosstalk-related indicators into the kernel ridge regression model. This avoids the subjectivity of manually setting fixed thresholds. At the same time, the regularization characteristics of kernel ridge regression can effectively suppress the overfitting risk caused by multicollinearity between features, ensuring the stability and generalization ability of the core indicator of crosstalk coupling degree under different users and different environments.

[0029] In a specific embodiment, such as Figure 3 As shown, in step S2, based on the crosstalk coupling characteristic analysis results and combined with the difficulty of the user's current action, the degree of difficulty in human-machine matching between the user and the exoskeleton is analyzed; specifically, the following steps are included: S213. Calculate the execution difficulty coefficient of the main body and the execution difficulty coefficient of the exoskeleton; perform a weighted summation of the execution difficulty coefficient of the main body and the execution difficulty coefficient of the exoskeleton to obtain the execution difficulty coefficient of the corresponding execution action within the corresponding time window, so as to simultaneously perceive the execution difficulties faced by the user's main body and the exoskeleton electromechanical system under the current action, avoiding the evaluation blind spot caused by focusing on only one aspect. The difficulty coefficient of the user's execution of the action represents the degree of load on the user's musculoskeletal system. The formula for calculating the difficulty coefficient of the user's execution of the action is as follows: ; In the formula, , where is the execution difficulty coefficient of the main body, 'a' is the net bio-torque weight, 'K' is the number of joints involved, 'Ti' is the absolute value of the net bio-torque of the i-th joint, 'T' is the maximum permissible bio-torque of the i-th joint, 'Wi' is the absolute value of the current angular velocity of the i-th joint, and 'W' is the maximum safe angular velocity of the i-th joint; where i is any one from 1 to K; the net bio-torque is obtained by subtracting the exoskeleton auxiliary torque from the joint torque sensor readings, the maximum permissible bio-torque is defined as the peak value of the isometric contraction torque of the joint, and the maximum safe angular velocity is defined as the upper limit of the speed of the exoskeleton joint actuator, used to describe the limit of how fast the joint rotates, and measures the maximum speed that the human joint or exoskeleton can physiologically or mechanically withstand; The exoskeleton execution difficulty coefficient represents the control precision required of the exoskeleton's electromechanical system for the current action. The formula for calculating the exoskeleton execution difficulty coefficient is: ; In the formula, Wn is the exoskeleton execution difficulty coefficient, b is the angular acceleration torque weight, Ai is the absolute value of the target angular acceleration of the i-th joint, A is the maximum angular acceleration of the i-th joint, Fi is the standard deviation of the external interaction force of the i-th joint, and F is the rated range of the torque sensor of the i-th joint; where the target angular acceleration is obtained by the target motion command through first-order difference, the maximum angular acceleration is the maximum angular acceleration capability of the joint actuator, and the standard deviation of the external interaction force is the standard deviation of the external interaction force on the i-th joint within the corresponding time window; S214. Extract error-related potential features from the frontal central region electrodes (in this embodiment, the frontal central region electrodes include Fz, FCz, and Cz electrodes) of the EEG signal. Error-related potentials are the negative EEG deflection components generated by the brain within a preset time window (e.g., 200-500ms) after the exoskeleton performs an action that does not match the user's movement intention. The extracted error-related potential features include the peak amplitude ratio of error-related potentials, the average power ratio within the error-related potential window, and the area under the waveform curve of the error-related potential. In this method, the average voltage of the Fz electrode within a 100ms window before the action is executed is used as the baseline. The minimum value of the signal after baseline removal within a 200-500ms window is calculated as the peak amplitude of the error-related potential. The absolute value of the peak amplitude of the error-related potential is divided by the reference amplitude (in this embodiment, the reference amplitude is based on the statistical average value of a healthy subject population, and is assumed to be 0.5%). ), to obtain the peak amplitude ratio of error-related potentials; The baseline-removed signal power within the corresponding time window is divided by the average power within a resting state time window of the same length (in this embodiment, the resting state time window of the same length is 300ms) to obtain the average power ratio within the error-related potential window, wherein the average power of the resting state window is taken from the power average of the same length window during the user's relaxed and inactive period. The absolute value of the baseline-removed signal within the corresponding time window is integrated and divided by the reference area value (in this embodiment, the reference area value is based on the statistical average of a healthy subject population and is assumed to be 0). ), to obtain the area under the error-related potential waveform curve; S215. Obtain the crosstalk coupling degree within the current time window, the execution difficulty coefficient of the corresponding action within the current time window, the electromyographic crosstalk index and the brain electrical signal-to-noise ratio, as well as the error-related potential characteristics within the current time window. Combine these into a human-machine matching feature vector, input it into a pre-trained human-machine matching difficulty analysis model, and obtain the human-machine matching difficulty between the user and the exoskeleton within the current time window. The human-machine matching difficulty analysis model uses a Gaussian process regression algorithm.

