A rehabilitation robot standard trajectory modeling method and a compliant interaction control method

Abstract

This invention discloses a standard trajectory modeling method and a compliant interactive control method for rehabilitation robots. The method includes: determining motion capture markers and surface electromyography (SEMG) acquisition locations based on a standardized upper limb motor function assessment task; acquiring kinematic data and multi-channel SEMG signals and preprocessing them to obtain preprocessed upper limb distal trajectory, velocity curves, and multi-channel normalized SEMG envelope mean curves; extracting effective movement segments by combining the velocity curves and SEMG envelope mean curves, extracting corresponding data segments from the preprocessed upper limb distal trajectory and performing time normalization to obtain effective movement segment data; constructing an initial standard trajectory template based on the effective movement segment data of each subject under the same task; performing piecewise polynomial fitting on the initial standard trajectory template, applying continuous constraints on position, velocity, and acceleration, as well as head-end boundary constraints, to adjacent segments to obtain a continuous standard trajectory model for use as an effective reference trajectory input for various rehabilitation robots.

Description

Technical Field

[0001] This invention relates to the field of upper limb rehabilitation training technology, specifically to a standard trajectory modeling method and a compliant interactive control method for a rehabilitation robot. Background Technology

[0002] Stroke is a leading cause of disability worldwide, often accompanied by varying degrees of upper limb motor dysfunction, severely impacting daily living abilities and quality of life. Rehabilitation training is a crucial pathway for restoring upper limb function; however, traditional rehabilitation training heavily relies on manual guidance from therapists, inherently limiting its effectiveness due to long training cycles, insufficient quantification of the process, and limited patient participation. Rehabilitation robots, offering repetitive, stable, and quantifiable training support, have become an important technological tool for upper limb functional rehabilitation in recent years.

[0003] However, in practical applications, the human-computer interaction of most current rehabilitation robots remains stiff, with low patient participation and potentially increasing the risk of secondary injuries. One of the core reasons for this bottleneck in human-computer interaction is the lack of reference motion trajectory templates that conform to the natural physiological movement patterns of the human body. Currently, many rehabilitation robots guide patients to perform tasks using simple point-to-point geometric straight lines, circles, figure-eights, or mathematically smooth curves. While these trajectories are easy to implement in engineering, they ignore the complex joint coordination patterns and velocity rhythms of the human upper limbs when performing real-life activities. When patients passively follow these unnatural, rigid trajectories, human-machine aggression is easily generated, severely weakening the smoothness of the interaction and the effectiveness of rehabilitation.

[0004] Regarding trajectory planning for rehabilitation robots, existing technologies mainly include the following categories: (1) collecting motion data from healthy individuals as reference trajectory templates and adjusting compliance using variable impedance control; (2) achieving automatic trajectory programming based on therapist-instructed actions; and (3) constructing standardized motion datasets and extracting temporal features to identify motion deviations and compensation patterns. In terms of interactive control, existing research has involved strategies such as adaptive control, impedance control, and singularity avoidance; however, the aforementioned existing technologies still have the following shortcomings:

[0005] (i) The reference trajectory lacks the characterization of the natural movement characteristics of the human body. The existing trajectories are mostly geometric straight lines, smooth curves or simple averages of data from healthy people. They are difficult to reflect the multi-joint coordination patterns and speed rhythms in upper limb functional tasks, which can easily lead to human-machine confrontation, affect the therapeutic effect and may increase the risk of injury.

[0006] (ii) The trajectory construction is disconnected from the clinical standardized assessment tools. The existing trajectory templates or motion datasets are not directly corresponding to standardized general motor function assessment scales such as the Wolf Motor Function Test (WMFT) and the Fugl-Meyer Assessment for Upper Extremity (FMA-UA). The functional tasks with clear start and end positions and operation criteria in the general assessment scales have not been directly converted into robot reference trajectories. The integration of assessment and training lacks technical support.

[0007] (iii) The modeling process does not integrate multimodal physiological signals. Existing methods rely on single kinematic data or manual teaching, and do not incorporate multimodal signals such as surface electromyography to assist in the identification of movement stages and temporal alignment, resulting in the trajectory being difficult to adapt to the movement differences of different patients.

[0008] (iv) The trajectory representation lacks a continuous parameterized form. Existing schemes mostly store it as discrete sampling points, which requires interpolation when calling it. This increases the amount of computation and makes it difficult to guarantee the high-order smoothness of velocity and acceleration, which is not conducive to compliant tracking control.

[0009] In summary, how to generate a reference trajectory that closely resembles the natural movement characteristics of the human body, connects to standardized tasks in a standardized general motor function assessment scale, integrates multimodal physiological information, and has continuous parameterized expression, while ensuring stable robot tracking, is a technical problem that urgently needs to be solved in the field of rehabilitation robots. Summary of the Invention

[0010] The purpose of this invention is to overcome the technical problems in the prior art where the reference trajectory of rehabilitation robots lacks task-oriented motion characteristics, is disconnected from standardized functional assessment tasks, and lacks continuous parameterized expression. This invention provides a standard trajectory modeling method and a compliant interactive control method for rehabilitation robots, which are used to generate standard reference trajectories that can be called by various rehabilitation robots and to perform compliant interactive control.

[0011] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution:

[0012] In a first aspect, the present invention provides a method for modeling the standard trajectory of a rehabilitation robot, comprising:

[0013] Based on the standardized upper limb motor function assessment task set, key markers were identified and motion capture markers were placed accordingly, as well as key muscle groups were identified and surface electromyography sensors were placed accordingly.

