A method, device and apparatus for estimating torque of a series viscoelastic actuator

Abstract

The present application discloses a torque estimation method, device and equipment for a series viscoelastic actuator, the method comprising: obtaining historical operating parameters of the series viscoelastic actuator; establishing an initial actuator torque estimation model based on a GMM model according to the historical operating parameters; optimizing the initial actuator torque estimation model using an EM algorithm to obtain an optimized actuator torque estimation model; obtaining real-time deformation data of the series viscoelastic actuator, inputting the deformation data into the fully trained actuator torque estimation model to obtain a torque estimation value. The present invention uses several Gaussian distributions to fit arbitrary data, and based on historical data, models the relationship between the deformation and output torque of the viscoelastic body. The parameters of the initial model are iteratively updated through the EM algorithm, effectively avoiding the influence of nonlinear effects on the torque estimation of the series viscoelastic actuator, and improving the accuracy of the torque estimation of the series viscoelastic body.

Description

Technical Field

[0001] The present invention relates to the technical field of serial viscoelastic actuators, and in particular to a torque estimation method, device, electronic equipment and computer-readable storage medium for serial viscoelastic actuators. Background Art

[0002] With the rapid development of science and technology, a variety of robots have emerged, and most of the application scenarios of robots require interaction with the external environment, such as walking, avoiding obstacles, grabbing soft objects, etc. In order to avoid damage to the robot or the interactive object due to excessive interaction force and ensure the safety of the robot in the process of interaction with the external environment, it is necessary to improve the flexibility between the robot and the environment; improving the torque perception ability of the robot is the basis for ensuring the friendliness and safety of robot interaction.

[0003] At present, the method to achieve compliant interaction between robots and the environment is usually to add elastic elements between the end of the robot actuator and the external environment. This type of drive is usually called a series elastic actuator (SEA). SEA has the advantages of low output impedance and convenient torque measurement, but the residual inertia generated by the elastic element and the robot end effector will cause SEA to produce undesirable vibration effects in high-frequency external environments. Replacing the elastic element in SEA with a viscoelastic body that has both elastic and damping links can effectively suppress the high-frequency oscillation caused by the impact of SEA. This type of drive is usually called a series viscoelastic actuator (SVA). SVA has the advantages of small size, light weight, and good flexibility. It is widely used in tactile, medical, collaborative robots and other fields. However, the viscoelastic body in SVA is easily affected by nonlinear factors such as stress relaxation, strain creep, and Mullins. The relationship between its deformation and torque is not a simple linear relationship, which seriously affects the accuracy of SVA torque estimation.

[0004] Therefore, it is necessary to provide a torque estimation method for a series viscoelastic actuator, which can simulate the relationship between the deformation of the viscoelastic body in the SVA and the output torque of the SVA, and accurately estimate the output torque of the SVA according to the deformation of the viscoelastic body in the SVA. Summary of the invention

[0005] In view of this, it is necessary to provide a torque estimation method, device and equipment for a series viscoelastic actuator to solve the problem in the prior art that there is no suitable model to analyze the deformation of the viscoelastic body in the series viscoelastic actuator and the output torque of the actuator, and the output torque of the actuator cannot be accurately estimated based on the deformation of the viscoelastic body.

[0006] In order to solve the above problems, the present invention provides a torque estimation method for a series viscoelastic actuator, comprising:

[0007] Obtaining historical operating parameters of the series viscoelastic actuator;

[0008] According to the historical operating parameters, an initial actuator torque estimation model is established based on the GMM model;

[0009] Optimizing the initial actuator torque estimation model using an EM algorithm to obtain an optimized actuator torque estimation model;

[0010] The real-time deformation data of the series viscoelastic actuator is acquired, and the deformation data is input into the optimized actuator torque estimation model to obtain a torque estimation value.

[0011] Furthermore, the historical operating parameters include: a historical deformation amount of the viscoelastic body in the actuator and an output torque of the actuator corresponding to the historical deformation amount.

