Exoskeleton man-machine motion intention cooperative control method and device and storage medium

Abstract

The invention provides an exoskeleton man-machine motion intention cooperative control method and device and a storage medium, and relates to the technical field of robot motion control signal processing. According to the method, signal phase differences are acquired through energy buttons at human joints, and an original phase difference sequence is formed and preprocessed; calculating a cross correlation coefficient of the target and the associated joint based on the preprocessed sequence, and constructing a motion feature set in combination with the joint speed and acceleration; constructing a prediction model by using a long-short-term memory network, and outputting motion intention and joint position prediction information through multi-feature fusion training; and analyzing the prediction information by adopting a second-order linear impedance model, generating a driving torque instruction and sending the driving torque instruction to the servo motor for execution. According to the method, signal phase difference analysis, LSTM prediction and impedance control are fused, the problems of response lag and poor adaptability of a traditional exoskeleton are solved, the control real-time performance, accuracy and cooperative fluency are improved, and the use requirements of different users in different scenes are met.

Description

Technical Field

[0001] This invention mainly relates to the field of robot motion control signal processing technology, specifically to a method, device, and storage medium for collaborative control of human-machine motion intentions in an exoskeleton. Background Technology

[0002] Exoskeleton robots, as human-machine collaborative equipment, are increasingly widely used in fields such as rehabilitation medicine and the military. Their core technological requirement lies in accurately capturing human movement intentions and outputting appropriate control commands to achieve smooth human-machine movement coordination. However, existing exoskeleton human-machine movement intention coordination control technologies still have many limitations and cannot meet the needs of practical applications. On the one hand, the accuracy of motion signal acquisition and processing is insufficient. Existing technologies mostly rely on traditional sensors to collect joint motion data, which are easily affected by human body shaking, electromagnetic interference, and sensor noise, resulting in a large number of missing values ​​and interference components in the raw signal; moreover, there is a lack of targeted data preprocessing strategies, which directly affects the reliability of subsequent feature extraction and makes it difficult to accurately characterize the human joint motion state.

[0003] On the other hand, the compatibility between motion intention prediction and control command generation is poor. Traditional methods often judge motion intention based on a single feature or a simple combination of features, ignoring the temporal correlation of human joint movements and the multi-joint coordination characteristics, resulting in lagging intention prediction and low accuracy. At the same time, control command generation often uses fixed parameter models, which cannot dynamically adjust the output torque according to the predicted motion intention and joint position, causing the exoskeleton response to be out of sync with the human motion intention, poor human-machine collaboration, and even the risk of equipment overload or human discomfort due to improper torque output.

[0004] Therefore, there is an urgent need for an exoskeleton human-machine motion intention collaborative control technology that can achieve precise signal processing, efficient intention prediction, and dynamic adaptation of control commands, in order to break through the existing technical bottlenecks and improve the practicality and safety of exoskeletons. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, device and storage medium for coordinated control of human-machine movement intention of exoskeleton, which addresses the shortcomings of the prior art.

[0006] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for coordinated control of exoskeleton human-machine motion intention, comprising: Multiple signal phase differences are collected by energy buttons placed at human joints to obtain the original phase difference sequence, and the original phase difference sequence is preprocessed. The cross-correlation coefficient between the target joint and the associated joint sequence is calculated based on the preprocessed phase difference sequence, and the human joint motion feature set is obtained based on the cross-correlation coefficient. A motion intention prediction model is constructed based on a long short-term memory network. The motion intention prediction model is trained by multi-feature fusion through human joint motion feature set, and the motion intention prediction information and joint position prediction information are output. Based on the second-order linear impedance model, the motion intention prediction information and joint position prediction information are analyzed to generate the driving torque command of the exoskeleton joint. The driving torque command is sent to the exoskeleton servo motor for execution.

