Method and system for calculating expected speed by using hidden Markov model and double-layer Bayes in teleoperation

Abstract

The invention belongs to the related technical field of robot teleoperation, and discloses a method and a system for calculating expected speed by using a hidden Markov model and double-layer Bayesian in teleoperation. The method comprises the following steps: acquiring an off-line motion track in off-line motion of a master-end robot, and performing off-line training on a hidden Markov model by utilizing the off-line motion track until the hidden Markov model is converged, so as to determine model parameters in the hidden Markov model; inputting the movement speed of the master-end robot at the current moment into the trained hidden Markov model to obtain hidden Markov model parameters; calculating the occurrence probability of each hidden state at the current moment and the next moment by using a Bayesian formula; and the expected speed of the slave end robot at the current moment is calculated by using the occurrence probability of each hidden state at the next moment. According to the invention, the problem of hand shake suppression in a complex track teleoperation task is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field related to robot teleoperation, and more specifically, relates to a method and system for calculating expected speed by using hidden Markov model and double-layer Bayesian in teleoperation. Background Art

[0002] Teleoperation technology relies on the operator's movements to control the movement of the slave robot. Usually, the slave robot strictly copies the trajectory of the master operator's hand. However, due to physiological reasons, the shaking of the human hand during movement is inevitable. If the instructions containing shaking interference are directly sent to the slave robot, the speed stability and trajectory smoothness will be seriously affected. Therefore, it is necessary to suppress the shaking of the human hand.

[0003] The Kalman filter can predict and filter the jitter signal of the human hand, but its prediction accuracy for nonlinear signals is not high. Machine learning methods such as support vector machines and neural networks can also be used to eliminate human hand jitter and are suitable for nonlinear systems, but they only filter the high-frequency and small-amplitude jitter signals of the human hand from the data level, and cannot suppress the low-frequency and large-amplitude signals caused by accumulated jitter from the level of the intention to complete the task. The fundamental reason why the operator still generates jitter signals when he is aware of the path required for the task is the randomness of human behavior, so this randomness needs to be characterized.

[0004] Hidden Markov models have been used to characterize the randomness of human behavior. Using hidden Markov models to identify human intentions, that is, the expected speed of human hand movement, can achieve the function of filtering jitter signals. At present, the existing use of jitter elimination is limited to the jitter elimination of straight line trajectories. In more application scenarios such as remote operation of glue coating, the task trajectories are diverse, and the use of hidden Markov models alone lacks adaptability to diverse trajectories. Therefore, there is an urgent need for a method that can be applied to remote operation jitter elimination. Summary of the invention

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a method and system for calculating the expected speed using a hidden Markov model and a double-layer Bayesian in teleoperation, so as to solve the problem of suppressing human hand jitter in complex trajectory teleoperation tasks.

[0006] To achieve the above object, according to one aspect of the present invention, a method for calculating expected speed using hidden Markov model and double-layer Bayesian in teleoperation is provided, the method comprising the following steps:

[0007] Obtaining an offline motion trajectory of the master robot in offline motion, and using the offline motion trajectory to offline train a hidden Markov model until convergence, thereby determining model parameters in the hidden Markov model, in which the speed of the master robot is an observable state and the expected speed of the slave robot is a hidden state;

[0008] The movement speed of the master robot at the current moment is input into the trained hidden Markov model to obtain the hidden Markov model parameters corresponding to the movement speed at the current moment; the probability of each hidden state in the hidden Markov model appearing at the current moment is calculated using the Bayesian formula; the probability of each hidden state appearing at the next moment is calculated using the Bayesian formula again in combination with the probability of each hidden state appearing at the current moment, and the expected speed of the slave robot at the current moment is calculated using the probability of each hidden state appearing at the next moment.

[0009] Further preferably, after offline training of the hidden Markov model, the offline trained hidden Markov model is trained online, and the process of online training is as follows:

[0010] The real-time motion speed of the master robot at each moment is obtained online, and then the real-time motion speed is input into the hidden Markov model after offline training for training, thereby performing an online training process.

