Consensus for Multiple Euler-Lagrange Dynamics with

Arbitrary Sampling Periods and Event-Triggered Strategy

and enforced only if the state error norm satisfies a
pre- scribed criterion. A theoretical framework for
Abstract— This paper investigates the agreement event-triggered accord protocol is studied in [10].
drawback of multi-agent systems with Euler- The sufficient conditions are derived to ensure
Lagrange dynamics, below uniform sampling and coordination by mistreatment the Input-to-State
absolute heterogeneous sampling schemes, severally. Stability (ISS) theory. its price noting that less
Firstly, we tend to propose a protocol mistreatment attention has been paid
each position and rate info and study the agreement to endeavor the accord drawback for Euler-Lagrange
conditions. Then we have a tendency to extend the dynamics based mostly on the event-triggered
results to the case of applying the event- triggered strategy. Euler-Lagrange equations are often used to
strategy and analyze the corresponding agreement model several sensible systems like rigid body
property. Simulation examples and comparisons dynamics and connected robotic manipulators [12].
verify the effectiveness of the planned ways. The accord drawback of Euler- Lagrange systems is
studied in [13] [14]. In [13], the authors propose a
distributed, leaderless accord formula for Euler-
INTRODUCTION Lagrange systems; the convergence analysis is
conduct- impotence supported the algebraical graph
In applications the networked management systems
theory, Lyapunov theory and Morozov’s Theorem.
and multi- agent systems are enforced beneath a
In lightweight of the higher than analysis, we've a
sampled-data frame- work [1]. inside
bent to are actuated to develop event-triggered accord
the literature, beneath the sampled-data frame-
protocols for the linearized Euler-Lagrange
work, accord protocols are investigated by solely
dynamics, by incorporating each position
mistreatment the position data of the agent and its
and rate data of the agents. the rest of this chapter is
neighbors [2], [3] and
organized as follows. Section II reviews some
[4]. once endeavor the accord drawback, the
background and necessary definitions and
commonly used strategies embrace the property
presents the matter formulation. Section III reports
analysis of infinite merchandise of random matrices,
the most results on making
linear matrix inequalities (LMIs) and Lyapunov
certain accord beneath the whimsical uniform
stability analysis [5], [6]. There are 2 main
and heterogeneous sampling ways, severally. In
schemes in line with the various sampling pattern:
Section IV, the centralized event-
Uniform sampling and heterogeneous sampling. In
triggered management protocol for the linearized
[7], the authors propose a accord protocol supported
Euler-Lagrange dynamics is projected. In Section V,
detectable position data of the agent and its
simulation results supported the Euler-Lagrange
neighbors beneath sampled-data settings;
model for a revolute joint arm are provided to verify
the projected protocol will increase the uniform
the effectiveness of 978-1-4799-5825-2/14/$31.00
sampling amount such the agreement is bonded.
©2014 IEEE 2596 the planned algorithms. we have a
The heterogeneous sampling mode is
tendency to conjointly gift the comparison of the
additionally studied in [7]. However, as a result of the
convergence speed between the planned algorithms
utilization of solely position data, the convergence
and people solely mistreatment position info. Finally,
rate of the accord might not be quick. In observe,
conclusions are offered in Section VI. " " − −
if each position and rate data are measurable, then
Notation: The superscript ‘T’ represents the matrix
intuitively and naturally, the convergence rate of
trans- create. E denotes the mathematical expectation.
the accord would be expected to be quicker if they
A matrix P > zero if and providing P is bilaterally
are at constant time incorporated into the
symmetrical and positive definite. ‘*’ during a matrix
protocol style. The “event-triggered”
stands for a term of block that's evoked by symmetry.
strategy basically permits the heterogeneous samplin
one denotes vector [1, 1..., 1] T, zero denotes zero
g [8]. The management action is enabled
matrix with correct dimension and that i is that the
unit matrix. . denotes the Euclidian norm. C denotes j, i ƒ= j. The graph Laplacian L ∈ Rn×n is defined as:
the complicated numbers and Rm×n stands m by n lij = −aij, ∀i ƒ= j; 52 =aij, i,j = 1, 2,..., n. j∈NiBy
matrices. Matrices are assumed to be compatible with definition, there's a novel corresponding L to any A.