[0007] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: an exoskeleton human-machine motion intention collaborative control device, comprising: The data preprocessing module is used to collect multiple signal phase differences through energy buttons set at human joints, obtain the original phase difference sequence, and perform data preprocessing on the original phase difference sequence; The instruction generation module is used to: calculate the cross-correlation coefficient between the target joint and the associated joint sequence based on the preprocessed phase difference sequence, and obtain the human joint motion feature set based on the cross-correlation coefficient; A motion intention prediction model is constructed based on a long short-term memory network. The motion intention prediction model is trained by multi-feature fusion through human joint motion feature set, and the motion intention prediction information and joint position prediction information are output. Based on the second-order linear impedance model, the motion intention prediction information and joint position prediction information are analyzed to generate the driving torque command of the exoskeleton joint. The instruction sending module is used to send the driving torque instruction to the exoskeleton servo motor for execution.

[0008] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: an exoskeleton human-machine motion intention collaborative control device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the exoskeleton human-machine motion intention collaborative control method as described above.

[0009] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the exoskeleton human-machine motion intention collaborative control method as described above.

[0010] The beneficial effects of this invention are: by deeply combining signal phase difference analysis, LSTM prediction model and second-order linear impedance control, it solves the problems of lag response and poor human-machine adaptability of traditional exoskeletons. It enables the exoskeleton to actively match the human body's movement state, significantly improving the real-time performance, accuracy and smoothness of human-machine collaboration of exoskeleton control, and can be widely adapted to the needs of rehabilitation training, military and other scenarios. Attached Figure Description

[0011] Figure 1 A flowchart of the exoskeleton human-machine motion intention collaborative control method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the module of the exoskeleton human-machine motion intention collaborative control device provided in an embodiment of the present invention. Detailed Implementation

[0012] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0013] Example 1: As Figure 1 As shown, this embodiment of the invention provides a method for coordinated control of exoskeleton human-machine motion intent, including: Multiple signal phase differences are collected by energy buttons placed at human joints to obtain the original phase difference sequence, and the original phase difference sequence is preprocessed. The cross-correlation coefficient between the target joint and the associated joint sequence is calculated based on the preprocessed phase difference sequence, and the human joint motion feature set is obtained based on the cross-correlation coefficient. A motion intention prediction model is constructed based on a long short-term memory network. The motion intention prediction model is trained by multi-feature fusion through human joint motion feature set, and the motion intention prediction information and joint position prediction information are output. Based on the second-order linear impedance model, the motion intention prediction information and joint position prediction information are analyzed to generate the driving torque command of the exoskeleton joint. The driving torque command is sent to the exoskeleton servo motor for execution.

[0014] In the above embodiments, the signal phase difference analysis, LSTM prediction model and second-order linear impedance control are deeply combined to solve the problems of lag response and poor human-machine adaptability of traditional exoskeletons. This allows the exoskeleton to actively match the human body's movement state, significantly improving the real-time performance, accuracy and smoothness of human-machine collaboration of exoskeleton control, and can be widely adapted to the needs of rehabilitation training, military and other scenarios.

[0015] Preferably, the data preprocessing of the original phase difference sequence includes: The lengths of multiple missing segments in the original phase difference sequence are counted. If the length of a missing segment is less than or equal to the preset number of sampling points, the missing segment is filled by linear interpolation. If the length of a missing segment is greater than the preset number of sampling points, the missing segment is marked as invalid and the invalid missing segment is filled by sliding window mean. The original phase difference sequence after padding is filtered by a minimum mean square error adaptive filter to obtain a preprocessed phase difference sequence.

[0016] In the above embodiments, a hierarchical data preprocessing strategy is provided to address the issues of missing and noisy original phase difference sequences: differential completion methods using linear interpolation and sliding window mean filling maximize data validity; furthermore, minimum mean square error adaptive filtering removes interference noise, ensuring the integrity and reliability of the preprocessed data. This solution addresses the pain point of sensor-acquired signals being susceptible to environmental and equipment influences, providing a high-quality data foundation for subsequent feature extraction and intent prediction, and avoiding model accuracy degradation or control command anomalies caused by poor-quality data.