[0011] Further preferably, the formula for calculating the probability of each hidden state in the hidden Markov model appearing at the current moment using the Bayesian formula is as follows:

[0012]

[0013] in, The hidden state at time n is z i The probability of The observed state at time n is {o1,o2,…,o n} and the hidden state is z i , i∈{1,2,…M} is the number of the hidden state, and M is the total number of hidden states.

[0014] Further preferably, the forward factor is calculated as follows:

[0015]

[0016] in, The observed state at time 1 is o1 and the hidden state is z i The probability of i The hidden state at time 1 is z i The probability of P(o1|s1=z i ) is the hidden state z at time 1 iUnder the condition of , the probability that the observable state is o1; The observed state at time n+1 is {o1,o2,…,o n ,o n+1} and the hidden state is z i The probability of ij =P(s n+1 =z j |s n =z i ) is the hidden state at time n is z i Under the condition of j The probability of n+1 |s n+1 =z i ) is the hidden state z at time n+1 i Under the condition of n+1 probability; The observed state at time n is {o1,o2,…,o n} and the hidden state is z i probability; i∈{1,2,…M} is the number of the hidden state at the current moment, j∈{1,2,…M} is the number of the hidden state at the next moment, and M is the total number of hidden states.

[0017] Further preferably, the Bayesian formula is used again in combination with the state transfer matrix at the current moment in the hidden Markov model parameters to calculate the probability of each hidden state appearing at the next moment, according to the following formula:

[0018]

[0019] in, At time n+1, the hidden state is z i The probability of ij =P(s n+1 =z j |s n =z i ) is the current hidden state z i Under the condition that j probability; At time n, the hidden state is z i , i∈{1,2,…M} is the number of the hidden state at the current moment, j∈{1,2,…M} is the number of the hidden state at the next moment, and M is the total number of hidden states.

[0020] Further preferably, the expected speed of the slave robot at the current moment is calculated by using the probability of each hidden state appearing at the next moment, according to the following formula:

[0021]

[0022] Among them, v n,intend is the expected speed of the slave robot at the current time n; At time n+1, the hidden state is z i The probability of At time n, the hidden state is z i Under the condition of n The mean of the Gaussian probability distribution satisfied by , i∈{1,2,…M} is the number of the hidden state at the current moment.

[0023] More preferably, the Calculate according to the following formula:

[0024]

[0025] Among them, μ i The hidden state at the current moment is z i Under the condition of n The Gaussian probability distribution P(o n |s n =z i ), is the first-order sufficient statistic of the Gaussian probability distribution, indicating that the observable state at the current moment is {o1,o2,...,o n}, the hidden state z i The mean number of occurrences at all times, is the second-order sufficient statistic of the Gaussian probability distribution, indicating that the observable state at the current moment is {o1,o2,...,o n}, the hidden state z appears i is the mean of the cumulative sum of the observable states corresponding to the time, i∈{1,2,…M} is the number of the hidden state at the current moment.

[0026] Further preferably, after obtaining the expected speed of the slave robot at the current moment, the joint space position of the slave robot at the current moment is calculated using the expected speed, so as to control the slave robot.

[0027] Further preferably, the joint space position of the slave robot at the current moment is obtained by using the expected speed calculation according to the following steps:

[0028] Integrate the expected speed of the robot at the current moment to obtain the expected position P of the slave robot at the current moment intend ;

[0029] The desired position of the slave robot is mapped to the base coordinate system of the slave robot using an incremental spatial mapping method to obtain the Cartesian spatial position P of the desired position of the slave robot. s ;

[0030] The Cartesian space position P is converted into s Transformed into the robot joint space position q intend .

[0031] According to another aspect of the present invention, a system for calculating the expected speed using a hidden Markov model and a double-layer Bayesian in remote operation is provided, the system comprising an actuator for executing the above-mentioned method for calculating the expected speed using a hidden Markov model and a double-layer Bayesian in remote operation.

[0032] In general, the above technical solutions conceived by the present invention have the following beneficial effects compared with the prior art:

[0033] 1. The present invention uses the hidden Markov model as the basic model combined with the double-layer Bayesian algorithm to identify the operator's intention to smoothly move the master robot during teleoperation, and calculates the expected trajectory that the operator subjectively intends to achieve hidden under the hand jitter trajectory; this trajectory is the smooth trajectory after eliminating the hand jitter, thereby achieving hand jitter suppression in teleoperation tasks.