[0012] Furthermore, according to the historical operating parameters, an initial actuator torque estimation model is established based on the GMM model, including:

[0013] Determining model parameters of the initial actuator torque estimation model according to the historical operating parameters, and establishing the initial actuator torque estimation model based on the model parameters and the GMM model;

[0014] Wherein, the mathematical expression of the initial actuator torque estimation model is:

[0015]

[0016] Among them, K means there are K Gaussian distribution models, α k Indicates the probability that the current sample x belongs to the kth Gaussian distribution, represents the probability density function of the kth Gaussian distribution, θ k =(μ k ,Σ k ) represents the kth Gaussian distribution, μ k represents the mean corresponding to the kth Gaussian distribution, Σ k Represents the covariance corresponding to the k-th Gaussian distribution.

[0017] Furthermore, the initial actuator torque estimation model is optimized using the EM algorithm to obtain an optimized actuator torque estimation model, including:

[0018] Optimizing the initial model parameters in the initial actuator torque estimation model using the EM algorithm to obtain optimized model parameters, and obtaining the optimized actuator torque estimation model based on the optimized model parameters;

[0019] The model parameters include the probability α of the kth Gaussian distributionk , the mean μ corresponding to the kth Gaussian distribution k The covariance Σ corresponding to the k-th Gaussian distribution k .

[0020] Further, the initial model parameters in the initial actuator torque estimation model are optimized using the EM algorithm to obtain optimized model parameters, including:

[0021] According to the initial actuator torque estimation model, a maximum likelihood function in the EM algorithm is obtained;

[0022] Calculating a partial derivative of the maximum likelihood function, and updating the initial model parameters according to the partial derivative result of the maximum likelihood function;

[0023] When the EM algorithm converges, optimized model parameters are obtained.

[0024] Furthermore, according to the initial actuator torque estimation model, a maximum likelihood function in the EM algorithm is obtained, including:

[0025] The mathematical expression of the maximum likelihood function in the EM algorithm is:

[0026]

[0027] Among them, X is the observed data, Z is the latent variable, P(Z|X,θ (t) ) is given the observed data X and the current parameter estimate θ (t) The conditional probability distribution of the latent variable Z; P(X,Z|θ) is the conditional probability of X and Z occurring simultaneously when the model parameter θ is known; Q(θ,θ(t+1)) is the parameter estimation function at the next moment.

[0028] Further, the partial derivative of the maximum likelihood function is obtained, and the initial model parameters are updated according to the partial derivative result of the maximum likelihood function, including:

[0029] Take the partial derivative of the maximum function and transform Q(θ,θ (t) ) is the maximum value of the parameter θ (t+1) As the model parameters at time t+1, the updated model parameters are obtained.

[0030] The present invention also provides a torque estimation device for a series viscoelastic actuator, comprising:

[0031] A data acquisition module, used for acquiring historical operating parameters of the series viscoelastic actuator;

[0032] A model building module, used for building an initial actuator torque estimation model based on the GMM model according to the historical operating parameters;

[0033] A model training module, used for optimizing the initial actuator torque estimation model by using an EM algorithm to obtain an optimized actuator torque estimation model;

[0034] The torque estimation module is used to obtain real-time deformation data of the series viscoelastic actuator, and input the deformation data into the optimized actuator torque estimation model to obtain a torque estimation value.

[0035] The present invention also provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, a torque estimation method for a series viscoelastic actuator as described in any of the above technical solutions is implemented.

[0036] The present invention also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the torque estimation method of a series viscoelastic actuator described in any of the above technical solutions is implemented.

