You are on page 1of 5

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCSII.2018.2884006, IEEE
Transactions on Circuits and Systems II: Express Briefs

Leader-follower Bipartite Consensus of Linear


Multiagent Systems over a Signed Directed Graph
Sourav Bhowmick∗ , Student Member, IEEE and Surajit Panja† , Member, IEEE
Indian Institute of Information Technology Guwahati, Assam-781001, India
Email: ∗ sourav@iiitg.ac.in, † surajit@iiitg.ac.in

Abstract—In this work, leader-follower bipartite consensus the other followers to track and reach consensus on a fixed
of a group of linear multiagent systems is studied over a value. One pertinent problem to bipartite consensus of leader-
signed directed graph where all the followers are subjected follower MAS over a directed signed graph containing the
to mismatched unknown bounded disturbances. To that end, a
distributed extended state observer (ESO) where the disturbances leader is the vanishing of M-matrix and symmetric matrix
are the extended state and a distributed relative output feedback properties for the Laplacian matrix of the graph which is one
consensus control law are designed to achieve bipartite consensus. of the main issues addressed in this work.
With the suitable design of scaler coupling gain parameters, and Most of the practical engineering systems are rarely ideal and
feedback gain and observer gain matrices obtained from linear free from external disturbances or parametric uncertainties.
matrix inequalities (LMI), it is shown that if the signed directed
graph contains a directed spanning tree rooted at the leader node, Therefore, consideration of these phenomena in the plants’
then the designed consensus protocol simultaneously stabilizes dynamics maintaining closed loop stability is challenging,
both the bipartite consensus error and observer estimation error yet more practical. [13] and [14] study bipartite consen-
with the errors converging to a small bounded set. The simulation sus problems of linear MAS subjected to bounded matched
results verify the effectiveness of the proposed method. uncertainties and matched disturbances with state feedback
Index Terms—Multiagent systems, bipartite consensus, signed
directed graph, mismatched disturbances. continuous/discontinuous and parameterized control laws, re-
spectively over signed undirected graphs. However, matched
uncertainties or disturbances are assumed to be acting via
I. I NTRODUCTION
the same channel as the control input, which is restrictive
MAS and their cooperative control have received a tremen- from practical viewpoint. Moreover, the comparison of state
dous amount of research impetus worldwide over the past feedback law [15] and output feedback law [16] is very well-
decade due to widespread applications in various fields such known in the literature. Compared to state feedback control
as robotics and unmanned aerial vehicles [1], surveillance and laws in [13], [14] or observer-based absolute output feedback
security [2], opinion dynamics and social networks [3], [4], laws in [17], [18], control laws using only relative output
etc. Most of the literature on MAS and its cooperative control information are more suitable in real-world applications. All
consider two interacting agents collaborating with each-other of these issues form the basis of motivation of this work.
to realize a control task. However, there exists certain practical In this paper, bipartite consensus of linear leader-follower
situations when the interests or objectives of agents differ MAS network over a signed directed graph is studied where
from one-another, e.g., social networks [4]. The interactions the followers are subjected to unknown mismatched bounded
between agents in these situations can either be cooperative or disturbances. To the best of the authors’ knowledge, bipartite
competitive. Instead of regular unsigned graphs, signed graphs consensus problem of linear MAS subjected to unknown,
are more suitable in such cases where a positive edge refers mismatched disturbances over a signed directed graph has not
to a cooperative interaction while a negative edge represents been studied yet. Based on ESO-based control [19] where
competitive interaction between two agents. In a structurally the disturbances are the extended state, this work proposes
balanced signed graph, [5] introduces ‘bipartite consensus’ a distributed ESO-based relative output feedback controller to
where the single integrator agents reach state consensus in achieve bipartite consensus. The contributions of this work
modulus but opposite in sign. Since then, bipartite consensus are threefold: (i) Compared to first or second order dynamics,
has been studied from various perspectives for single or this work considers more generalized linear MAS dynamics
double-integrator agents [6]–[9] and general linear and higher- where the followers are also subjected to more practically
order agents [10]–[14]. However, barring [9]–[11] which are plausible mismatched lumped disturbances which can consti-
leaderless MAS, most of these works have considered undi- tute unmodelled dynamics, parametric uncertainties, external
rected signed graph as the underlying communication network. disturbances, etc. By designing a suitable disturbance gain
Compared to undirected graphs, directed graphs are more compensator, the effects of disturbances are reduced; (ii)
generalized and cost-effective in the sense that edge weights The underlying communication graph containing the leader
can be arbitrary between two agents. Moreover, MAS are real- considered in this work is a signed directed graph where
world plants, and therefore, it is imperative that there should the Laplacian matrix is neither an M-matrix nor a symmetric
be an ideal reference trajectory generator or the leader for matrix to use the properties. By exploiting diagonal domi-