[0014] Perform each task in the standardized upper limb motor function assessment task set, and collect kinematic data and multi-channel surface electromyography signals based on the motion capture markers and surface electromyography sensors;

[0015] The kinematic data and multi-channel surface electromyography (EMG) signals are preprocessed to obtain the preprocessed upper limb distal trajectory, velocity curves, and multi-channel normalized surface EMG envelope mean curves.

[0016] The effective action segments are extracted based on the velocity curve and the mean curve of the multi-channel normalized surface electromyography envelope. The corresponding data segments in the preprocessed upper limb end trajectory are then extracted based on the effective action segments and subjected to time normalization to obtain the effective action segment data.

[0017] The effective action segment data of each subject belonging to the same task are averaged point by point at each normalized time point to construct the initial standard trajectory template for the task.

[0018] The initial standard trajectory template for each task is divided into multiple segments on the normalized time axis. A piecewise polynomial fitting model is established for each segment trajectory, and position continuity constraints, velocity continuity constraints, acceleration continuity constraints, and head and tail boundary constraints are applied at the segment connection points. The least squares method with equality constraints is used to solve the fitting coefficients of each piecewise polynomial fitting model to obtain the continuous standard trajectory model for each task.

[0019] Optionally, the tasks in the standardized upper limb motor function assessment task set are selected from tasks in the Wolf motor function test, the Fugl-Meyer upper limb motor function rating scale, or other standardized upper limb motor function rating scales;

[0020] The key markers include trunk key markers for constructing the trunk local coordinate system and at least two distal upper limb markers for determining the distal end of the upper limb;

[0021] The key muscle groups are determined by the joint movements and muscle functions involved in performing the task;

[0022] The kinematic data includes the spatial positions of the motion capture markers at each sampling time.

[0023] Optionally, the preprocessing of the kinematic data and multi-channel surface electromyography (EMG) signals to obtain the preprocessed upper limb distal trajectory, velocity curve, and multi-channel normalized EMG envelope mean curve includes:

[0024] Based on the spatial positions of key trunk markers in the kinematic data at each sampling time, a local orthogonal coordinate system is constructed that is dynamically updated with trunk posture.

[0025] Based on the spatial position of the distal upper limb markers in the kinematic data at each sampling time, the distal upper limb point is determined and the distal upper limb trajectory is generated.

[0026] The trajectory of the upper limb end is repaired and filtered, and then transformed from the global coordinate system to the local orthogonal coordinate system to obtain the preprocessed trajectory of the upper limb end.

[0027] The velocity curve is extracted based on the preprocessed upper limb distal trajectory.

[0028] The multi-channel surface electromyography (EMG) signals are sequentially subjected to DC removal, bandpass filtering, notch filtering, full-wave rectification, and sliding root mean square envelope extraction to obtain the EMG envelope signal of each channel.

[0029] The electromyography envelope signals of each channel are interpolated to the kinematic time axis and robustly normalized to obtain multi-channel normalized surface electromyography envelope signals.

[0030] At the same sampling time, the multi-channel normalized surface electromyography envelope signal is arithmetically averaged along the channel dimension to obtain the mean curve of the multi-channel normalized surface electromyography envelope.

[0031] Optionally, the step of extracting the effective movement segment based on the velocity curve and the multi-channel normalized surface electromyography envelope mean curve, and then extracting the corresponding data segment from the preprocessed upper limb distal trajectory based on the effective movement segment and performing time normalization processing to obtain the effective movement segment data, including:

[0032] The velocity curve and the mean value curve of the multi-channel normalized surface electromyography envelope are smoothed respectively, the peak value of each curve is calculated, and the velocity threshold and electromyography threshold are set according to a preset ratio.

[0033] Frame by frame, determine whether each sampling point simultaneously satisfies both the velocity greater than the velocity threshold and the electromyography (EMG) greater than the EMG threshold. Mark the sampling points that simultaneously satisfy the conditions as active and the sampling points that do not satisfy the conditions as inactive, thus obtaining the binary joint activation mask.

[0034] The sequence of sampling points marked as active and arranged consecutively in the binarized joint activation mask is used as a candidate segment to obtain a candidate segment set.

[0035] Perform a dual-channel joint threshold determination based on the candidate segment set:

[0036] If the duration of the longest candidate segment in the candidate segment set is greater than or equal to the preset minimum action duration threshold, the longest candidate segment is taken as the valid action segment.

[0037] If the duration of the longest candidate segment is less than the preset minimum action duration threshold, the segment is discarded, and the longest candidate segment in the remaining candidate segments is traversed for dual-channel joint threshold determination until a candidate segment that meets the duration condition is selected as the valid action segment.

[0038] If the duration of all candidate segments is less than the minimum action duration threshold, then switch to single-channel speed threshold determination mode, and take the longest candidate segment that satisfies both speed greater than speed threshold and duration greater than the minimum action duration threshold as the valid action segment.

[0039] Based on the start and end times of the effective segment of the action on the original time axis, the data segment corresponding to the time interval in the preprocessed upper limb end trajectory is extracted to obtain the trajectory data segment.

[0040] The trajectory data segments are time-normalized and mapped to a preset time axis of uniform length through interpolation to obtain effective action segment data.

[0041] Optionally, the calculation formula for the initial standard trajectory template is as follows:

[0042] ,

[0043] in, This indicates the total number of sampling points in the trajectory data segment. Indicates the first Normalized time for each sampling point Indicates the task index. Indicates task The number of subjects, Indicates task The In the effective action segment data of the subjects Spatial location at a given time Indicates task The initial standard trajectory template in Spatial location at any given moment.

