Gait detection method for exoskeleton robot and related device thereof

Abstract

The application discloses a gait detection method of an exoskeleton robot and related equipment thereof, and the gait detection method comprises the following steps: acquiring an angle signal and an angular velocity of a motor for driving a hip joint of the exoskeleton robot to move; determining a first event occurrence time based on an intensity mutation feature of the angle signal, wherein the first event occurrence time is a time when a k+1 heel touch-down event of a gait event of the exoskeleton robot occurs; in a current gait cycle, determining a plurality of first candidate times of toe-off events based on the angle signal, and taking the first candidate time closest to a time when the angular velocity crosses zero as a second event occurrence time of the toe-off event; and outputting a control signal according to the first event occurrence time and the second event occurrence time. The application solves the problems of insufficient accuracy and high hardware complexity in detecting the gait event of the exoskeleton robot.

Description

Technical Field

[0001] This application relates to the field of robotics technology, and more specifically, to a gait detection method and related equipment for an exoskeleton robot. Background Technology

[0002] This section is intended to provide background or context for the embodiments of the invention set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section.

[0003] Hip exoskeleton robots, as wearable smart devices, provide users with movement assistance or rehabilitation support by simulating the movement patterns of the human hip joint and combining sensors, drive systems, and intelligent algorithms. The recognition of gait events (such as heel strike (HS) and toe lift (TO)) in exoskeleton robots is crucial for their control and gait analysis.

[0004] Existing gait event detection methods mostly rely on plantar pressure or inertial sensors, using thresholds or simple features for event detection. While these methods perform reasonably well on flat terrain, they are susceptible to noise and individual differences in complex environments such as slopes, stairs, etc., leading to false positives or false negatives and insufficient accuracy. Furthermore, multi-sensor fusion increases hardware complexity, hindering wearable applications. Therefore, current gait event detection methods for exoskeleton robots suffer from insufficient accuracy and high hardware complexity. Summary of the Invention

[0005] This application provides a gait detection method and related equipment for exoskeleton robots, which at least solves the problems of insufficient accuracy and high hardware complexity in the detection of gait events of exoskeleton robots in the prior art.

[0006] According to one aspect of this application, a gait detection method for an exoskeleton robot is provided, comprising:

[0007] Acquire the angle signal and angular velocity of the motor that drives the hip joint movement of the exoskeleton robot;

[0008] The time of occurrence of the first event is determined based on the intensity abrupt change characteristics of the angle signal. The time of occurrence of the first event is the time when the (k+1)th heel strike event of the gait event occurs. The intensity abrupt change characteristics include features that reflect the change in the angle signal when the heel strike event occurs, where k is a positive integer.

[0009] Within the current gait cycle, multiple first candidate moments of the toe-off event are determined based on the angle signal, and the first candidate moment closest to the moment when the angular velocity crosses zero is taken as the second event occurrence moment of the toe-off event. The gait cycle is the time period between the first event occurrence moment and the moment when the kth heel-off event has occurred.

[0010] Based on the occurrence times of the first event and the second event, a control signal is output, which is used to control the movement of the exoskeleton robot.

[0011] Furthermore, the step of determining the time of occurrence of the first event based on the intensity abrupt change characteristics of the angle signal includes:

[0012] The angle signal is subjected to wavelet transform processing to obtain the transform result, and the detail coefficients in the transform result are reconstructed to obtain a reconstructed signal that reflects the intensity abrupt change feature;

[0013] The time of occurrence of the first event is determined based on the negative peak of the reconstructed signal.

[0014] Furthermore, the step of reconstructing the high-frequency detail coefficients in the transformation result includes:

[0015] Using a specified wavelet function and a preset number of decomposition layers, the angle signal is subjected to multi-scale convolution processing to obtain a wavelet coefficient vector and a length vector. The wavelet coefficient vector includes the detail coefficients and approximation coefficients of each decomposition layer, and the length vector includes the lengths of the detail coefficients and the approximation coefficients.

[0016] Based on the wavelet coefficient vector and the length vector, the reconstructed signal is obtained by reconstructing the detail coefficients of the predetermined decomposition layer using a wavelet reconstruction function.

[0017] Furthermore, before determining the time of the first event based on the negative peak value of the reconstructed signal, the method further includes:

[0018] Obtain multiple amplitude values ​​of the reconstructed signal at different times t;

[0019] Calculate the root mean square value of multiple amplitude values, normalize the reconstructed signal based on the root mean square value, and determine the time of occurrence of the first event based on the normalized reconstructed signal.

[0020] Furthermore, the step of determining the occurrence time of the first event based on the negative peak of the reconstructed signal includes:

[0021] Filter one or more second candidate moments corresponding to the minimum point located at the negative peak of the reconstructed signal to obtain a candidate set consisting of one or more second candidate moments;

[0022] Within a set first time search window, find the second candidate moment located at the deepest negative peak of the reconstructed signal in the candidate set, so as to determine the second candidate moment at the deepest negative peak as the time of occurrence of the first event. The first time search window is located within a predetermined time period after the kth heel-landing event occurs.

[0023] Furthermore, before finding the second candidate time point located at the deepest negative peak of the reconstructed signal in the candidate set within the set within the set first time search window, the method further includes:

[0024] If the average gait period of a predetermined number of gait events in the past is T, then the sum of the occurrence time of the kth heel strike event and ρ1T is taken as the starting point of the first time search window, and the sum of the occurrence time of the kth heel strike event and ρ2T is taken as the ending point of the first time search window, where ρ1 and ρ2 are preset proportional coefficients.

[0025] The ρ2 is greater than the ρ1; and / or,

[0026] The relationship between ρ1 and ρ2 is: 0 < ρ1 < 1 < ρ2 < 2.