pure mathematics operations. drawback Euler-Lagrange systems are diagrammatic by ∈
Mi(qi)q¨i + Ci(qi, q˙i)q˙i = τi,i = 1,..., n.(1) wherever
FORMULATION
qiRp is that the vector of generalised coordinates, ∈
We have a tendency to denote a weighed graph with Mi(qi)Rp×p is that the inertia matrix that is
n nodes by G = (V , E , A). V = represents the nodes bilaterally symmetrical ∈ and positive definite. Ci(qi,
set and E ⊆ V × V is that the edge set. a footing (vj, q˙i)q˙i Rp is that the vector of Coriolis and
vi) ∈ E is that the and rate info of every agent and its centrifugal torques. τi is that the vector of the torques
neighbors at the same time at a sequence of your time exerted to ith agent. we've the subsequent
instants zero, t1, t2,.. ., with tk < tk+1,k = 1, 2,.. .. assumptions: − "" ≤"" ≤" " ≤ There exist kilometer,
Considering a gaggle of n agents with Euler- km¯ and kHz > zero, for any agent, zero kilometer
Lagrange dynamics (1), we've the subsequent <Mi(qi)km¯ ; (2)Ci(qi, q˙i)kc q˙i ; (3) M˙ i(qi)2Ci(qi,
protocol: q˙i) is skew bilaterally symmetrical. during this
chapter, firstly we have a tendency to propose a
τi = Ci (qi , q̇i )q̇I + Mi (qi )−1(−k1 (q̇i (t) − q̂˙i (tk ) − k2 (qi (t) bearing protocol by incorporating each position and
− q̂i (tk ))), rate info. Assume that the system measures and
obtains the position j∈Ni G ij1 − ρ aij, i = j
where
Remark 1: Compared with Σ the algorithmic rule in [7],
q̂i (tk ) = qi (tk ) + ρaij (qj (tk ) − qi (tk )), j∈Ni Σ
the pro- display protocol has incorporated the rate
q̂˙i (tk ) = q̇i (tk ) + ρ aij (q̇j (tk ) − q̇i (tk )). j∈Ni
feedback in (3). below the planned protocol, the
solvability of agreement drawback are going to be
Here, k1, k2 are positive management gains with studied within the sequel. we are going to study each
relevance rate and position info of the agent and its the uniform and therefore the heterogeneous
neighbors, info flow from vj to vi. sampling schemes. Note that the properties of W are
going to be applied within the following analysis and
The plus nearness and zero < ρ < 2max{Σ one ΣΣ Σ discussion. Lemma one in [7] shows that if A is
ΣΣΣ j∈Ni aij,i=1,2,...,n} . By work (2) ∈≥∀ matrix A bilaterally symmetrical, so is W , and W may be a
= [aij]Rn×n, (aij0, i, j = 1, 2,..., n), models the double random matrix and diagonally dominant. If
communication topology among the agents. If into the interaction topology includes a spanning tree, one
(1), the general system dynamics will be written is that the algebraically straightforward eigenvalue of
within the following form: there's an immediate link W . Besides, there exists Associate in Nursing
from agent j to agent i, which suggests agent i orthogonal matrix P specified PWP T = diag(μ1,
receives info from agent j, then aij ƒ= zero; q˙i(t) μ2,..., μn), wherever μ1 = one, μ2,..., μn are eigen-
q¨i(t) =01 −k2 −k1 qi(t) q˙i(t) alternatively aij = 0. values of W in decreasing order. diag(μ1, μ2,..., μn)
Associate in Nursing rudderless graph implies a rep- resents the square matrix with μi because the ith
graph of Σ00 ΣΣ qˆ (t) Σ that the link from i to j still entry on the diagonal.