[0030] When training the human-computer matching difficulty analysis model, the training label of the human-computer matching difficulty is used as the training objective; the specific formula for calculating the human-computer matching difficulty training label is as follows: ; In the formula, P is the training label for the difficulty of human-machine matching; These represent the angle deviation weight, the settling time weight, and the overshoot count weight, respectively. The difference between the actual angle of a joint of the exoskeleton at time t and the target angle issued by the control system. The angle tolerance represents the preset allowable deviation value of the angle. When the absolute value of the angle deviation does not exceed this value, the exoskeleton is considered to be performing actions within an acceptable range. ts is the duration of a single action test; ts is the settling time, which represents the time required from the moment the action command is issued until the exoskeleton's actual angle first enters and remains within the allowable deviation range. The settling time ends when it exceeds the allowable deviation range. The maximum allowable stabilization time is used to represent the preset upper limit of the stabilization time. When the actual stabilization time exceeds this value, the system response is considered too slow and there are significant difficulties in human-machine matching. Overshoot count, used to indicate the number of times the actual angle exceeds the allowable deviation range of the target angle during a single action test. Specifically, when... The absolute value is greater than If this situation occurs again after the previous entry into the allowable deviation range, it is counted as one overshoot; furthermore, the method for determining the number of overshoots is: monitoring Fluctuations in absolute value. Each time from The absolute value is less than or equal to The state becomes The absolute value is greater than If the initial state is already out of bounds, then the process from the start of the action until the first entry into the allowable range is not counted as an overshoot, but is counted as part of the settling time. In a specific embodiment, the construction process of the human-machine matching difficulty analysis model includes: Acquire historical exoskeleton operation experiments under different crosstalk environments and different movement difficulty conditions; In each group of exoskeleton operation experiments, the brain-to-noise ratio, electromyographic crosstalk index, crosstalk coupling degree, difficulty coefficient of the action to be performed, and error-related potential characteristics were calculated. The brain-to-noise ratio, electromyographic crosstalk index, crosstalk coupling degree, difficulty coefficient of the action to be performed, and error-related potential characteristics of each exoskeleton operation experiment were combined into a historical human-machine matching feature vector. Obtain training labels for the difficulty of human-machine matching in each group of exoskeleton operation experiments; combine the historical human-machine matching feature vectors of all exoskeleton operation experiments and the corresponding training labels for the difficulty of human-machine matching into a human-machine matching dataset, and divide the human-machine matching dataset into a human-machine matching training set and a human-machine matching validation set in a 7:3 ratio. A Gaussian process regression model is constructed, with the covariance function chosen as a combination of a radial basis function kernel and a white noise kernel. The initial length scale parameter of the radial basis function kernel is set to 1, and the initial noise variance of the white noise kernel is set to 0.1. Based on the human-machine matching training set, the human-machine matching feature vector is used as the input feature of the Gaussian process regression model, and the human-machine matching difficulty training label is used as the output target of the Gaussian process regression model. The Gaussian process regression model is trained, and the kernel function hyperparameter is determined by maximizing the log marginal likelihood. The optimizer adopts the L-BFGS algorithm, and the maximum number of iterations is set to 100. Convergence occurs when the increment of the log marginal likelihood is less than one part per million, resulting in the initial human-machine matching difficulty analysis model. The initial human-machine matching difficulty analysis model is validated using a human-machine matching validation set. The root mean square error on the validation set is calculated, and the initial human-machine matching difficulty analysis model whose root mean square error on the validation set is less than or equal to a preset error threshold (the error threshold in this embodiment is 0.15 by default) is used as the human-machine matching difficulty analysis model.