[0017] Preferably, the cross-correlation coefficient between the target joint and the associated joint sequence is calculated based on the preprocessed phase difference sequence, and the human joint motion feature set is obtained based on the cross-correlation coefficient, including: The joint motion velocity is obtained by performing a first-order difference calculation on the preprocessed phase difference sequence based on the joint motion velocity formula, which is: , in, Let be the joint motion velocity corresponding to the i-th sampling time. Let i be the phase difference value at the i-th sampling point in the phase difference sequence of the target joint or associated joint. This represents the phase difference value at the (i+1)th sampling point in the phase difference sequence of the target joint or associated joint. The sampling time interval; The joint motion velocity is calculated using a first-order difference based on the joint motion acceleration formula to obtain the joint motion acceleration. The joint motion acceleration formula is as follows: , in, To represent the joint motion acceleration corresponding to the i-th sampling time, Let be the joint motion velocity at the (i+1)th sampling time. The target joint θ is calculated using the synergy calculation formula and the preprocessed phase difference sequence. k With associated joint θ h The cross-relationship coefficient, and the formula for calculating the degree of collaboration is: , Where r is the cross-correlation coefficient and n is the number of data points in the phase difference sequence. Let be the phase difference value of the i-th sampling point in the target joint phase difference sequence. The mean of the target joint phase difference sequence. , This represents the phase difference value at the i-th sampling point in the associated joint phase difference sequence. The mean of the associated joint phase difference sequence. ; Each cross-correlation coefficient is compared with a preset feature threshold. If the cross-correlation coefficient is greater than the preset feature threshold, the target joint corresponding to the cross-correlation coefficient is removed. Associated joints Features based on the removed target joints Associated joints The feature set is used to obtain the human joint motion feature set.

[0018] In the above embodiments, a multi-dimensional human joint motion feature set consisting of "dynamic features + collaborative features" was constructed. The dynamic attributes of joint motion were quantified using velocity and acceleration formulas, and the cross-correlation coefficient was used to represent the collaborative relationships among multiple joints. Redundant features were also eliminated through threshold filtering. This approach overcomes the limitations of traditional single-feature extraction, comprehensively capturing both the "individual dynamics" and "joint collaboration" characteristics of human motion. It provides more discriminative input features for the LSTM model, effectively improving the accuracy of motion intent classification and joint position prediction, and is particularly suitable for feature expression needs in complex motion scenarios.

[0019] Preferably, a motion intention prediction model is constructed based on a long short-term memory network. The model is then trained using a human joint motion feature set through multi-feature fusion to predict motion intention, outputting motion intention prediction information and joint position prediction information, including: A motion intention prediction model is constructed based on a long short-term memory network. The motion intention prediction model includes a feature preprocessing layer, an LSTM fusion layer, and a dual-task output layer. The feature preprocessing layer is used to concatenate the joint motion velocity, joint motion acceleration and cross-correlation coefficient of the continuously sampled time moments in the human joint motion feature set according to time steps to form an LSTM temporal feature matrix with a dimension of {number of time steps × feature dimension}. It should be understood that the core objective of the feature preprocessing layer is to convert scattered multi-dimensional features into a temporal feature matrix that can be processed by the LSTM network, thereby eliminating the impact of differences in feature scale and inconsistent formats. 1. Feature dimension alignment: Human joint movement characteristics are concentrated, joint speed (Unit: radians / second) Acceleration (Unit: radians / second) 2 The physical meaning and numerical range of the cross-correlation coefficient r (dimensionless, taking values ​​[-1, 1]) differ greatly. Within each layer, scale interference is first eliminated through Min-Max normalization (mapping the eigenvalues ​​to the [0, 1] interval).

[0020] 2. Timing window splicing: The normalized features are concatenated in units of "time steps". For example, if the number of time steps is set to 20 (corresponding to a 200ms sampling duration and 20 sampling points at a 100Hz sampling rate), and the feature dimension is 3 (corresponding to three types of features: velocity, acceleration, and cross-correlation coefficient), then the [v] values ​​from 20 consecutive sampling times will be concatenated. i a i The feature vectors [r] are stacked row by row to form an LSTM temporal feature matrix with dimension {20×3}.

[0021] By transforming discrete sampling point features into continuous motion segment features, LSTM can make predictions based on "a motion process" rather than "a single moment", thus improving the reliability of the results.