[0034] 2. After obtaining the parameters of the hidden Markov model using the actual motion speed, the present invention first uses the Bayesian algorithm to calculate the probability of each hidden state appearing, and then uses the Bayesian algorithm again to calculate the probability of the hidden state appearing at the next moment. Since humans have reaction time, the jitter at the current moment will be adjusted in the next time step, that is, the real expected speed at the current moment will be reflected at the next moment. Therefore, the probability of the hidden state appearing at the next moment is used to calculate the expected speed at the current moment, so as to achieve more accurate expected speed calculation.

[0035] 3. After the present invention performs offline training on the hidden Markov model, the offline trained model is used as the initial model for online training, realizing two training processes of offline training and online training of the hidden Markov model, so that the parameters of the hidden Markov model are updated while executing the remote operation task, thereby identifying new operator intentions different from those in the offline training trajectory, and realizing the suppression of human hand jitter for any complex trajectory not limited to the offline trajectory. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a control framework diagram of an operator expected speed estimation method based on an online hidden Markov model and a double-layer Bayesian method constructed according to a preferred embodiment of the present invention.

[0037] Figure 2 is a structural diagram of a hidden Markov model according to a preferred embodiment of the present invention.

[0038] Figure 3 is a flow chart of an expectation-maximization algorithm according to a preferred embodiment of the present invention.

[0039] Figure 4 is a flow chart of an online expectation-maximization algorithm according to a preferred embodiment of the present invention.

[0040] Figure 5 It is a double-layer Bayesian module framework diagram constructed according to the preferred embodiment of the present invention.

[0041] Figure 6 It is the estimation result of the expected speed of different human hand movement trajectories according to the preferred embodiment of the present invention, wherein (a1), (a2) and (a3) ​​are the estimation results of the expected speed in the three directions of x, y and z for the straight line trajectory respectively, (b) is the displacement result obtained by integrating the speed of the straight line trajectory; (c1), (c2) and (c3) are the estimation results of the expected speed in the three directions of the circular trajectory respectively, (d) is the displacement result obtained by integrating the speed of the circular trajectory, (e1), (e2) and (e3) are the estimation results of the expected speed in the three directions of the "S"-shaped trajectory respectively, and (f) is the displacement result obtained by integrating the speed of the "S"-shaped trajectory.

[0042] Figure 7 The invention is a remote-controlled gluing task experimental platform according to a preferred embodiment of the present invention.

[0043] Figure 8 1 is a diagram of the results of remote-controlled gluing according to a preferred embodiment of the present invention, wherein (a) and (b) are respectively the results of remote-controlled gluing of a straight trajectory without and with the method of the present invention, (c) and (d) are respectively the results of remote-controlled gluing of a circular trajectory without and with the method of the present invention, and (e) and (f) are respectively the results of remote-controlled gluing of an "S"-shaped trajectory without and with the method of the present invention.

[0044] Fig. 9 The figures are result diagrams of remote-controlled gluing on an engine oil pan according to a preferred embodiment of the present invention, wherein (a) is the result of remote-controlled gluing of a straight-line trajectory without and with the method of the present invention, and (b) is the result of remote-controlled gluing of a straight-line trajectory with the method of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0046] like Figure 1 As shown, a method for estimating operator expected speed based on an online hidden Markov model and a double-layer Bayesian method comprises the following steps:

[0047] Training of S1 Hidden Markov Model

[0048] (1) Offline training

[0049] A hidden Markov model is established with the actual speed of the human hand as the observable state and the expected speed as the hidden state. The original offline trajectory data of the human hand along the random straight line direction is collected by the motion tracking device. The offline trajectory refers to the situation where the master robot does not drive the slave robot to move. The model parameters are obtained by offline training using the classic hidden Markov model method, such as Figure 3 shown.