1549-7747 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCSII.2018.2884006, IEEE
Transactions on Circuits and Systems II: Express Briefs

nance, positive stability, and gauge transformation related to is called degree matrix. A matrix H ∈ RN ×N is defined such
the signed Laplacian, this problem is solved in this work by that H = L + Dp . Let, ∆ = (H + H T ) and ∆ = HH T .
proposing the Lemma 1; (iii) This work considers an ESO Structurally balanced signed graphs are important for bipartite
and a distributed relative output feedback control law by consensus [5]. A signed graph G(V, E, As ) is said to be
considering the disturbances as the extended state. Observer- structurally balanced if its node set V can be divided into
based relative output feedback control laws are more suitable two disjoint groups V1 and V2 , i.e., V1 ∪ V2 = V, and
than state feedback or absolute output feedback laws in real- V1 ∩ V2 = ∅, such that aij ≥ 0, ∀vi , vj ∈ Vq , or ∀vi , vj ∈ Vr ,
world applications. and aij ≤ 0, ∀vi ∈ Vq , ∀vj ∈ Vr where q ̸= r, and
The rest of the paper is organized as follows. Section II q, r ∈ {1, 2}. Signature matrix S is defined in such a way
provides notations and relevant basics of graph theory followed that S = diag{s1 , s2 , . . . , sn } where si = 1, ∀vi ∈ Vq and
by system model and problem statement in Section III. Section si = −1, ∀vi ∈ Vr where q ̸= r.
IV provides main results of this work followed by simulation
III. S YSTEM M ODEL AND P ROBLEM F ORMULATION
results in Section V, and concluding remarks and future scope
of work in Section VI. A. Agent dynamics
The N followers subjected to external mismatched bounded
II. N OTATIONS AND BASICS OF G RAPH T HEORY
disturbances in the MAS are as follows:
A. Notations
ẋi (t) = Axi (t) + Bc ui (t) + Bd wi (t)
Let, ℑn = {1, 2, . . . , n}. Rn and Rm×n denote the set
of the n-dimensional Euclidean space and the m × n real yi (t) = Cxi (t) (1)
matrices, respectively. AT denotes the transpose of matrix A. The leader in the MAS is given as below:
A > 0 (A ≥ 0) and A < 0 (A ≤ 0) denote positive (semi)-
definite and negative (semi)-definite matrices, respectively. ẋ0 (t) = Ax0 (t)
λmax (A) and λmin (A) are maximum and minimum eigenvalues y0 (t) = Cx0 (t) (2)
of matrix A, respectively. I represents a real identity matrix
of appropriate dimension. 1n ∈ Rn and 0 denote a vector of where xi (t) ∈ Rn , ui (t) ∈ Rm , wi (t) ∈ Rm , and yi ∈ Rm are
1s and a matrix of 0s of appropriate dimension, respectively. the state, control input, external mismatched bounded distur-
⊗ denotes the Kronecker product operation between two bances, and output vectors for the ith follower, respectively,
matrices. ||.||2 denotes the Euclidean norm of a real vector. while x0 (t) ∈ Rn and y0 (t) ∈ Rm are the state and output
A diagonal matrix is denoted by diag{f1 , f2 , . . . , fn } with vectors for the leader, respectively.
fi (i ∈ ℑn ) being diagonal entries. sign(.) represents the Assumption 1: The signed digraph G is structurally balanced,
signum function. abs(.) denotes absolute values of elements and augmented graph G has a directed spanning tree rooted
of a matrix. max(.) and min(.) represent the maximum and at the leader. This ensures that the unpinned agents can also
minimum between two scalers, respectively. ∅ denotes the receive relayed leader information over time.
null set. Assumption 2: The unknown mismatched external distur-
bances wi (t) and ẇi (t) (i ∈ ℑN ) are bounded, i.e.,
B. Graph Theory ||wi (t)||2 ≤ ω and ||ẇi (t)||2 ≤ ω where ω, ω > 0.
A weighted signed directed graph (digraph) G = (V, E, As ) The following is the definition of bipartite consensus:
is considered where V = {v1 , v2 , · · · , vN } is node set, E ⊆
lim ||xi (t) − si x0 (t)||2 ≤ γ, i ∈ ℑN (3)
{(vi , vj ) : vi , vj ∈ V, j ̸= i} is edge set, and As = [aij ]N ×N t→∞
is weighted signed adjacency matrix. If (vj , vi ) ∈ E, then where γ > 0 is a scaler, and si is as defined in Section 1.
aij ̸= 0; aij = 0, otherwise. Moreover, aij ≥ 0 and aij ≤ 0 Let, ϵxi (t) = xi (t) − si x0 (t) be the bipartite consensus error.
represent cooperative interaction and competitive interaction Then, (3) can be equivalently written as:
between two agents, respectively. Ni = {vj : (vj , vi ) ∈ E}
represents the set of all the neighbouring nodes of ith node. lim ||ϵxi (t)||2 ≤ γ, i ∈ ℑN (4)
t→∞
The graph contains a directed path from node vp1 to vpm if
B. Problem statement
(vpk , vpk+1 ) ∈ E (k ∈ ℑm−1 ). A digraph is said to contain
a directed spanning tree if there exists atleast a node, called For any initial conditions of MAS (1) and (2), the objective
the root which has a directed path to every other node in the is to find distributed ESO-based relative output feedback
graph. The signed directed graph containing the leader v0 is control protocol ui (t) (∀i ∈ ℑN ) such that (4) is achieved.
the augmented graph G = (V, E) where V = V ∪ {v0 } and IV. M AIN W ORK
E ⊆ {(vi , vj ) : vi , vj ∈ V}, j ̸= i}. A diagonal matrix Dp =
diag{b1 , b2 , . . . , bN } is called pinning matrix where bi > 0, A. Useful lemmas
if ith follower (pinned agent) can receive direct information Lemma 1. Under Assumption 1, there exists a diagonal matrix
from the leader, otherwise bi = 0 (unpinned agent). The G = diag{g1 , g2 , . . . , gN } > 0 such that GH + H T G > 0
[ ]T
Laplacian matrix of∑ the signed graph
∑ is given as ∑ L = Ds − As where G can be found out from g = g1 g2 . . . gN =
where Ds = diag{ j∈N1 a1j , j∈N2 a2j , . . . , j∈NN aN j } abs((H T )−1 )1N .

