Fully Distributed Event-Triggered Observer-Based Leader-Follower Consensus

Control for Heterogeneous Multi-Agent Systems

Xin Wang Dongsheng Yang Weihua Li

College of Information Science College of Information Science Space Engineering University
and Engineering and Engineering Beijing, 101416, China
Northeastern University, Northeastern University, Email: wh li [email protected]
Shenyang, Liaoning, 110819, China Shenyang, Liaoning, 110819, China
Email: dnsy [email protected] Email: [email protected]

Zhanchao Ma
College of Information Science
and Engineering
Northeastern University,
Shenyang, Liaoning, 110819, China
Email: [email protected]

Abstract—In this paper, the leader-follower output consensus [6], [7]. Moreover, the authors in [8] summarized some
problem for linear heterogeneous multi-agent systems (MASs) kinds of event-triggering mechanisms (event-based sampling
is considered. Firstly, in order for all followers to obtain the schemes, model-based event-triggered schemes, sample-
matrices and state of leader, the adaptive observer is designed data-based event-triggered schemes, self-triggered sampling
for each follower. Considering that the actual system has schemes) in the existing works, and analyzed the advantages
limited resources to communicate, event-triggering mechanism and disadvantages of the corresponding triggering mecha-
is designed to realize discrete-time communication between nisms. See [8] for details.
neighbor agents. Secondly, the regulator equation is solved The development of event-triggered control protocol has
online, then, based on the solution of regulator equation and three stages: centralized, distributed and fully distributed.
the designed observers, the new controller is designed to solve Centralized event-triggered control protocol indicates that
the leader-follower consensus problem. Finally, a simulation is the design of controller for each agent is relevant to all
given, and the results show the effectiveness of the designed agents, and it is only suitable for small MASs. See [9]
method. and its references. Distributed control protocol means that
the designed controller is connected to the information
Index Terms—Output consensus, heterogeneous multiagent sys- of neighbors. See [4], [10] and so on. Fully distributed
tems, adaptive observer, event-triggered control. control protocol means that the construction of distributed
controllers does not depend on the global information of
1. Introduction the whole communication graph, and it is suitable for large
MASs with large-scale networks. See [1], [2] and so on.
With the rise of artificial intelligence, MASs have be- Existing leader-follower output consensus results for het-
come an increasingly popular research object, which in- erogeneous linear MASs mostly depended on the assump-
cludes the consensus control, containment control, formation tion that all followers know the leader system matrix such as
control for MASs and so on. The consensus control problem [1], [2], [10]. However, in some operational applications, not
is a basis research topic, which aims to design controller every follower can obtain the leader matrix. For example, if
to make all agents achieve average consistency or leader- the leader signal is given in sinusoidal form, it is impossible
follower consistency. There are many impressive works for each follower to know the leader matrix since it means
on consensus problem such as [1]–[4] and the references that each follower knows the frequency of sinusoidal signal.
therein. The authors in [4] considered the cooperative output regu-
Since each individual agent system is equipped with lation issue of heterogeneous MASs, where only followers
limited resources, it is very important to design control connected with leader could obtain the leader matrix. In or-
algorithms on the premise of saving system resources. For der to remove assumption that all followers know the leader
this reason, the event-triggered control protocol was de- matrix, adaptive observers were designed in [4]. However,
signed to replace the continuous-time control protocol in continuous communication between neighbor agents was

979-8-3503-8778-0/24/$31.00 ©2024 IEEE 4087

needed and the designed control algorithm depended on the In order to make the problem solvable, some necessary
graph’s global information. assumptions are needed.
Inspired by these mentioned works, this conference pa- Assumption 1. The connected graph Ḡ has a directed s-
per considers the leader-follower output consensus prob- panning tree and the root node is the leader. And the
lem of heterogeneous MASs by designing adaptive event- subgraph G for all followers is undirected.
triggered observers. Overall, the main contributions of this
conference paper are summarized as three aspects. Assumption 2. The pair (Ai , Bi ), i = 1, · · · , N is stabiliz-
able.
1) The fully distributed event-triggered observers are
proposed to observe the leader state and matri- Assumption 3. There exist solution pairs (Xi , Ui ) for the
ces, respectively, where neighbor agents realize regulator equation:
event-triggered communication and save system re-
sources. Ai Xi + Bi Ui = Xi A0
2) Compared with [1], [2], [10], we consider a more Ci Xi − C0 = 0, i = 1, · · · , N. (3)
general situation, that is, not all followers know the
leader system matrix. Remark 1. These assumptions are mild and standard for
3) Compared with [4], [9], [10], our method is de- the solvability of the leader-follower output consensus
pendent of communication topology, which is fully problem, which can be found in [1], [4], [10] and so on.
distributed. Moreover, the book [11] gave the detailed explanation
about the solvability of the regulator equation (3).
Notation: For B ∈ Rm×n , vec(B) =
col(B1 , · · · , Bn ) ∈ Rmn×1 , in which Bj , j = 1, 2, · · · , n,
is the j th column for matrix B . For y = col(y1 , · · · , yq ) ∈ 3. Design the observers and controller
RN q , in which yi ∈ RN , i = 1, · · · , N, is column vector,
q
MN (y) = [y1 , · · · , yq ]. 3.1. Design adaptive event-triggered observers