because the link from j to i exist and disappear
synchronously during a graph. Otherwise the graph is AGREEMENT ANALYSIS OF THE MASS
directed. A path from vertex i to a vertex j may be a WITH ABSOLUTE SAMPLING PERIODS
−−k2 −k1 iqˆ˙i(t) ,t ∈ [tk, tk+1). (3) sequence of
During this section, we are going to study the
distinct vertices beginning with i and ending with j,
agreement of the multiple Euler-Lagrange dynamics
specified consecutive vertices are adjacent [15]. we
in (1) with the planned protocol below uniform
have a tendency to use Next, we have a tendency to
sampling and heterogeneous sampling cases,
define the matrix W = [wij]: nickel to denote the
severally. Define xi(t) = [qi(t), q˙i(t)]T. when
neighbor set of agent i. The neighbors of indicate the
resolution the equation in (3), we have a tendency to
agents from that i receives info. Thus, w = .Σρaij,i ƒ=
acquire to analyze the eigenvalues of Di, we've det(sI
j .j∈Ni Σ N i = vj : (vj, vi). Assume that Associate in
− D ) = (1 − μ )(s − eλ1h)(s − eλ2h) iix (t) =
Nursing agent doesn't receive info from itself. With
eK(t−tk)x (t ) Σ ii k ΣΣΣ λ2eλ1h − λ1eλ2h + (I −
this assumption, we've aii = zero and aij ≥ 0 for all i,
eK(t−tk)) 01 −k2 −k1 qˆi(tk) qˆ˙i(tk) , (4) + μi(s − a diagonal block matrix with Di Σ eλ1h zero on the
1)(s − Next, we want to indicate ).(6) λ2 − λ1 diagonal line. Σ ⎡ D1 ⎢ x(tk) = (P ⊗ T )−1 ⎣ . . . k ⎦⎥
(P ⊗ T )x(0). Di = 0eλ2h .Σ + i1 one two λ1 1 μΣ −λ
Define x(t) = Σx (t)T,x (t)T,...,x (t)TΣT, we've λ2 − (1 − eλ h) − λ2 (1 − eλ h) Σ 2 Dn(8) λ2 − λ1 λ1 (1 −
λ1 wherever K = Σ01 Σ is Hurwitz stable. λ eλ1h − λ eλ2h)λ2(1 − eλ1h) Since D1 includes a 1 eigenvalue
eλ2h two −k2 −k1 max < one< 1.(7) λ2−λ1 λ2−λ1 λ λ and therefore the alternative eigenvalue in −
one two n 1st, contemplate the case λ < λ . it's pronto μiλ2(1−eλ1 h) stable. From (8), we have a tendency
obtained x(tk + one) = I ⊗ e k k that λ2e 1 λ1e two > to conclude that limk→+∞ ⎡ ⎤ λ2max, one eλ1h −
λ2e 2 λ1e 2 = eλ2 > eλ1 . To prove the λ λ .K(t +1−t ) μiλ1(1−eλ1 h) two and Di,i = 2, 3,...,n are Schur x(tk)
−− twelve K(t +1−t ) Σ 1k1 ΣΣ right-hand aspect of
asymp- μiλ1(1−eλ2 h) λ2(λ2−λ1) = ⎣2 − λ2−λ1 two
the difference in (7), we've +W ⊗ (I − ek k )k2 00
eλ2h + μiλ2(1−eλ2 h) λ2−λ1 λ1(λ2−λ1)⎦ .x(tk)
x(tk). (5) λ2eλ1h − λ1eλ2h λ2 − λ1 − one = λ2(eλ1h
asymp- totically converges to some steady state. Case
− 1) − λ1(eλ2h − 1) . λ2 − λ1 we have a tendency to
2: Dapsang = 4k2. one Case 2: Dapsang = 4k2. two 2
currently gift the agreement leads to the subsequent
k1 2 one 2 2 Let λ = − k1 , T = Σ 0−1 Σ and y (t) = (P
Theorem. G Theorem 1: If the communication
⊗ T ). within the interval (−1, 1). Therefore, we will
topologyhas a span- ning tree, by applying the
show that, with absolute sampling periods, the
protocol in (2), the agreement of The difference in (7)
planned protocol in (2) will equally, we have a
holds if and providing eλ1 h−1 < eλ2 h−1 . λλ xh12 x
tendency to acquire the evolution of y2(t) as follows
Let f (x) = e −1. Next we have a tendency to show f
y2(tk+1) = Ξy(tk), ⎡⎢ ⎦ wherever Ξ is additionally the