[0044] Optionally, it also includes evaluating the in-sample representativeness of the initial standard trajectory template based on the spatial deviation of each subject's effective action segment data relative to the initial standard trajectory template:

[0045] For each task, the Euclidean distance between the effective action segment data of all subjects in that task and the initial standard trajectory template at each normalized time step is calculated, and an evaluation index is calculated based on the Euclidean distance; wherein, the evaluation index includes any one of the root mean square deviation, 95th percentile deviation and maximum deviation;

[0046] If the evaluation index is less than the preset threshold, the initial standard trajectory template for the task is determined to be usable for generating the continuous trajectory model.

[0047] If the evaluation index is greater than or equal to the preset threshold, it is determined that the initial standard trajectory template for the task has large individual differences within the sample, and the initial standard trajectory template for the task is unusable. It is necessary to reacquire the kinematic data for the task and generate the initial standard trajectory template for the task.

[0048] Optionally, it also includes evaluating the in-sample representativeness of the initial standard trajectory template based on the spatial deviation of each subject's effective action segment data relative to the initial standard trajectory template:

[0049] For each task, calculate the Euclidean distance between the effective action segment data of all subjects in that task and the initial standard trajectory template at each normalized time step, and calculate the root mean square deviation, 95th percentile deviation and maximum deviation based on the Euclidean distance;

[0050] The root mean square deviation, 95th percentile deviation, and maximum deviation are weighted and summed based on preset weights to obtain the fusion deviation;

[0051] If the fusion deviation is less than a preset threshold, the initial standard trajectory template for the task is determined to be usable for generating the continuous trajectory model.

[0052] If the fusion deviation is greater than or equal to the preset threshold, it is determined that the initial standard trajectory template for the task has large individual differences within the sample, and the initial standard trajectory template for the task is unusable. It is necessary to reacquire the kinematic data for the task and generate the initial standard trajectory template for the task.

[0053] Optionally, the root mean square deviation is obtained by the following formula:

[0054] ,

[0055] in, Indicates task The The root mean square deviation of each subject from the initial standard trajectory template This represents the total number of sampling points for the valid action segment data. Indicates the first Normalized time for each sampling point Indicates task The Each subject and the initial standard trajectory template Euclidean distance at time;

[0056] The 95th percentile deviation is obtained by the following formula:

[0057] ,

[0058] in, Indicates task The The 95th percentile deviation of each subject from the initial standard trajectory template. This represents the 95th percentile deviation function;

[0059] The maximum deviation is obtained by the following formula:

[0060] ,

[0061] in, Indicates task The The maximum deviation of a subject from the initial standard trajectory template, where max represents the maximum value.

[0062] Optionally, the expression for the piecewise polynomial fitting model is as follows:

[0063] ,

[0064] in, Indicates the segment number. express of Power of 1 Indicates the first Segmented time variables, As a preset value, Indicates the first Segmentation Spatial location at a given time Indicates the first Piecewise fitting coefficients;

[0065] The positional continuity constraint is achieved through the following formula:

[0066] ,

[0067] in, Indicates the first Segmentation Spatial location at any given time;

[0068] The velocity continuity constraint is achieved through the following formula:

[0069] ,

[0070] in, express right The first derivative, express right The first derivative;

[0071] The acceleration continuity constraint is achieved through the following formula:

[0072] ,

[0073] in, express right The second derivative, express right The second derivative;

[0074] The initial and final boundary constraints are implemented using the following formula:

[0075] ,

[0076] ,

[0077] in, This indicates the start time of the initial standard trajectory template. Indicates the termination time of the initial standard trajectory template. Indicates the starting segment number. Indicates the end segment number. Indicates task The initial standard trajectory template at the start time Spatial location, Indicates task The starting segment At the start time Spatial location, Indicates task The initial standard trajectory template at the termination time Spatial location at a given time Indicates task Termination segment At the end time Spatial location at any given moment.

[0078] Secondly, the present invention provides a compliant interactive control method for a rehabilitation robot, comprising:

[0079] The continuous standard trajectory model of each task is used as the reference trajectory input of the end effector of the corresponding type of rehabilitation robot; wherein, the continuous standard trajectory model of each task is obtained by any of the rehabilitation robot standard trajectory modeling methods in the first aspect;

[0080] Real-time acquisition of force information between various rehabilitation robots and the patient's upper limbs;

[0081] Based on the interaction force information, the reference trajectory is corrected by admittance control to generate a compliant corrected trajectory;

[0082] The compliance correction trajectory is input as a motion control command to the servo driver of various rehabilitation robots, driving the rehabilitation robot to move according to the compliance correction trajectory, so as to guide the patient's upper limb to achieve active training of human-machine compliance control.

[0083] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0084] 1. The standard trajectory modeling method for rehabilitation robots proposed in this invention uses specific functional tasks in standardized upper limb motor function assessment scales such as WMFT and FMA-UA as the direct basis for trajectory construction. This establishes a correspondence between the reference trajectory and the clinical assessment task at the task level, forming a standard trajectory template that conforms to the natural physiological movement law of the human upper limb. This provides a technical foundation for the integrated connection between clinical assessment and rehabilitation training. The method employs a combined criterion of velocity and electromyography to identify action segments, improving the accuracy of determining the start and end times of movements. Through cross-subject averaging and piecewise polynomial parameterization, the discrete trajectory is transformed into a continuously adjustable analytical model, which helps improve the smoothness of trajectory tracking and patient wearing comfort.