[0031] In a specific embodiment, such as Figure 4 As shown, step S3, based on the analysis results of the difficulty of human-machine matching, and combined with the exoskeleton motion state data and target motion commands, predicts and analyzes the risk of exoskeleton motion deviation caused by human-machine matching difficulties; including the following specific steps: S31. Compare the exoskeleton motion state data within the corresponding time window with the target motion command joint by joint, and calculate the degree of angular deviation, angular velocity deviation and motion synchronization delay of each joint within the corresponding time window. The degree of angle deviation of each joint is obtained by dividing the angle deviation of each joint by the angle tolerance. The angle deviation is the difference between the target angle and the actual angle. The degree of angular velocity deviation of each joint is obtained by dividing the angular velocity deviation of each joint by the angular velocity tolerance, where the angular velocity deviation is the difference between the target angular velocity and the actual angular velocity. The motion synchronization delay of each joint is obtained by dividing the motion synchronization delay of each joint by the reference delay (the reference delay is 100ms by default in this embodiment). The motion synchronization delay is defined as the time delay corresponding to the maximum value of the normalized cross-correlation function of the target angle and the actual angle time sequence signals. The value of the cross-correlation function at the time delay is equal to the product of the covariance of the target angle and the actual angle time sequence signals divided by the standard deviation of the target angle and the actual angle time sequence signals. The cross-correlation value corresponding to all time delays is calculated by traversing within the preset time delay search range of -500ms to 500ms with a sampling interval of 4ms. The time delay that makes the cross-correlation function reach its maximum value is taken as the motion synchronization delay of the joint. S32. Take the root mean square values ​​of the degree of angular deviation, the degree of angular velocity deviation, and the motion synchronization delay of all joints within the corresponding time window to obtain the risk of angular deviation, angular velocity deviation, and motion synchronization delay caused by human-machine matching difficulties within the corresponding time window. In this embodiment, the squaring operation in the root mean square operation eliminates the directional influence of the deviation sign, so that the comprehensive index only reflects the magnitude of the deviation and is not affected by the direction of the deviation. S33. The degree of difficulty in human-machine matching between the user and the exoskeleton in the current time window, the peak amplitude ratio of error-related potentials in the error-related potential characteristics, the electromyographic crosstalk index, and the risk of angle deviation, angular velocity deviation, and motion synchronization delay caused by human-machine matching difficulties in the current time window, as well as the first-order difference of the risk of angle deviation caused by human-machine matching difficulties in the current time window, are time-aligned to construct a propagation feature vector. S34. Input the propagation feature vector into the pre-trained motion deviation risk prediction model to obtain the risk of exoskeleton motion deviation caused by human-machine matching difficulties in the current time window. The larger the value, the higher the risk of motion deviation exceeding the limit event in the future preset time window. Among them, the motion deviation risk prediction model adopts a long short-term memory recurrent neural network. Its training label is a binary deviation over-limit label. When the absolute value of the angle deviation of any joint within the future preset time window is greater than 1, the label is set to 1, otherwise it is set to 0. A momentum regularization term is introduced during model training to improve the stability of time series prediction. The process of building a motion deviation risk prediction model includes: The time-series records of the exoskeleton robot running continuously under various historical working conditions are obtained. Each time-series record contains the propagation feature vector within the corresponding historical time window and the deviation exceeding the limit label. The rule for determining the deviation exceeding the limit label is: if the absolute value of the angle deviation of any joint within a future preset time window (the default length of the future preset time window in this embodiment is 500ms) is greater than 1, then the label is 1; otherwise, the label is 0. The propagation feature vectors of all time-series records and their corresponding deviation exceedance labels are combined to form a deviation risk dataset, which is then divided into a deviation risk training set and a deviation risk validation set according to time order. Specifically, after arranging the data in time order, the first 80% of the data is used as the deviation risk training set, and the last 20% is used as the deviation risk validation set. A long short-term memory recurrent neural network model is constructed. The network structure includes an input layer, two long short-term memory hidden layers, and an output layer. The number of hidden units in each layer is a preset value, and the output layer uses the sigmoid activation function. In a specific embodiment, the network structure is set as follows: the input layer receives a seven-dimensional propagation feature vector, followed by two long short-term memory hidden layers, each with 64 hidden units. The first layer returns the complete time series output to the second layer, and the second layer only returns the hidden state of the last time step to the output layer. The output layer is a fully connected layer containing only one unit and uses the sigmoid activation function to map the output to a numerical range of 0-1. The output value is the action deviation risk. The larger the value, the higher the risk of an action deviation exceeding the limit event within the next 500 millisecond time window. Based on the bias risk training set, the propagation feature vector of each historical time window is used as the input feature of the Long Short-Term Memory Recurrent Neural Network (LSTM), and the bias exceedance label corresponding to the historical time window is used as the output target of the LSTM LSTM model. The LSTM LSTM model is trained using the binary cross-entropy loss function and the Adam optimizer. The initial learning rate is set to 0.001, the batch size is set to 32 time window samples, the training epoch limit is set to 200 epochs, and an early stopping mechanism is introduced. If the validation set loss does not decrease within 15 consecutive epochs, the training is terminated early and the model parameters with the lowest validation set loss are saved. To improve the stability of the model in time-series prediction under crosstalk conditions, this embodiment can also introduce a momentum regularization term during training to constrain the fluctuation of weight updates between adjacent time steps. At the same time, to prevent overfitting, a dropout regularization with a dropout rate of 0.3 is applied after each long short-term memory hidden layer, and a dropout regularization with a dropout rate of 0.5 is applied before the fully connected layer; thus obtaining the initial action deviation risk prediction model; the initial action deviation risk prediction model is validated through a deviation risk validation set, the binary classification accuracy on the validation set is calculated, and the initial action deviation risk prediction model with a binary classification accuracy on the validation set greater than or equal to a preset accuracy threshold (the accuracy threshold is 0.85 by default in this embodiment) is output as the action deviation risk prediction model.