[0027] Furthermore, the step of determining the first candidate moment of the toe-off event based on the angle signal includes:

[0028] Based on the angle signal, multiple first candidate moments corresponding to multiple minimum angle values ​​are found within a set second time search window. The starting point of the second time search window is the moment when the k-th heel-landing event occurs, and the ending point of the second time search window is the moment when the first event occurs.

[0029] Furthermore, when selecting the first candidate moment closest to the moment when the angular velocity crosses zero as the moment when the second event of the toe-off event occurs, the method includes:

[0030] Find one or more velocity zero-crossing moments when the angular velocity crosses zero from negative to positive.

[0031] Calculate the time difference between the velocity zero-crossing moment and the first candidate moment, and take the first candidate moment corresponding to the smallest time difference as the moment when the second event occurs.

[0032] Furthermore, the time of occurrence of the kth heel-landing event is set as T(HS).k The second event occurs at time T(TO). k The gait period is T. k The method further includes:

[0033] If the next toe-off event occurs at time T(TO) k+1 If T(TO) is set, then T(TO) is set. k+1 ) and the T(TO) k The following relationship is satisfied:

[0034] T(TO k+1 )-T(TO k )≥0.8×T k ; and / or,

[0035] If the first event occurs at time T(HS) k+1 ), then the T(HS) k+1 ) and the T(HS) k The following relationship is satisfied:

[0036] T(HS k+1 )-T(HS k )≥0.8×T k ; and / or,

[0037] If the time of the (k-1)th heel strike event is T(HS) k-1 ), then the T(HS) k-1 ) and the T(TO) k The following relationship is satisfied:

[0038] T(TO k )≥T(HS k-1 )+0.1*T k ; and / or,

[0039] The T(HS) k ) and the T(TO) k The following relationship is satisfied:

[0040] T(HS k )≥T(HO k )+0.22*T k .

[0041] Furthermore, before determining the time of the first event based on the intensity abrupt change characteristics of the angle signal, the method further includes:

[0042] At least one of the angle signal and the angular velocity is smoothed using a predetermined filter to obtain the angle signal after eliminating signal noise; and / or,

[0043] The angular velocity is calculated based on the angle signal and the preset angle sampling time interval.

[0044] According to another aspect of this application, a gait detection device for an exoskeleton robot is also provided, which is used to implement the gait detection method for the exoskeleton robot.

[0045] According to another aspect of this application, an exoskeleton robot control system is also provided, which includes a gait detection device for the exoskeleton robot.

[0046] According to another aspect of this application, an exoskeleton robot is also provided, the exoskeleton robot including the exoskeleton robot control system.

[0047] According to another aspect of this application, a readable storage medium is also provided, on which computer instructions are stored, wherein the computer instructions, when executed by a processor, implement the gait detection method of the exoskeleton robot.

[0048] The gait detection method for exoskeleton robots provided in this application can determine the timing of the first event based on the intensity abrupt change characteristics of the angle signal. These intensity abrupt change characteristics include features reflecting changes in the angle signal when the heel strike occurs. This is because the heel strike event is a significant abrupt change in the gait cycle, causing a rapid change in the hip joint angle signal. This intensity abrupt change characteristic (such as the angle abrupt change point) corresponds to this rapid change and is reflected in the angle signal, thus achieving accurate detection of the timing of the first event. Simultaneously, this application also considers the first candidate moment closest to the moment when the angular velocity crosses zero as the timing of the second event (the moment the toe lifts off the ground), improving the accuracy of detecting the timing of the second event. Based on the above, this application can still guarantee the accuracy of the timing of the first and second events under complex gait conditions, improving the robustness of gait detection and thus enhancing the timeliness and accuracy of subsequent gait analysis and motion control. Furthermore, the method provided in this application relies on the operating parameters of the motor itself, eliminating the need for the installation and calibration of additional sensors (such as foot or inertial sensors), thereby reducing the complexity and cost of the exoskeleton robot system. Attached Figure Description

[0049] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0050] Figure 1This is a flowchart of a gait detection method for an exoskeleton robot disclosed in an embodiment of this application;

[0051] Figure 2 for Figure 1 The specific execution flowchart of step S12;

[0052] Figure 3 for Figure 2 The detailed execution flowchart of step S121;

[0053] Figure 4 This is a flowchart for normalizing the reconstructed signal;

[0054] Figure 5 The flowchart shows the process of detecting the moment of occurrence of a unique first event by applying step constraints through a defined first-time search window;

[0055] Figure 6 This is a flowchart of a gait detection method for an exoskeleton robot disclosed in an application embodiment of this application;

[0056] Figure 7 This is a timing diagram of gait events disclosed in an embodiment of this application. Detailed Implementation

[0057] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0058] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0059] First, some nouns or terms that appear in the description of the embodiments of this application shall be interpreted as follows:

[0060] Wavelet transform is a commonly used time-frequency analysis method that can simultaneously characterize signal features (such as the angle signal mentioned below in this application) in both the time and frequency domains. Wavelet decomposition can effectively extract local abrupt changes in signal intensity from the motor angle signal in this application, such as the negative peak of the heel-to-spot event in the gait events of an exoskeleton robot, for accurate identification of the timing of the heel-to-spot event.

[0061] Zero-crossing of velocity refers to the moment when a velocity signal crosses zero from positive to negative or vice versa. This invention utilizes this zero-crossing of velocity to precisely correct the occurrence time of candidate toe-off events in gait events, thereby obtaining reliable toe-off occurrence times.

[0062] Motor angle and angular velocity signals: The angle and angular velocity signals of the hip joint motors of the exoskeleton robot are collected, which reflect the periodic changes of the hip joint in gait movement and are the core input data for gait event detection.