[0050] The actual hand movement speed o={v x ,v y ,v z} is regarded as the observable state of the hidden Markov model, and a two-dimensional matrix O = {o1,o2,...,o n} (n represents the time), each column of the two-dimensional matrix is ​​the actual movement speed of the hand o = {v x ,v y ,v z}.

[0051] The human intention is defined as the desired speed of the operator's hand movement, which should be smooth. As the hidden state of the hidden Markov model, a two-dimensional matrix S = {s1, s2, …s n}, each column of the two-dimensional matrix is ​​the expected movement speed of the human hand Hidden state s n It must be from a fixed set Z = {z1,z2,…z M}, where M is the number of hidden states.

[0052] The parameters of the hidden Markov model are λ={π,A,B}, where π={π1,π2,…π M} is the initial state matrix, π i=P(s1=z i ) represents the initial hidden state z i The probability of (i∈{1,2,…M}); A M×M is the state transfer matrix, where each element a ij It is called the state transition probability, which represents the probability a from the current hidden state to the next hidden state ij =P(s n+1 =z j |s n =z i )(i,j∈{1,2,…M}); B M×1 is the emission matrix, where each row represents an emission probability, and the probability distribution P(o) of observing the observable state under the current hidden state n |s n =z i )(i∈{1,2,…M}). The emission probability is considered to satisfy the multivariate Gaussian distribution as follows:

[0053]

[0054] μ,Σ represent the mean and standard deviation matrices of the multivariate Gaussian distribution, respectively, so the emission matrix The relationship diagram of each parameter is as follows Figure 2 As shown.

[0055] The master robot collects the original offline trajectory data of the human hand along the random straight line direction as an observable sequence O = {o1, o2, ..., o n}.

[0056] First, the parameters of the hidden Markov model are initialized to λ with arbitrary values init ={π init ,A init ,B init}, and then calculate the forward factor and backward factor according to the properties of the hidden Markov model and the forward and backward algorithm:

[0057]

[0058]

[0059] The forward factor Represents the hidden state at time n is z i And the observable sequence is (o1,o2,...,o n ), the backward factor Represents the hidden state at time n is z i And the observable sequence is (o n+1 ,o n+2 ,...,o N ) probability.

[0060] Calculate the intermediate variable γ n (i),ξ n (i,j):

[0061]

[0062]

[0063] The parameters λ = {π, A, B} that need to be trained are expressed by the following iterative formula:

[0064]

[0065] Where D is the number of observation sequence samples used to train the parameters. The above calculation is performed iteratively until the parameters λ = {π, A, B} converge.

[0066] The final converged parameter λ con Sent to the online hidden Markov model module as initial values ​​for parameter update.

[0067] (2) Online training

[0068] An online hidden Markov model module is added to the teleoperation framework, and real-time training is performed using the online trajectory data collected by the master robot.

[0069] During teleoperation, the observable state of the speed of the human hand movement is o n = {v x ,v y ,v z} are sent to the online hidden Markov model module in sequence. And the initial parameters of the online hidden Markov model module are set to the λ calculated by the offline hidden Markov model con .

[0070] Further preferably, in step S3, each time an observable state o is received n = {v x ,v y ,v z}, the online expectation maximization algorithm is used to update the initial parameters calculated by the offline hidden Markov model to λ new ={A n ,μ n ,Σ n According to the established hidden Markov model, the emission probability B is Gaussian distribution, and four sufficient statistics of Gaussian distribution are selected:

[0071]

[0072] in Represents the indicator function, which takes 1 if the value in the brackets is true and 0 if the value in the brackets is false; i,j,k∈{1,2,3,4,5} is the index of the hidden state.

[0073] Then calculate the intermediate variable q n ,φ n :

[0074]

[0075]

[0076] For sufficient statistics To update:

[0077]

[0078] where {χ1,χ2…χ n} is a sequence of decreasing step length, which meets the requirements of random approximation: ∑ n≥1 χ n =∞,∑ n≥1 χ n 2 <∞. The above expression uses only one sample point for iterative update each time.

[0079] When n≥n min After that, the model parameters and sufficient statistics of the hidden Markov model are updated according to:

[0080]

[0081] Among them, n min The role of is to ensure that there is enough data for parameter update, because when the sample points are very small, the estimation of the expectation maximization algorithm is inaccurate. By looping the above steps, you can achieve online every time you receive an observable variable o n = {v x ,v y ,v z}Update the hidden Markov model parameters once.