[0022] The LSTM fusion layer includes a forget gate, an input gate, and an output gate. The forget gate is used to perform redundant filtering on the input LSTM temporal feature matrix. The input gate is used to enhance the filtered LSTM temporal feature matrix with temporal features related to motion intent. The output gate is used to capture the long-short-term feature relationship in the enhanced temporal features to obtain a temporal feature vector. It should be understood that LSTM (Long Short-Term Memory Network) solves the problem of "vanishing gradients in long sequences" in traditional recurrent neural networks (RNNs) through a unique gating mechanism (forget gate, input gate, output gate), and can accurately capture the correlation between short-term instantaneous changes in motion features and long-term trends.

[0023] The purpose of the forget gate is to filter valuable historical features and discard noise or redundant information. It outputs the "forget probability" (with values ​​[0,1]) through the sigmoid activation function, expressed as: , in, Here is the forget gate weight matrix. This is the output of the hidden layer from the previous time step (historical feature memory). The feature vector at the current time step. For bias terms, This is the sigmoid function.

[0024] For example, a sudden increase in acceleration that occurs accidentally during exercise (such as an unexpected jolt of the body) will be removed from historical memory by the forget gate with a probability close to 0; while periodic speed peaks during walking (such as those occurring every 0.5 seconds) will be removed from historical memory by the forget gate with a probability close to 1, thus preserving their long-term regularity.

[0025] The input gate is responsible for "writing the core features of the current time step into the memory unit", which is divided into two steps: The first step is to determine "which features need to be updated" (update probability) using the sigmoid function; The second step is to generate "candidate feature values ​​to be updated" using the tanh function; Finally, the "update probability × candidate value" is added to the memory cell, as shown in the formula: , , , For example, when going up or down stairs, the hip-knee cross-correlation coefficient r increases from 0.6 to 0.9 (high coordination state). The input gate strengthens the weight of this feature and writes it into the memory unit as a key basis for judging the "intention to go upstairs".

[0026] The output gate is responsible for "combining the current memory unit with the state of the historical hidden layer to generate the fused temporal feature vector", the formula of which is: , , in, The output feature vector for the current time step contains "key features of the current moment" (such as the current velocity). Furthermore, it incorporates "correlational features of historical moments" (such as the acceleration trends of the previous 5 moments). For example, during walking, the output gate can capture "acceleration..." After the peak occurs two time steps later, the speed... The long-term and short-term correlation of "reaching the maximum value" provides temporal logic support for subsequent location prediction.

[0027] The dual-task output layer includes a motion intent classification branch and a joint position prediction branch. The motion intent classification branch maps the temporal feature vector output by the LSTM fusion layer to a dimension matching the number of motion intent categories, resulting in multiple motion intent categories. A softmax activation layer is then used to calculate the probability distribution of each motion intent category, and the category with the highest probability is taken as the motion intent prediction information. The joint position prediction branch includes a regression layer, which learns joint movement velocities. Joint motion acceleration Based on the intrinsic relationship with the phase difference change, the predicted joint position is output.

[0028] It should be understood that the dual-task output layer adopts a "parallel branch" design, simultaneously realizing "motion intention classification" (decision-making task) and "joint position prediction" (quantization task) based on the temporal feature vector fused by LSTM. The two are mutually constrained and collaboratively optimized, and the specific mechanism is as follows: Motion intention classification branch.

[0029] Step 1: Fully Connected Layer Mapping. The temporal feature vector (e.g., 64-dimensional) output by the LSTM is mapped to a dimension that matches the number of motion intent categories (e.g., 4-dimensional vectors for 4 intent categories), thus achieving the dimensional transformation from features to intents.

[0030] Step 2: Softmax probability calculation. The mapped vector is converted into a probability distribution through a Softmax activation layer.

[0031] Step 3: Intent Decision. The category corresponding to the highest probability is taken as the motion intent prediction information, and the probability is output as the "confidence level" (e.g., the probability of "going upstairs" is 0.92, so the confidence level is 0.92), which is used by the exoskeleton to judge the reliability of the prediction results (a secondary prediction is triggered when the confidence level is <0.8).

[0032] Joint position prediction branch.

