Exoskeleton control method, device, equipment and storage medium

Abstract

The application belongs to the technical field of exoskeleton control, and discloses an exoskeleton control method, device, equipment and storage medium, comprising the following steps: in response to the end of measurement of an auxiliary walking device, acquiring IMU data collected in the movement process of the auxiliary walking device, the auxiliary walking device being provided with an IMU inertial sensor, and the auxiliary walking device marking an initial position, a target position and the initial position again in sequence during measurement; calculating a three-dimensional relative position relationship between the initial position and the target position according to the IMU data; and controlling the exoskeleton to move to the target position according to the three-dimensional relative position relationship. The exoskeleton trajectory estimation precision in a scenario requiring accurate trajectory estimation can be improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of exoskeleton control, and particularly relates to an exoskeleton control method, device, equipment and storage medium. BACKGROUND

[0002] Exoskeleton (i.e. external skeleton) is a kind of bionic device coupled with human body through mechanical structure, which is used to assist or enhance human body movement function, so that the movement function obstacle group can regain the basic moving ability such as standing and walking.

[0003] Controlling exoskeleton movement is a complex human-machine cooperation process, which needs to combine human body movement intention recognition, dynamics modeling, real-time feedback control and other links. IMU (inertial measurement unit) as one of the core sensors plays a key role in perceiving human body movement state and assisting control decision, and position estimation is obtained by double integration of free acceleration. However, IMU attitude estimation and free acceleration estimation problems are coupled with each other, which brings difficulty to trajectory estimation, and double integration will bring significant position estimation inaccuracy. When precise trajectory estimation is needed, for example: walking in complex walking scenes such as climbing stairs, stepping over a puddle and walking on a pile of stones, the inaccuracy of trajectory estimation may cause the landing point of the controlled exoskeleton to deviate, which affects walking safety. SUMMARY

[0004] The purpose of the present application is to provide an exoskeleton control method, device, equipment and storage medium, which is used to improve the trajectory estimation accuracy of exoskeleton in the scene requiring precise trajectory estimation.

[0005] The first aspect of the present application discloses an exoskeleton control method, comprising:

[0006] In response to the end of measurement of the walking aid device, IMU data collected during movement of the walking aid device is obtained, the walking aid device is installed with an IMU inertial sensor, and the walking aid device sequentially marks an initial position, a target position and then marks the initial position again during measurement;

[0007] According to the IMU data, a three-dimensional relative position relationship between the initial position and the target position is calculated;

[0008] According to the three-dimensional relative position relationship, the exoskeleton is controlled to move to the target position.

[0009] In some embodiments, the walking aid device is a walking stick, and the measurement process of the walking aid device comprises:

[0010] Starting measurement, clicking the initial position with the end of the walking stick, clicking the target position with the end of the walking stick, clicking the initial position again with the end of the walking stick, and ending measurement.

[0011] In some embodiments, the calculating the three-dimensional relative position relationship between the initial position and the target position according to the IMU data comprises:

[0012] dividing the IMU data into first IMU data corresponding to a movement from the initial position to the target position and second IMU data corresponding to a movement from the target position to the initial position;

[0013] performing attitude estimation based on the first IMU data and the second IMU data by using a multi-kernel entropy extended Kalman filter to obtain acceleration information;

[0014] performing trajectory estimation according to the acceleration information to obtain the three-dimensional relative position relationship.

[0015] In some embodiments, the performing attitude estimation based on the first IMU data and the second IMU data by using a multi-kernel entropy extended Kalman filter to obtain acceleration information comprises:

[0016] constructing a nonlinear state space model;

[0017] performing forward filtering and backward filtering on the first IMU data and the second IMU data based on the multi-kernel entropy extended Kalman filter, solving the nonlinear state space model, and obtaining the acceleration information.

[0018] In some embodiments, the multi-kernel entropy extended Kalman filter is constructed based on an extended Kalman filter, and the constructing the multi-kernel entropy extended Kalman filter comprises:

[0019] replacing a traditional mean square error with a multi-kernel correlation entropy to construct an optimization problem based on the extended Kalman filter; in the forward transmission and the backward transmission: performing whitening on a prediction covariance and an observation noise covariance through Cholesky decomposition; introducing a diagonal weight matrix, and adaptively weighting a prediction residual and an observation residual based on a kernel function; reconstructing a Kalman gain by kernel weighting to adjust the prediction covariance and the observation noise covariance; and adding a robust compensation term in updating a posteriori error covariance.

[0020] In some embodiments, the performing trajectory estimation according to the acceleration information to obtain the three-dimensional relative position relationship comprises:

[0021] performing velocity estimation based on the acceleration information by using a forward-backward Kalman smoothing strategy to obtain a velocity estimation value;

[0022] performing trajectory estimation according to the velocity estimation value to obtain a position estimation value;

[0023] calculating a three-dimensional relative position relationship of the target position relative to the initial position according to the position estimation value.