[0063] Time search window: A method of extracting subsequences from a continuous signal according to a fixed time or number of samples for segmented analysis and feature extraction. This invention uses a time search window for wavelet decomposition and zero-crossing detection to ensure the real-time performance of the algorithm.

[0064] Gait events (HS / TO): The heel strike event (HS) is the starting point of the gait cycle, and the toe-off event (TO) is the important ending point of the gait cycle. This invention extracts the occurrence time of the heel strike event (HS) using wavelet negative peaks and combines it with velocity zero-crossing correction to obtain the occurrence time of the TO event.

[0065] Savitzky-Golay filter: The core idea is to smooth data through local polynomial least squares fitting. This method can effectively preserve the high-frequency variation characteristics of the signal (such as peak value and width) while removing noise.

[0066] Gait events in exoskeleton robots include heel strike events (HS) and toe lift events (TO). Addressing the issues of insufficient accuracy and high hardware complexity in exoskeleton robot gait event detection, the inventors of this invention have discovered that existing methods for gait event detection largely rely on foot pressure sensors or simple threshold detection. The former is hardware-complex and costly, while the latter is prone to false detections in complex scenarios such as slopes and stairs, failing to meet the requirements of wearable devices for lightweight design, real-time performance, and stability. Furthermore, it is prone to multiple false heel strikes or toe lifts within a single cycle, leading to exoskeleton assistive control failure. Sensor-dependent solutions also suffer from fixed feature extraction, failing to adapt to individual differences and resulting in decreased accuracy in complex environments (such as slopes and stairs). Moreover, the lack of stride distance prediction and rolling constraints easily leads to multiple redundant events within a single cycle; simultaneously, TO determination relies solely on angle troughs without incorporating speed correction, making it sensitive to noise and lacking in real-time performance and robustness.

[0067] To address the aforementioned problems, the first embodiment of this invention provides a gait detection method for an exoskeleton robot. Please refer to [link to relevant documentation]. Figure 1 The method includes the following steps:

[0068] Step S11: Acquire the angle signal and angular velocity of the motor that drives the hip joint movement of the exoskeleton robot. The angle signal can be acquired in real time by an encoder set on the motor.

[0069] Step S12: Determine the time of occurrence of the first event based on the intensity change characteristics of the angle signal. The time of occurrence of the first event is the time when the (k+1)th heel strike event of the gait event occurs. The intensity change characteristics include features that reflect the change in the angle signal when the heel strike event occurs, where k is a positive integer.

[0070] By extracting the intensity abrupt change features of the angle signal to detect the timing of heel strike events, the accuracy of detecting the timing of heel strike events in complex scenarios such as uphill and downhill slopes and stairs can be improved.

[0071] Step S13: Within the current gait cycle, determine multiple first candidate moments of the toe-off event based on the angle signal, and take the first candidate moment closest to the moment when the angular velocity crosses zero as the second event moment when the toe-off event occurs. The gait cycle is the time interval between the moment when the first event occurs and the moment when the kth heel-off event has occurred.

[0072] Since the trough of the angle signal corresponds to the moment of toe-off event, this is because the rapid movement during a toe-off event creates a significant low point in the angle signal, which is the trough of the angle signal change curve. This embodiment can obtain multiple first candidate moments through the troughs of the angle signal. To obtain a unique and accurate second event moment of the toe-off event, this application can correct the actual toe-off event moment based on the moment the angular velocity crosses zero. The first candidate moment closest to the moment the angular velocity crosses zero is taken as the second event moment of the toe-off event. This allows for precise identification of the unique second event moment of the actual toe-off event from one or more first candidate moments, avoiding the situation where multiple toe-off events occur within a single gait cycle.

[0073] Step S14: Output control signals based on the occurrence times of the first and second events. These control signals are used to control the movement of the exoskeleton robot. The output control signals may include gait cycle, step frequency, etc., thereby providing these control signals to the controller of the exoskeleton robot for phase switching and assist control, ensuring the effectiveness of the exoskeleton robot's assist control.

[0074] As can be seen, the gait detection method for exoskeleton robots provided in this application can determine the timing of the first event based on the intensity abrupt change characteristics of the angle signal. These intensity abrupt change characteristics include features reflecting changes in the angle signal when the heel strike occurs. This is because the heel strike event is a significant abrupt change in the gait cycle, causing a rapid change in the hip joint angle signal. This intensity abrupt change characteristic (such as the angle abrupt change point) corresponds to this rapid change and is reflected in the angle signal, thus achieving accurate detection of the timing of the first event. Simultaneously, this application also uses the first candidate moment closest to the moment when the angular velocity crosses zero as the timing of the second event (the moment the toe lifts off the ground), improving the accuracy of detecting the timing of the second event. Based on the above, this application can still guarantee the accuracy of the timing of the first and second events under complex gait conditions, improving the robustness of gait detection and thus enhancing the timeliness and accuracy of subsequent gait analysis and motion control. Furthermore, the method provided in this application relies on the operating parameters of the motor itself, eliminating the need for the installation and calibration of additional sensors (such as foot or inertial sensors), thereby reducing the complexity and cost of the exoskeleton robot system.

[0075] Please see Figure 2 In step S12, the step of determining the time of occurrence of the first event based on the intensity abrupt change characteristics of the angle signal includes:

[0076] Step S121: Perform wavelet transform on the angle signal to obtain the transform result, and reconstruct the detail coefficients in the transform result to obtain a reconstructed signal that reflects the intensity abrupt change characteristics.

[0077] Among them, the detail coefficients obtained after wavelet transform decomposition are mathematical representations of the high-frequency components of the angle signal at one scale, reflecting the intensity and location of local abrupt or transient features of the angle signal. At different scales, the detail coefficients represent the high-frequency components of the angle signal (such as details, rapid changes, and noise), which can reveal the local features of the angle signal at different resolutions, thus facilitating the accurate detection and extraction of changes in the angle signal that occur when the heel strikes the ground.