[0033] This branch addresses the need for precise control of joint movement trajectories in exoskeletons, outputting predicted joint positions for future moments. The workflow is as follows: Step 1: Regression Layer Modeling. A linear regression layer (with ReLU activation function) is used, based on the temporal feature vector output by LSTM, to learn the "velocity". acceleration Phase difference change The inherent relationship between "".

[0034] Step 2: Position Prediction. Output the predicted joint phase difference value for the next time step. .

[0035] Step 3: Result Calibration. The predicted values ​​are corrected based on the cross-correlation coefficient r: if r > 0.8 for the target joint and related joints, the adjustment is made by referring to the phase difference trend of the related joints. This avoids excessive prediction errors for a single joint.

[0036] In the above embodiments, the feature preprocessing layer regularizes temporal data, the LSTM fusion layer captures long-term and short-term feature dependencies through a gating mechanism, and the dual-task output layer simultaneously achieves intent classification and position prediction. This structure is specifically adapted to the temporal and multi-dimensional nature of human motion features, solving the problem that traditional models struggle to balance "category judgment" and "quantitative prediction." It can simultaneously output highly reliable motion intent and accurate joint position information, providing crucial support for exoskeleton early assistance and trajectory planning.

[0037] Preferably, the method further includes co-training the motion intent classification branch and the joint position prediction branch of the dual-task output layer using a joint loss function, wherein the joint loss function is: , in, Cross-entropy loss for classifying motion intent. This represents the mean squared error loss for joint position prediction. These are the weighting coefficients.

[0038] In the above embodiments, the dual-task output layers are trained collaboratively using a joint loss function, and the cross-entropy loss and mean squared error loss are weighted and fused. This approach avoids the model's unbalanced performance caused by single-task training, enabling the "discriminativeness" of motion intent classification and the "quantitative accuracy" of joint position prediction to mutually reinforce each other—intent classification provides scene constraints for position prediction, and position prediction provides dynamic basis for intent classification, ultimately achieving simultaneous optimization of dual-task performance and improving the overall robustness and generalization ability of the model.

[0039] Preferably, based on a second-order linear impedance model, command analysis is performed on the motion intention prediction information and joint position prediction information to generate driving torque commands for the exoskeleton joints, including: Get the current actual position of the exoskeleton joints and the predicted joint position value Perform difference calculation to obtain positional deviation. ; Extract the maximum probability from the motion intent prediction information, compare the maximum probability with a preset confidence level, and if it is greater than or equal to the preset confidence level, then the motion intent category is determined to be valid. Based on the effective motion intention category, preset basic impedance parameters are invoked, based on the correction formula and position deviation. The preset basic impedance parameters are corrected using the following formula: , , Where K is the corrected stiffness. This corresponds to the maximum range of motion of the joint. B represents the preset basic stiffness, and B represents the corrected damping. To preset the basic damping, This represents the upper limit of safe speed under the corresponding motion intention; A second-order linear impedance model is used to construct the torque calculation equation. Based on the torque calculation equation and position deviation... The driving torque is calculated using the corrected stiffness K and the corrected damping B. The torque calculation equation is as follows: , in, This refers to the driving torque that the exoskeleton joints need to output. This represents the stiffness moment corresponding to the positional deviation. The damping torque corresponding to the velocity. The compensating torque for the human body's exertion; The calculated With maximum torque and minimum torque If compared, Then take the maximum torque. ,like Then take the minimum torque. It converts the driving torque command into a servo motor signal and sends it.

[0040] In the above embodiments, the validity of the input is ensured through confidence and deviation verification, the impedance parameters are dynamically adjusted according to the motion intention, the accurate torque is calculated by combining human-computer interaction force, and safety is ensured by limit verification. This solves the problems of fixed impedance control parameters and poor adaptability in traditional systems. It enables the exoskeleton to dynamically output assist torque according to the human body's motion intention and real-time status, which not only meets the assistance needs in different scenarios, but also avoids equipment overload or human injury through torque limit constraints, thus balancing control accuracy and safety.