1549-7747 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCSII.2018.2884006, IEEE
Transactions on Circuits and Systems II: Express Briefs

Proof. From Assumption 1, H is a positive stable matrix [20]. where Kc ∈ Rm×n and Kd ∈ Rm×m are feedback gain and
Therefore, H −1 exists. Moreover, from the definition of H, H disturbance compensation gain, respectively which are to be
is a diagonally dominant matrix. Thus from [21], there exists designed later. The following proposed Theorem underscores
a diagonal matrix Q ∈ RN ×N > 0 such that QH + H T Q > the principle works of this paper.
0. Furthermore, through gauge transformation [5] of signed
Theorem 1. MAS (1) and (2) interacting over a structurally
Laplacian L, L = SLS is obtained where S is signature matrix
balanced signed directed graph having its root at the leader
defined in Section 1, and L is the Laplacian matrix of the
achieve bipartite consensus in the sense of (4) with ESO (6)
corresponding unsigned graph. Then, H = SHS = L + Dp
and distributed relative output feedback control law (7) if the
where H is a nonsingular M-matrix [22]. Therefore, there
T designed parameters are such that:
exists a diagonal matrix G > 0 such that GH + H G > 0. 1+λmax (∆) ′
Pre- and post-multiplying this inequality by SS = I yields 1) scaler coupling gains κ ≥ λmin (∆) > 0 and κ ≥
λmax (G)
S(GH + H T G)S > 0 which implies GH + H T G > 0 where δ > 0 where ∆ = (H + H ), ∆ = HH T , and T