[0078] Step S122: Determine the time of occurrence of the first event based on the negative peak of the reconstructed signal.

[0079] The reconstructed signal, as a time-domain signal, describes the change of the angle signal over time. Its waveform has time as the independent variable on the horizontal axis and the amplitude of the angle signal change on the vertical axis. The negative peaks here are obvious downward spikes (i.e., local minima) appearing in the reconstructed signal waveform. These spikes correspond to the instantaneous impact or deceleration generated when the heel strikes (HS) event occurs in the gait event, manifesting as a rapid decline in the reconstructed signal. On the waveform, this rapid decline forms a downward peak, hence the term negative peak. Based on this principle, this application aims to improve the accuracy of the detection result at the moment of the first event by finding the negative peaks of the reconstructed signal.

[0080] Please see Figure 3 In step S121 above, the step of reconstructing the high-frequency detail coefficients in the transformation result includes:

[0081] Step S1211: Using the specified wavelet function and a preset number of decomposition layers (hereinafter referred to as levelHS), perform multi-scale convolution processing on the angle signal to obtain a wavelet coefficient vector (hereinafter referred to as C) and a length vector (hereinafter referred to as L). The wavelet coefficient vector includes the detail coefficients and approximation coefficients of each decomposition layer, and the length vector includes the lengths of the detail coefficients and approximation coefficients. The length vector is used for accurate segmentation coefficients during reconstruction. The wavelet function can include one of the following: Haar wavelet, Daubechies wavelet ('db1'), Symlets wavelet (such as 'sym2'), etc. The optimal wavelet function (i.e., wavelet basis) can be selected according to specific task requirements (such as real-time performance, smoothness, and noise type).

[0082] Step S1212: Based on the wavelet coefficient vector and length vector, the detailed coefficients of the predetermined decomposition layer are reconstructed using the wavelet reconstruction function to obtain the reconstructed signal.

[0083] In step S1211, the number of decomposition layers refers to the number of layers in which the angle signal is decomposed during wavelet decomposition. For example, when the number of decomposition layers is 3 (i.e., levelHS = 3), the decomposition yields 3 layers of detail coefficients for the high-frequency part and 1 layer of approximation coefficients for the low-frequency part. At this time, the wavelet coefficient vector C = [3rd layer approximation coefficient A3, 3rd layer detail coefficient D3, 2nd layer detail coefficient D2, 1st layer detail coefficient D1].

[0084] Please see Figure 4 In step S122, before determining the time of the first event based on the negative peak value of the reconstructed signal, the method provided in this application further includes the following steps:

[0085] Step S21: Obtain multiple amplitude values ​​of the reconstructed signal at different times t.

[0086] Step S22: Calculate the root mean square (RMS) values ​​of multiple amplitude values, and normalize the reconstructed signal based on the RMS values ​​to determine the time of the first event. This normalization method is called RMS normalization, which involves dividing each sample point of the reconstructed signal by its RMS value.

[0087] Since the amplitude of the same signal may vary greatly at different time periods, this application normalizes the reconstructed signal based on the above steps S21 to S22, scaling the amplitude of the reconstructed signal to the root mean square scale, so that the reconstructed signal has the same energy scale, which is convenient for subsequent comparison, analysis or processing.

[0088] Please see Figure 5 In step S122, the step of determining the time of occurrence of the first event based on the negative peak of the reconstructed signal includes:

[0089] Step S1221: Filter one or more second candidate moments corresponding to the minimum point located at the negative peak of the reconstructed signal to obtain a candidate set consisting of one or more second candidate moments.

[0090] Step S1222: Find the second candidate moment located at the deepest negative peak of the reconstructed signal in the candidate set within the set first time search window, so as to determine the second candidate moment at the deepest negative peak as the time of occurrence of the first event. The first time search window is located within the predetermined time period after the occurrence of the kth heel-to-toe landing event.

[0091] After completing steps S1221 to S1222, this application, after finding the second candidate time corresponding to the minimum point of the reconstructed signal and obtaining the candidate set, can accurately find the time of occurrence of the first event within the set first time search window, improving the accuracy, real-time performance, and robustness of heel strike event detection. Furthermore, based on the set first time search window, stride constraints are implemented, ensuring that the detected heel strike event matches the actual gait process and guaranteeing that only one heel strike event is retained in each gait cycle.

[0092] In some embodiments, before identifying the second candidate time point located at the deepest negative peak of the reconstructed signal within a set first time search window, the method provided in this application further includes the following steps:

[0093] If the average gait period of a predetermined number of gait events in the past is T, then the sum of the occurrence time of the kth heel strike event and ρ1T is taken as the starting point of the first time search window, and the sum of the occurrence time of the kth heel strike event and ρ2T is taken as the ending point of the first time search window. ρ1 and ρ2 are pre-set proportional coefficients, where;

[0094] ρ2 is greater than ρ1; and / or,

[0095] ρ1 and ρ2 satisfy the relationship: 0 < ρ1 < 1 < ρ2 < 2. The value of ρ1 can be one of 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, etc. The value of ρ2 can be one of 1.1, 1.2, 1.3, 1.5, 1.6, 1.8, etc.

[0096] In other words, the first-time search window begins at time ρ1T after the k-th heel strike event and ends at time ρ2T after the k-th heel strike event. The predetermined number can include five, six, seven, etc. By setting the start and end points of the first-time search window, this application can improve the real-time performance and robustness of detecting the first event. This rolling stride prediction update mechanism can adapt to changes in gait frequency among different individuals and in different environments, avoiding the maladaptive effects of fixed parameters, thereby improving the generalization ability of gait detection.

