CN104503229A - Wave integral bilateral teleoperation control method based on LS-SVM (least square support vector machine) delay predication - Google Patents

Abstract

The invention provides a wave integral bilateral teleoperation control method based on LS-SVM (least square support vector machine) delay predication, which can be applied to a teleoperation system under time-varying delay. On the basis of a passivity wave variable method, a delay predication step and a wave integral step for the LS-SVM are added, the LS-SVM method is used for delay predication, predication delay which has small error with actual delay is obtained, the wave integral method can be used for enabling the communication step to have passivity, and thus instability caused to the teleoperation system due to time variable delay can be overcome. The invention relates to a control method in the teleoperation system under time-varying delay; at the main end, the force applied by a hand and the contact force fed back to the main end from a slave end together act on the main hand, speed and displacement are generated by the main hand, the speed signals of the main hand are transmitted to the slave end via a communication channel, and the signals serve as a control command to control movement of the slave hand.

Description

Based on the ripple integration bilateral teleoperation control method of LS-SVM latency prediction

Technical field

The present invention relates to a kind of ripple integration bilateral teleoperation control method based on LS-SVM latency prediction, the method is on the basis of ordinary passive wave variables method, add LS-SVM latency prediction link and ripple integral element, the remote control system under making it be applicable to time-vary delay system condition.Belong to remote operating control technology field.

Background technology

Teleoperation has been deep into a lot of field, as space industry, deep ocean work and tele-medicine etc., is with a wide range of applications.Force feedback bilateral teleoperation system, can make operator produce telepresenc.For the stability problem of bilateral teleoperation system, at present existing a lot of research method, has wherein become the development trend of bilateral teleoperation control system based on the control strategy of modern control theory.Bilateral control method owing to not needing to predict from hands movement, so be applicable to non-structured circumstances not known, simultaneously in bilateral control system, operator can according to feed back to main side from hand and environmental exposure power, and then determine next step action.This method projects the intelligence of people from end robot by communication port, makes robot can complete the task of relative complex.

Hold the restriction of the factors such as distant, signaling rate, channel width due to principal and subordinate, in remote operating, certainly exist time delay, and very little time delay just likely causes the instability of system.In order to processing delay problem, researchers propose a lot of control strategies, and at present, the bilateral teleoperation control method overcoming communication delay mainly contains: (1) Passivity Theory method; (2) based on the method for event control; (3) based on the control method of sliding formwork; (4) based on Lyapunove stability control method; (5) based on the control method of H ∞ theory.In remote control system under fixed delay condition, these control methods can make system have good stability and operating performance, but under time-vary delay system condition, the stability of remote control system and operating performance will be destroyed.

Document " the bilateral force-feedback control algorithm research based on passivity " (Zhang Zhiqing. Harbin Institute of Technology, 2011.) in, the wave variables method that Anderson and Spong proposes based on Passivity Theory and scattering theory, this method can ensure the stability of system under any fixed delay.But, many times, time delay often time become, such as Space teleoperation, network tele-operation.For the situation that communication delay is change, at document " A Delay Prediction Approach for Teleoperation over the Intemet " (T.Mirfakhrai and S.Payandeh.Proc.IEEE Int.Conf.On Robotics and Automation, PP.2178-2183, 2002.) in, Niemeyer and Sbtine proposes wave variables Integral Thought, namely in wave variables controls, first wave variables signal is carried out integration, then be transferred to from end, wave filter is being utilized to be reconstructed integrated signal from end, make communication link passive, but the latency prediction method accuracy in the method is lower, cause Dynamic System performance lower.At document " Bilateral Teleoperation under Time-VaryingDelay using Wave Variables " (Satler M, Avizzano C.A etc.IEEE/RSJ Int.Conf onIntelligent Robots and Systems, 2009:4596-4602.) in, author improves Yokokohji compensator, make to which overcome packet loss problem, but too increase the computing time of system simultaneously, thus reduce the operating performance of remote control system.At document " Bilateral Teleoperation With Time-Varying Delay:ACommunication Channel Passification Approach " (Yongqiang Ye, Ya-JunPan, andTrent Hilliard.IEEE/ASME Transactions on Mechatronics, 18 (4), 2013:1431-1434.) in, author proposes the time domain passive control method based on power monitor (PO) and power controller (PC), this method overcomes stability problem, but reduces the operating performance of system.