H is associated with the signed graph. Therefore, one natural GH + H T G ≥ δI


choice of Q is G, i.e., Q = G. Furthermore, for nonsingular 2) λmin (G)ρ2 I ≥ P1 Bc BcT P1 , λmin (G)η 2 I ≥ BdT Bd ,
M-matrix H, G is found out as G = diag{g1 , g2 , . . . , gN } R = diag{ρ2 κ2 I, η 2 I}, where ρ, η > 0, β ≥
[ ]T T
where g = g1 g2 . . . gN = (H )−1 1N [23]. It is then max( λmax1(P1 ) , λmax1(P2 ) ) > 0, feedback gain matrix

Kc = −κBcT P1 , observer gain matrix L = κ P2−1 C,
T
sufficient to prove that (H )−1 = abs((H T )−1 ). To that
T disturbance compensation gain Kd = (C(A +
end, gauge transformation yields H = SH T S which implies
T
(H )−1 = S((H T )−1 )S = abs((H T )−1 ). This completes the Bc Kc )−1 Bc )−1 C(A+Bc Kc )−1 Bd where P1−1 = W >
proof. 0 and P2 > 0 are the solution of the following LMIs,
respectively:
Remark 1: Without any loss of generality, it is considered
that GH + H T G ≥ δI where δ > 0. AW + W AT − Bc BcT + βW + I < 0 (8)
[ ]
Lemma 2. ∀x, y ∈ Rn , 2xT ΩΣy ≤ xT ΩP ΩT x + T T
A P2 + P2 A − C C + βP2 + R P2 < 0
y T ΣT P −1 Σy, where Ω and Σ are any matrices of compatible (9)
P2 −I
dimensions and P > 0.
B. ESO and relative output feedback control protocol Proof. Let, exi (t) = xi (t)−zxi (t) and ewi (t) = wi (t)−zwi (t)
be the state estimation error and disturbance estimation error
Considering wi (t) to be an extended state variable, the
for ith follower, respectively. Moreover, ei (t) = xi (t) −
following is obtained from (1): [ ]T
zi (t) = eTxi (t) eTwi (t) be the stacked estimation error
ẋi (t) = Axi (t) + Bui (t) + Ddi (t) vector for the ith follower. Therefore, estimation error for ith
yi (t) = Cxi (t) (5) follower can be obtained using (5)-(7) as:
[ T ]T ∑
where xi](t) = [xi (t)
[ ] wiT (t) , di (t) = ẇ[i (t) ] and A = ėi (t) = Aei (t) + Ddi (t) − LC( |aij |(ei (t) − sign(aij )
A Bd Bc [ ] 0 j∈Ni
,B= , C = C 0 and D = .
0 0 0 I × ej (t)) + bi ei (t)) (10)
Assumption 3: The pair (A, Bc ) is controllable and (A, C) is
observable. This ensures that the distributed ESO exists. In a more compact form, the following estimation error
Considering zxi (t) ∈ Rn and zwi (t) ∈ Rm to be the trajectory is obtained for the entire MAS:
estimation of xi (t) and wi (t) for ith follower respectively,
ė(t) = (IN ⊗ A − H ⊗ LC)e(t) + (IN ⊗ D)d(t) (11)
the following ESO based on relative output information of
neighbour agents is proposed for each follower: [ ]T [ ]T
where e(t) = eTx (t) eTw (t) = eT1 (t) eT2 (t) . . . eTN (t
∑ [ T ] T
żi (t) = Azi (t) + Bui (t) + L( |aij |((yi (t) − sign(aij ) and d(t) = d1 (t) dT2 (t) . . . dTN (t . Furthermore, using (1),
j∈Ni (2), and (7), bipartite consensus error ϵxi (t) for ith follower
× yj (t)) − C(zi (t) − sign(aij )zj (t))) + bi (yi − Czi (t))) can be written as:
(6) ∑
ϵ̇xi (t) = Aϵxi (t) + Bc Kc ( |aij |(ϵxi (t) − sign(aij )ϵxj (t))
[ T T
]T
where zi (t) = zxi (t) zwi (t) is the extended observer j∈N