[0097] In some implementations, ρ1≈0.5 and ρ2≈1.5 can be set, that is, to search for candidate moments within a time range of half a gait cycle to one and a half gait cycles to obtain the moment when the first event occurs.

[0098] Step S13, which involves determining the first candidate moment of the toe-off event based on the angle signal, includes:

[0099] Based on the angle signal, multiple first candidate moments corresponding to multiple minimum angle values ​​are found within the set second time search window. The starting point of the second time search window is the moment when the k-th heel touches down, and the ending point of the second time search window is the moment when the first event occurs.

[0100] Therefore, this embodiment finds multiple first candidate moments in the second time search window to accurately identify the moment of occurrence of the second event from these first candidate moments. This not only ensures high detection accuracy and avoids generating multiple redundant events, but also improves the real-time performance of detecting toe-off events, ensuring that the detected toe-off events match the actual gait process and improving the effectiveness of subsequent control.

[0101] Secondly, when the first candidate moment closest to the moment when the angular velocity crosses zero is taken as the moment when the second event of toe-off occurs, the method provided in this application includes the following steps:

[0102] Find one or more velocity zero-crossing moments when the angular velocity crosses zero from negative to positive. Calculate the time difference between the velocity zero-crossing moment and the first candidate moment, and select the first candidate moment with the smallest time difference as the moment when the second event occurs.

[0103] Therefore, this application calculates the time difference between the velocity zero-crossing moment and each first candidate moment, and uses the first candidate moment corresponding to the minimum time difference as the second event occurrence moment, thus accurately detecting toe-off events. This detection method combines the velocity zero-crossing moment of angular velocity to accurately correct the toe-off event occurrence moment, improving real-time performance and robustness. Moreover, by introducing a second time search window as a time constraint and fusing it with the feature of angular velocity crossing zero, the false detection rate is significantly reduced, achieving high-precision gait event recognition under low-cost conditions, making it suitable for application in hip exoskeleton robot control systems that can achieve real-time gait control.

[0104] By selecting one or more velocity zero-crossing moments when the angular velocity crosses zero from negative to positive, the direction of motion of the hip joint toe-off event can be accurately reflected, thereby precisely capturing the moment when the toe-off event occurs.

[0105] In addition, to ensure the physiological rationality of the detection results, this application has imposed the following constraints on the size range of each time point, so that the detection results are more consistent with the actual gait process and conform to physiological characteristics:

[0106] First, let the time of the kth heel landing event be T(HS). k The second event occurs at time T(TO). k The current detected gait period is T. k The method provided in this application also includes the following:

[0107] If the next toe-off event occurs at time T(TO) k+1 If ), then set T(TO) k+1 ) and T(TO k The following relationship is satisfied:

[0108] T(TO k+1 )-T(TO k )≥0.8×T k And / or,

[0109] If the time of the first event is T(HS) k+1 ), then T(HS) k+1 ) and T(HS) k The following relationship is satisfied:

[0110] T(HS k+1 )-T(HS k )≥0.8×T k And / or,

[0111] If the time of the (k-1)th heel landing event is T(HS)k-1 ), then T(HS) k-1 ) and T(TO k The following relationship is satisfied:

[0112] T(TO k )≥T(HS k-1 )+0.1*T k And / or,

[0113] T(HS k ) and T(TO k The following relationship is satisfied:

[0114] T(HS k )≥T(HO k )+0.22*T k .

[0115] Thus, by imposing the aforementioned constraints on each time point, this application can better guarantee the physiological rationality of the test results.

[0116] Before determining the time of the first event based on the intensity abrupt change characteristics of the angle signal, the method provided in this application also includes the following:

[0117] A predetermined filter is used to smooth at least one of the angle signal and angular velocity to obtain an angle signal with noise removed. When this noise-removed angle signal is subsequently used to detect the occurrence times of the first and second events, the detection results will not be affected by noise, resulting in more accurate and reliable detection.

[0118] Secondly, the angular velocity is calculated based on the angle signal and a pre-set angle sampling time interval. Therefore, there is no need to add an extra sensor to detect the angular velocity; the angular velocity can be obtained based on the angle signal from the motor, further reducing cost and the structural complexity of the exoskeleton robot.

[0119] As shown above, the method provided in this application integrates wavelet transform and the zero-crossing moment of angular velocity to detect heel strike and toe-off events. The core idea is as follows: First, the hip joint angle signal is preprocessed with smoothing filtering. Then, wavelet transform is used to decompose and extract detail coefficients of the angle signal at a specific scale. These detail coefficients are then reconstructed to obtain a reconstructed signal. The negative peak of the reconstructed signal is detected to obtain the second candidate moment of the heel strike event (also known as heel strike HS), and a stride prediction mechanism is used to ensure that only one heel strike event (HS) is retained in each gait cycle. Within the gait cycle where the heel strike event occurs, based on the waveform of the angle signal, the moment of the second event is refined by searching for angle troughs and utilizing the zero-crossing moment of angular velocity from negative to positive, resulting in a unique toe-off event (also known as toe-off TO). Furthermore, by incorporating time constraints such as the first time search window and the second time search window with zero crossover techniques of angular velocity, the false detection rate of gait events is significantly reduced, achieving high-precision gait event recognition under low-cost sensor conditions. This is suitable for application in the control system of hip joint exoskeleton robots and can stably and accurately identify heel strike events and toe lift events in complex environments in real time.

[0120] The second embodiment of this application also provides a gait detection device for an exoskeleton robot, which is used to implement a gait detection method for the exoskeleton robot. For details of the gait detection method for the exoskeleton robot, please refer to the content provided in the first embodiment of this application.