(x) is monotonically decreasing once x < zero. the
block diagonal system matrix. we've make sure the
multiple Euler-Lagrange dynamics system in (1) will
agreement. within the next section, we are going to
be achieved with any given sampling amount h once
propose a replacement centralized event-triggered
Dapsang ≥ 4k2. f j(x) = 1+ (xh1)exh− . one 1 Proof:
protocol. and D1j ⎣ Ξ = . . . Dnj ⎤⎥ , (9)
we have a tendency to contemplate the cases once
Dapsang > 4k 4k2, severally. one two and Dapsang = CENTRALIZED EVENT-TRIGGERED
x2 Let z(x) = (xh − 1)exh, and zj(x) = xh2exh. it's STRATEGY
shown 1ΣΣ Case 1: Dapsang > 4k2. The Hurwitz
matrix K has 2 negativeλ−1 that z(x) is In sensible networked dynamic systems or networked
monotonically decreasing once x < zero. we have a multi-agent systems, the network information
tendency to haveminz(x) > z(0) = −1. Thus, f j(x) > measure and resources are restricted. to form full use
zero. With the sameeigenvalues λ1 and λ2. we have a of the restricted resources, if the concerned variables
tendency to define T1 = and a pair of −λ11 argument, or signals don't vary significantly, one might opt for
for the case λ1 > λ2, the difference in (7) holds. ⊗ to not sample the signals sporadically with a Then Dij
y1(t) = (PT1)x(t), wherever P is that the orthogonal Σ = eλh − μiλheλh −λheλh + μiλheλhΣ μ (eλh 1)eλh + μ
matrix mentioned in Remark one . Thus, we've T KT (1eλh) Σ .i−i − quick frequency; instead, one will
−1 = From the on top of analysis, it's shown that D1 sample or update the signals providing some pre-
has 2 that2− 1D ,i = 2, 3,..., n eigenvalues, i.e., one scribed live are going to be satisfied. Here, we have a
and λ2eλ1 −λ1eλ2 . From (6) we have a tendency to tendency to propose a replacement protocol
conclude λ1 zero 0λ2 Σ and 11λ λ ..ΣΣ.Σ∪ i, has 2 supported the centralized event- det(sI − Dij) = (1 −
eigenvalues within the interval- smin eλ1h, eλ2h), μi)(s − eλh)2 λhλh+ μi(s − 1)(s − e+ λhe ).(10) Next,
max(eλ1h, eλ2hλ2eλ1 −λ1eλ2 , 1 . Therefore, Di is we want to indicate zero < eλh − λheλh < one. Let f
Schur stable. .λ2−λ1 I ⊗ eK(tk+1−tk) √ y1(tk+1) = (x) = exh − xhexh, we've f j(x) = −xh2exh < zero,
(P ⊗ T1) +W ⊗ (I − eK(t k+1 k1 −tk))k2 00 P is once x < zero. so f (x) is monotonically decreasing.
Associate in Nursing orthogonal matrix with the first limx→−∞ f (x) = zero and limx→0 f (x) = one. D1j ,
row (1/ Σ one ΣΣ we've n)1T. × (P ⊗ T1)−1y1(tk). Σ has 2 eigenvalues: one and triggered strategy
It will be any written as ⎡ D1 .⎢ . . y1(tk) = (P ⊗ T incorporating each position and rate info of the agent
)x(tk) = ⎣ k k ⎤ Dn ⎥⎦ y1(0) y1(tk + 1) = .I ⊗ eλ1h and its neighbors. for every agent i (i = one, 2,..., n),
zero 0eλ2h Σ + PWP T ⎡ D1⎤ ⊗ Σ− 1eλ1h 0 λ h Σ T1 we have a tendency to contemplate the second- order
dynamics delineate by (3). Define the new variable
k1 Σ Dapsang T1−1 y1(tk), = ⎢⎣ . . . ⎦⎥ (P ⊗ T )x(0).
X(t) = [q1(t), q2(t),..., qn(t), q˙1(t), q˙2(t),..., q˙n(t)]T
01 − e 200DnΣ one − ⎤ Σ wherever h = tk+1 tk,k = 0,
. Then we've eλh − λheλh. Dij,i = 2, 3,...,n has 2
1,.. .. is that the uniform samplingTherefore, period.