[0037] Compared with the prior art, the beneficial effects of the present invention include: first, obtaining the historical operation data of the series viscoelastic actuator, and establishing an initial actuator torque estimation model based on the historical operation data and the GMM model; second, optimizing the initial actuator torque estimation model using the EM algorithm to obtain an optimized actuator torque estimation model; finally, obtaining the real-time deformation data of the actuator, and obtaining the torque estimation value corresponding to the real-time deformation data using the fully trained actuator torque estimation model. The present invention establishes an initial torque estimation model based on the GMM model, uses several Gaussian distributions to fit arbitrary data, does not consider the difference in the initial conditions of the viscoelastic material, and models the relationship between the deformation and output torque of the viscoelastic body based on historical data. The parameters of the initial model are iteratively updated using the EM algorithm to obtain an optimized torque estimation model. The GMM model based on the EM algorithm combines the parameter optimization algorithm with the hybrid model, providing a new modeling method for scientific research and practical engineering. By training a complete torque estimation model, the relationship between the deformation and torque output of the series viscoelastic actuator was analyzed, which effectively avoided the influence of nonlinear effects such as hysteresis, stress relaxation and strain creep effect of the viscoelastic body on the torque estimation of the series viscoelastic actuator, improved the accuracy of the torque estimation of the series viscoelastic body, and provided a theoretical basis for realizing the compliant interaction between SVA and the external environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic structural diagram of an embodiment of a series viscoelastic actuator provided by the present invention;

[0039] Figure 2A schematic diagram of an embodiment of a control process of a series viscoelastic actuator provided by the present invention;

[0040] Figure 3 A schematic flow chart of an embodiment of a torque estimation method for a series viscoelastic actuator provided by the present invention;

[0041] Figure 4 A schematic structural diagram of an embodiment of an experimental bench for a series viscoelastic actuator provided by the present invention;

[0042] Figure 5 A schematic structural diagram of an embodiment of a torque estimation system provided by the present invention;

[0043] Figure 6 A comparative analysis diagram of an embodiment of the estimated torque and the real torque provided by the present invention;

[0044] Figure 7 A schematic structural diagram of an embodiment of a torque estimation device for a series viscoelastic actuator provided by the present invention;

[0045] Figure 8 A schematic structural diagram of an electronic device according to an embodiment of the present invention;

[0046] In the figure, 1-driving motor, 2-harmonic reducer, 3-rubber body, 4-upper surface orthogonal photoelectric encoder, 5-lower surface orthogonal photoelectric encoder, 6-transmission gear, 7-series viscoelastic actuator, 8-connector, 9-dynamic torque sensor, 10-magnetic powder brake. DETAILED DESCRIPTION

[0047] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0048] Before describing the embodiments, the relevant terms are explained:

[0049] Series viscoelastic actuator (SVA): In order to reduce the stiffness of the robot, a viscoelastic body with elastic and damping links is added between the end of the robot actuator and the external environment. This type of actuator is usually called a series viscoelastic actuator (SVA). SVA has the advantages of small size, light weight, good flexibility, etc., and is widely used in tactile, medical, collaborative robots and other fields.

[0050] GMM model: Gaussian mixture model, which uses Gaussian probability density function (normal distribution curve) to accurately quantify things and decomposes one thing into several models based on Gaussian probability density function (normal distribution curve). GMM model is widely used in numerical approximation, speech recognition, image classification, image denoising, image reconstruction, fault diagnosis, video analysis, email filtering, density estimation, target recognition and tracking and other fields.

[0051] EM algorithm: Maximum expectation algorithm is a type of optimization algorithm that performs maximum likelihood estimation (MLE) through iteration. The standard computational framework of the EM algorithm consists of alternating E-step (Expectation-step) and M-step (Maximization step). The convergence of the algorithm can ensure that the iteration at least approaches the local maximum. The EM algorithm and its improved versions are often used to solve the parameters of machine learning algorithms.