To sum up analyze, in remote operating field, existing bilateral control method is only applicable to the remote control system under fixed delay condition mostly.Therefore, research is applicable to time-vary delay system condition and the bilateral teleoperation control method with high operating performance is significant.

Summary of the invention

The technical matters solved:

In time-vary delay system remote operating, existing bilateral control method is only applicable to the remote control system of fixed delay condition mostly.In order to overcome the impact that time-vary delay system brings to remote control system stability and operability, the present invention proposes a kind of ripple integration bilateral teleoperation control method of carrying out latency prediction based on LS-SVM, in order to solve the stability problem of time-vary delay system remote control system, and ensure that system can follow the tracks of the motion of main hand well from hand.

Technical scheme:

For the design of ripple integration bilateral teleoperation control method carrying out latency prediction based on LS-SVM, the technical scheme taked:

1, remote control system theory diagram is built:

(1) theory diagram of the ripple integration bilateral teleoperation control method of latency prediction is carried out in design based on LS-SVM, mainly comprise main hand, main side controller, communication link, from side controller, from hand, LS-SVM latency prediction link, ripple integral element;

(2) meaning of parameters and the relation between them in clear and definite block diagram, mainly comprise the σ of speed, power, master-slave controller parameter, ripple integral element;

2, latency prediction link is designed based on least square support vector machines (LS-SVM)

(1) build based on the latency prediction model of SVM: utilize SVM theoretical by there is nonlinear characteristic time series by non-linear conversion to high-dimensional feature space, then in this space, extract necessary information, for realizing latency prediction;

(2) LS-SVM utilizing counting yield higher than SVM realizes latency prediction;

3, ripple integral element is designed:

(1) based on ripple integral principle design ripple integral element;

(2) according to latency prediction result, the value function of design ripple integral parameter σ, wherein that control system should be made to disturb to external world is insensitive in the selection of σ, makes system be it often fully compensated simultaneously.In order to make the error in each moment minimum, the value of σ should change along with the change of time delay, and concrete rule is: if prediction time delay larger than current time delay, then σ is negative, and large must be more, σ will become less; On the contrary, if prediction time delay less than current time delay, then σ just be, and little must be more, σ will become larger, but while σ can not be allowed again too large.Its value expression formula is as follows:

σ _k=-a Δ T and (a > 0)

Wherein, Δ T=T _k+1-T _k, namely predict the difference of time delay and current time delay, a is constant factor.

4, the ripple integration bilateral teleoperation system of carrying out latency prediction based on LS-SVM is built, the rationality of checking the method under time-vary delay system condition and validity:

(1) the ripple integration bilateral teleoperation system of carrying out latency prediction based on LS-SVM is built;

(2) verify that LS-SVM is to the accuracy of latency prediction

(3) analytic system stability, and verify by experiment.

Beneficial effect:

The present invention proposes a kind of ripple integration bilateral teleoperation control method of carrying out latency prediction based on LS-SVM, on ordinary passive wave variables method basis, add LS-SVM latency prediction link and ripple integral element, LS-SVM latency prediction link is to the passivity of the accuracy of latency prediction and ripple integral element, ensure that the stability of remote control system under time-vary delay system condition, and make system from hand to the motion following the tracks of main hand well.

Accompanying drawing explanation

Fig. 1 is ordinary passive wave variables schematic diagram;

Fig. 2 is the ripple integration bilateral teleoperation control method schematic diagram carrying out latency prediction based on LS-SVM;

Fig. 3 is ripple integral principle figure;

Fig. 4 is the comparison diagram of latency prediction value and actual value;

Fig. 5 is that remote control system follows the tracks of the experiment effect figure of main hand from hand;

Embodiment

The present invention proposes a kind of ripple integration bilateral teleoperation control method of carrying out latency prediction based on LS-SVM, describes in detail below in conjunction with summary of the invention and accompanying drawing.