[0082] The updated parameter λ new ={A n ,μ n ,Σ n}Sent to the two-layer Bayesian module.

[0083] S2 adds a double-layer Bayesian module to the teleoperation framework, and uses the online trained hidden Markov model and the double-layer Bayesian method to calculate the operator's expected speed; the identified expected speed is used for the motion control of the slave robot.

[0084] Calculate the known observable sequence O = {o1,o2,...,o n}, each hidden state z i The probability of occurring at the current moment This probability represents the possibility that multiple possible expected speeds of human hand movements actually exist at the current moment:

[0085]

[0086]

[0087] Using the Bayesian formula and the total probability formula, ζ n i It can be expressed by the forward factor as:

[0088]

[0089] Using the current state transfer matrix A and the total probability formula, we can calculate the probability of the known observable sequence O = {o1, o2, ..., o n}, each hidden state z i The probability of appearing at the next moment ε i =P(s n+1 =z i |o1,o2,...,o n ), which represents the possibility that multiple possible expected speeds of hand movements actually exist at the next moment:

[0090]

[0091] For each hidden state z i , the mean μ of the Gaussian distribution it satisfies i The actual expected speed is represented by the weighted average of the basic expected moving speed factors. However, due to human reaction time, it takes a certain amount of time to adjust in the right direction after realizing the deviation caused by jitter. In other words, the actual expected speed at the current moment will be reflected at the next moment, so the final expected speed at the current moment v intend Calculated by the following formula

[0092]

[0093] in, Represents the expected speed at the next moment.

[0094] Finally, the desired velocity is integrated to obtain the desired position P of the end of the slave robot: intend={x,y,z,r,p,y}, where x,y,z represent the desired displacement, and r,p,y=0 means keeping the hand posture unchanged

[0095] The desired position of the human hand is mapped to the base coordinate system of the slave robot through the incremental space mapping method to obtain P s , the Cartesian space position P is converted to s Transform the joint space position q intend ={q1,q2,...,q n}, where n is the number of slave robot joints.

[0096] Q intend It is sent as the actual control signal to the slave robot to control its movement.

[0097] The present invention will be further described below with reference to specific embodiments.

[0098] Training of S1 Hidden Markov Model

[0099] (1) Offline training

[0100] In this embodiment, eight pieces of offline training data are collected. Taking one piece O as an example, the matrix of 3 rows and 782 columns is as follows, with the unit being mm / s:

[0101]

[0102] The expectation maximization algorithm is used to train the collected data. First, the parameters of the hidden Markov model are initialized with arbitrary values ​​as λ init ={π init ,A init ,B init},in:

[0103] π init =[0.2 0.2 0.2 0.2 0.2]

[0104]

[0105]

[0106]

[0107] Then the forward factor and backward factor are calculated according to the properties of the hidden Markov model and the forward and backward algorithm:

[0108]

[0109]

[0110] The forward factor Represents the hidden state at time n is z i And the observable sequence is (o1,o2,...,o n ), the backward factor Represents the hidden state at time n is z i And the observable sequence is (o n+1 ,o n+2 ,...,o N ) probability.

[0111] Calculate the intermediate variable γ n (i),ξ n (i,j):

[0112]

[0113] The parameters λ = {π, A, B} that need to be trained are expressed by the following iterative formula:

[0114]

[0115] Where D is the number of observation sequence samples used to train the parameters. In this embodiment, D = 8. The above calculation is iterated until the parameter λ = {π, A, B} converges. In this embodiment, the offline parameter λ finally converged is con ={π,A,B} is as follows:

[0116] π=[0.433 0.148 0.313 0.106 0.000]

[0117]

[0118]

[0119]

[0120]

[0121] The final converged model parameter λ con Sent to the online hidden Markov model module as initial values ​​for parameter update.

[0122] (2) Online training

[0123] An online hidden Markov model module is added to the teleoperation framework, and the online movement trajectory data of the human hand collected by the master robot is used as an observable sequence for real-time training. The training process is as follows: Figure 4 shown.