[0121] Based on the first and second embodiments, in order to make the content of the present invention clearer, the third embodiment of this application provides a specific application embodiment of gait detection.

[0122] The gait detection device for the exoskeleton robot in this application may include a signal acquisition module, a signal preprocessing module, a wavelet feature extraction module, a gait event detection module, and an output feedback module.

[0123] Signal acquisition module: Acquires angle signals in real time through the hip joint motor encoder and calculates angular velocity using a differential method.

[0124] Signal preprocessing module: Savitzky-Golay filtering is used to smooth the angle and angular velocity and eliminate noise.

[0125] Wavelet feature extraction module: Performs wavelet decomposition on the smoothed angle signal to extract negative peak features related to heel strike (HS).

[0126] Gait event detection module: Based on wavelet negative peak candidate points, HS is screened, and combined with angle wave trough and velocity zero crossover to detect toe off the ground TO, ensuring that only one HS and one TO are retained in each cycle.

[0127] Output feedback module: Outputs HS and TO time points, stride length and stride frequency results, which are provided to the exoskeleton controller for phase switching and assist control.

[0128] 3.1 The signal preprocessing module performs signal preprocessing.

[0129] When the input contains only angle, the angular velocity ω(t) is calculated by difference:

[0130]

[0131] Where θ(t) is the motor angle signal, Δt is the angle sampling time interval, and θ(t-1) is the angle sampled at intervals Δt. Therefore, the angular velocity ω(t) can be obtained based on the motor angle signal θ(t), further reducing cost and the structural complexity of the exoskeleton robot.

[0132] The angle signal θ(t) is then smoothed using a Savitzky-Golay filter, resulting in the filtered angle signal θ. f (t):

[0133]

[0134] In formula (2), θ(t) is the original angle signal, θ f (t) represents the filtered angle signal. n is the width of half of the filter window, which defines how many data points are looked at on both the left and right sides when calculating the midpoint smoothing value. The window width is w = 2*n+1.

[0135] c m For the convolution kernel or filter coefficients, this is a set of pre-calculated weight values ​​for each position m within the window. The smoothing process is essentially a weighted sum of the original data θ(t+m) within the window and these weight coefficients Cm (i.e., discrete convolution). In application, the convolution kernel c can be obtained by looking up a table or calling a library function based on the window width w and the polynomial order k. m parameter.

[0136] 3.2 The wavelet feature extraction module performs wavelet decomposition and feature extraction on the filtered angle signal.

[0137] Perform wavelet decomposition on the filtered angle signal:

[0138] C,L=wavede c(θ f ,levelHS,wvHS) (3)

[0139] In formula (3):

[0140] θ f The input signal is the filtered angle signal to be decomposed.

[0141] levelHS: This refers to the number of decomposition levels or the decomposition level. For example, when the number of decomposition levels is 3, it can be 3 high-frequency detail coefficients + 1 low-frequency approximation coefficient.

[0142] wvHS: is the name of the wavelet basis (wavelet function) to be used (e.g., 'db1', 'haar', 'sym2', etc.), and specifies the mother wavelet to be used for decomposition.

[0143] C: Output wavelet coefficient vector, for example, when levelHS=3: C=[A3,D3,D2,D1]. Dk: High-frequency detail coefficients of the k-th layer (local abrupt change information), Ak: Low-frequency approximation coefficients of the k-th layer (signal trend).

[0144] L: Records the length vector of the coefficients at each decomposition level, identifies the length of each component in C, and is used to accurately segment the coefficients during reconstruction.

[0145] By using the wavelet reconstruction function wrcoef, detail coefficients at specific levels are extracted as HS features of heel strike events in gait events.

[0146] d HS (t)=wrcoef(′d′,C,L,wvHS,levelHS) (4)

[0147] In formula (4):

[0148] d HS (t) is the reconstructed signal obtained from the reconstruction, which belongs to the time domain.

[0149] 'd' represents the reconstruction detail coefficients (high-frequency components), while 'a' represents the reconstruction approximation coefficients (low-frequency components). This application aims to extract the heel-to-ground (HS) feature from the angle signal, therefore reconstructing the detail coefficients of the high-frequency component.

[0150] C: Output wavelet coefficient vector.

[0151] L: A length vector that records the length of the coefficients at each level.

[0152] wvHS: The name of the wavelet basis to use, specifying the wavelet function to be used.

[0153] levelHS: The specific level to be reconstructed, the level of detail coefficients to be extracted (e.g., levelHS=3 indicates the 3rd level of detail coefficients). The higher the level, the lower the frequency resolution and the higher the temporal resolution.

[0154] The reconstructed signal is then normalized because the amplitude of the same signal can vary significantly across different time periods. Normalization ensures that all signals have the same "scale."

[0155]

[0156] In formula (5), the denominator rm s(d HS The denominator d represents the root mean square value obtained above. HS (t) represents the original signal for reconstructing the signal, and d′ HS (t) is the reconstructed signal after normalization.

[0157] 3.3 Gait event detection module detects gait events.

[0158] 3.3.1 HS Candidate Points (Wavelet Negative Peaks):

[0159] HS cand ={t|d′ HS} (6)

[0160] In formula (6), HS cand It is the candidate set mentioned above, {t|d′ HS Let} represent the set of all second candidate timestamps that satisfy the condition that the second candidate timetamp is a local minimum of dHSi(t) (i.e., the negative peak of the wavelet coefficients in the reconstructed signal).

[0161] 3.3.2 Filtering the occurrence time of the unique heel strike event HS (deepest negative peak + stride constraint), and recording the occurrence time T(HS) of the first event obtained after filtering. k+1 ):

[0162]

[0163] In formula (7):

[0164] HS k : Represents the time when the k-th confirmed heel strike event occurs.