[0024] In some embodiments, the velocity estimation is performed based on the acceleration information by using a forward-backward Kalman smoothing strategy to obtain a velocity estimation value, including:

[0025] According to the first velocity of the initial position, the second velocity of the target position and the acceleration information, the velocity estimation is performed by using a forward Kalman smoothing to obtain a forward smoothing velocity, and the velocity estimation is performed by using a backward Kalman smoothing to obtain a backward smoothing velocity;

[0026] The forward smoothing velocity and the backward smoothing velocity are fused to obtain the velocity estimation value.

[0027] The second aspect of the present application discloses an exoskeleton control device, including:

[0028] The data acquisition module is configured to acquire IMU data collected during movement of the auxiliary walking device in response to the auxiliary walking device ending measurement, the auxiliary walking device being provided with an IMU inertial sensor, and the auxiliary walking device being configured to sequentially mark an initial position, a target position and then mark the initial position again during measurement.

[0029] The trajectory calculation module is configured to calculate a three-dimensional relative position relationship between the initial position and the target position based on the IMU data.

[0030] The control module is configured to send the three-dimensional relative position relationship to the exoskeleton to control the exoskeleton to move to the target position.

[0031] The third aspect of the present application discloses an electronic device including a memory storing executable program code and a processor coupled to the memory; the processor invokes the executable program code stored in the memory to execute the exoskeleton control method disclosed in the first aspect.

[0032] The fourth aspect of the present application discloses a computer readable storage medium storing a computer program, wherein the computer program causes a computer to execute the exoskeleton control method disclosed in the first aspect.

[0033] The present application has the beneficial effect that in a scenario requiring accurate trajectory estimation, the auxiliary walking device is used to sequentially mark an initial position, a target position and then mark the initial position again, and when the IMU data collected by the IMU inertial sensor installed on the auxiliary walking device is calculated, the influence of double integration can be eliminated, thereby improving the trajectory estimation accuracy of the exoskeleton and accurately controlling the movement of the exoskeleton. BRIEF DESCRIPTION OF DRAWINGS

[0034] The drawings here show the specific examples of the technical solutions of the present application, and constitute a part of the specification together with the specific embodiments, for explaining the technical solutions, principles and effects of the present application.

[0035] Unless specifically stated or defined otherwise, the same reference signs in different drawings represent the same or similar technical features, and different reference signs can also be used to represent the same or similar technical features.

[0036] Figure 1 is a flow chart of an exoskeleton control method disclosed by the embodiment of the present application;

[0037] Figure 2 is a schematic diagram of a three-dimensional motion estimation function module of the embodiment of the present application;

[0038] Figure 3 is a flow chart of three-dimensional motion estimation based on a multi-core entropy expansion Kalman filter of the embodiment of the present application;

[0039] Figure 4 is a schematic diagram of a probability density function of the embodiment of the present application;

[0040] Figure 5 is a structural schematic diagram of an exoskeleton control device of the embodiment of the present application;

[0041] Figure 6 is a structural schematic diagram of an electronic device of the embodiment of the present application. DETAILED DESCRIPTION

[0042] Unless specifically stated or defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the technical solutions of the present application belong. In the case of combining the technical solutions of the present application with the real scene, all technical and scientific terms used herein can also have the meaning corresponding to the purpose of implementing the technical solutions of the present application. The terms "first, second" used herein are only used for distinguishing the names, and do not represent the specific quantity or order. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0043] It should be noted that when an element is considered to be "fixed" to another element, it can be directly fixed to the other element or there can be a middle element; when an element is considered to be "connected" to another element, it can be directly connected to the other element or there can be a middle element; when an element is considered to be "mounted" to another element, it can be directly mounted to the other element or there can be a middle element. When an element is considered to be "provided" to another element, it can be directly provided to the other element or there can be a middle element.

[0044] Unless specifically stated or otherwise as can be apparent from the preceding discussion, it is intended that all features described herein are applicable to every embodiment of the application described herein, separately or in combination with one another. Moreover, the terms "comprise", "have" and "include" and variations thereof, as used in the specification and throughout the claims are intended to cover both the singular and plural unless the context otherwise requires. For example, a process, method, object, article, composition or apparatus that comprises a list of steps or elements is not necessarily limited to only those steps or elements but can include other steps or elements not expressly listed or inherent to such process, method, object, article, composition or apparatus.