[0124] During teleoperation, the hand movement speed o is used as the observable state. n = {v x,v y ,v z} are sent to the online hidden Markov model module in sequence, and the initial parameters of the online hidden Markov model module are set to λ calculated by the offline hidden Markov model con .

[0125] Each time an observable state o is received n = {v x ,v y ,v z}, then the online expectation maximization algorithm is used to update the initial value of the offline hidden Markov model to λ new ={π,A n ,B n}.

[0126] First, it is determined that the emission probability B in the established hidden Markov model is a Gaussian distribution, which meets the prerequisite of the online expectation maximization algorithm, and four sufficient statistics of the Gaussian distribution are selected as follows:

[0127]

[0128] in Represents the indicator function, which takes 1 if the value in the brackets is true and 0 if the value in the brackets is false; i,j,k∈{1,2,3,4,5} is the index of the hidden state.

[0129] Then calculate the intermediate variable q n ,φ n :

[0130]

[0131]

[0132] For sufficient statistics To update:

[0133]

[0134] where {χ1,χ2…χ n} is a sequence of decreasing step length, which meets the requirements of random approximation: ∑ n≥1 χ n =∞,∑ n≥1 χ n 2 <∞. The above expression uses only one sample point for iterative update each time.

[0135] When n≥n min After that, the model parameters and sufficient statistics of the hidden Markov model are updated according to:

[0136]

[0137] Among them, n min The role of is to ensure that there is enough data for parameter update, because when the sample points are very small, the estimation of the expectation maximization algorithm is inaccurate. By looping the above steps, you can achieve online every time you receive an observable variable o n = {v x ,v y ,v z}Update the hidden Markov model parameters once.

[0138] In this embodiment, x n =n -0.7 , n min = 10. When n = 1, the received observable state o1 = {-0.349 1.053 0.197}, which is calculated as follows:

[0139] First, calculate the initial parameter λ con The emission probability of each hidden state under i ):

[0140]

[0141] Then calculate the intermediate variable φ1:

[0142]

[0143] Then calculate the sufficient statistic S A (i,j),S B (i), taking i=1 as an example for calculation:

[0144]

[0145] Loop through the above steps to calculate q at each step n ,φ n , Until n = 10, the received observable state o 10 =[-0.498 0.717 0.948], using the above intermediate variables to update the model parameters, the specific calculation is as follows:

[0146] First, calculate the initial parameter λ con The emission probability of each hidden state under i ):

[0147]

[0148] Then calculate the intermediate variable q 10 (i, j) (taking i=j=1 as an example):

[0149]

[0150] Calculate the intermediate variable φ 10 (i) (taking i = 1 as an example, φ9(i) is known) is calculated as follows:

[0151]

[0152] Then calculate the sufficient statistic S 10 A (i,j),S 10 B (i) (Take i=j=1 as an example, S9 A (i,j),S9 B (i) Known):

[0153] Finally, calculate the parameter a 11 (i,j),μ 10 (i),Σ 10 (i) and update (taking i=j=1 as an example):

[0154]

[0155] The final parameter update is:

[0156] λ new ={π,A n ,B n}:

[0157] π=[0.433 0.148 0.313 0.106 0.000]

[0158]

[0159]

[0160]

[0161]

[0162] The updated parameter λ new ={A n ,μ n ,Σ n}Sent to the two-layer Bayesian module.

[0163] S2 adds a double-layer Bayesian module to the teleoperation framework, and uses the online trained hidden Markov model parameters and the double-layer Bayesian method to calculate the operator's expected speed from the original online hand movement trajectory. The framework of the double-layer Bayesian method is as follows Figure 5shown.