[0165] HS k+1 : Represents the next (k+1) heel-landing event.

[0166] T: Average gait cycle time, which is the average of the past five gait cycles, with an initial value of 1.

[0167] ρ1, ρ2: proportionality coefficients, usually set to 0 < ρ1 < 1 < ρ2 < 2.

[0168] argmin: The parameter used to find the minimum value of the function. This is used to determine the second candidate time point at the deepest negative peak as the time point T(HS) of the first event. k+1 The value of ).

[0169] Detailed explanation:

[0170] [HS k +ρ1T,HS k [+ρ2T] defines the first time-space search window, which begins at time ρ1T after the previous HS point and ends at time ρ2T after the previous HS point. Typically, ρ1≈0.5, ρ2≈1.5 (i.e., searching for a time range of half a cycle to one and a half cycles).

[0171] 3.3.3, Toe-off event TO candidate point (angle trough):

[0172]

[0173] Detailed explanation of formula (8):

[0174] [HS k HS k+1 A second time search window is defined, and the minimum joint angle value (i.e., the minimum angle value mentioned above) and the corresponding first candidate time are obtained within this second time search window. cand This is the set of the first candidate moments selected.

[0175] 3.3.4. Based on the velocity zero-crossing moment at the zero-crossing point of the angular velocity (t in the formula below), confirm and record the current second event occurrence time T(TO). k ):

[0176]

[0177] In formula (9), |t-TO cand | represents the time difference between the velocity zero-crossing moment and the first candidate moment.

[0178] Formula (9) is used to calculate the minimum time difference T(TO) between the moment the toe leaves the ground (TO) event occurs and the moment the velocity crosses zero. k The value of ) is used to accurately determine the precise moment when the toes leave the ground in the current gait cycle. t∈{zc|ωt-1<0,ωt≥0}: This represents the moment when the velocity zero-crossing point (zc) is found that satisfies the condition that "the angular velocity at the previous moment is less than zero and the angular velocity at the current moment is greater than or equal to zero".

[0179] Formula (9) can accurately identify and record the precise moment when the toes leave the ground during the current gait cycle, providing reliable data support for subsequent gait analysis and control.

[0180] 3.4 Step Frequency Calculation

[0181] Gait cycle:

[0182] T k =T(HS) k+1 )-T(HS k (10)

[0183] To ensure the physiological rationality of the test results, this application sets the following constraints:

[0184] If the next toe-off event occurs at time T(TO) k+1 If ), then set T(TO) k+1 ) and T(TO k The following relationship is satisfied:

[0185] T(TO k+1 )-T(TO k )≥0.8×T k .

[0186] If the time of the first event is T(HS) k+1 ), then T(HS) k+1 ) and T(HS) k The following relationship is satisfied:

[0187] T(HS k+1 )-T(HS k )≥0.8×T k .

[0188] If the time of the (k-1)th heel landing event is T(HS) k-1 ), then T(HS) k-1 ) and T(TO k The following relationship is satisfied:

[0189] T(TO k )≥T(HS k-1 )+0.1*T k .

[0190] T(HS k ) and T(TO k The following relationship is satisfied:

[0191] T(HS k )≥T(HO k )+0.22*T k .

[0192] Thus, by imposing the aforementioned constraints on each moment, this application can better guarantee the physiological rationality of the detection results and avoid false event detection.

[0193] In conclusion, as Figure 6 As shown, the gait detection method of this application can be described as follows: signal acquisition → signal preprocessing → wavelet decomposition → HS candidate extraction → step distance prediction based on time search window → HS candidate time selection → TO detection and correction → HS / TO pairing → output results.

[0194] like Figure 7 As shown, within a gait cycle, HS appears near the start of the gait cycle, and TO appears at the trough and is corrected by the zero-crossing moment of velocity. The detection results are more accurate, and it ensures that there is only one HS event and one TO event within a gait cycle.

[0195] The gait detection method provided in this application uses wavelet decomposition to extract the abrupt change characteristics of the motor angle signal when the heel strikes the ground, and combines the zero-crossing of angular velocity to correct the timing of the toe-off event. Compared with traditional threshold or single feature methods, this application can still ensure the accuracy of detecting the timing of the first event of HS and the second event of TO under complex gait conditions, thus improving robustness.

[0196] By introducing update mechanisms such as rolling stride prediction (with a set corresponding time search window) and normalized preprocessing, the gait detection method can adapt to stride frequency changes in different individuals and environments, avoiding the maladaptive effects of fixed parameters and thus improving generalization ability.

[0197] Because it relies solely on the angle data of the motor itself, the installation and calibration of additional sensors are eliminated, reducing system complexity and cost, while improving system real-time performance and embedded deployability.

[0198] By incorporating causal logic between HS and TO into the temporal constraints (e.g., HS must be ≥ a certain period proportion after the previous TO), false event detection is effectively avoided, making the recognition results more consistent with the laws of human movement.

[0199] The third embodiment of this application also provides an exoskeleton robot control system, which includes a gait detection device for the exoskeleton robot. For details on the gait detection device for the exoskeleton robot, please refer to the content provided in the second embodiment of this application, which will not be repeated here.

[0200] The fourth embodiment of this application also provides an exoskeleton robot, which includes the exoskeleton robot control system provided in the third embodiment of this application.

[0201] The fifth embodiment of this application also provides a readable storage medium storing computer instructions thereon, wherein the computer instructions, when executed by a processor, implement a gait detection method for an exoskeleton robot. For details of the gait detection method for the exoskeleton robot, please refer to the content provided in the first embodiment of this application, which will not be repeated here.