[0045] The core process of trajectory calculation according to IMU data is to estimate the attitude, velocity and position information of the object by processing and integrating the accelerometer and gyroscope data, combined with error compensation and sensor fusion technology. The specific process is as follows: the IMU raw data usually contains noise and offset error, first, pre-processing such as filtering, normalization and denoising is carried out to improve the stability of subsequent processing; then, Kalman filter is used to calibrate acceleration and angular velocity data to eliminate sensor noise and drift, wherein Kalman filter can combine prior model and real-time observation to optimize the estimation result; then, the position is calculated by double integration of acceleration, and the attitude angle is obtained by integrating the angular velocity, and the trajectory estimation is carried out.

[0046] Since the attitude estimation and free acceleration estimation problems are coupled in the above calculation process, the position estimation is not accurate, and the trajectory estimation accuracy is not high, which will affect the walking safety in the scene where accurate trajectory estimation is required. Therefore, the present application can eliminate the influence of double integration and improve the accuracy of trajectory estimation in complex walking scenes by marking the initial position, the target position and then marking the initial position again.

[0047] In order to facilitate the understanding of the present application, the specific embodiments of the present application will be described in more detail below with reference to the accompanying drawings of the specification.

[0048] The embodiment of the present application discloses a kind of exoskeleton control method, can be realized by computer programming.The execution main body of this method can be embedded in the electronic equipment of auxiliary walking device or the control chip of exoskeleton, the present application does not limit this.

[0049] As Figure 1 As shown in the figure, the method comprises the following steps:

[0050] Step S100: in response to the end of measurement of the auxiliary walking device, the IMU data collected during the movement of the auxiliary walking device is acquired, the IMU inertial sensor is installed on the auxiliary walking device, and the auxiliary walking device marks the initial position, the target position and then marks the initial position in sequence during measurement;

[0051] The exoskeleton is matched with an auxiliary walking device, which is a walking stick in this embodiment, but is not limited to a walking stick. The end of the walking stick (the end close to the bottom surface) is provided with an IMU inertial sensor for collecting IMU data during movement of the walking stick. The IMU inertial sensor of this embodiment can collect three-axis acceleration, three-axis angular velocity and three-axis geomagnetic field data.

[0052] The initial position is the current foot position, and the target position is the desired foot position. The specific process of measurement using the walking stick is as follows:

[0053] A button is installed at the handle of the walking stick, and the button is pressed to start measurement. The initial position is clicked by the end of the walking stick, the target position is clicked by the end of the walking stick, and the initial position is clicked again by the end of the walking stick. Finally, the button is released to end the measurement. During movement of the walking stick, the collected IMU data is obtained from the IMU inertial sensor. It should be noted that the measurement process is not limited to using a button. Other sensors (such as force sensors) can also be used to determine the state as a clicking action when the data characteristics (such as accelerometer characteristics) meet certain conditions.

[0054] Step S200: calculating the three-dimensional relative position relationship between the initial position and the target position according to the IMU data;

[0055] Specifically, three-dimensional motion estimation is performed according to the IMU data to obtain the three-dimensional relative position relationship between the initial position and the target position.

[0056] The three-dimensional motion estimation is as shown in Figure 2 The calculation steps mainly include: (1) data interval segmentation: the IMU data is segmented, and the initial velocity and the terminal velocity of the segmented interval are zero. (2) attitude estimation: the attitude estimation is obtained using a smoother in the motion interval, and then the free acceleration is obtained. (3) trajectory estimation: the motion estimation is performed using a smoother.

[0057] Although the attitude estimation of the IMU is conceptually simple to describe, the attitude estimation problem of the IMU itself is a non-Gaussian nonlinear problem, and the traditional Gaussian-based assumption does not hold in motion estimation. Some methods give trajectory estimation, but do not give the distribution to which the trajectory and related acceleration, velocity and other information are subject. This is not conducive to understanding the accuracy of the current trajectory or further fusion with other sensor data (such as ultrasonic, ultra-wideband, or camera).

[0058] To solve the above problems, this embodiment proposes a multi-core entropy extended Kalman filter, which can effectively deal with non-Gaussian noise. The three-dimensional motion estimation based on the multi-core entropy extended Kalman filter is as shown in Figure 3 The specific steps include:

[0059] Step S210: dividing the IMU data into first IMU data corresponding to the movement from the initial position to the target position and second IMU data corresponding to the movement from the target position to the initial position;

[0060] In the IMU-based exoskeleton motion trajectory estimation, in order to improve the positioning accuracy between the initial position and the target position, the sensor data needs to be processed separately according to the movement direction characteristics.

[0061] According to the movement of the auxiliary walking device from the initial position to the target position and the movement back to the initial position, the continuous IMU data stream is divided into two categories: first IMU data and second IMU data.