1, remote control system theory diagram is built

(1) schematic diagram is built

With reference to ordinary passive wave variables Method And Principle (see Fig. 1), the theory diagram of the ripple integration bilateral teleoperation control method of latency prediction is carried out in design based on LS-SVM, mainly comprise main hand, main side controller, communication link, from side controller, from hand, LS-SVM latency prediction link, ripple integral element (see Fig. 2);

(2) meaning of parameters and the relation between them in clear and definite theory diagram (see Fig. 2):

In figure, staff applies power f _hwith from holding the contact force f feeding back to main side _eacting in conjunction, in main hand, makes main hand produce displacement x _mand speed main hand rate signal obtains through wave conversion the ripple u that moves ahead _m, the ripple that moves ahead obtains from end incoming wave u through communication channel and firstorder filter _s, then through backward wave conversion, obtain the speed control commands from end in order to make the actual speed from hand follow the tracks of from end control command pI speed control is being designed from end.PI controller is according to from hand velocity error calculate the control f from hand _sact on from hand, thus make to carry out alternately from hand track desirably and speed and environment.F _efor the interaction force from end and environment.From end by f _ewith speed be converted to close echo v _s, return through communication channel and firstorder filter obtains returning incoming wave v _m, then obtain feedback force f through ripple inverse transformation _mc, and act on main hand, enable operator's perception from the contact situation of hand and environment.

2, latency prediction link is designed based on least square support vector machines (LS-SVM)

(1) the latency prediction model based on SVM is built

Time delay in remote control system can be regarded as the time series with nonlinear characteristic, therefore SVM can be utilized theoretical, by nonlinear transformation, the time series of the input space is mapped to high-dimensional feature space, then in this space, extract necessary information, be used for realizing the prediction of network delay.

If the network delay sequence before current time be d (k), k=1,2,3 ..., N}, can find a suitable embedding dimension to carry out prediction to former time series in state space and reconstruct.Be that the input vector of N number of seasonal effect in time series SVM forecast model of m can be expressed as based on Embedded dimensions:

D _k＝[d(k),d(k-1),…,d(k-m)] (1)

Thus, can obtain, latency prediction model is:

d(k+1)＝F(D _k) (2)

Wherein, F () is a nonlinear function based on the delay data reconstructs prediction model before the k+1 moment, and namely SVM needs the function of matching.Define grid time-delay series fitting function is:

kind, select radial basis function as kernel function herein:

k(D _i,D _j)＝exp(-||D _i-D _j|| ²/2σ ²) (4)

Wherein σ is the hyper-function of kernel function.

(2) LS-SVM utilizing counting yield higher than SVM realizes latency prediction

Previous step has established based on SVM latency prediction model, utilizes LS-SVM to predict time delay below.

Suppose there is following time delay sample set:

{(D ₁,y ₁),…,(D _N,y _N)},D∈R ⁿ,y∈R (5)

Then the LS-SVM recurrence expression of this sample is:

$\min J=\frac{1}{2}{\mathrm{\ω\ω}}^T+C\underset{i}{\overset{N}{\mathrm{\Σ}}}{e}_i^2---\left(6\right)$

Wherein, C is regularization parameter, and ω is weight coefficient vector, e _itraining error.For asking above-mentioned constrained optimization problem, build as bright in pull-down analog device day function:

Ask extreme value to Lagrangian function, the problems referred to above are converted into and solve following system of linear equations:

$\left[\begin{array}{cccc}0& 1& ...& 1\\ 1& k(D_1,D_1)+\frac{1}{C}& ...& k(D_1,D_N)\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ 1& k\left(D_N{,D}_1\right)& ...& k(D_N,D_N)+\frac{1}{C}\end{array}\right]\left[\begin{array}{c}b\\ {\mathrm{\α}}_1\\ \·\\ \·\\ \·\\ {\mathrm{\α}}_N\end{array}\right]=\left[\begin{array}{c}0\\ y_1\\ \·\\ \·\\ \·\\ y_N\end{array}\right]---\left(9\right)$