eigenvalues within the − ˙Σ0IΣ one interval (eλh
The on top of state matrix within the brackets may be
λheλh, 1). Dij is Schur stable. Then with the similar negative. we have a tendency to claim that R may be
analysis, once Dapsang = 4k2, the planned protocol a negative definite matrix. −∈ Define a blunder term
in (2) will make sure the agreement. wherever X(t) = e(t) = X(tk)X(t),t[tk, tk+1). For the event-triggered
k2W˜ k1W˜ X(t),(11) Remark 2: In light-weight of strategy, the management input are going to be ···−
the elegant leads to [7], we have a tendency to unbroken as a relentless between any 2 triggering
analyze the agreement conditions for the linearized time instants t0, t1, . If let R˜ = R, then R˜ is positive
multiple Euler- Lagrange dynamics system below the definite. the general system will be delineate as X˙ (t)
planned protocol that W˜ =⎡ − Σj∈N1 w1j = −R˜X(tk), t ∈ [tk, tk+1).(12) r ImRe Fig. 1. an
w12···w1n⎤ ⎢ incorporates not solely the position indication of Gerˇsgorin Theorem applied to W˜
however conjointly the rate infor- w21− . j∈N2 w2j .Consider the subsequent Lyapunov perform
···w2n⎥ .Σ . . wn1wn2···− mation, whereas the lead willdidate can conclude that the velocities of the
to [7] considers double-integrator⎢⎣. .. . ..⎥⎦ j∈Nn wnj agents converge to zero. Further, the positions are
Σ systems and solely uses the position info. on an adequate to a gradual state with the weighted average
analogous line, we have a tendency to study the case with relevance W˜
once management gains k1 and Dapsang satisfy some
condition specified the planned control protocol in ILLUSTRATIVE EXAMPLES
(2) will make sure the agreement. Since the extra rate During this section, 2 numerical examples are going
info has been incorporated into the protocol Lemma to be given to verify the planned management
1: [16] (Gersgorin Theorem) Let ij=1, jƒ=iˇtt = [gij] ∈ protocols for a gaggle of revolute joint arms. within
≤ R (tt). Rn×n, and let Rt(tt) ≡ Σn|gij|, 1 ≤ i ≤ n. Then the first set of simulation, we have a tendency to
all the eigenvalues of tt are settled within the union of show that by applying the protocol in (2), agreement
n discs: rate info has been incorporated into the will be reached and therefore the convergence rate is
protocol style, unsurprisingly, the convergence speed improved compared to the algorithms given in [7].
are going to be quicker than the leads to [7]. The The second set of simulation is conferred to
comparison study is provided within the simulations. demonstrate the event-triggered strategy: agreement
National Trust Σ i=1iLemma 2: [16] If M =AB ×− will be achieved with quicker speed than the strategy
CD Σ, where A, B, C, D are Remark 3: For the by solely mistreatment the position info [10]. ··· it's
heterogeneous sampling theme, Theorem two in [7] assumed that for each joint arm the dynamics
proves that the management protocol while not rate parameters are the same: The mass of the link is
feedback will guarantee the agreement if any zero.6kg, the length of the link is zero.5m, the
sampling amount hk is larger than a given bound. the distances from the joint to the middle of mass of the
most technique nn matrices and CD = DC, then link is zero.25m, the instant of inertia of the link is
det(M ) = det(AD ΣΣ BC). Σ k2W˜ k1W˜ currently zero.2kgm2. Six joint arms begin from random Then
we have a tendency to study the eigenvalues of R we've V (t) = one XT two (t)R˜X(t). initial conditions
=0I. of the proof lies within the following: One block qi(0) and q˙i(0), i = 1, 2, , 6. Then the states of agents
square matrix of the evolution equations for y(tk) is evolve below the management protocol in (2) within
with eigenvalues one FromLemma2,weobtaindet(sI − the following cases. Figure two shows the fixed
R)= sI−I Σ det = det(s2 − k sW˜ − k W˜ ). of the system topology 2 three one four vi five X˙ (t) =
evolution equations for y(tk) is with eigenvalues one −R˜X(t ) = −R˜(X(t)+ e(t)),t ∈ [t ,t), with a spanning
and another real worth within the interval (−; tree. k V˙ = −XT(t)R˜R˜(X(t)+ e(t)) = −"R˜X(t)"2 −
alternative block −k2W˜sI − k1W˜1 two 1, 1) XT(t)R˜R˜e(t) k k+1 ≤ −"R˜X(t)"2 +
diagonal matrices are all Schur stable. Then it will be "R˜X(t)""R˜""e(t)".(13) so, if we have a tendency to
tested that infinite merchandise of such system set ""≤ e(t)"R˜X(t)" ,(14) "R˜" we will enforce the
matrices asymptomatically converge to bound values. Lyapunov perform to remain negative. How- ever, it
below the planned protocol mistreatment Lemma can be sensitive and risky by setting the triggering ""
one, we tend to acquire that W˜ is negative definite time directly supported e(t) = "R˜X(t)". Sometimes,
(see Figure 1). Let ui < zero,i = 1, 2, ·· · , n be the once "R˜" "" e(t) satisfies (14), however gets terribly
eigenvalues of two 2 √ W˜ . so we've s = k1ui± k u on the point of "R˜X(t)", consequent "R˜" Fig. 2.