i

state and L ∈ R(n+m)×n is the observer gain. The following + bi ϵxi (t)) − Bc Kc ( |aij |(exi (t) − sign(aij )exj (t))
distributed relative output feedback control law is proposed j∈Ni
for each follower agent: + bi exi (t)) + Bd ewi (t) (12)

ui (t) = Kc ( |aij |(zxi (t) − sign(aij )zxj (t)) + bi (zxi (t)
where disturbance compensation gain Kd = (C(A +
Bc Kc )−1 Bc )−1 C(A + Bc Kc )−1 Bd is designed and
j∈Ni
− si x0 (t))) − Kd zwi (t) (7) sign(a )s = s is used. In addition, in a more compact
ij i j

1549-7747 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCSII.2018.2884006, IEEE
Transactions on Circuits and Systems II: Express Briefs

form, bipartite consensus error trajectory for the entire MAS by pre- and post-multiplying on both sides by P −1 = W ,
T T
can be written as: and A P2 + P2 A − C C + βP2 + P22 + R < 0 which
[ ]T is equivalent to (9) (from Schur complement lemma [24]).
−(H ⊗ Bc Kc )T
ϵ̇x (t) = (IN ⊗A+H ⊗Bc Kc )ϵx (t)+ e(t) Moreover, β ≥ max( λmax1(P1 ) , λmax1(P2 ) ) is chosen. Clearly,
(IN ⊗ Bd )T
[ ]T (13) V̇ (t) < 0 if ||ϵx (t)||22 +||e(t)||22 ≥ max(λmin (P1 ),λmin
σ
(G)λmin (P2 ))β .
where ϵx (t) = ϵTx1 (t) ϵTx2 (t) . . . ϵTxN (t) and e(t) = Moreover, from comparison lemma [25], one has:
[ T ]T
ex (t) eTw (t) are used. The following Lyapunov function σ σ
V (t) ≤ (V (0) − ) exp(−βt) + (18)
is considered: β β
V (t) = ϵTx (t)(I ⊗ P1 )ϵx (t) + eT (t)(G ⊗ P2 )e(t) Since, V (t) ≥ λmin (P1 )||ϵx (t)||22 + λmin (G)λmin (P2 )||e(t)||22 ,
then it follows from (18) that the bipartite consensus error
= V1 (t) + V2 (t) (14)
ϵx (t) and estimation error e(t) converges to the bounded set
where V1 (t) = ϵTx (t)(I ⊗ P1 )ϵx (t) and V2 (t) = eT (t)(G ⊗
P2 )e(t). Then, considering ∆ = (H + H T ), ∆ = HH T , { σ }
[ ]T U = ϵx (t), e(t) : V (t) ≤
Kc = −κBcT P1 , e(t) = eTx (t) eTw (t) , and making use of min(λmin (P1 ), λmin (G)λmin (P2 ))β
Lemma 5 in [11], the derivative of V1 (t) along the trajectories (19)
of (13) yields the following inequality: with a convergence rate greater than or equal to exp(−βt).
V̇1 (t) ≤ ϵTx (I ⊗ (AT P1 + P1 A − κλmin (∆)P1 Bc BcT P1 ))ϵx (t) This completes the proof.
+ ϵTx (t)(κH ⊗ P1 Bc BcT P1 )ex (t) + eTx (t)(κH T ⊗ P1 Bc V. E XAMPLE AND S IMULATION
× BcT P1 )ϵx (t) + ϵTx (t)(I ⊗ P1 Bd )ew (t) + eTw (t)(I ⊗ BdT A MAS of 4 followers and a leader is considered over
× P1 )ϵx (t) a signed directed graph as shown in Fig. 1. The solid
lines represent cooperative interaction while the dotted
≤ ϵTx (I ⊗ (AT P1 + P1 A − (κλmin (∆) − λmax (∆))P1 Bc lines represent competitive interaction between two agents.
× BcT P1 + P12 ))ϵx (t) + κ2 eTx (t)(I ⊗ P1 Bc BcT P1 )ex (t) Obviously, the graph is structurally balanced with node sets
+ eTw (t)(I ⊗ BdT Bd )ew (t) V1 = {1, 2} and V2 = {3, 4}. The leader is assumed to
be cooperative with V1 and competitive with V2[. Various ]
≤ ϵTx (I ⊗ (AT P1 + P1 A − (κλmin (∆) − λmax (∆))P1 Bc 0 2
× BcT P1 + P12 ))ϵx (t) + eT (t)(I ⊗ R)e(t) (15) matrices for the MAS (1) and (2) are: A = −2 0 ,
[ ]T [ ]T [ ]
where (15) is obtained by making use of Lemma 2. Again, Bc = 1 1 , Bd = 0.7 0.5 , and C = 1 0 .
taking derivative of V2 (t) along the trajectories of (11) yields: The external bounded disturbance considered is di (t) =
T ′ T exp(−0.001t) sin(t) (i ∈ ℑ4 ). xi0 = ([0.9 0.2]T , [0.2 −
V̇2 (t) = eT (t)(G ⊗ (A P2 + P2 A) − κ (GH + H T G) ⊗ C 0.4]T , [−0.8 0.9]T , [−0.6 − 0.7]T , [0.6 0.5]T )T
× C)e(t) + e (t)(G ⊗ P2 D)d(t) + d (t)(G ⊗ D P2 )
T T T (i = 0, . . . , 4) and zi0 = ([−0.2 0.6 − 0.7]T , [0.4 −
0.9 0.5]T , [0.6 − 0.3 − 0.4]T , [−0.6 − 0.4 0.8]T )T
× e(t)