[0164] In this embodiment, when n=10, the received observable state o 10 =[-0.498 0.717 0.948], forward factor Therefore 11 i The calculation is as follows (taking i=1 as an example):

[0165]

[0166] ξ 10 =[0.011,0.027,0.000,0.00,0.961]

[0167] Using the state transfer matrix A and the total probability formula, we can calculate the probability of the known observable sequence O = {o1, o2, ..., o n}, each hidden state z i The probability of appearing at the next moment ε i =P(s n+1 =z i |o1,o2,...,o n ), which represents the possibility that multiple possible expected speeds of hand movements actually exist at the next moment:

[0168]

[0169] In this embodiment, when n=10, The calculation is as follows (taking i=1 as an example):

[0170]

[0171] ε 10 =[0.068,0.159,0.002,0.009,0.761]

[0172] In this embodiment, when n=10, v intend The calculation is as follows:

[0173]

[0174] The expected speed estimation results for different trajectories using the above method are as follows: Figure 6As shown in the figure, (a1), (a2), (a3), (b) are the expected speed estimation results in the x, y, and z directions of the linear trajectory, and (b) is the displacement result obtained by integrating the speed; (c1), (c2), (c3), (d) are the estimation results for the circular trajectory, and (e1), (e2), (e3), (f) are the estimation results for the "S"-shaped trajectory. It can be seen from the figure that whether it is a linear, circular or "S"-shaped trajectory, the expected speed estimation result is obviously smoother than the original speed, and the displacement obtained by integrating the expected speed estimation result is also smoother than the original displacement, indicating that the hand shake has been effectively suppressed. …

[0175] The identified desired velocity is used for motion control of the slave robot.

[0176] Integrate the desired velocity to get the desired position P of the end of the slave robot intend ={x,y,z,r,p,y}, where r,p,y=0 means keeping the posture of the slave robot unchanged

[0177] The Cartesian space position P is converted to intend = {x, y, z, r, p, y} converted joint space position q intend ={q1,q2,…,q n}, where n is the number of slave robot joints.

[0178] Q intend Sent to the slave robot as the actual control signal.

[0179] In order to verify the effectiveness of the online matching method proposed in the present invention, the accuracy is verified by using an example. In this example, a remote-operated glue application task platform is built for experimental verification. Figure 7 As shown, it includes a master robot Virtuose 6D TAO, a slave robot Franka, a controller and a gluing device, and the control frequency of the robots is 20HZ. Figure 8 The results of teleoperated gluing tasks on straight line, circle and “S” trajectories are shown with and without the method of the present invention. Fig. 9 The results of the remote-controlled gluing task on the engine oil pan are shown in Figure 1 and 2 respectively. The appearance of the sealant applied by the remote-controlled gluing method of the present invention is smoother than that of the sealant applied by ordinary remote-controlled gluing, which shows the effectiveness of the method of the present invention in suppressing hand shaking.

[0180] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for calculating expected speed using hidden Markov model and double-layer Bayesian in teleoperation, characterized in that: The method comprises the following steps: Obtaining an offline motion trajectory of the master robot in offline motion, and using the offline motion trajectory to offline train a hidden Markov model until convergence, thereby determining model parameters in the hidden Markov model, in which the speed of the master robot is an observable state and the expected speed of the slave robot is a hidden state; Input the current movement speed of the master robot into the trained hidden Markov model to obtain the hidden Markov model parameters corresponding to the current movement speed; The probability of each hidden state in the hidden Markov model appearing at the current moment is calculated using the Bayesian formula; The Bayesian formula is used again and combined with the probability of each hidden state appearing at the current moment to calculate the probability of each hidden state appearing at the next moment. The probability of each hidden state appearing at the next moment is used to calculate the expected speed of the slave robot at the current moment.

2. The method for calculating the expected speed using a hidden Markov model and a double-layer Bayesian in teleoperation according to claim 1, characterized in that: After offline training of the hidden Markov model, the offline trained hidden Markov model is trained online, and the process of online training is as follows: The real-time motion speed of the master robot at each moment is obtained online, and then the real-time motion speed is input into the hidden Markov model after offline training for training, thereby performing an online training process.

3. A method for calculating expected speed using hidden Markov model and double-layer Bayesian in teleoperation as claimed in claim 1 or 2, characterized in that: The formula for calculating the probability of each hidden state in the hidden Markov model appearing at the current moment using the Bayesian formula is as follows: in, The hidden state at time n is z i The probability of The observed state at time n is {o1,o2,…,o n } and the hidden state is z i , i∈{1,2,…M} is the number of the hidden state, and M is the total number of hidden states.