[0085] 2. The compliant interactive control method for rehabilitation robots proposed in this invention can, on the one hand, enable the rehabilitation robot to guide the patient's upper limbs to achieve trajectory tracking training that conforms to the natural physiological movement law of the human body; on the other hand, it can correct the end trajectory in real time through admittance control, drive the robot to move according to the compliant correction trajectory, realize compliant interaction between the robot and the patient, enhance the patient's active participation, and facilitate the realization of task-oriented active training by the rehabilitation robot guiding the patient's upper limbs. Attached Figure Description

[0086] Figure 1 This is a flowchart of a standard trajectory modeling method for rehabilitation robots provided according to an embodiment of the present invention;

[0087] Figure 2 This is a classification diagram of the 17 standard movement assessment test tasks of the Wolf motor function test provided in an embodiment of the present invention;

[0088] Figure 3 This is a schematic diagram of a data acquisition experimental scenario provided by an embodiment of the present invention;

[0089] Figure 4 This is a schematic diagram illustrating a scenario where a hand is placed face down on a table, based on a human-computer compliant interaction simulation model built on the MuJoCo simulation platform according to an embodiment of the present invention. Detailed Implementation

[0090] 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 and specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations thereof. In the absence of conflict, the embodiments and technical features in the embodiments can be combined with each other.

[0091] It should be noted that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0092] Example 1

[0093] This invention discloses a method for modeling the standard trajectory of a rehabilitation robot based on an upper limb motor function assessment task, with reference to... Figure 1 As shown, the specific steps include the following:

[0094] S1. Based on the standardized upper limb motor function assessment task set, key markers are identified and motion capture markers are placed accordingly, and key muscle blocks are identified and surface electromyography sensors are placed accordingly.

[0095] S2, perform each task in the standardized upper limb motor function assessment task set, and collect kinematic data and multi-channel surface electromyography signals based on the motion capture markers and surface electromyography sensors;

[0096] S3, preprocess the kinematic data and multi-channel surface electromyography signals to obtain the preprocessed upper limb end trajectory, velocity curve and multi-channel normalized surface electromyography envelope mean curve.

[0097] S4. Extract the effective segment of the action based on the velocity curve and the mean curve of the multi-channel normalized surface electromyography envelope. Extract the corresponding data segment from the preprocessed upper limb end trajectory based on the effective segment of the action and perform time normalization processing to obtain the effective action segment data.

[0098] S5, average the effective action segment data of each subject belonging to the same task at each normalized time point to construct the initial standard trajectory template for the task.

[0099] S6. Divide the normalized time axis of the initial standard trajectory template of each task into multiple segments, establish a piecewise polynomial fitting model for each segment trajectory, and apply position continuity constraints, velocity continuity constraints, acceleration continuity constraints, and head and tail boundary constraints at the segment connection points. Use the least squares method with equality constraints to solve the fitting coefficients of each piecewise polynomial fitting model to obtain the continuous standard trajectory model of each task.

[0100] This embodiment selects the internationally recognized Wolf Motor Function Test (WMFT) as the research object; refer to Figure 2 As shown, based on movement characteristics and functional focus, the 17 standard movement assessment test tasks of WMFT are summarized into three categories: the first category is shoulder-elbow compound movements, including proximal dominant movements such as forearm placement, shoulder abduction, shoulder flexion, elbow extension, and elbow flexion; the second category is fine hand manipulation, including tasks such as picking up a can, picking up a pencil, and folding a towel; the third category is comprehensive movement testing, including grip strength testing, weight-bearing strength tasks, and lifting a basket. This embodiment focuses on the eight tasks in the first category of shoulder-elbow compound movements.

[0101] In step S1, for the eight WMFT upper limb motor function assessment tasks selected in this embodiment, key markers and key muscle blocks related to upper limb movement are determined based on the movement forms of the standardized upper limb motor function assessment tasks and the joint movements and muscle functions involved. The key markers include trunk key markers used to construct the trunk local coordinate system and at least two upper limb distal markers used to determine the upper limb distal points.

[0102] In this embodiment, four key markers are selected: the base of the posterior neck, the clavicle, the sternum, and the symmetrical point on the back of the sternum. These are numbered Marker1 to Marker4 and used to construct a dynamic local coordinate system for the trunk. The geometric midpoints of the markers on the inner and outer sides of the right hand and the inner and outer sides of the right wrist are used to construct key points for the hand and wrist, serving as distal markers for the upper limbs. These are numbered Marker5 to Marker8. Based on the joint movements and main muscle groups involved in each task, the anterior deltoid, middle deltoid, biceps brachii, triceps brachii, brachioradialis, extensor carpi radialis, extensor carpi ulnaris, and flexor carpi radialis are identified as muscles for surface electromyography (SEMG) acquisition, and the corresponding sensor placement positions are determined.

[0103] In step S2, this embodiment uses, as follows Figure 3 The data acquisition experiment shown in the figure is conducted according to the positions of motion capture markers and surface electromyography (SEMG) sensors. The three-dimensional spatial coordinates of key markers at each sampling time during the subject's performance of various tasks are obtained, as well as the 8-channel SEMG signals corresponding to the 8 key muscles.

[0104] In this embodiment, a total of 20 healthy subjects were recruited, including 10 males and 10 females, none of whom had a history of neurological diseases or upper limb musculoskeletal injuries. Each subject completed the 8 tasks in a predetermined order, and each task was collected repeatedly, with 3 to 5 valid collections per task.