[0052] The structural diagram of the series viscoelastic actuator (SVA) involved in this application is as follows Figure 1 As shown, the series viscoelastic actuator includes a driving motor 1, a harmonic reducer 2, a rubber body 3, an upper surface orthogonal photoelectric encoder 4, a lower surface orthogonal photoelectric encoder 5 and a transmission gear 6. The output end of the lower surface of the rubber body 3 is fixed on the output shaft of the actuator, and the input end of the rubber body 3 is bonded to the transmission gear 6. The torque generated by the driving motor 1 is amplified by the harmonic reducer 2, and then squeezes the rubber body 3 through the transmission gear 6 to produce deformation. The upper surface orthogonal photoelectric encoder 4 and the lower surface orthogonal photoelectric encoder 5 respectively measure the angle difference between the upper and lower surfaces of the rubber body 3 as the deformation amount of the rubber body 3.

[0053] The control flow diagram of the above-mentioned series viscoelastic actuator (SVA) is as follows: Figure 2 As shown: the host computer sends the position, speed, and torque instructions of the desired SVA output to the main control chip through CAN communication. The main control chip uses the STM32 single-chip microcomputer; the main control chip converts the above instructions into current and applies it to the driver; the driver generates the corresponding vector voltage through the FOC algorithm to control the movement of the drive motor in the SVA, and the drive motor drives the harmonic reducer to rotate. The harmonic reducer amplifies the output torque of the drive motor and transmits the torque to the upper surface of the rubber body (i.e., the viscoelastic body) through the gears, squeezing the rubber to deform. The deformation θ of the upper and lower surfaces of the rubber body can be measured by the photoelectric orthogonal encoder fixed on the upper and lower surfaces of the rubber body b and θ a , and θ a As the deformation of SVA. In addition, the driver can also sample the three-phase current of the drive motor in real time, and feed the sampling results back to the main control chip to achieve closed-loop control.

[0054] The embodiment of the present invention provides a torque estimation method for a series viscoelastic actuator. Figure 3 A schematic flow chart of an embodiment of a method for estimating torque of a series viscoelastic actuator provided by the present invention includes:

[0055] Step S101: obtaining historical operating parameters of the series viscoelastic actuator;

[0056] Step S102: establishing an initial actuator torque estimation model based on the GMM model according to the historical operating parameters;

[0057] Step S103: optimizing the initial actuator torque estimation model using the EM algorithm to obtain an optimized actuator torque estimation model;

[0058] Step S104: acquiring real-time deformation data of the series viscoelastic actuator, and inputting the deformation data into the optimized actuator torque estimation model to obtain a torque estimation value.

[0059] The torque estimation method of the series viscoelastic actuator provided in this embodiment first obtains the historical operation data of the series viscoelastic actuator, and establishes the initial actuator torque estimation model based on the historical operation data and the GMM model; secondly, the initial actuator torque estimation model is optimized by using the EM algorithm to obtain the optimized actuator torque estimation model; finally, the real-time deformation data of the actuator is obtained, and the torque estimation value corresponding to the real-time deformation data is obtained by using the fully trained actuator torque estimation model. This embodiment establishes the initial torque estimation model based on the GMM model, uses several Gaussian distributions to fit arbitrary data, does not consider the difference in the initial conditions of the viscoelastic material, and models the relationship between the deformation and output torque of the viscoelastic body based on historical data. The parameters of the initial model are iteratively updated by the EM algorithm to obtain the optimized torque estimation model. The GMM model based on the EM algorithm combines the parameter optimization algorithm with the hybrid model, providing a new modeling method for scientific research and practical engineering. By training a complete torque estimation model, the relationship between the deformation variable and torque output of the series viscoelastic actuator is analyzed. The output torque can be estimated more accurately according to the deformation variable of the actuator, which provides a theoretical basis for achieving compliant interaction between SVA and the external environment.

[0060] As a preferred embodiment, in step S101, the historical operating parameters include: the historical deformation amount of the viscoelastic body in the actuator and the output torque of the actuator corresponding to the historical deformation amount.