[0062] Step S220: based on the first IMU data and the second IMU data, using a multi-kernel entropy extended Kalman filter to perform attitude estimation to obtain acceleration information;

[0063] The multi-kernel entropy extended Kalman filter is constructed based on the extended Kalman filter, and in the extended Kalman filter, the traditional mean square error is replaced by the multi-kernel correlation entropy to construct an optimization problem; in the forward transmission and backward transmission: the prediction covariance and the observation noise covariance are whitened through Cholesky decomposition; a diagonal weight matrix is introduced, and the prediction residual and the observation residual are adaptively weighted based on the kernel function; the Kalman gain is reconstructed by kernel weighting to adjust the prediction covariance and the observation noise covariance; a robust compensation term is added in the update of the posterior error covariance.

[0064] Specifically, the process of constructing the multi-kernel entropy extended Kalman filter is as follows:

[0065] The variables used in the subsequent formulas are defined as follows:

[0066] F k-1 is the state transition matrix, H k is the observation matrix, μ is the mean, R k is the observation noise covariance matrix, Q k is the process noise covariance matrix, P k|k is the prediction covariance matrix, is the forward Kalman gain, I is the identity matrix, B p,k , B r,k is the whitening matrix, is the diagonal weight matrix, is the adjusted Kalman gain, is the adjusted prediction covariance matrix, is the adjusted observation noise covariance matrix, D p,k and D q,k are lower triangular matrices, D k is a matrix obtained by Dp,k and D q,k a block diagonal matrix composed of is a kernel function, is an adjusted process noise covariance matrix, is a fused velocity, is a mean of velocities,

[0067] First, the IMU inertial sensor model includes a gyroscope model, an accelerometer model, and a magnetometer model. Among them, the gyroscope model is:

[0068] y G,k = w k + b k + v G,k (1)

[0069] where y G,k ∈ R 3 is the sensor reading at time step k, w k is the angular velocity, b k is the slowly varying bias, and v G,k is the noise.

[0070] The accelerometer model is:

[0071]

[0072] where y A,k ∈ R 3 is the sensor reading, S g k and E g = [0, 0, 9.81] T are the gravity vectors in the sensor coordinate system and the earth coordinate system, S a k is the free acceleration in the sensor coordinate system, is the direction from the earth coordinate system to the sensor coordinate system (in quaternion form), the operator R( s q k ) converts the quaternion to a rotation matrix, and v A,k is the noise.

[0073] The magnetometer model is:

[0074]

[0075] where y M,k is the sensor reading, S m k and E m is the geomagnetic vector in the sensor coordinate system and the earth coordinate system (since E m is a constant, the subscript k is omitted), Sd k is the magnetic disturbance on the sensor frame, v M,k is the noise.

[0076] quaternion of the current step can be obtained from the quaternion of the previous step and the current angular velocity, which is given by:

[0077]

[0078] where Δt is the sampling time, w q = [0, w T ] T is the quaternion of the true angular velocity w k In practical applications, y G,k is used to approximate w k , which gives the following equation:

[0079]

[0080] where, v G,k is the related noise due to the approximation of w k . Let Equation (5) can be generalized as:

[0081] x k = f(x k-1 , y G,k ) + v G,k (6)

[0082] where,

[0083] Similarly, equations (2) and (3) can be rewritten as:

[0084] y k = h(x k ) + v AM,k (7)

[0085] and

[0086]

[0087] where h A (x k ) = -R(x k ) E g, h M (x k ) = R(x k ) E m, v A,k = S a k + vA,k ,v M,k = S d k +v M,k .

[0088] Equations (6) and (7) provide a general nonlinear state-space model for the IMU. An extended Kalman filter (such as the Rauch-Tung-Striebel smoother: RTS smoother) is used for solving the problem.

[0089] Traditional orientation estimation algorithms heavily rely on the Gaussian assumption. This assumption is sensitive to gyroscope noise v. G,k It is possible, but due to the existence of unknown acceleration. S a k and magnetic disturbance S d k Therefore, the accelerometer noise v A,k and magnetometer noise v M,k Not valid. A,k and v M,k A better density representation would be a heavy-tailed distribution. To illustrate this, consider v. A,k and v M,k The probability density function (hereinafter referred to as pdf) was studied under both undisturbed and perturbed conditions: first, the IMU was kept stationary and undisturbed, and the corresponding v A,k and v M,k The probability density function is as follows Figure 4 Part a and Figure 4 As shown in section c. Then, external acceleration and magnetic disturbances are manually generated, and the corresponding results are as follows. Figure 4 Part b and Figure 4 As shown in part d. Unsurprisingly, when there is no disturbance, v A,k and v M,k The density of the magnetic field exhibits a Gaussian distribution, but with a thick tail when perturbed. Unfortunately, in many practical applications of IMUs, external acceleration and magnetic perturbations are unavoidable, which significantly degrades the performance of traditional algorithms.