[0105] In step S3, the kinematic data and multi-channel surface electromyography (EMG) signals are preprocessed to obtain the preprocessed upper limb distal trajectory, velocity curve, and multi-channel normalized surface EMG envelope mean curve, including:

[0106] S3.1, Based on the spatial positions of the key trunk markers Marker1 to Marker4 at each sampling time, construct a local orthogonal coordinate system that is dynamically updated with the trunk posture;

[0107] S3.2, Based on the spatial position of the distal marker point of the upper limb at each sampling time, determine the distal point of the upper limb and generate the trajectory of the distal limb;

[0108] S3.3, perform missing data repair on the upper limb distal trajectory, including: repairing the missing coordinates in the upper limb distal trajectory using segmented cubic Hermite interpolation, and setting the maximum number of consecutive missing frames (10 frames) as the quality control threshold; when the number of consecutive missing frames exceeds 10 frames, the corresponding trial is determined to be invalid; calculate the initial velocity based on the displacement between adjacent sampling frames, and use the sum of the median of the initial velocity and 8 times the median absolute deviation as the flying point determination threshold, set abnormal frames exceeding the flying point determination threshold to null values, and then perform segmented cubic Hermite interpolation again; filter the missing data repaired upper limb distal trajectory using a fourth-order Butterworth zero-phase low-pass filter with a cutoff frequency of 6 Hz, and transform the filtered upper limb distal trajectory from the global coordinate system to the local orthogonal coordinate system to obtain the preprocessed upper limb distal trajectory;

[0109] S3.4, differentiate the preprocessed upper limb end trajectory with respect to time, and calculate the magnitude of the triaxial velocity vector to obtain the velocity curve;

[0110] S3.5, the surface electromyography (EMG) signals of the eight channels corresponding to the anterior deltoid, middle deltoid, biceps brachii, triceps brachii, brachioradialis, extensor carpi radialis, extensor carpi ulnaris, and flexor carpi radialis are sequentially processed by DC removal, 20-450Hz fourth-order Butterworth zero-phase bandpass filtering, 50Hz notch filtering, and full-wave rectification, and the surface EMG envelope signal of each channel is extracted using a 50ms sliding root mean square window;

[0111] S3.6, linearly interpolate the surface electromyography (EMG) envelope signals of each channel to a 100Hz kinematic time axis, and perform robust normalization using the 95th percentile of the corresponding channel's EMG envelope signal as the normalization scale; when the 95th percentile is less than the machine accuracy, set the normalization scale of the corresponding channel to 1; at each sampling time, take the arithmetic mean of the 8-channel normalized EMG envelope signals to generate the 8-channel normalized EMG envelope mean curve.

[0112] In step S4, the effective movement segment is extracted based on the velocity curve and the mean curve of the multi-channel normalized surface electromyography envelope. The corresponding data segment in the preprocessed upper limb distal trajectory is then extracted based on the effective movement segment and subjected to time normalization to obtain the effective movement segment data, including:

[0113] S4.1, the velocity curve and the 8-channel normalized surface electromyography envelope mean curve are smoothed using a 0.10s moving average window respectively; 8% of the peak value of the smoothed velocity curve is set as the velocity threshold, and 15% of the peak value of the smoothed 8-channel normalized surface electromyography envelope mean curve is set as the electromyography threshold.

[0114] S4.2, determine frame by frame whether each sampling point simultaneously satisfies the conditions that the velocity is greater than the velocity threshold and the mean value of the 8-channel normalized surface electromyography envelope is greater than the electromyography threshold; mark the sampling points that simultaneously satisfy the two conditions as active, and mark the sampling points that do not simultaneously satisfy the two conditions as inactive, to obtain the binary joint activation mask;

[0115] S4.3, the sequence of sampling points marked as active and arranged continuously in the binarized joint activation mask is taken as a candidate segment to obtain a candidate segment set;

[0116] S4.4, Perform dual-channel joint threshold determination based on the candidate segment set:

[0117] S4.4.1 If the duration of the longest candidate segment in the candidate segment set is greater than or equal to the preset minimum action duration threshold, the longest candidate segment is taken as the valid action segment.

[0118] S4.4.2 If the duration of the longest candidate segment is less than the preset minimum action duration threshold, then the segment is discarded, and the longest candidate segment in the remaining candidate segments is traversed to perform dual-channel joint threshold determination until a candidate segment that meets the duration condition is selected as the valid action segment.

[0119] S4.4.3 If the duration of all candidate segments is less than the minimum action duration threshold, then switch to single-channel speed threshold determination mode, and take the longest candidate segment that satisfies the condition that the speed is greater than the speed threshold and the duration is greater than the minimum action duration threshold as the valid action segment.

[0120] S4.5, extend the start and end times of the main action segment outward by 0.08 s, and limit the extended start and end times within the original record range of the corresponding trial to obtain the effective action segment;

[0121] S4.6, Based on the start and end times of the effective action segment on the original time axis, extract the data segment of the corresponding time interval in the preprocessed upper limb end trajectory to obtain the trajectory data segment;

[0122] S4.7, the original time of the trajectory data segment is mapped to a relative time range of 0 to 1, and then mapped to a preset uniform length time axis by using segmented cubic Hermite interpolation to obtain effective action segment data; uniform length time axis u = linspace(0, 1, N), where N = 200.

[0123] In step S5, the effective action segment data of 20 subjects under the same task are directly averaged point by point at each normalization time; after traversing N=200 sampling points, an initial standard trajectory template for the task is formed, and the cross-subject standard deviation of each coordinate direction is calculated. The specific calculation formula is as follows:

[0124] ,

[0125] in, This indicates the total number of sampling points in the trajectory data segment. Indicates the first Normalized time for each sampling point Indicates the task index. Indicates task The number of subjects, Indicates task The In the effective action segment data of the subjects Spatial location at a given time Indicates task The initial standard trajectory template in Spatial location at any given moment.

[0126] To evaluate the in-sample representativeness of the constructed initial standard trajectory template, this embodiment calculates the Euclidean distance between each subject's trajectory data segment and the initial standard trajectory template at each normalized time step for each task, and further calculates the root mean square deviation, 95th percentile deviation, and maximum deviation. The statistical results of these indicators across subjects are used to evaluate the template's generalization of the common movement trends of the samples, and to explain the template's applicability. This evaluation is not intended to simply attribute individual differences to data collection issues.