[0061] As a specific embodiment, in order to obtain the historical operating parameters of the series viscoelastic actuator, a Figure 4The test bench shown is used to measure the operating parameters. Figure 4 In the experiment, the dynamic torque sensor 9 is used to measure the magnitude of the interaction force between the series viscoelastic actuator 7 and the external environment; the magnetic powder brake 10 is used as the load when the motor is working. Through the experimental bench, the actual data of the viscoelastic body shape variable and output torque in the series viscoelastic actuator are obtained, providing a data basis for subsequent modeling and analysis.

[0062] As a preferred embodiment, in step S102, according to the historical operating parameters, an initial actuator torque estimation model is established based on the GMM model, including:

[0063] Determining model parameters of the initial actuator torque estimation model according to the historical operating parameters, and establishing the initial actuator torque estimation model based on the model parameters and the GMM model;

[0064] Wherein, the mathematical expression of the initial actuator torque estimation model is:

[0065]

[0066] Among them, K means there are K Gaussian distribution models, α k Indicates the probability that the current sample x belongs to the kth Gaussian distribution, represents the probability density function of the kth Gaussian distribution, θ k =(μ k ,Σ k ) represents the kth Gaussian distribution, μ k represents the mean corresponding to the kth Gaussian distribution, Σ k Represents the covariance corresponding to the k-th Gaussian distribution.

[0067] As a preferred embodiment, the initial actuator torque estimation model is optimized using the EM algorithm to obtain an optimized actuator torque estimation model, including:

[0068] Optimizing the initial model parameters in the initial actuator torque estimation model using the EM algorithm to obtain optimized model parameters, and obtaining the optimized actuator torque estimation model based on the optimized model parameters;

[0069] The model parameters include the probability α of the kth Gaussian distribution k , the mean μ corresponding to the kth Gaussian distribution k The covariance Σ corresponding to the k-th Gaussian distribution k .

[0070] As a specific embodiment, GMM can be regarded as a linear superposition of several Gaussian mixture models; EM is an iterative algorithm and a powerful tool for parameter estimation. Parameter estimation of the GMM model is an important application of the EM algorithm.

[0071] As a preferred embodiment, in step S103, the initial model parameters in the initial actuator torque estimation model are optimized using the EM algorithm to obtain optimized model parameters, including:

[0072] According to the initial actuator torque estimation model, a maximum likelihood function in the EM algorithm is obtained;

[0073] Calculating a partial derivative of the maximum likelihood function, and updating the initial model parameters according to the partial derivative result of the maximum likelihood function;

[0074] When the EM algorithm converges, optimized model parameters are obtained.

[0075] As a preferred embodiment, according to the initial actuator torque estimation model, a maximum likelihood function in the EM algorithm is obtained, including:

[0076] The mathematical expression of the maximum likelihood function in the EM algorithm is:

[0077]

[0078] Among them, X is the observed data, Z is the latent variable, P(Z|X,θ (t) ) is given the observed data X and the current parameter estimate θ (t) The conditional probability distribution of the latent variable Z; P(X,Z|θ) is the conditional probability of X and Z occurring simultaneously when the model parameter θ is known; Q(θ,θ(t+1)) is the parameter estimation function at the next moment.

[0079] As a preferred embodiment, the partial derivative of the maximum likelihood function is obtained, and the initial model parameters are updated according to the partial derivative result of the maximum likelihood function, including:

[0080] Take the partial derivative of the maximum function and transform Q(θ,θ (t) ) is the maximum value of the parameter θ (t+1) As the model parameters at time t+1, the updated model parameters are obtained.

[0081] The process of optimizing the model parameters by using the EM algorithm is described in detail below with reference to a specific embodiment.

[0082] As a specific embodiment, the iteration of the EM algorithm includes two steps: E-step and M-step. The process of optimizing the parameters of the above-mentioned actuator torque estimation model through the EM algorithm is as follows: first, initialize the model parameters such as the mean and covariance of each Gaussian distribution in the torque estimation model, and calculate the maximum likelihood function of the E-step in the EM based on the initialized model parameters; then, calculate the partial derivative of the maximum likelihood function of the E-step, and update the M-step in the EM algorithm; finally, determine whether the EM algorithm converges. If it converges, the optimized model parameters of the actuator torque estimation model can be obtained; if it does not converge, repeat the update of the E-step and the M-step until the function converges.