4. A method for calculating expected speed using hidden Markov model and double-layer Bayesian in teleoperation as claimed in claim 3, characterized in that: The forward factor is calculated as follows: in, The observed state at time 1 is o1 and the hidden state is z i The probability of i The hidden state at time 1 is z i The probability of P(o1|s1=z i ) is the hidden state z at time 1 i Under the condition of , the probability that the observable state is o1; The observed state at time n+1 is {o1,o2,…,o n ,o n+1 } and the hidden state is z i The probability of ij =P(s n+1 =z j |s n =z i ) is the hidden state at time n is z i Under the condition of j The probability of n+1 |s n+1 =z i ) is the hidden state z at time n+1 i Under the condition of n+1 probability; The observed state at time n is {o1,o2,…,o n } and the hidden state is z i probability; i∈{1,2,…M} is the number of the hidden state at the current moment, j∈{1,2,…M} is the number of the hidden state at the next moment, and M is the total number of hidden states.

5. The method for calculating the expected speed using a hidden Markov model and a double-layer Bayesian in teleoperation as claimed in claim 3 or 4, characterized in that: The Bayesian formula is used again in combination with the state transfer matrix at the current moment in the hidden Markov model parameters to calculate the probability of each hidden state appearing at the next moment, according to the following formula: in, At time n+1, the hidden state is z i The probability of ij =P(s n+1 =z j |s n =z i ) is the current hidden state z i Under the condition that j probability; At time n, the hidden state is z i , i∈{1,2,…M} is the number of the hidden state at the current moment, j∈{1,2,…M} is the number of the hidden state at the next moment, and M is the total number of hidden states.

6. A method for calculating expected speed using hidden Markov model and double-layer Bayesian in teleoperation as claimed in claim 5, characterized in that: The expected speed of the slave robot at the current moment is calculated by using the probability of each hidden state appearing at the next moment, according to the following formula: Among them, v n,intend is the expected speed of the slave robot at the current time n; At time n+1, the hidden state is z i The probability of At time n, the hidden state is z i Under the condition of n The mean of the Gaussian probability distribution satisfied by , i∈{1,2,…M} is the number of the hidden state at the current moment.

7. A method for calculating expected speed using hidden Markov model and double-layer Bayesian in teleoperation as claimed in claim 6, characterized in that: Said Calculate according to the following formula: Among them, μ i The hidden state at the current moment is z i Under the condition of n The Gaussian probability distribution P(o n |s n =z i ), is the first-order sufficient statistic of the Gaussian probability distribution, indicating that the observable state at the current moment is {o1,o2,...,o n }, the hidden state z i The mean number of occurrences at all times, is the second-order sufficient statistic of the Gaussian probability distribution, indicating that the observable state at the current moment is {o1,o2,...,o n }, the hidden state z appears i is the mean of the cumulative sum of the observable states corresponding to the time, i∈{1,2,…M} is the number of the hidden state at the current moment.

8. A method for calculating expected speed using hidden Markov model and double-layer Bayesian in teleoperation as claimed in claim 1 or 2, characterized in that: After obtaining the expected speed of the slave robot at the current moment, the joint space position of the slave robot at the current moment is calculated using the expected speed, so as to control the slave robot.

9. A method for calculating expected speed using hidden Markov model and double-layer Bayesian in teleoperation as claimed in claim 8, characterized in that: The expected speed is used to calculate the joint space position of the slave robot at the current moment according to the following steps: Integrate the expected speed of the robot at the current moment to obtain the expected position P of the slave robot at the current moment intend ; The desired position of the slave robot is mapped to the base coordinate system of the slave robot using an incremental spatial mapping method to obtain the Cartesian spatial position P of the desired position of the slave robot. s ; The Cartesian space position P is converted into s Transformed into the robot joint space position q intend .

10. A system for calculating expected speed using hidden Markov model and double-layer Bayesian in teleoperation, characterized in that: The system includes an actuator, which is used to execute the method of calculating the expected speed using a hidden Markov model and a double-layer Bayesian in a remote operation as described in any one of claims 9.

Images