[0127] The root mean square deviation is obtained by the following formula:

[0128] ,

[0129] in, Indicates task The The root mean square deviation of each subject from the initial standard trajectory template This represents the total number of sampling points for the valid action segment data. Indicates the first Normalized time for each sampling point Indicates task The Each subject and the initial standard trajectory template The Euclidean distance at time.

[0130] The 95th percentile deviation is obtained using the following formula:

[0131] ,

[0132] in, Indicates task The The 95th percentile deviation of each subject from the initial standard trajectory template. This represents the 95th percentile deviation function.

[0133] The maximum deviation is obtained by the following formula:

[0134] ,

[0135] in, Indicates task The The maximum deviation of each subject from the initial standard trajectory template This indicates taking the maximum value.

[0136] In this embodiment, the root mean square deviation is used as the main evaluation index, and the 95th percentile deviation and maximum deviation are used as auxiliary evaluation indexes. The statistical distribution of the three indicators among the subjects is combined to comprehensively evaluate the in-sample representativeness of the initial standard trajectory template.

[0137] Overall, the standard trajectory templates for different movements showed significantly different levels of representativeness within the sample. The standard trajectory templates constructed in this embodiment were able to extract the main movement patterns at the sample level well for most movements, but their representativeness was not completely consistent across all movements. The cross-subject template construction method used in this embodiment aims to obtain a group template that reflects the common movement trends of healthy individuals, rather than to establish personalized trajectories for specific subjects.

[0138] In step S6, the expression for the piecewise polynomial fitting model is as follows:

[0139] ,

[0140] in, Indicates the segment number. express of Power of 1 Indicates the first Segmented time variables, As a preset value, Indicates the first Segmentation Spatial location at a given time Indicates the first Piecewise fitting coefficients;

[0141] The positional continuity constraint is achieved through the following formula:

[0142] ,

[0143] in, Indicates the first Segmentation Spatial location at any given time;

[0144] The velocity continuity constraint is achieved through the following formula:

[0145] ,

[0146] in, express right The first derivative, express right The first derivative;

[0147] The acceleration continuity constraint is achieved through the following formula:

[0148] ,

[0149] in, express right The second derivative, express right The second derivative;

[0150] The initial and final boundary constraints are implemented using the following formula:

[0151] ,

[0152] ,

[0153] in, This indicates the start time of the initial standard trajectory template. Indicates the termination time of the initial standard trajectory template. Indicates the starting segment number. Indicates the end segment number. Indicates task The initial standard trajectory template at the start time Spatial location, Indicates task The starting segment At the start time Spatial location, Indicates task The initial standard trajectory template at the termination time Spatial location at a given time Indicates task Termination segment At the end time Spatial location at any given moment.

[0154] Example 2

[0155] Based on the same inventive concept as Embodiment 1, this embodiment of the invention discloses a compliant interactive control method for a rehabilitation robot, comprising:

[0156] The continuous standard trajectory models of various standardized upper limb motor function assessment tasks are used as reference trajectory inputs for the end effector of the corresponding type of rehabilitation robot; wherein, the continuous standard trajectory models of each task are obtained through the method of Example 1;

[0157] Real-time acquisition of force information between various rehabilitation robots and the patient's upper limbs;

[0158] Based on the interaction force information, the reference trajectory is corrected by admittance control to generate a compliant corrected trajectory;

[0159] The compliance correction trajectory is input as a motion control command to the servo driver of various rehabilitation robots, driving the rehabilitation robot to move according to the compliance correction trajectory, so as to guide the patient's upper limb to achieve active training of human-machine compliance control.

[0160] This embodiment establishes an end-effector traction scenario for a rehabilitation robot in the MuJoCo simulation platform, constructing a human-machine compliant interactive control model that includes a virtual force sensor, an admittance correction loop, and an equivalent end-effector servo loop; specifically, it includes the following steps:

[0161] refer to Figure 4 As shown, taking the typical task of placing the hand face up on the table in WMFT as an example, a simulation scene including the table, chair, seated human body, right upper limb and robotic arm is built in MuJoCo; the right upper limb adopts a rigid body geometry model, which consists of the upper arm, forearm and hand; the robotic arm adopts a simplified serial structure, and the end effector establishes an equivalent connection with the right wrist through the wrist rest; the continuous standard trajectory model is mapped to the wrist rest traction point as the end effector reference input.

[0162] The end effector of the robotic arm is simplified into a three-dimensional motion unit with mass and damping characteristics, and its dynamic equation is as follows:

[0163] ,

[0164] in, Indicates spatial location, This represents velocity, which is the first derivative of spatial position with respect to time. This represents acceleration, specifically the second derivative of spatial position with respect to time. To normalize equivalent quality, To normalize the equivalent damping, For external disturbance force, To control the input, it can be represented as:

[0165] ,

[0166] in, For position feedback gain, For speed feedback gain, For the amplitude limiting function, Indicates the reference acceleration. Indicates reference speed. Indicates positional error. The speed error is represented by the control input limit of [-300, 300] in each coordinate direction. The continuous standard trajectory model obtained by the standard trajectory modeling method of rehabilitation robot in Example 1 is input into the model in the form of time series, the simulation is run and the output trajectory is recorded.

[0167] A three-dimensional virtual interactive force is constructed at the wrist support traction point during the task of placing the hand face down on the table: a sinusoidal perturbation with an amplitude of 4.0N is applied in the X direction within the normalized execution time of 0.30~0.72, a pulse perturbation with an amplitude of -5.2N is applied in the Y direction within the time of 0.46~0.58, and a sinusoidal perturbation with an amplitude of 1.2N is applied in the Z direction within the time of 0.38~0.64; zero-point compensation is performed by superimposing zero-point bias and Gaussian noise on the original force signal, and a first-order low-pass filter with a time constant of 0.060s is applied after zero-point compensation, and amplitude is limited in each direction.