[0032] In a specific embodiment, the use of a long short-term memory network, through its gating mechanism, is naturally suitable for capturing the temporal dependence and nonlinear dynamic characteristics of the propagation process from human-machine matching difficulties to action deviations. It can issue risk warnings in advance before the deviation actually exceeds the limit, thus gaining valuable response time for subsequent collaborative safety assessment and control strategy adjustment. At the same time, the combined use of momentum regularization and multi-layer dropout effectively suppresses the temporal overfitting and prediction oscillation problems that are prone to occur in recurrent networks under crosstalk noise environments.

[0033] In one specific embodiment, step S4 integrates the motion deviation risk prediction and analysis results with the exoskeleton structural status data to comprehensively assess the collaborative safety of the exoskeleton human-machine system; this includes the following specific steps: S41. From the exoskeleton structural state data, extract the absolute value of the actual joint torque of each joint within the current time window, the actual stress of each link, the joint design torque safety threshold, and the link material yield strength. The joint design torque safety threshold is obtained from the torque value corresponding to the material yield limit in the mechanical design drawings of the exoskeleton joint drive system, and the link material yield strength is determined according to the national standard yield strength parameter of the alloy material selected for the exoskeleton structural components. S42. Divide the absolute value of the actual joint torque of each joint by the safety threshold of the joint design torque, and subtract 1 from the result to obtain the safety margin of the joint torque of each joint; divide the actual stress of each link by the yield strength of the link material, and subtract 1 from the result to obtain the safety redundancy of the link stress of each link. S43. Extract the minimum value of the joint torque safety margin of all joints as the torque safety margin of the weakest joint; extract the minimum value of the link stress safety redundancy of all links as the stress safety redundancy of the weakest link. S44. Extract the risk of exoskeleton execution deviation caused by human-machine matching difficulties within the current time window; subtract the risk of exoskeleton execution deviation caused by human-machine matching difficulties within the current time window from 1 to obtain the safety margin of exoskeleton execution deviation within and outside the current time window. S45. The safety margin of the skeleton's execution action deviation, the safety margin of the weakest joint torque, and the safety redundancy of the weakest link stress are weighted and summed to obtain the collaborative safety status of the skeleton human-machine system inside and outside the current time window. The advantage of this embodiment is that by comprehensively evaluating the risk of motor deviation at the neurocognitive level and the structural safety margin at the physical and mechanical level under a unified weighted fusion framework, the soft safety (i.e., control coordination) and hard safety (i.e., structural integrity) of the exoskeleton human-machine system are jointly quantified. This avoids the fragmented safety assessment problem caused by focusing on only a single dimension in traditional solutions. At the same time, the introduction of the weakest link effect ensures that this embodiment can respond in a timely manner when any single link deteriorates, without being averaged out by the good performance of other links.