[0041] like Figure 2 As shown, another embodiment of the present invention provides an exoskeleton human-machine motion intention collaborative control device, comprising: The data preprocessing module is used to collect multiple signal phase differences through energy buttons set at human joints, obtain the original phase difference sequence, and perform data preprocessing on the original phase difference sequence; The instruction generation module is used to: calculate the cross-correlation coefficient between the target joint and the associated joint sequence based on the preprocessed phase difference sequence, and obtain the human joint motion feature set based on the cross-correlation coefficient; A motion intention prediction model is constructed based on a long short-term memory network. The motion intention prediction model is trained by multi-feature fusion through human joint motion feature set, and the motion intention prediction information and joint position prediction information are output. Based on the second-order linear impedance model, the motion intention prediction information and joint position prediction information are analyzed to generate the driving torque command of the exoskeleton joint. The instruction sending module is used to send the driving torque instruction to the exoskeleton servo motor for execution.

[0042] Preferably, the cross-correlation coefficient between the target joint and the associated joint sequence is calculated based on the preprocessed phase difference sequence, and the human joint motion feature set is obtained based on the cross-correlation coefficient, including: The joint motion velocity is obtained by performing a first-order difference calculation on the preprocessed phase difference sequence based on the joint motion velocity formula, which is: , in, Let be the joint motion velocity corresponding to the i-th sampling time. Let i be the phase difference value at the i-th sampling point in the phase difference sequence of the target joint or associated joint. This represents the phase difference value at the (i+1)th sampling point in the phase difference sequence of the target joint or associated joint. The sampling time interval; The joint motion velocity is calculated using a first-order difference based on the joint motion acceleration formula to obtain the joint motion acceleration. The joint motion acceleration formula is as follows: , in, To represent the joint motion acceleration corresponding to the i-th sampling time, Let be the joint motion velocity at the (i+1)th sampling time. The target joint θ is calculated using the synergy calculation formula and the preprocessed phase difference sequence. k With associated joint θ h The cross-relationship coefficient, and the formula for calculating the degree of collaboration is: , Where r is the cross-correlation coefficient and n is the number of data points in the phase difference sequence. Let be the phase difference value of the i-th sampling point in the target joint phase difference sequence. The mean of the target joint phase difference sequence. , The phase difference value at the i-th sampling point in the associated joint phase difference sequence reflects the reference phase level of the target joint motion. The mean of the associated joint phase difference sequence. ; Each cross-correlation coefficient is compared with a preset feature threshold. If the cross-correlation coefficient is greater than the preset feature threshold, the target joint corresponding to the cross-correlation coefficient is removed. Associated joints Features based on the removed target joints Associated joints The feature set is used to obtain the human joint motion feature set.

[0043] Example 3: Another embodiment of the present invention provides an exoskeleton human-machine motion intention collaborative control device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the exoskeleton human-machine motion intention collaborative control method as described above.

[0044] Example 4: Another embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the exoskeleton human-machine motion intention collaborative control method as described above.

[0045] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises 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.

[0046] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0047] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0048] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0049] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An exoskeleton motion intention collaborative control method, characterized in that, The method comprises the following steps: Collecting multiple signal phase differences through energy buttons arranged at human joints to obtain an original phase difference sequence, and performing data preprocessing on the original phase difference sequence; Calculating the cross-correlation coefficients of the target joint and the associated joint sequence based on the preprocessed phase difference sequence, and obtaining a human joint motion feature set based on the cross-correlation coefficients; Building a motion intention prediction model based on a long short-term memory network, performing multi-feature fusion motion intention prediction training on the motion intention prediction model through the human joint motion feature set, and outputting motion intention prediction information and joint position prediction information; Performing instruction analysis on the motion intention prediction information and the joint position prediction information based on a second-order linear impedance model to generate a driving torque instruction of an exoskeleton joint; Sending the driving torque instruction to an exoskeleton servo motor for execution.

2. The exoskeleton co-motion intention control method according to claim 1, wherein The data preprocessing on the original phase difference sequence comprises the following steps: Statistically analyzing the lengths of multiple single-segment missing lengths in the original phase difference sequence, if the length of a single-segment missing length is less than or equal to a preset sampling point number, using linear interpolation to complete the single-segment missing length, if the length of a single-segment missing length is greater than the preset sampling point number, marking the single-segment missing length as invalid, and using a sliding window mean to fill the invalid single-segment missing length; Performing filtering processing on the filled original phase difference sequence through a least mean square error adaptive filter to obtain a preprocessed phase difference sequence.