[0168] The filtered interaction force is processed by the admittance controller. Converted to admittance displacement correction :

[0169] ,

[0170] in, This represents the admittance equivalent quality. This represents the admittance equivalent damping. Indicates the admittance equivalent stiffness. This represents the scaled filter interaction force. , Represents the scaling factor, admittance acceleration Admittance velocity and admittance displacement All are limited to control the admittance displacement. Superimposed on the original reference trajectory Obtain the command trajectory .

[0171] With the command trajectory As input, the acceleration command is constructed using position error feedback and velocity damping term: ,in, This represents the desired acceleration command at the current moment. This indicates the position tracking error at the current moment. This represents the actual velocity at the end of the previous time step. , This indicates the command trajectory at the current moment. This indicates the position of the last output at the previous time step.

[0172] 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.

[0173] 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.

[0174] 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.

[0175] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0176] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for modeling the standard trajectory of a rehabilitation robot, characterized in that, include: Based on the standardized upper limb motor function assessment task set, key markers were identified and motion capture markers were placed accordingly, as well as key muscle groups were identified and surface electromyography sensors were placed accordingly. Perform each task in the standardized upper limb motor function assessment task set, and collect kinematic data and multi-channel surface electromyography signals based on the motion capture markers and surface electromyography sensors; The kinematic data and multi-channel surface electromyography (EMG) signals are preprocessed to obtain the preprocessed upper limb distal trajectory, velocity curves, and multi-channel normalized surface EMG envelope mean curves. The effective action segments are extracted based on the velocity curve and the mean curve of the multi-channel normalized surface electromyography envelope. The corresponding data segments in the preprocessed upper limb end trajectory are then extracted based on the effective action segments and subjected to time normalization to obtain the effective action segment data. The effective action segment data of each subject belonging to the same task are averaged point by point at each normalized time point to construct the initial standard trajectory template for the task. The initial standard trajectory template for each task is divided into multiple segments on the normalized time axis. A piecewise polynomial fitting model is established for each segment trajectory, and position continuity constraints, velocity continuity constraints, acceleration continuity constraints, and head and tail boundary constraints are applied at the segment connection points. The least squares method with equality constraints is used to solve the fitting coefficients of each piecewise polynomial fitting model to obtain the continuous standard trajectory model for each task.

2. The standard trajectory modeling method for rehabilitation robots according to claim 1, characterized in that, The tasks in the standardized upper limb motor function assessment task set are selected from tasks in the Wolf motor function test, the Fugl-Meyer upper limb motor function rating scale, or other standardized upper limb motor function rating scales; The key markers include trunk key markers for constructing the trunk local coordinate system and at least two distal upper limb markers for determining the distal end of the upper limb; The key muscle groups are determined by the joint movements and muscle functions involved in performing the task; The kinematic data includes the spatial positions of the motion capture markers at each sampling time.

3. The standard trajectory modeling method for rehabilitation robots according to claim 1, characterized in that, The preprocessing of the kinematic data and multi-channel surface electromyography (EMG) signals to obtain the preprocessed upper limb distal trajectory, velocity curve, and multi-channel normalized EMG envelope mean curve includes: Based on the spatial positions of key trunk markers in the kinematic data at each sampling time, a local orthogonal coordinate system is constructed that is dynamically updated with trunk posture. Based on the spatial position of the distal upper limb markers in the kinematic data at each sampling time, the distal upper limb point is determined and the distal upper limb trajectory is generated. The trajectory of the upper limb end is repaired and filtered, and then transformed from the global coordinate system to the local orthogonal coordinate system to obtain the preprocessed trajectory of the upper limb end. The velocity curve is extracted based on the preprocessed upper limb distal trajectory. The multi-channel surface electromyography (EMG) signals are sequentially subjected to DC removal, bandpass filtering, notch filtering, full-wave rectification, and sliding root mean square envelope extraction to obtain the EMG envelope signal of each channel. The electromyography envelope signals of each channel are interpolated to the kinematic time axis and robustly normalized to obtain multi-channel normalized surface electromyography envelope signals. At the same sampling time, the multi-channel normalized surface electromyography envelope signal is arithmetically averaged along the channel dimension to obtain the mean curve of the multi-channel normalized surface electromyography envelope.

4. The standard trajectory modeling method for rehabilitation robots according to claim 1, characterized in that, The effective movement segment is extracted based on the velocity curve and the mean curve of the multi-channel normalized surface electromyography envelope. The corresponding data segment in the preprocessed upper limb distal trajectory is then extracted from the effective movement segment and subjected to time normalization to obtain the effective movement segment data, including: The velocity curve and the mean value curve of the multi-channel normalized surface electromyography envelope are smoothed respectively, the peak value of each curve is calculated, and the velocity threshold and electromyography threshold are set according to a preset ratio. Frame by frame, determine whether each sampling point simultaneously satisfies both the velocity greater than the velocity threshold and the electromyography (EMG) greater than the EMG threshold. Mark the sampling points that simultaneously satisfy the conditions as active and the sampling points that do not satisfy the conditions as inactive, thus obtaining the binary joint activation mask. The sequence of sampling points marked as active and arranged consecutively in the binarized joint activation mask is used as a candidate segment to obtain a candidate segment set. Perform a dual-channel joint threshold determination based on the candidate segment set: If the duration of the longest candidate segment in the candidate segment set is greater than or equal to the preset minimum action duration threshold, the longest candidate segment is taken as the valid action segment. If the duration of the longest candidate segment is less than the preset minimum action duration threshold, the segment is discarded, and the longest candidate segment in the remaining candidate segments is traversed for dual-channel joint threshold determination until a candidate segment that meets the duration condition is selected as the valid action segment. If the duration of all candidate segments is less than the minimum action duration threshold, then switch to single-channel speed threshold determination mode, and take the longest candidate segment that satisfies both speed greater than speed threshold and duration greater than the minimum action duration threshold as the valid action segment. Based on the start and end times of the effective segment of the action on the original time axis, the data segment corresponding to the time interval in the preprocessed upper limb end trajectory is extracted to obtain the trajectory data segment. The trajectory data segments are time-normalized and mapped to a preset time axis of uniform length through interpolation to obtain effective action segment data.