[0034] In one specific embodiment, step S5 involves executing a differentiated exoskeleton collaborative control scheme based on the assessment results of the collaborative safety of the exoskeleton human-machine system; this includes the following specific steps: S51. Extract the collaborative safety status of the human-machine skeletal system inside and outside the current time window; S52. A preset collaborative safety threshold is established. When the collaborative safety status of the exoskeleton human-machine system within and outside the current time window is greater than the collaborative safety threshold, the exoskeleton human-machine system is determined to be in a safe state, and the conventional collaborative control scheme is executed. When the collaborative safety status of the exoskeleton human-machine system within and outside the current time window is less than or equal to the collaborative safety threshold, the exoskeleton human-machine system is determined to be in a dangerous state, and a protective collaborative control scheme is executed. Specifically, the conventional collaborative control scheme includes the following: Keep the target motion commands of each joint of the exoskeleton unchanged, and execute the motion normally according to the preset speed range corresponding to the execution motion difficulty coefficient calculated in step S2. The torque limit of each joint and the stress limit of the link are maintained at the design default value.

[0035] Protective collaborative control schemes include the following: The exoskeleton immediately stops executing the current target movement command and switches to assisted holding mode. In assisted holding mode, each joint maintains its current actual position with minimal resistance, providing only gravity compensation torque to prevent the exoskeleton from falling due to its own weight and injuring the user. At the same time, it provides a small damping torque to prevent uncontrollable drift of the joints in the absence of commands. The torque limit threshold of each joint is reduced to 0.5 times the design default value to prevent instantaneous torque overload due to command errors or environmental interference. A safety warning is issued through the sound and light prompt devices deployed on the shoulders and waist of the exoskeleton to remind the user that the system is in a protective state.

[0036] The methods for determining the weights and thresholds not specifically described in this embodiment are as follows: Data on the entire exoskeleton operation process was collected from multiple groups of users under different crosstalk environments and movement difficulties. Following the aforementioned steps, the corresponding comprehensive crosstalk coupling feature vector, human-machine matching feature vector, propagation feature vector, and all intermediate indicators, including the torque safety margin of each joint and the stress safety redundancy of the linkage, were calculated. Simultaneously, sensors objectively recorded safety event labels for each experiment, such as whether the final movement deviation exceeded limits and whether the joint torque was overloaded, which served as the true benchmark for the fitting.

[0037] Import all the intermediate indicators and corresponding security event labels obtained above into the fitting software. In this embodiment, the fitting software can be MATLAB's Curve Fitting Toolbox or Python's Scikit-learn library.

[0038] The software uses either grid search or random search to traverse all candidate combinations with all weight coefficients between 0 and 1 and all thresholds between 0 and 1. For each candidate parameter combination, it calculates the corresponding crosstalk coupling degree, human-machine matching difficulty index, collaborative security index, and final security status judgment result, and compares them with real security event labels to calculate the determination coefficient for each parameter combination.

[0039] The output is the combination of weights and thresholds that gives the highest coefficient of determination, which is then used as the final parameter values ​​in the model.

[0040] Example 2 like Figure 5 As shown, this embodiment provides a brain-computer interface-based exoskeleton collaborative control system, including: The data acquisition module is used to acquire exoskeleton structural state data and simultaneously acquire the user's EEG signal data, exoskeleton motion state data and environmental crosstalk characteristic data through multimodal brain-computer interface sensors, motion state sensors and environmental perception sensors deployed on the exoskeleton robot. The human-machine matching analysis module is used to analyze the crosstalk coupling characteristics of brain-computer interface signals by combining EEG signal data and environmental crosstalk feature data; and based on the crosstalk coupling feature analysis results, combined with the difficulty of the user's current performance action, it analyzes the degree of difficulty of human-machine matching between the user and the exoskeleton. The execution status prediction module is used to predict and analyze the risk of exoskeleton execution deviation caused by human-machine matching difficulties, based on the analysis results of the difficulty of human-machine matching, combined with exoskeleton motion status data and target motion commands. The collaborative safety monitoring module is used to integrate the results of motion deviation risk prediction and analysis with exoskeleton structural status data to comprehensively assess the collaborative safety of the exoskeleton human-machine system. The collaborative control module is used to execute differentiated exoskeleton collaborative control schemes based on the assessment results of the collaborative safety of the exoskeleton human-machine system.

[0041] The parameters and steps for each unit module in the brain-computer interface-based exoskeleton collaborative control system of the present invention to achieve their respective functions can be referred to the parameters and steps in the embodiments of the brain-computer interface-based exoskeleton collaborative control method above, and will not be repeated here.