[0083] Specifically, assuming that the observed variable data is X, the latent variable is Z, the joint distribution probability is P(X, Z|θ), the conditional distribution probability is P(Z|X,θ), and the GMM parameter is θ, then:

[0084] The E step of the EM algorithm is as follows:

[0085]

[0086] Among them, P(Z|X,θ(t)) is the conditional probability distribution of the latent variable Z given the observed data X and the current parameter estimate θ(t), and P(X,Z|θ) is the conditional probability of X and Z occurring simultaneously when the model parameter θ is known. Q(θ,θ(t+1)) is the parameter estimation function at the next moment or the Q function.

[0087] The M step of the EM algorithm is to find the partial derivative of the estimated parameter θ(t) of the Q function at time t, and take the maximum value θ(t+1) of the parameter in the Q function at time t as the model parameter at time t+1. The formula is:

[0088]

[0089] Repeat the above steps E and M until the EM algorithm converges or meets the conditions, and then the parameters estimated by the EM algorithm can be obtained.

[0090] The Q function is Under the condition of k , μ k and Σ k By taking partial derivatives, we can obtain the parameter values ​​of the actuator torque estimation model in each iteration of the EM algorithm:

[0091]

[0092]

[0093]

[0094] where γ ik Represents the responsiveness, which represents the sample x i Probability from the kth model:

[0095]

[0096] As a specific embodiment, since the estimation of GMM parameters is related to the initial value setting of α, μ and Σ in the model, repeated experiments show that the final estimated model parameters are indeed different with different initial value settings. In order to solve this problem, when estimating the parameters of the moment model, α, μ and Σ are set to 1 / K, random values ​​and unit matrix respectively, where K is the number of Gaussian models. The average value of the results of several repeated experiments is taken as the final estimated output parameters of the moment model, thereby eliminating the interference of accidental factors.

[0097] In order to verify the effect of the above method, a Figure 5 The torque estimation system shown. The system is mainly composed of three parts: a driving module 501, a synchronous data acquisition module 502 and a data analysis and processing module 503. The driving module 501 is composed of a motor and a reducer, and is mainly used to provide a power source for the system. The synchronous data acquisition module 502 is composed of a position sensor and a torque sensor, which collects the rubber body deformation and torque data in real time through the sensor, thereby obtaining a hysteresis curve between the real deformation of the rubber body and the real output torque. The data analysis and processing module 503 mainly analyzes the hysteresis curve of the rubber body deformation and torque through the torque estimation model established based on the GMM model, and compares and analyzes the estimated torque and the actual output torque, thereby verifying the feasibility of the above scheme.

[0098] Use during verification Figure 4 The rubber body torque measurement experiment was carried out on the experimental platform shown in the figure: First, a sinusoidal torque input signal with a frequency of 400Hz and an amplitude of 4N·m was applied to the motor. During the test, the right side of the rubber body was fixed on the static torque sensor, and the torque sensor was used to measure the actual output torque of the rubber. At the same time, the torque estimated by the torque estimation model was used as the estimated torque for a comparative experiment. Finally, the experimental results were analyzed. The test results are shown in the figure. Figure 6 As shown, from Figure 6 It can be seen that the estimated torque is basically consistent with the true torque, which verifies the feasibility of the rubber body torque method obtained by the actuator torque estimation model.