5. The standard trajectory modeling method for rehabilitation robots according to claim 1, characterized in that, The calculation formula for the initial standard trajectory template is as follows: ， in, This indicates the total number of sampling points in the trajectory data segment. Indicates the first Normalized time for each sampling point Indicates the task index. Indicates task The number of subjects, Indicates task The In the effective action segment data of the subjects Spatial location at a given time Indicates task The initial standard trajectory template in Spatial location at any given moment.

6. The standard trajectory modeling method for rehabilitation robots according to claim 1, characterized in that, It also includes evaluating the in-sample representativeness of the initial standard trajectory template based on the spatial deviation of each subject's effective action segment data relative to the initial standard trajectory template: For each task, the Euclidean distance between the effective action segment data of all subjects in that task and the initial standard trajectory template at each normalized time step is calculated, and an evaluation index is calculated based on the Euclidean distance; wherein, the evaluation index includes any one of the root mean square deviation, 95th percentile deviation and maximum deviation; If the evaluation index is less than the preset threshold, the initial standard trajectory template for the task is determined to be usable for generating the continuous trajectory model. If the evaluation index is greater than or equal to the preset threshold, it is determined that the initial standard trajectory template for the task has large individual differences within the sample, and the initial standard trajectory template for the task is unusable. It is necessary to reacquire the kinematic data for the task and generate the initial standard trajectory template for the task.

7. The standard trajectory modeling method for rehabilitation robots according to claim 1, characterized in that, It also includes evaluating the in-sample representativeness of the initial standard trajectory template based on the spatial deviation of each subject's effective action segment data relative to the initial standard trajectory template: For each task, calculate the Euclidean distance between the effective action segment data of all subjects in that task and the initial standard trajectory template at each normalized time step, and calculate the root mean square deviation, 95th percentile deviation and maximum deviation based on the Euclidean distance; The root mean square deviation, 95th percentile deviation, and maximum deviation are weighted and summed based on preset weights to obtain the fusion deviation; If the fusion deviation is less than a preset threshold, the initial standard trajectory template for the task is determined to be usable for generating the continuous trajectory model. If the fusion deviation is greater than or equal to the preset threshold, it is determined that the initial standard trajectory template for the task has large individual differences within the sample, and the initial standard trajectory template for the task is unusable. It is necessary to reacquire the kinematic data for the task and generate the initial standard trajectory template for the task.

8. The standard trajectory modeling method for rehabilitation robots according to claim 6 or 7, characterized in that, The root mean square deviation is obtained by the following formula: ， in, Indicates task The The root mean square deviation of each subject from the initial standard trajectory template This represents the total number of sampling points for the valid action segment data. Indicates the first Normalized time for each sampling point Indicates task The Each subject and the initial standard trajectory template Euclidean distance at time; The 95th percentile deviation is obtained by the following formula: ， in, Indicates task The The 95th percentile deviation of each subject from the initial standard trajectory template. This represents the 95th percentile deviation function; The maximum deviation is obtained by the following formula: ， in, Indicates task The The maximum deviation of a subject from the initial standard trajectory template, where max represents the maximum value.

9. The standard trajectory modeling method for rehabilitation robots according to claim 1, characterized in that, The expression for the piecewise polynomial fitting model is as follows: ， in, Indicates the segment number. express of Power of 1 Indicates the first Segmented time variables, As a preset value, Indicates the first Segmentation Spatial location at a given time Indicates the first Piecewise fitting coefficients; The positional continuity constraint is achieved through the following formula: ， in, Indicates the first Segmentation Spatial location at any given time; The velocity continuity constraint is achieved through the following formula: ， in, express right The first derivative, express right The first derivative; The acceleration continuity constraint is achieved through the following formula: ， in, express right The second derivative, express right The second derivative; The initial and final boundary constraints are implemented using the following formula: ， ， in, This indicates the start time of the initial standard trajectory template. Indicates the termination time of the initial standard trajectory template. Indicates the starting segment number. Indicates the end segment number. Indicates task The initial standard trajectory template at the start time Spatial location, Indicates task The starting segment At the start time Spatial location, Indicates task The initial standard trajectory template at the termination time Spatial location at a given time Indicates task Termination segment At the end time Spatial location at any given moment.

10. A compliant interactive control method for a rehabilitation robot, characterized in that, include: The continuous standard trajectory model of each task is used as the reference trajectory input of the end effector of the corresponding type of rehabilitation robot; wherein, the continuous standard trajectory model of each task is obtained by the rehabilitation robot standard trajectory modeling method of any one of claims 1-9; Real-time acquisition of force information between various rehabilitation robots and the patient's upper limbs; Based on the interaction force information, the reference trajectory is corrected by admittance control to generate a compliant corrected trajectory; The compliance correction trajectory is input as a motion control command to the servo driver of various rehabilitation robots, driving the rehabilitation robot to move according to the compliance correction trajectory, so as to guide the patient's upper limb to achieve active training of human-machine compliance control.

Images