[0042] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, the embodiments for IoT devices and media are relatively simple in description because they are fundamentally similar to the method embodiments; relevant parts can be referred to the descriptions in the method embodiments.

[0043] The systems, media, and methods provided in the embodiments of the present invention are in one-to-one correspondence. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0044] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0045] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0046] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0047] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0048] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0049] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0050] It should also be noted that the terms include, encompass, or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitations, the inclusion of an element by a statement that includes a… does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0051] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A brain-computer interface-based exoskeleton collaborative control method, characterized in that, Includes the following steps: S1. Acquire exoskeleton structural state data, and simultaneously collect user's EEG signal data, exoskeleton motion state data, and environmental crosstalk characteristic data through multimodal brain-computer interface sensors, motion state sensors, and environmental perception sensors deployed on the exoskeleton robot. S2. Combining EEG signal data and environmental crosstalk characteristic data, analyze the crosstalk coupling characteristics of brain-computer interface signals; and based on the crosstalk coupling characteristic analysis results, combined with the difficulty of the user's current performance, analyze the degree of human-machine matching difficulty between the user and the exoskeleton. S3. Based on the analysis results of the difficulty of human-machine matching, and combined with the exoskeleton motion state data and target motion commands, predict and analyze the risk of exoskeleton execution deviation caused by human-machine matching difficulties. S4. Integrate the results of motion deviation risk prediction and analysis with exoskeleton structural status data to comprehensively assess the collaborative safety of the exoskeleton human-machine system. S5. Based on the assessment results of the collaborative safety of the exoskeleton human-machine system, implement differentiated exoskeleton collaborative control schemes.

2. The brain-computer interface-based exoskeleton collaborative control method according to claim 1, characterized in that, In step S2, the crosstalk coupling characteristics of the brain-computer interface signals are analyzed by combining EEG signal data and environmental crosstalk feature data; specifically, this includes the following: S21. Perform bandpass filtering on the EEG signal data to separate out the motion-related signals. Rhythm band and Rhythm band and resting baseline band; S22. Calculate the energy of the signal in each frequency band window by window using short-time Fourier transform; perform windowing and fast Fourier transform on the EEG signal in the corresponding time window to obtain the power spectrum in the corresponding time window. S23. On the power spectrum within the corresponding time window, integrate and sum the power values ​​within the frequency band range to obtain the task-state frequency band power of the specified frequency band within the current time window; S24. Divide the difference between the task-state frequency band power and the resting-state frequency band average power within the corresponding time window by the resting-state frequency band average power to obtain the event-related desynchronization degree within the corresponding time window; wherein, the resting-state frequency band average power is the average value of the specified frequency band power within the last preset resting-state duration; S25. Obtain the average value of the signal power of the task frequency band and the signal power of the high-frequency noise frequency band within the corresponding time window; S26. Divide the signal power of the task frequency band within the corresponding time window by the mean of the signal power of the high-frequency noise frequency band to obtain the brain-to-electrical signal-to-noise ratio within the corresponding time window. S27. From the environmental crosstalk feature data, extract the preset narrowband frequency range containing the power frequency within the corresponding time window as the power frequency band power and the number of electrooculogram artifact events detected within the corresponding time window; simultaneously, calculate the interaction between each electromyography channel and the electroencephalogram signal. Coherence coefficient within the frequency band; S28. Divide the power frequency band power near the power frequency in the corresponding time window by the signal power of the task frequency band to obtain the power frequency interference intensity in the corresponding time window. S29. Divide the number of electrooculogram artifact events detected in the corresponding time window by the product of the sampling frequency and the duration of the time window to obtain the electrooculogram artifact density in the corresponding time window. S210, Combine all electromyographic channels and electroencephalogram signals within the corresponding time window. The absolute values ​​of the coherence coefficients within the frequency band are arithmetic averaged to obtain the electromyographic crosstalk index within the corresponding time window; S211. Obtain the standard deviation of EEG signal amplitude and the mean of the standard deviation of resting EEG signal amplitude within the corresponding time window; subtract the mean of the standard deviation of EEG signal amplitude within the corresponding time window and the mean of the standard deviation of resting EEG signal amplitude; divide the result of the subtraction by the mean of the standard deviation of resting EEG signal amplitude to obtain the intensity of motion artifacts within the corresponding time window. S212. Combine the event-related desynchronization degree, brain-to-electrical-to-noise ratio, power frequency interference intensity, electrooculogram artifact density, electromyography crosstalk index, and motion artifact intensity within the corresponding time window into a comprehensive crosstalk coupling feature vector, input it into the pre-trained crosstalk coupling degree calculation model, and obtain the crosstalk coupling degree within the corresponding time window.