(i ∈ ℑ4 ) are the initial state vectors and observer state
T κδ T vectors, respectively. Moreover, β = 2.7, δ = 0.2,
≤ e (t)(G ⊗ (A P2 + P2 A −
T 2
C C + P2 ))
λmax (G) ρ = 2. η = 1.5, λmax (G) = 2.7 and λmin (G) = 0.65
× e(t) + σ (16) are considered. All the simulations are implemented in
YALMIP toolbox [26] in MATLAB and SIMULINK [27].
where (16) is obtained by considering Lemmas 1-2, and σ > Using MATLAB, from Theorem 1, the following are obtained:
T [ ]
0 such that σ ≥ λmax (G)λmax (D D)||d(t)||22 where d(t) is ′ 0.1859 −0.0430
bounded according to Assumption 2. Then, considering κ ≥ κ = 58.9277 and κ = 13.6000, P1 = −0.0430 0.0586 ,
1+λmax (∆) ′ λmax (G)  
λmin (∆) and κ ≥ δ , (14)-(16) yields: 1.5112 −0.5690 −0.1640
P2 = −0.5690 0.3792 0.0743 , Kc =
V̇ (t) ≤ ϵTx (I ⊗ (AT P1 + P1 A − P1 Bc BcT P1 + P12 ))ϵx (t) −0.1640 0.0743 0.0209
[ ]
T T
+ eT (t)(G ⊗ (A P2 + P2 A − C C + P22 + R))e(t) −8.4231 −0.9228 , Kd = 0.5923, and L =
[ ]T
+σ 61.4861 −7.6112 509.9674 . Fig. 2 (a) − (c) show
the state trajectories with (c) being the disturbance states
= −βV (t) + ϵx (I ⊗ (A P1 + P1 A − P1 Bc Bc P1 + βP1
T T T
and Fig. 2 (d) − (f ) show the corresponding observer states
T T
+ P12 ))ϵx (t) + eT (t)(G ⊗ (A P2 + P2 A − C C + βP2 with (f ) being the observed state of the unknown bounded
+ P22 + R))e(t) + σ disturbances. Fig. 3 (a), (b) show the bipartite consensus error
trajectories while Fig. 3 (c) − (e) estimation error trajectories.
≤ −βV (t) + σ (17)
From Fig. 2 and Fig. 3 it can be observed that bipartite
where (17) is obtained by considering AT P1 + P1 A − consensus is achieved with bounded bipartite consensus and
P1 Bc BcT P1 + βP1 + P12 < 0 which is equivalent to (8) estimation errors with the designed ESO and the control law.