[0090] Therefore, in order to adapt to non-Gaussian distributions, the multi-kernel entropy RTS smoother is first derived, and the specific process is as follows:

[0091] Consider linear problems:

[0092]

[0093] Where, x k ∈R n ,y k ∈R m The purpose is to estimate. With measurement set

[0094] In RTS smoother, it is assumed that the noise follows a Gaussian distribution, i.e. Moreover, let x 0|0 , w k and v k be mutually independent. Based on MAP decomposition, it can be derived that

[0095]

[0096] Rauch et al. proved that the above batch optimization method can be solved in an iterative way by forward pass optimization and backward pass optimization, as follows:

[0097]

[0098] where the forward pass optimization is shown as follows:

[0099] Given the initial guess and k = 1,..., N, perform:

[0100]

[0101] The backward pass optimization is shown as follows:

[0102] Initialize and k = N - 1,..., 0, perform:

[0103]

[0104] The calculation process of RTS smoother is as follows: the input is and the output is the state The forward pass process is as follows: for k = 1, 2,..., N, calculate and P k|k using formula (11). The backward pass process is as follows: for k = N - 1,..., 0, calculate

[0105] The existing RTS smoother is derived under the assumption that x0, w k and v k follow a Gaussian distribution. However, this is not always true in practical applications. Therefore, a MKCL (Multi-Kernel Correntropy Learning) based optimization problem is constructed as follows:

[0106]

[0107] where JCL (·) represents the multi-core correlation entropy. Analogous to the identity relationship between formulas (9) and (10), formula (13) can be solved iteratively using the following objective function:

[0108]

[0109] The solutions for formulas (14a) and (14b) are as follows:

[0110] Forward propagation process:

[0111] First, the state prediction is performed as follows:

[0112]

[0113] Then, consider the following equation:

[0114]

[0115] and

[0116]

[0117] Among them, B p,k and B r,k This can be obtained through Cholesky decomposition. Multiply both sides of equation (16) by B on the left. -1 ,get:

[0118]

[0119] and

[0120]

[0121] Then, consider the following basic MKCL optimization problem:

[0122]

[0123] in, yes The i-th element, yes The i-th row. Note that after some transformations, formula (20) is the same as formula (14a).

[0124] right Relative to x k Taking partial derivatives And set it to zero. Given:

[0125]

[0126] in, Equation (21) is a fixed-point function that can be solved by a fixed-point algorithm. Equation (21) can be written as:

[0127]

[0128] and

[0129]

[0130] Substitute the expression of w k into equation (22) to obtain:

[0131]

[0132] Then, using the matrix inversion lemma, we obtain

[0133] In addition,

[0134]

[0135] Substitute equation (25) and equation (26) into equation (22) and use to represent x k , to obtain:

[0136]

[0137] and

[0138]

[0139] Finally, update the posterior error covariance:

[0140]

[0141] The forward pass process of MKCRTS (Multi-kernel entropy Kalman filter) is summarized as follows:

[0142] Step 1: Initialization and and σ i , i = 1, …, n + m

[0143] Step 2: Perform the following equations for state prediction

[0144]

[0145] Calculate B p,k and B r,k by Cholesky decomposition.

[0146] Step 3: Let where x k|k,tThis represents the estimated state when t starts fixed-point iteration from 1, and then the following fixed-point algorithm is executed.

[0147]

[0148] and

[0149]

[0150] Until the predetermined stop threshold is reached

[0151]

[0152] Step 4: Update the posterior error covariance and return to step 2.

[0153]

[0154] Backpropagation process:

[0155] For k = N-1,…,0, consider the following equation:

[0156]

[0157] and

[0158]

[0159] Among them, D p,k and D q,k This can be obtained through Cholesky decomposition. Multiply both sides of equation (35) by D on the left. k We can obtain:

[0160]

[0161] in, and,

[0162]

[0163] Then, consider the following fundamental optimization problem of MKCL:

[0164]

[0165] in, yes The i-th element, yes The i-th row.

[0166] right Relative to x k Taking partial derivatives And set it to zero. Given:

[0167]

[0168] Note Equation (40) is a fixed-point function that can be solved by fixed-point algorithm. Equation (40) can be written as:

[0169]

[0170] and

[0171]

[0172] Substitute the expression of into equation (41), we get

[0173]

[0174] Then, using the matrix inversion lemma, we get

[0175]

[0176] In addition, we have

[0177]

[0178] Substitute equation (44) and equation (45) into equation (41), and let x k be expressed as We can get:

[0179]

[0180] and

[0181]

[0182] Based on the lemma (A+B) -1 = A -1 -A -1 B(A+B) -1 and proof:

[0183]

[0184] We can get:

[0185]

[0186] Where 1) is obtained by using the symmetry of and Therefore, we have:

[0187]

[0188] Where 2) again uses and The symmetry is obtained. Substituting equation (50) into equation (46), we have

[0189]

[0190] and,

[0191]

[0192] Finally, update the posterior error covariance, which is given by

[0193]

[0194] The backward pass of MKCRTS is summarized as follows:

[0195] Step 1: Initialization and σ i where i = 1,..., 2n

[0196] Step 2, k = N - 1,..., 0, first use Cholesky decomposition to obtain D k Let where x k|k,t denotes the estimated state at iteration t starting from 1. Then, the following equation is executed:

[0197]

[0198] and,

[0199]

[0200] until a predetermined stopping threshold

[0201]

[0202] Step 3: Update the posterior error covariance, go back to Step 2.