[0099] This embodiment also provides a torque estimation device for a series viscoelastic actuator, and its structural block diagram is as follows: Figure 7 As shown, the torque estimation device 700 of a series viscoelastic actuator comprises:

[0100] The data acquisition module 701 is used to acquire the historical operating parameters of the series viscoelastic actuator;

[0101] A model building module 702, used to build an initial actuator torque estimation model based on the GMM model according to the historical operating parameters;

[0102] A model training module 703 is used to optimize the initial actuator torque estimation model using an EM algorithm to obtain an optimized actuator torque estimation model;

[0103] The torque estimation module 704 is used to obtain the real-time deformation data of the series viscoelastic actuator, and input the deformation data into the optimized actuator torque estimation model to obtain a torque estimation value.

[0104] like Figure 8 As shown in the above-mentioned torque estimation method of a series viscoelastic actuator, the present invention also provides an electronic device 800, which can be a computing device such as a mobile terminal, a desktop computer, a notebook, a palm computer, and a server. The electronic device includes a processor 801, a memory 802, and a display 803.

[0105] In some embodiments, the memory 802 may be an internal storage unit of a computer device, such as a hard disk or memory of the computer device. In other embodiments, the memory 802 may also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Further, the memory 802 may also include both an internal storage unit of the computer device and an external storage device. The memory 802 is used to store application software and various data installed in the computer device, such as program codes installed in the computer device. The memory 802 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a torque estimation method program 1104 for a series viscoelastic actuator is stored on the memory 802, and the torque estimation method program 1104 for a series viscoelastic actuator can be executed by the processor 801, thereby realizing a torque estimation method for a series viscoelastic actuator in each embodiment of the present invention.

[0106] In some embodiments, the processor 801 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 802, such as executing a torque estimation program for a series viscoelastic actuator.

[0107] In some embodiments, the display 803 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 803 is used to display information on the computer device and to display a visual user interface. The components 801-803 of the computer device communicate with each other through a system bus.

[0108] This embodiment further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for estimating the torque of a series viscoelastic actuator described in any of the above technical solutions is implemented.

[0109] The computer-readable storage medium and computing device provided according to the above-mentioned embodiments of the present invention can be implemented with reference to the specific description of the torque estimation method for a series viscoelastic actuator as described above according to the present invention, and have similar beneficial effects as the torque estimation method for a series viscoelastic actuator as described above, which will not be repeated here.

[0110] The present invention discloses a torque estimation method, device, electronic device and computer-readable storage medium for a series viscoelastic actuator. First, historical operation data of the series viscoelastic actuator is obtained, and an initial actuator torque estimation model is established based on the historical operation data and a GMM model; secondly, the initial actuator torque estimation model is optimized using an EM algorithm to obtain an optimized actuator torque estimation model; finally, real-time deformation data of the actuator is obtained, and a torque estimation value corresponding to the real-time deformation data is obtained using a well-trained actuator torque estimation model.

[0111] The present invention establishes an initial torque estimation model based on the GMM model, uses several Gaussian distributions to fit arbitrary data, does not consider the difference in the initial conditions of the viscoelastic material, and models the relationship between the deformation and the output torque of the viscoelastic body based on historical data. The parameters of the initial model are iteratively updated by the EM algorithm to obtain an optimized torque estimation model. The GMM model based on the EM algorithm combines the parameter optimization algorithm with the hybrid model, providing a new modeling method for scientific research and practical engineering. It effectively avoids the influence of nonlinear effects such as hysteresis, stress relaxation and strain creep effect of the viscoelastic body on the torque estimation of the series viscoelastic actuator, improves the accuracy of the torque estimation of the series viscoelastic body, and provides a theoretical basis for realizing the compliant interaction between SVA and the external environment.

[0112] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for estimating the torque of a series viscoelastic actuator, characterized in that: include: Acquiring historical operating parameters of the series viscoelastic actuator, the historical operating parameters comprising: a historical deformation of the viscoelastic body in the actuator and an output torque of the actuator corresponding to the historical deformation; According to the historical operating parameters, an initial actuator torque estimation model is established based on the GMM model; Optimizing the initial actuator torque estimation model using an EM algorithm to obtain an optimized actuator torque estimation model; The real-time deformation data of the series viscoelastic actuator is acquired, and the deformation data is input into the optimized actuator torque estimation model to obtain a torque estimation value.