1549-7747 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCSII.2018.2884006, IEEE
Transactions on Circuits and Systems II: Express Briefs

[3] H. Liang, H. Su, Y. Wang, C. Peng, M. Fei, and X. Wang,


“Continuous-time opinion dynamics with stochastic multiplicative
noises,” IEEE Trans. Circuits Syst. II, Exp. Briefs, to be published,
doi:10.1109/TCSII.2018.2867944.
[4] A. V. Proskurnikov, A. S. Matveev, and M. Cao, “Opinion dynamics in
social networks with hostile camps: consensus vs. polarization,” IEEE
Trans. Autom. Control, vol. 61, no. 6, pp. 1524-1536, 2016.
[5] C. Altafini, “Consensus problems on networks with antagonistic interac-
tions,” IEEE Trans. Autom. Control, vol. 58, no. 4, pp. 935-946, 2013.
Fig. 1: Signed directed graph topology of MAS (1) and leader [6] C. Ma and Z. Qin, “Bipartite consensus on networks of agents with
antagonistic interactions and measurement noises,” IET Control Theory
(2) Appl., vol. 10, no. 17, pp. 2306-2313, 2016.
[7] L. Zhao, Y. Jia, and J. Yu, “Adaptive finite-time bipartite consensus for
2 2 2 second-order multi-agent systems with antagonistic interactions,” Syst.
x i1 (t)

x i2 (t)

w i (t)
Control Lett., vol. 102, pp. 22-31, 2017.
0 0 0 [8] Y. Zhu, S. Li, J. Ma, and Y. Zheng, “Bipartite consensus in networks
-2 -2 -2 of agents with antagonistic interactions and quantization,” IEEE Trans.
0 Time5(sec) 10 0 0 Time5(sec) 10 Circuits Syst. II, Exp. Briefs, vol. 65, no. 12, pp. 2012-2016, 2018.
Time5(sec) 10
[9] X. Guo, J. Liang, and J. Lu, “Asymmetric bipartite consensus over
2 (a) 2 (b) 2 (c)
directed networks with antagonistic interactions,” IET Control Theory
(t)

(t)

zw (t)

Appl., vol. 12, no. 17, pp. 2295-2301, 2018.


0 0 0
i1

i2

[10] H. Zhang and J. Chen, “Bipartite consensus of multi-agent systems over


zx

zx

-2 -2 -2 signed graphs: state feedback and output feedback control approaches,”


0 5 10 0 0 5 10 Int. J. Robust Nonlinear Control, vol. 27, no. 1, pp. 3-14, 2017.
Time (sec) Time5(sec) 10 Time(sec)
[11] J. Qin, W. Fu, W. X. Zheng, and H. Gao, “On the bipartite consensus
(d) (e) (f)
for generic linear multiagent systems with input saturation,” IEEE Trans.
Fig. 2: (a) − (c) state xi (t) and (d) − (f ) observer state zi (t) Cybern., vol. 47, no. 8, pp. 1948-1958, 2017.
trajectories of MAS (1) and (2) under control protocol (7) [12] G. Wen, H. Wang, X. Yu, and W. Yu, “Bipartite tracking consensus of
linear multi-agent systems with a dynamic leader,” IEEE Trans. Circuits
1
Syst. II, Exp. Briefs, vol. 65, no. 9, pp. 1204-1208, 2018.
1
[13] M. Liu, X. Wang, and Z. Li, “Robust bipartite consensus and tracking
ǫ x (t)

ǫ x (t)

0 0 control of high-order multiagent systems with matching uncertainties


i1

i2

and antagonistic interactions,” IEEE Trans. Syst., Man, Cybern., Syst.,


-1 -1 to be published, doi:10.1109/TSMC.2018.2821181.
0 5 10 0 5 10
Time (sec) Time (sec) [14] Y. Wu, Y. Zhao, and J. Hu, “Bipartite consensus control of high-order
(a) (b) multiagent systems with unknown disturbances,” IEEE Trans. Syst.,
1 1 5
Man, Cybern., Syst., to be published, doi:10.1109/TSMC.2017.2761362.
ex (t)

ex (t)

ew (t)

0 0 0 [15] K. Sun, S. Sui, and S. Tong, “Fuzzy adaptive decentralized optimal


i1

i2

control for strict feedback nonlinear large-scale systems,” IEEE Trans.