[0203]

[0204] The MKCRTS smoother can be summarized as the following pseudo code:

[0205] The input is The output is the state Forward pass:

[0206] Fork = 1, 2,..., N, do

[0207] Use equation (30) to make state prediction

[0208] While t < iter max and equation (33) is not satisfied do

[0209] Recursively use equations (31) and (32) for state update

[0210] t←t+1

[0211] End While

[0212] Update error covariance using equation (34)

[0213] End for

[0214] Backward pass:

[0215] For k = N - 1, N - 2,..., 0 do

[0216] While t≤iter max and equation (56) is not satisfied do

[0217] Update state using equations (54) and (55)

[0218] t←t+1

[0219] End While

[0220] Update error covariance using equation (57)

[0221] End for

[0222] Then, on top of MKCRTS, consider the following nonlinear state space dynamics:

[0223]

[0224] By first order linearization, the state transition matrix can be expressed as The observation matrix can be expressed as Moreover, the state prediction becomes x k|k-1 = f(x k-1|k-1 ), and the velocity estimate is y k = g(x k|k-1 ), using these substitutions, we can obtain MKCERTS (multi-kernel entropy extended RTS smoother), which is similar to MKCRTS (multi-kernel entropy RTS smoother), but differs in the following equations.

[0225] 1. In the forward pass, the state prediction of equation (30) can be replaced by

[0226]

[0227] where,

[0228] 2. In the forward pass, the update equation of equation (31) is replaced by

[0229]

[0230] In addition, in equation (32) use and the last line of equation (32) is replaced by:

[0231]

[0232] 3. In the backward pass, the last line of equation (55) is replaced by equation (62).

[0233]

[0234] The MKCERTS smoother can be summarized in the following pseudo code:

[0235] The input is The output is the state

[0236] Forward pass:

[0237] For k = 1,2…,N do

[0238] Predict the state using equation (59)

[0239] While t < iter max and equation (33) is not satisfied do

[0240] Update the state using equation (60) and equation (32), but replace the last line of equation (32) by equation (61)

[0241] t <— t + 1

[0242] End While

[0243] Update the error covariance using equation (34)

[0244] End for

[0245] Backward pass:

[0246] For k = N-1,N-2…,0 do

[0247] While t < iter max and equation (56) is not satisfied do

[0248] Update the state using equation (54) and equation (55), but replace the last line of equation (55) by equation (62)

[0249] t <— t + 1

[0250] End While

[0251] Update the error covariance with formula (57)

[0252] End for

[0253] On the basis of the above-mentioned multi-core entropy extended Kalman filter is constructed, the multi-core entropy extended Kalman filter is used for attitude estimation, and the process of obtaining acceleration information includes: constructing a nonlinear state space model; based on the multi-core entropy extended Kalman filter, both the first IMU data and the second IMU data are subjected to forward filtering and backward filtering, the nonlinear state space model is solved, and acceleration information is obtained.

[0254] Step S230: trajectory estimation is performed according to the acceleration information, and a three-dimensional relative position relationship is obtained.

[0255] A forward-backward smoother for velocity and position estimation is used in trajectory estimation. Based on the acceleration information, the forward-backward smoother is used to estimate the velocity by using the forward-backward Kalman smoothing strategy, and a velocity estimation value is obtained; the trajectory is estimated according to the velocity estimation value, and a position estimation value is obtained; and the three-dimensional relative position relationship of the target position relative to the initial position is calculated according to the position estimation value.

[0256] Specifically, first, some useful lemmas are proposed as follows:

[0257] Lemma 1 can obtain smoothed state estimation under the forward-backward smoothing framework in the following way:

[0258]

[0259] And

[0260]

[0261] Where P f,k is the forward-passed error covariance, P b,k is the backward-passed error covariance

[0262] Lemma 2 considers the velocity estimation problem:

[0263]

[0264] Where v k is the velocity, a k-1 is the measured acceleration, is the ground true acceleration, ΔT is the sampling time, is white noise, and the initial velocity v1 is assumed to be accurately known. Then there is

[0265]

[0266] Based on equation (65), we have

[0267]

[0268] where and k = (N - 1)AT 2 R a The derivation in equation (1) exploits the affine property of Gaussian distribution and the summation property Thus the proof is completed.