3. The exoskeleton motion intention co-control method according to claim 1, characterized in that, Calculating the cross-correlation coefficients of the target joint and the associated joint sequence based on the preprocessed phase difference sequence, and obtaining a human joint motion feature set based on the cross-correlation coefficients, which comprises the following steps: Performing first-order difference calculation on the preprocessed phase difference sequence based on a joint motion speed formula to obtain joint motion speed, the joint motion speed formula being: ， wherein, is the joint motion velocity at the i-th sampling time, is the phase difference value of the i-th sampling point in the target joint or associated joint phase difference sequence, is the phase difference value of the i+1-th sampling point in the target joint or associated joint phase difference sequence, is the sampling time interval; Performing first-order difference calculation on the joint motion speed based on a joint motion acceleration formula to obtain joint motion acceleration, the joint motion acceleration formula being: ， wherein, is the joint motion acceleration at the i-th sampling time, is the joint motion velocity at the i+1-th sampling time; The target joint θ is calculated by a synergy degree calculation formula and a preprocessed phase difference sequence k and a cross-correlation coefficient of the associated joint θ h , the synergy degree calculation formula being: ， wherein r is a cross-correlation coefficient, n is a number of data points of the phase difference sequence, is a phase difference value of an i th sampling point in the target joint phase difference sequence, is a mean value of the target joint phase difference sequence, , is a phase difference value of an i th sampling point in the associated joint phase difference sequence, is a mean value of the associated joint phase difference sequence, ; The cross-correlation coefficient is compared with a preset feature threshold value, and if the cross-correlation coefficient is greater than the preset feature threshold value, the target joint corresponding to the cross-correlation coefficient is removed The features of the associated joints Based on the feature set of the target joints after removal The features of the associated joints The human joint motion feature set is obtained.

4. The exoskeleton co-motion intention control method according to claim 1, wherein Building a motion intention prediction model based on a long short-term memory network, performing multi-feature fusion motion intention prediction training on the motion intention prediction model through the human joint motion feature set, and outputting motion intention prediction information and joint position prediction information, which comprises the following steps: Building a motion intention prediction model based on a long short-term memory network, the motion intention prediction model comprising a feature preprocessing layer, an LSTM fusion layer and a double-task output layer; The feature preprocessing layer is used to splice the joint motion speed and joint motion acceleration at consecutive sampling time points in the human joint motion feature set and the cross-correlation coefficients according to time steps to form an LSTM time sequence feature matrix with a dimension of {number of time steps x feature dimension}; The LSTM fusion layer comprises a forgetting gate, an input gate and an output gate, the forgetting gate is used to perform redundant filtering processing on the input LSTM time sequence feature matrix, the input gate is used to perform strengthening processing on the filtered LSTM time sequence feature matrix and motion intention related time sequence features, and the output gate is used to capture long short-term feature relationships in the strengthened time sequence features to obtain a time sequence feature vector; The dual-task output layer includes a motion intention classification branch and a joint position prediction branch, the motion intention classification branch is used for mapping a time sequence feature vector output by the LSTM fusion layer to a dimension matching a number of motion intention categories, obtaining a plurality of motion intention categories, and calculating a probability distribution of each motion intention category through a Softmax activation layer, and taking a category corresponding to a maximum probability as motion intention prediction information; the joint position prediction branch includes a regression layer, the regression layer is used for outputting a joint position prediction value by learning an internal relationship between a joint motion speed , a joint motion acceleration and a phase difference change amount.

5. The exoskeleton motion intention co-control method according to claim 4, characterized in that, The motion intention classification branch and the joint position prediction branch of the double-task output layer are collaboratively trained by a joint loss function, and the joint loss function is: ， wherein, is a cross-entropy loss for motion intention classification, is a mean squared error loss for joint position prediction, is a weight coefficient.