3. The brain-computer interface-based exoskeleton collaborative control method according to claim 2, characterized in that, In step S2, based on the crosstalk coupling characteristic analysis results and the difficulty of the user's current action, the degree of difficulty in human-machine matching between the user and the exoskeleton is analyzed; specifically, the following steps are included: S213. Calculate the execution difficulty coefficient of the main body and the execution difficulty coefficient of the exoskeleton; perform a weighted sum of the execution difficulty coefficient of the main body and the execution difficulty coefficient of the exoskeleton to obtain the execution difficulty coefficient of the corresponding execution action within the corresponding time window; S214. Extract error-related potential features from the frontal central region electrodes of the EEG signal. The error-related potential is the negative deflection component of the brain generated in the brain within a preset time window after the action is performed when the exoskeleton's action does not match the user's movement intention. The extracted error-related potential features include the peak amplitude ratio of error-related potentials, the average power ratio within the error-related potential window, and the area under the waveform curve of error-related potentials. S215. Obtain the crosstalk coupling degree within the current time window, the execution difficulty coefficient of the corresponding action within the current time window, the electromyographic crosstalk index and the brain electrical signal-to-noise ratio, as well as the error-related potential characteristics within the current time window, combine them into a human-machine matching feature vector, input it into a pre-trained human-machine matching difficulty analysis model, and obtain the human-machine matching difficulty between the user and the exoskeleton within the current time window; the human-machine matching difficulty analysis model adopts the Gaussian process regression algorithm.

4. The brain-computer interface-based exoskeleton collaborative control method according to claim 3, characterized in that, The construction process of the human-machine matching difficulty analysis model includes: Acquire historical exoskeleton operation experiments under different crosstalk environments and different movement difficulty conditions; In each group of exoskeleton operation experiments, the brain-to-noise ratio, electromyographic crosstalk index, crosstalk coupling degree, difficulty coefficient of the action to be performed, and error-related potential characteristics were calculated. The brain-to-noise ratio, electromyographic crosstalk index, crosstalk coupling degree, difficulty coefficient of the action to be performed, and error-related potential characteristics of each exoskeleton operation experiment were combined into a historical human-machine matching feature vector. Obtain training labels for the difficulty of human-machine matching in each group of exoskeleton operation experiments; combine the historical human-machine matching feature vectors of all exoskeleton operation experiments and the corresponding training labels for the difficulty of human-machine matching into a human-machine matching dataset, and divide the human-machine matching dataset into a human-machine matching training set and a human-machine matching validation set. A Gaussian process regression model is constructed, with the covariance function chosen as a combination of radial basis function kernel and white noise kernel. Based on the human-machine matching training set, the human-machine matching feature vector is used as the input feature of the Gaussian process regression model, and the training label of human-machine matching difficulty is used as the output target of the Gaussian process regression model. The Gaussian process regression model is trained, and the kernel function hyperparameter is determined by maximizing the log marginal likelihood, resulting in an initial human-machine matching difficulty analysis model. The initial human-machine matching difficulty analysis model is validated using a human-machine matching validation set. The root mean square error on the validation set is calculated, and the initial human-machine matching difficulty analysis model whose root mean square error on the validation set is less than or equal to a preset error threshold is output as the final human-machine matching difficulty analysis model.