[0269] Based on equation (65), given the initial velocity v1(i.e., the first velocity at the start position) and the final velocity v N (i.e., the second velocity at the target position), as well as a series of measured accelerations (i.e., the acceleration information), the velocity estimation is performed using forward Kalman smoothing to obtain the forward smoothed velocity, using backward Kalman smoothing to obtain the backward smoothed velocity, and then fusing the forward smoothed velocity and the backward smoothed velocity to obtain the velocity estimate.

[0270] Specifically, performing the forward pass of equation (65), we have

[0271]

[0272] where

[0273] Performing the backward pass of equation (65) (i.e., v k-1 = v K - a k-1 AT), we have

[0274]

[0275] where Then, based on Lemma 1, fusing the states we have

[0276]

[0277] and

[0278]

[0279] It can be observed that this velocity smoother is the same as the zero-velocity update (ZUPT) strategy but with an extra error covariance. Accordingly, the position can be obtained by the following equation (assuming P1is known)

[0280]

[0281] and

[0282]

[0283] From the above, the embodiment first gives the IMU inertial sensor model, which shows that the accelerometer and geomagnetic meter data are heavy tails in the case of disturbance. Secondly, the entropy-based RTS smoother is derived, which can effectively deal with non-Gaussian noise, improve the accuracy of attitude estimation, and further obtain more accurate free acceleration estimation. Finally, the forward-backward Kalman smoother is used for velocity estimation. The forward-backward velocity estimator used is proved to be equivalent to ZUPT (zero velocity update), but can provide the corresponding velocity error covariance, and by integrating the velocity, the probability-based 3D trajectory estimation is finally obtained.

[0284] Step S300: controlling the exoskeleton to move to the target position according to the three-dimensional relative position relationship.

[0285] After obtaining the three-dimensional relative position relationship, the exoskeleton can determine the foot movement trajectory of the exoskeleton according to the three-dimensional relative position relationship, and then control the exoskeleton to move to the target position according to the foot movement trajectory. It should be noted that the foot movement trajectory of the exoskeleton can also be calculated on the electronic device of the walking aid according to the three-dimensional relative position relationship, and then the foot movement trajectory is sent to the control chip to control the exoskeleton to move to the target position.

[0286] In summary, the embodiment obtains the IMU attitude estimation through the gyroscope, accelerometer, and geomagnetic meter data. The multi-core entropy extended Kalman filter is used in the segmented interval to obtain the IMU attitude estimation and free acceleration. The forward-backward Kalman smoothing strategy is used to obtain the motion velocity, and then the velocity is integrated to obtain the motion estimation based on the probability distribution.

[0287] As shown in Figure 5 Based on the above exoskeleton control method, the embodiment of the application discloses an exoskeleton control device, which comprises:

[0288] The data acquisition module 600 is configured to, in response to the end of measurement of the auxiliary walking device, acquire IMU data collected during movement of the auxiliary walking device, wherein the auxiliary walking device is provided with an IMU inertial sensor, and the auxiliary walking device is configured to sequentially mark an initial position, a target position, and then mark the initial position again during measurement.

[0289] The trajectory calculation module 610 is configured to calculate a three-dimensional relative position relationship between the initial position and the target position according to the IMU data.

[0290] The control module 620 is configured to control the exoskeleton to move to the target position according to the three-dimensional relative position relationship.

[0291] As shown in Figure 6 An electronic device is disclosed, which comprises a memory 401 storing executable program codes and a processor 402 coupled with the memory 401.

[0292] The processor 402 invokes the executable program codes stored in the memory 401 to execute the exoskeleton control method described in each of the above embodiments.

[0293] The embodiment of the present application also discloses a computer readable storage medium storing a computer program, wherein the computer program enables a computer to execute the exoskeleton control method described in each of the above embodiments.

[0294] The above embodiments are intended to exemplarily reproduce and deduce the technical solutions of the present application, and to completely describe the technical solutions, objects and effects of the present application, so as to make the public more thoroughly and comprehensively understand the disclosed content of the present application, and not to limit the protection scope of the present application.

[0295] The above embodiments are not based on the exhaustive enumeration of the present application, and there can be a plurality of other embodiments not listed. Any replacement and improvement made without violating the concept of the present application shall fall within the protection scope of the present application.