6. The exoskeleton motion intention cooperative control method according to claim 4, wherein, The motion intention prediction information and the joint position prediction information are analyzed based on a second-order linear impedance model to generate a driving torque instruction of an exoskeleton joint, including: Acquiring current actual position of exoskeleton joint and joint position prediction value to obtain position deviation ; A probability maximum value is extracted from the motion intention prediction information, and the probability maximum value is compared with a preset confidence. calling preset base impedance parameters based on the valid motion intention category, based on a correction formula and a position deviation correcting the preset base impedance parameters, the correction formula being ， ， K is the corrected stiffness, is the maximum active angle of the corresponding joint, is the preset basic stiffness, and B is the corrected damping, is the preset basic damping, is the upper limit of the safety speed under the corresponding motion intention; A second order linear impedance model is used to construct a moment calculation equation, and a driving moment is calculated based on a position deviation , a corrected stiffness K, and a corrected damping B, the moment calculation equation being: ， wherein, is the driving torque required to be output by the exoskeleton joint, is the stiffness torque corresponding to the position deviation, is the damping torque corresponding to the velocity, is the compensation torque for the human body force; The calculated maximum torque and minimum torque are compared, and if the maximum torque is taken, and if the minimum torque is taken. The drive torque command is converted into a servo motor signal and transmitted.

7. An exoskeleton motion intention cooperative control device characterized by comprising: The method comprises: A data preprocessing module is configured to collect a plurality of signal phase differences through an energy button arranged at a human joint to obtain an original phase difference sequence, and to perform data preprocessing on the original phase difference sequence. An instruction generation module is configured to: calculate a cross-correlation coefficient of a target joint and a sequence of associated joints based on the preprocessed phase difference sequence, and obtain a human joint motion feature set based on the cross-correlation coefficient. A motion intention prediction model is constructed based on a long short-term memory network, and the motion intention prediction model is trained for motion intention prediction based on the human joint motion feature set to output motion intention prediction information and joint position prediction information. The motion intention prediction information and the joint position prediction information are analyzed based on a second-order linear impedance model to generate a driving torque instruction of an exoskeleton joint. An instruction sending module is configured to send the driving torque instruction to an exoskeleton servo motor for execution.

8. The exoskeleton motion-intent co-control device according to claim 7, characterized in that, The cross-correlation coefficient of the target joint and the sequence of associated joints is calculated based on the preprocessed phase difference sequence, and the human joint motion feature set is obtained based on the cross-correlation coefficient, including: The preprocessed phase difference sequence is calculated based on a joint motion velocity formula to obtain a joint motion velocity, and the joint motion velocity formula is: ， wherein, is the joint motion velocity at the i-th sampling time, is the phase difference value of the i-th sampling point in the target joint or associated joint phase difference sequence, is the phase difference value of the i+1-th sampling point in the target joint or associated joint phase difference sequence, is the sampling time interval; The joint motion velocity is calculated based on a joint motion acceleration formula to obtain a joint motion acceleration, and the joint motion acceleration formula is: ， wherein, is the joint motion acceleration at the i-th sampling time, is the joint motion velocity at the i+1-th sampling time; The target joint θ is calculated by a synergy degree calculation formula and a preprocessed phase difference sequence k and a cross-correlation coefficient of the associated joint θ h , and the synergy degree calculation formula is: ， wherein r is a cross-correlation coefficient, n is a number of data points of the phase difference sequence, is a phase difference value of an i-th sampling point in the target joint phase difference sequence, is a mean value of the target joint phase difference sequence, , is a phase difference value of an i-th sampling point in the associated joint phase difference sequence, is a mean value of the associated joint phase difference sequence, ; The cross-correlation coefficient is compared with a preset feature threshold value, and if the cross-correlation coefficient is greater than the preset feature threshold value, the target joint corresponding to the cross-correlation coefficient is removed The features of the associated joints Based on the feature set of the target joints after removal The features of the associated joints The human joint motion feature set is obtained.

9. An exoskeleton motion intention cooperative control device characterized by comprising: The computer program is stored in the memory and executable on the processor, and when the processor executes the computer program, the exoskeleton human-machine motion intention collaborative control method of any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. When the computer program is executed by the processor, the exoskeleton human-machine motion intention collaborative control method of any one of claims 1 to 6 is implemented.

Images