In formula, for kernel function, it is Nonlinear Mapping inner product.Solve linear equations, can obtain the recurrence weight coefficient α in k moment _kwith off-set value b, then the Predicting Internet Delay value in k+1 moment is:

$\hat{d}(k+1)=y\left(D_k\right)=\underset{i}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_{i}k(D_i,D_k)+b---\left(10\right)$

3, ripple integral element is designed

(1) based on ripple integral principle design ripple integral element

Had by ripple integral principle (see Fig. 3) and document " Determing strange attractors in turbulence " (TAKENS F.Lecture notes in mathematics, 1981,1:366.):

$\mathrm{\Δ}\left(t\right)={\∫u}_m\left(t\right)\mathrm{dt}-{\∫\hat{u}}_s\left(t\right)\mathrm{dt}---\left(11\right)$

Wherein, Δ can be regarded as the measured value of energy signal, and is multiplied by certain gain σ feeds back to as the energy dissipated so can be expressed as from the wave conversion of end:

$\begin{array}{c}{u}_s\left(t\right){\hat{u}}_s\left(t\right)-\mathrm{\σ\Δ}\left(t\right)\\ ={\hat{u}}_s\left(t\right)-\mathrm{\σ}(\∫u_m\left(t\right)\mathrm{dt}-\∫{\hat{u}}_s\left(t\right)\mathrm{dt})\end{array}---\left(12\right)$

(2) according to latency prediction result, the value function of design ripple integral parameter σ, wherein that control system should be made to disturb to external world is insensitive in the selection of σ, makes system be it often fully compensated simultaneously.In order to make the error in each moment minimum, the value of σ should change along with the change of time delay, and concrete rule is: if prediction time delay larger than current time delay, then σ is negative, and large must be more, σ will become less; On the contrary, if prediction time delay less than current time delay, then σ just be, and little must be more, σ will become larger, but while σ can not be allowed again too large.Its value expression formula is as follows:

σ _k=-a Δ T and (a > 0) (13)

Wherein, Δ T=T _k+1-T _k, namely predict the difference of time delay and current time delay, a is constant factor.

4, the ripple integration bilateral teleoperation system of carrying out latency prediction based on LS-SVM is built, the stability of checking the method under time-vary delay system condition and tracing property

(1) the ripple integration bilateral teleoperation system of carrying out latency prediction based on LS-SVM is built

According to the bilateral control method combined based on LS-SVM latency prediction and ripple integration that Section 4 proposes, development one dimension force feedback bilateral control remote control system.System is mainly divided into 3 parts: main fingerprint block, communication module and from fingerprint block.Main fingerprint block accepts the motion command of operator and the power fed back from hand is rendered to operator, communication module, at main hand and from transmitting position and force information between hand and producing time delay, is followed the motion of main hand from hand and the interaction force with environment is fed back to main hand.In this article, main hand getting Three-degree-of-freetranslation translation force feedback hand controller, from hand getting falcon hand controls, the terminal position of the two is mutually corresponding.

(2) verify that LS-SVM is to the accuracy of latency prediction

Based on said system, setting unidirectional time-vary delay system is fixed delay (2s)+random delay (-1 ~ 1s), predicts, predict the outcome see Fig. 4 to time delay.As can be seen from the figure, utilize least square vector (LS-SVM) to predict time-varying delay, the value very little with true time delay error can be obtained.In order to have more cogency, calculating true time delay with the variance of prediction time delay is 0.03670262, and from variance, latency prediction has higher accuracy.

(3) analytic system stability, and verify by experiment.