[0202] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A gait detection method for an exoskeleton robot, characterized in that, include: Acquire the angle signal and angular velocity of the motor that drives the hip joint movement of the exoskeleton robot; The time of occurrence of the first event is determined based on the intensity abrupt change characteristics of the angle signal. The time of occurrence of the first event is the time when the (k+1)th heel strike event of the gait event occurs. The intensity abrupt change characteristics include features that reflect the change in the angle signal when the heel strike event occurs, where k is a positive integer. Within the current gait cycle, multiple first candidate moments of the toe-off event are determined based on the angle signal, and the first candidate moment closest to the moment when the angular velocity crosses zero is taken as the second event occurrence moment of the toe-off event. The gait cycle is the time period between the first event occurrence moment and the moment when the kth heel-off event has occurred. Based on the occurrence times of the first event and the second event, a control signal is output, and the control signal is used to control the movement of the exoskeleton robot. The steps for determining the time of occurrence of the first event based on the intensity abrupt change characteristics of the angle signal include: The angle signal is subjected to wavelet transform processing to obtain the transform result, and the detail coefficients in the transform result are reconstructed to obtain a reconstructed signal that reflects the intensity abrupt change feature; The time of occurrence of the first event is determined based on the negative peak of the reconstructed signal; The steps for reconstructing the high-frequency detail coefficients in the transformation result include: Using a specified wavelet function and a preset number of decomposition layers, the angle signal is subjected to multi-scale convolution processing to obtain a wavelet coefficient vector and a length vector. The wavelet coefficient vector includes the detail coefficients and approximation coefficients of each decomposition layer, and the length vector includes the lengths of the detail coefficients and the approximation coefficients. Based on the wavelet coefficient vector and the length vector, the reconstructed signal is obtained by reconstructing the detail coefficients of the predetermined decomposition layer using a wavelet reconstruction function.

2. The method according to claim 1, characterized in that, Before determining the time of the first event based on the negative peak value of the reconstructed signal, the method further includes: Obtain multiple amplitude values ​​of the reconstructed signal at different times t; Calculate the root mean square value of multiple amplitude values, normalize the reconstructed signal based on the root mean square value, and determine the time of occurrence of the first event based on the normalized reconstructed signal.

3. The method according to claim 1, characterized in that, The step of determining the occurrence time of the first event based on the negative peak of the reconstructed signal includes: Filter one or more second candidate moments corresponding to the minimum point located at the negative peak of the reconstructed signal to obtain a candidate set consisting of one or more second candidate moments; Within a set first time search window, find the second candidate moment located at the deepest negative peak of the reconstructed signal in the candidate set, so as to determine the second candidate moment at the deepest negative peak as the time of occurrence of the first event. The first time search window is located within a predetermined time period after the kth heel-landing event occurs.

4. The method according to claim 3, characterized in that, Before finding the second candidate time point located at the deepest negative peak of the reconstructed signal in the candidate set within the set within the first time search window, the method further includes: If the average gait period of a predetermined number of gait events in the past is T, then the occurrence time of the kth heel strike event is compared with... The sum of T is used as the starting point of the first time search window, and the occurrence time of the kth heel-landing event is compared with... The sum of T serves as the endpoint of the first time-based search window. and stated The pre-set scaling factor, where; The Greater than the ; and / or, The and stated Satisfying the relation: 0 < <1< <2.

5. The method according to claim 1, characterized in that, The steps for determining the first candidate moment of the toe-off event based on the angle signal include: Based on the angle signal, multiple first candidate moments corresponding to multiple minimum angle values ​​are found within a set second time search window. The starting point of the second time search window is the moment when the k-th heel-landing event occurs, and the ending point of the second time search window is the moment when the first event occurs.

6. The method according to claim 1, characterized in that, When selecting the first candidate moment closest to the moment when the angular velocity crosses zero as the moment when the second event of the toe-off event occurs, the method includes: Find one or more velocity zero-crossing moments when the angular velocity crosses zero from negative to positive. Calculate the time difference between the velocity zero-crossing moment and the first candidate moment, and take the first candidate moment corresponding to the smallest time difference as the moment when the second event occurs.

7. The method according to any one of claims 1 to 6, characterized in that, Let the time of occurrence of the k-th heel strike event be T( The second event occurs at time T( The gait period is The method further includes: If the next toe-off event occurs at time T( ), then set the T( ) and the T( The following relationship is satisfied: T( -T( ≥0.8× ; and / or, If the first event occurs at time T( ), then the T( ) and the T( The following relationship is satisfied: T( )-T( )≥0.8× ; and / or, If the time of the (k-1)th heel-landing event is T( ), then the T( ) and the T( The following relationship is satisfied: T( ) ≥T( ) + 0.1* ; and / or, The T( ) and the T( The following relationship is satisfied: T( ) ≥T( ) +0.22* 。 8. The method according to any one of claims 1 to 6, characterized in that, Before determining the time of the first event based on the intensity abrupt change characteristics of the angle signal, the method further includes: At least one of the angle signal and the angular velocity is smoothed using a predetermined filter to obtain the angle signal after eliminating signal noise; and / or, The angular velocity is calculated based on the angle signal and the preset angle sampling time interval.

9. A gait detection device for an exoskeleton robot, characterized in that, The gait detection device of the exoskeleton robot is used to implement the gait detection method of the exoskeleton robot according to any one of claims 1 to 8.

10. A control system for an exoskeleton robot, characterized in that, The device includes the gait detection device for the exoskeleton robot as described in claim 9.

11. An exoskeleton robot, characterized in that, The exoskeleton robot includes the exoskeleton robot control system as described in claim 10.

12. A readable storage medium having computer instructions stored thereon, wherein, When executed by the processor, the computer instructions implement the gait detection method for the exoskeleton robot according to any one of claims 1 to 8.

Images