-1 -1 -5
0 5 10 0 5 10 0 5 10 Cybern., vol. 48, no. 4, pp. 1326-1339, 2018.
Time (sec) Time (sec) Time (sec) [16] K. Sun, Y. Li, and S. Tong, “Fuzzy adaptive output feedback optimal
(c) (d) (e) control design for strict-feedback nonlinear systems,” IEEE Trans. Syst.,
Fig. 3: (a), (b) bipartite consensus error ϵxi (t) and (c) − (e) Man, Cybern., Syst., vol. 47, no. 1, pp. 33-44, 2017.
estimation error ei (t) for MAS (1) and (2) under control [17] Z. Li, Z. Duan, G. Chen, and L. Huang, “Consensus of multiagent
systems and synchronization of complex networks: A unified viewpoint,”
protocol (7) IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 57, no. 1, pp. 213-224,
2010.
[18] H. Su, M. Z. Q. Chen, X. Wang, and J. Lam, “Semiglobal observer-
VI. C ONCLUSION based leader-following consensus with input saturation,” IEEE Trans.
Ind. Electron, vol. 61, no. 6, pp. 2842-2850, 2014.
To summarize the main features of the work, bipartite [19] J. Han, “From PID to active disturbance rejection control,” IEEE Trans.
consensus problem is investigated for linear MAS subjected to Ind. Electron., vol. 56, no. 3, pp. 900-906, 2009.
[20] J. Hu and Y. Hong, “Leader-following coordination of multi-agent
mismatched disturbances over a signed directed graph having systems with coupling time delays,” Physica A, vol. 374, pp. 853-863,
a directed spanning tree rooted at the leader. A distributed 2007.
ESO and a distributed relative output feedback control law [21] A. Bhaya, E. Kaszkurewicz, and R. Santos, “Characterizations of classes
of stable matrices,” Linear Algebra Appl., vol. 374, pp. 159-174, 2003.
are designed to achieve the leader-follower bipartite consensus. [22] A. Berman and R. J. Plemmons, Nonnegative Matrices in the Mathe-
The effect of mismatched disturbances is reduced by a distur- matical Sciences. Philadelphia, PA, USA: SIAM, 1994.
bance compensator, while the consensus and estimation errors [23] H. Zhang, Z. Li, Z. Qu, and F. L. Lewis, “On constructing Lyapunov
functions for multi-agent systems,” Automatica, vol. 58, no. 5, pp. 39-42,
converge to a small bounded set by suitably designed scaler 2015.
gains and feedback gain matrices parameters. Future research [24] S. Boyd, L. E. Ghaoui, E. Feron, and V. Balakrishnan, Linear Matrix
will focus on designing fully distributed adaptive consensus Inequalities in System and Control Theory. Philadelphia, PA, USA:
SIAM, 1994.
protocol independent of graph spectral information. [25] H. K. Khalil, Nonlinear Systems, 3rd ed. Upper Saddle River, NJ, USA:
Prentice-Hall, Inc, 2002.
R EFERENCES [26] (2004-2016) YALMIP Toolbox website. [Online]. Available:
[1] X. Dong, B. Yu, Z. Shi, and Y. Zhong, “Time-varying formation control https://yalmip.github.io/
for unmanned aerial vehicles: Theories and applications,” IEEE Trans. [27] (1994-2015) The MATLAB website. [Online]. Available:
Control Syst. Technol., vol. 23, no. 1, pp. 340-348, 2015. http://www.mathworks.com
[2] S. Du, X. Sun, M. Cao, and W. Wang, “Pursuing an evader through
cooperative relaying in multi-agent surveillance networks,” Automatica,
vol. 83, pp. 155-161, 2017.

1549-7747 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like