5. The brain-computer interface-based exoskeleton collaborative control method according to claim 4, characterized in that, In step S3, based on the analysis results of the difficulty of human-machine matching, and combined with the exoskeleton motion state data and target motion commands, the risk of exoskeleton execution deviation caused by human-machine matching difficulty is predicted and analyzed. The specific steps include the following: S31. Compare the exoskeleton motion state data within the corresponding time window with the target motion command joint by joint, and calculate the degree of angular deviation, angular velocity deviation and motion synchronization delay of each joint within the corresponding time window. S32. Take the root mean square values ​​of the degree of angular deviation, the degree of angular velocity deviation, and the motion synchronization delay of all joints within the corresponding time window to obtain the risk of angular deviation, angular velocity deviation, and motion synchronization delay caused by human-machine matching difficulties within the corresponding time window. S33. The degree of difficulty in human-machine matching between the user and the exoskeleton in the current time window, the peak amplitude ratio of error-related potentials in the error-related potential characteristics, the electromyographic crosstalk index, and the risk of angle deviation, angular velocity deviation, and motion synchronization delay caused by human-machine matching difficulties in the current time window, as well as the first-order difference of the risk of angle deviation caused by human-machine matching difficulties in the current time window, are time-aligned to construct a propagation feature vector. S34. Input the propagation feature vector into the pre-trained motion deviation risk prediction model to obtain the risk of exoskeleton motion deviation caused by human-machine matching difficulties within the current time window.

6. The brain-computer interface-based exoskeleton collaborative control method according to claim 3, characterized in that, In step S4, the results of motion deviation risk prediction and analysis and the exoskeleton structural status data are integrated to comprehensively assess the collaborative safety of the exoskeleton human-machine system. The specific steps include the following: S41. Extract the absolute value of the actual joint torque of each joint, the actual stress of each link, the joint design torque safety threshold, and the yield strength of the link material from the exoskeleton structural state data within the current time window. S42. Divide the absolute value of the actual joint torque of each joint by the joint design torque safety threshold, and subtract the result of the division from 1 to obtain the joint torque safety margin of each joint. Divide the actual stress of each link by the yield strength of the link material, and subtract 1 from the result to obtain the stress safety redundancy of each link. S43. Extract the minimum value among the joint torque safety margins of all joints as the torque safety margin of the weakest joint. Extract the minimum value of the stress safety redundancy of all links as the stress safety redundancy of the weakest link; S44. Extract the risk of exoskeleton execution deviation caused by human-machine matching difficulties within the current time window; subtract the risk of exoskeleton execution deviation caused by human-machine matching difficulties within the current time window from 1 to obtain the safety margin of exoskeleton execution deviation within and outside the current time window. S45. The safety margin of the skeleton's execution action deviation, the safety margin of the weakest joint torque, and the safety redundancy of the weakest link stress are weighted and summed to obtain the collaborative safety status of the skeleton human-machine system inside and outside the current time window.

7. The brain-computer interface-based exoskeleton collaborative control method according to claim 6, characterized in that, In step S5, a differentiated exoskeleton collaborative control scheme is executed based on the assessment results of the collaborative safety of the exoskeleton human-machine system. The specific steps include the following: S51. Extract the collaborative safety status of the human-machine skeletal system inside and outside the current time window; S52. Preset a collaborative safety threshold. When the collaborative safety status of the exoskeleton human-machine system inside and outside the current time window is greater than the collaborative safety threshold, the exoskeleton human-machine system is judged to be in a safe state and the conventional collaborative control scheme is executed. When the collaborative safety status of the exoskeleton human-machine system inside and outside the current time window is less than or equal to the collaborative safety threshold, the exoskeleton human-machine system is judged to be in a dangerous state and the protective collaborative control scheme is executed.

8. A brain-computer interface-based exoskeleton collaborative control system, implemented based on any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to acquire exoskeleton structural state data and simultaneously acquire the user's EEG signal data, exoskeleton motion state data and environmental crosstalk characteristic data through multimodal brain-computer interface sensors, motion state sensors and environmental perception sensors deployed on the exoskeleton robot. The human-machine matching analysis module is used to analyze the crosstalk coupling characteristics of brain-computer interface signals by combining EEG signal data and environmental crosstalk feature data; and based on the crosstalk coupling feature analysis results, combined with the difficulty of the user's current performance action, it analyzes the degree of difficulty of human-machine matching between the user and the exoskeleton. The execution status prediction module is used to predict and analyze the risk of exoskeleton execution deviation caused by human-machine matching difficulties, based on the analysis results of the difficulty of human-machine matching, combined with exoskeleton motion status data and target motion commands. The collaborative safety monitoring module is used to integrate the results of motion deviation risk prediction and analysis with exoskeleton structural status data to comprehensively assess the collaborative safety of the exoskeleton human-machine system. The collaborative control module is used to execute differentiated exoskeleton collaborative control schemes based on the assessment results of the collaborative safety of the exoskeleton human-machine system.

Images