Stability analysis:

Conventional wave variable method is not only passive, but also be harmless, when time-vary delay system condition, we add LS-SVM latency prediction and ripple integral element, from the definition of passivity, as long as the link added is passive, so whole system is exactly passive, as long as and add link and do not increase system capacity, it is exactly passive for so adding link, thus whole system is exactly stable.Make to add link passive, then the following condition of demand fulfillment (namely input energy be more than or equal to export energy), as follows:

$\left\{\begin{array}{c}{\∫}_0^t\frac{1}{2}{u}_m^2\left(\mathrm{\τ}\right)\mathrm{d\τ}\≥{\∫}_0^t\frac{1}{2}{u}_s^2\left(\mathrm{\τ}\right)\mathrm{d\τ}\\ {\∫}_0^t\frac{1}{2}{v}_m^2\left(\mathrm{\τ}\right)\mathrm{d\τ}\≥{\∫}_0^t\frac{1}{2}{v}_s^2\left(\mathrm{\τ}\right)\mathrm{d\τ}\end{array}\right.---\left(14\right)$

Had by formula (14) and systematic schematic diagram (see Fig. 2):

$\left\{\begin{array}{c}{u}_m\left(t\right)+\mathrm{\σ}{\∫}_0^t{u}_m\left(\mathrm{\τ}\right)\mathrm{d\τ}\≥u_m(t+T)+\mathrm{\σ}{\∫}_0^t{u}_m(\mathrm{\τ}+T)\mathrm{d\τ}\\ v_s\left(t\right)+\mathrm{\σ}{\∫}_0^t{v}_s\left(\mathrm{\τ}\right)\mathrm{d\τ}\≥v_s(t+T)+\mathrm{\σ}{\∫}_0^t{v}_s(\mathrm{\τ}+T)\mathrm{d\τ}\end{array}\right.---\left(15\right)$

From above formula, the stability of system depends on σ, and the situation of Delay Variation is answered in the value negate of σ, the stability problem that so just can bring by regulating the value of σ to solve time-vary delay system, makes system stable under time-vary delay system condition.To sum up analyze, carry out latency prediction based on LS-SVM and the stability of the remote operating bilateral control method combined with ripple integration depends on the accuracy of latency prediction and the value of constant a.

Based on said system, carry out position tracking test, experimental result is see Fig. 5, as seen from the figure, the ripple integration bilateral teleoperation control method of carrying out latency prediction based on LS-SVM can solve the stability problem of time-vary delay system remote control system, and ensures that remote control system can follow the tracks of the motion of main hand well from hand.

Claims (1)

1., based on the ripple integration bilateral teleoperation control method of LS-SVM latency prediction, it is characterized in that following steps:

Step 1, structure remote control system theory diagram:

(1) design is based on the theory diagram of the ripple integration bilateral teleoperation control method of LS-SVM latency prediction, mainly comprise main hand, main side controller, communication link, from side controller, from hand, LS-SVM latency prediction link, ripple integral element;

(2) meaning of parameters and the relation between them in clear and definite block diagram, mainly comprise the σ of speed, power, master-slave controller parameter, ripple integral element;

Step 2, design latency prediction link based on least square support vector machines (LS-SVM)

(1) build based on the latency prediction model of SVM: utilize SVM theoretical by there is nonlinear characteristic time series by non-linear conversion to high-dimensional feature space, then in this space, extract necessary information, for realizing latency prediction;

(2) LS-SVM utilizing counting yield higher than SVM realizes latency prediction;

Step 3, design ripple integral element:

(1) based on ripple integral principle design ripple integral element;

(2) according to latency prediction result, the value function of design ripple integral parameter σ, wherein that control system should be made to disturb to external world is insensitive in the selection of σ, makes system be it often fully compensated simultaneously.In order to make the error in each moment minimum, the value of σ should change along with the change of time delay, and concrete rule is: if prediction time delay larger than current time delay, then σ is negative, and large must be more, σ will become less; On the contrary, if prediction time delay less than current time delay, then σ just be, and little must be more, σ will become larger, but while σ can not be allowed again too large.Its value expression formula is as follows:

σ _k=-a Δ T and (a > 0)

Wherein, Δ T=T _k+1-T _k, namely predict the difference of time delay and current time delay, a is constant factor.