−4uik2 . the $64000 a part of one i2 with further rate Communication topology with a directed spanning
info, we've tested that the system matrices of y1(tk) tree. Fig. three shows the comparison results
and y2(tk) are each block diagonal matrices with one concerning the conver- gence speed between
eigenvalue one and alternative eigenvalues being s is
management protocols with and while not the rate we have a tendency to planned a protocol
info [7] once Dapsang > 4k . It will be discovered ≤ incorporating the position and rate info, below
checking of "e(t)" might fail specified the spinoff of uniform and heterogeneous sampling schemes,
the12 Lyapunov perform might not be negative. severally. Then, we tend to plan a replacement
Thus, we have a tendency to set the triggering time centralized event-triggered strategy for the linearized
once "e(t)" satisfies ""≥ e(t)σ "R˜X(t)" ,(15) "R˜" Euler-Lagrange dynamics. Simulation results verified
≤−"" wherever zero < σ < one and V˙ (σ 1) R˜X(t) the effectiveness of the planned ways.
two. Once the triggering is activated, the error is
about to zero and therefore the management input REFERENCES
remains constant as planned in (12) till consequent [1]. T. Chen and B. Francis, optimum Sampled-data
triggering is activated. currently we have a tendency management Systems. New York: Springer, 1995.
to study the agreement property below the planned Control, vol. 57, no. 12, pp. 3230–3235, Dec. 2012.
protocol and therefore the triggering mechanism.
Theorem 2: If the communication topology is
connected, by applying the planned protocol and [2]. Y. Cao and W. Ren, “Sampled-data discrete-
presumptuous zero < σ < one, all the agents will time coordination algo- rithms for double-
asymptotically reach the weighted average initial integrator dynamics under dynamic directed
position and therefore the velocities converge to zero. interac- tion,” International Journal of Control,
Proof: Since V˙ ≤ (σ − 1)"R˜X(t)"2, we will see that vol. 53, no. 9, pp. 2142–2148, Oct. 2008.
link→∞ R˜X(tk) = zero. From the structure of R˜ and
(12) we have a tendency to that: once the uniform [3]. J. Wu and Y. Shi, “Consensus in multi-agent
systems with random delays governed by a markov
sampling periods are little (h 1), the protocol (2)
chain,” System & Control Letters, vol. 60, no. 11,
demonstrates higher performance; once the sampling
pp. 863–870, Jun. 2011.
periods are massive, there's no obvious distinction on
the convergence speed between these 2 protocols. It [4]. H. Li and Y. Shi, “Robust distributed model
will be seen that the convergence speed has been predictive control of constrained continuous-time
improved. The similar trend will be discovered for nonlinear systems: A robustness con- straint
the case Dapsang = 4k2; it's omitted here. one The approach,” IEEE Transactions on Automatic
second set of simulation is to verify the event- Control, Dec. 2013, “Distributed receding horizon
triggered strategy as shown in (12). we elect σ = control of large-scale nonlinear systems: Handling
zero.6. Fig. four shows that the agreement is reached communication delays and disturbances,” Automat-
supported the planned protocol (12) and therefore the ica, vol. 50, no. 4, pp. 1264–1271, Apr. 2014.
one while not rate info ([10]), whereas the previous
converges quicker that the latter case. Fig. five shows
the "e(t)" and therefore the certain of triggering.

CONCLUSION

During this paper, we've investigated the agreement

prob- lunar excursion module for multi-agent systems
with Euler-Lagrange dynamics,