2. The torque estimation method of a series viscoelastic actuator according to claim 1, characterized in that: According to the historical operating parameters, an initial actuator torque estimation model is established based on the GMM model, including: Determining model parameters of the initial actuator torque estimation model according to the historical operating parameters, and establishing the initial actuator torque estimation model based on the model parameters and the GMM model; Wherein, the mathematical expression of the initial actuator torque estimation model is: Among them, K means there are K Gaussian distribution models, α k Indicates the current sample x Belong to k The probability of a Gaussian distribution, φ ( y |θ k ) indicates the k The probability density function of a Gaussian distribution is θ k (μ k ,Σ k ) indicates the k Gaussian distribution, μ k Indicates k The mean corresponding to the Gaussian distribution, Σ k Indicates k The covariance corresponding to the Gaussian distribution.

3. The torque estimation method of a series viscoelastic actuator according to claim 2, characterized in that: The initial actuator torque estimation model is optimized by using the EM algorithm to obtain an optimized actuator torque estimation model, including: Optimizing the initial model parameters in the initial actuator torque estimation model using the EM algorithm to obtain optimized model parameters, and obtaining the optimized actuator torque estimation model based on the optimized model parameters; The model parameters include k The probability of a Gaussian distribution k , No. k The mean μ corresponding to the Gaussian distribution k and k The covariance Σ corresponding to the Gaussian distribution k .

4. The torque estimation method of a series viscoelastic actuator according to claim 3, characterized in that: The initial model parameters in the initial actuator torque estimation model are optimized using the EM algorithm to obtain optimized model parameters, including: According to the initial actuator torque estimation model, a maximum likelihood function in the EM algorithm is obtained; Calculating a partial derivative of the maximum likelihood function, and updating the initial model parameters according to the partial derivative result of the maximum likelihood function; When the EM algorithm converges, optimized model parameters are obtained.

5. The torque estimation method of a series viscoelastic actuator according to claim 4, characterized in that: According to the initial actuator torque estimation model, the maximum likelihood function in the EM algorithm is obtained, including: The mathematical expression of the maximum likelihood function in the EM algorithm is: in, X is the observed data, Z is a hidden variable, P ( Z | X , θ (t) ) is given the observed data X and the current parameter estimates θ (t) Hidden variables Z The conditional probability distribution of ; P ( X , Z | θ ) is the model parameter θ In known cases, X and Z Conditional probability of simultaneous occurrence; Q ( θ , θ ( t +1)) is the parameter estimation function for the next moment.

6. The torque estimation method of a series viscoelastic actuator according to claim 4, characterized in that: Calculating a partial derivative of the maximum likelihood function, and updating the initial model parameters according to the partial derivative result of the maximum likelihood function, including: Take the partial derivative of the maximum likelihood function and put t Time Q( θ , θ (t) ) is the maximum value of the parameter θ (t+1) As t +1 time point model parameters to obtain updated model parameters.

7. A torque estimation device for a series viscoelastic actuator, characterized in that: include: A data acquisition module, used to acquire historical operating parameters of the series viscoelastic actuator, wherein the historical operating parameters include: a historical deformation amount of the viscoelastic body in the actuator and an output torque of the actuator corresponding to the historical deformation amount; A model building module, used for building an initial actuator torque estimation model based on the GMM model according to the historical operating parameters; A model training module, used for optimizing the initial actuator torque estimation model by using an EM algorithm to obtain an optimized actuator torque estimation model; The torque estimation module is used to obtain real-time deformation data of the series viscoelastic actuator, and input the deformation data into the optimized actuator torque estimation model to obtain a torque estimation value.

8. An electronic device, characterized in that: The invention comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, a torque estimation method for a series viscoelastic actuator as claimed in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for estimating the torque of a series viscoelastic actuator according to any one of claims 1 to 6 is implemented.

Images