Claims

1. An exoskeleton control method, characterized by, The method comprises the following steps: in response to the end of the measurement of the walking aid, acquiring IMU data collected during the movement of the walking aid, the walking aid being provided with an IMU inertial sensor, and the walking aid marking an initial position, a target position and the initial position again in sequence during the measurement; calculating a three-dimensional relative position relationship between the initial position and the target position according to the IMU data; controlling the exoskeleton to move to the target position according to the three-dimensional relative position relationship; the calculation of the three-dimensional relative position relationship between the initial position and the target position according to the IMU data comprises: dividing the IMU data into first IMU data corresponding to the movement from the initial position to the target position and second IMU data corresponding to the movement from the target position to the initial position; based on the first IMU data and the second IMU data, performing attitude estimation by using a multi-core entropy extended Kalman filter to obtain acceleration information; performing trajectory estimation according to the acceleration information to obtain the three-dimensional relative position relationship; the attitude estimation by using the multi-core entropy extended Kalman filter based on the first IMU data and the second IMU data to obtain the acceleration information comprises: constructing a nonlinear state space model; based on the multi-core entropy extended Kalman filter, performing forward filtering and reverse filtering on the first IMU data and the second IMU data, solving the nonlinear state space model, and obtaining the acceleration information; the multi-core entropy extended Kalman filter is constructed based on an extended Kalman filter, and when the multi-core entropy extended Kalman filter is constructed, the following steps are included: based on the extended Kalman filter, a multi-core correlation entropy is used to replace a traditional mean square error to construct an optimization problem; during the forward transmission and the backward transmission: the prediction covariance and the observation noise covariance are whitened by Cholesky decomposition; a diagonal weight matrix is introduced, and the prediction residual and the observation residual are adaptively weighted based on a kernel function; the prediction covariance and the observation noise covariance are adjusted by kernel weighting to reconstruct the Kalman gain; a robust compensation term is added during the update of the posterior error covariance.

2. The exoskeleton control method of claim 1, wherein, the walking aid is a walking stick, and the process of the measurement of the walking aid comprises: starting the measurement, clicking the initial position with the end of the walking stick, clicking the target position with the end of the walking stick, and finally clicking the initial position with the end of the walking stick to end the measurement.

3. The exoskeleton control method of claim 1, wherein, the trajectory estimation according to the acceleration information to obtain the three-dimensional relative position relationship comprises: based on the acceleration information, performing velocity estimation by using a forward-backward Kalman smoothing strategy to obtain a velocity estimation value; performing trajectory estimation according to the velocity estimation value to obtain a position estimation value; calculating the three-dimensional relative position relationship of the target position relative to the initial position according to the position estimation value.

4. The exoskeleton control method of claim 3, wherein, the velocity estimation based on the acceleration information by using the forward-backward Kalman smoothing strategy to obtain the velocity estimation value comprises: performing velocity estimation by using forward Kalman smoothing to obtain a forward smoothing velocity and performing velocity estimation by using backward Kalman smoothing to obtain a backward smoothing velocity according to a first velocity of the initial position, a second velocity of the target position and the acceleration information; fusing the forward smoothed velocity and the backward smoothed velocity to obtain the velocity estimate.

5. An exoskeleton control device characterized by comprising: Comprise: A data acquisition module for acquiring IMU data collected during movement of an auxiliary walking device in response to the auxiliary walking device ending measurement, an IMU inertial sensor being installed on the auxiliary walking device, the auxiliary walking device sequentially marking an initial position, a target position and then marking the initial position again during measurement; A trajectory solving module for calculating a three-dimensional relative position relationship between the initial position and the target position according to the IMU data; A control module for controlling an exoskeleton to move to the target position according to the three-dimensional relative position relationship; The calculation of the three-dimensional relative position relationship between the initial position and the target position according to the IMU data comprises: Dividing the IMU data into first IMU data corresponding to the initial position to the target position and second IMU data corresponding to the target position to the initial position; Based on the first IMU data and the second IMU data, using a multi-core entropy extended Kalman filter to perform attitude estimation to obtain acceleration information; According to the acceleration information, trajectory estimation is performed to obtain the three-dimensional relative position relationship; The multi-core entropy extended Kalman filter is constructed based on an extended Kalman filter, and when constructing the multi-core entropy extended Kalman filter, it comprises: On the basis of the extended Kalman filter, the traditional mean square error is replaced by the multi-core related entropy to construct an optimization problem; in forward transmission and backward transmission: the prediction covariance and the observation noise covariance are whitened by Cholesky decomposition; a diagonal weight matrix is introduced, and the prediction residual and the observation residual are adaptively weighted based on the kernel function; the Kalman gain is reconstructed by kernel weighting adjustment of the prediction covariance and the observation noise covariance; a robust compensation term is added when updating the posterior error covariance. The computer readable storage medium stores a computer program, wherein the computer program causes the computer to execute the exoskeleton control method of any one of claims 1 to 4. The computer readable storage medium stores a computer program, wherein the computer program causes the computer to execute the exoskeleton control method of any one of claims 1 to 4. ​ 6. An electronic device, characterized by ​ 7. A computer readable storage medium characterized by, ​

Images