2. Graph theory and problem description As in [4], this paper considers the more general case
where the leader system matrix is known only to followers
2.1. Graph theory directly connected to the Leader. Inspired by [4], we de-
sign the adaptive event-triggered observers (4), which can
In this paper, the communication between followers is estimate not only the leader’s state but also the leader’s
described by G = {V, E, W}, where V = {1, · · · , N } is the matrix. Compared with (4), the observer designed in this
node set, E ∈ V × V is the edge set, W = [wij ] ∈ RN ×N paper makes neighbors avoid continuous communication,
is the weight matrix. wij > 0 if node i and node j are saves system resources and is independent of the whole
connected, otherwise wij = 0. The graph G is undirected communication graph structure.
if (i, j) ∈ E ⇔ (j, i) for any i, j = 1, · · · , N. The Design the following fully-distributed event-triggered
communication for whole MAS is described by Ḡ = {V̄, Ē}, observer for ith follower:
in which V̄ = V ∪{0}, and Ē ∈ V̄ × V̄ . wi0 > 0 if leader can N

transmit the information to follower i, otherwise wi0 = 0. Ṡi = − csij wij (S̃i − S̃j )
j=0
2.2. Problem description N

ξ˙i =Āi ξi − cξij wij (ξ˜i − ξ˜j ) (4)
This paper considers a leader-follower MAS, where the j=0
dynamics of leader and followers are modeled by where Si and ξi are the observers’ states to estimate
ẋ0 =A0 x0 , y0 = C0 x0 (1) vec([A0 ; C0 ]) and leader state x0 , S̃0 = vec([A0 ; C0 ]), ξ˜0 =
x0 , Āi represents a matrix consisting od the first n0 rows of
and Mnn00+o (Si ), S̃i = Si (tis ), t ∈ [tis , tis+1 ) and ξ˜i = ξi (tik ), t ∈
[tik , tik+1 ), csij and dsij are adaptive weights satisfying that
ẋi = Ai xi + Bi ui
yi = Ci xi , i = 1, · · · , N (2) ċsij =ιsij wij (S̃i − S̃j )T (S̃i − S̃j )
where x0 (t) ∈ Rn0 , y0 (t) ∈ Ro are the state and output ċξ =ιξ wij (ξ˜i − ξ˜j )T (ξ˜i − ξ˜j )
ij ij (5)
for leader and xi ∈ Rni , ui (t) ∈ Rmi , yi (t) ∈ Ro are the
state, control input, output for follower i, respectively. with ιsij = ιji > 0, ιi0 > 0 and cij (0) = cji (0) ≥ 0.
The purpose of this paper is to design the controller so Define the measurement errors eSi and eξi as
that the leader-follower output consensus problem for MAS eSi =Si (tiα ) − Si , t ∈ [tiα , tiα+1 )
composed of (1) and (2) is solvable for any initial states
x0 (0), xi (0), namely, limt→∞ yi − y0 = 0, i = 1, · · · , N. eξi =ξi (tiβ ) − ξi , t ∈ [tiβ , tiβ+1 ).

4088 2024 36th Chinese Control and Decision Conference (CCDC)

 N
The triggering instants tiα , tiβ are given by (csi0 − γs )
+ wi0 ŜiT Ŝi . (11)
2
ETM − α : tiα+1 = inf{t > tα |fiα > 0} (6) i=1
 N Further, consider the following inequalities by Young’s
where fiα = 12 (1 + μcsi0 )wi0 eTSi eSi + j=1 (1 +
s T 1
N T 1 inequality
μcij )wij eSi eSi − 2 j=1 wij S̃i (S̃i − S̃j ) − 2 wi0 (S̃i −
N
N 
S0 )T (S̃i − S0 ) − σe−ψt with μ, σ, ψ being positive con- 
stants, and − csij wij S̄iT (Ŝi − Ŝj ) (12)
i=1 j=1
ETM − β : tiβ+1 = inf{t > tβ |fiβ > 0} (7) N N N N
 1  s   s
N ≤− cij wij ŜiT Ŝi − Ŝj + cij wij eTSi eSi
where fiβ = 12 (1 + μ̄cξi0 )wi0 eTξi eξi + j=1 (1 + 2 i=1 j=1
 i=1 j=1
μ̄cij )wij eTξi eξi − 2 j=1 wij ξ˜iT (ξ˜i − ξ˜j )− 2 wi0 (ξ˜i −x0 )T ×
ξ 1 N 1

(ξ˜i − x0 ) − σ̄e−ψ̄t with μ̄, σ̄, ψ̄ being positive constants. and

Define the global observers’ errors as N

S̄ = col(S̄1 , · · · , S̄N ) with S̄i = Si − S0 and − csi0 wi0 S̄iT Ŝi
ξ¯ = col(ξ¯1 , · · · , ξ¯N ) with ξ¯i = ξi − x0 . Set Ŝi = S̃i − S0 i=1
and ξˆi = ξ˜i − x0 for i = 1, · · · , N. 1 N N
1 s
Then, the following dynamics can be obtained from (4), ≤− csi0 wi0 ŜiT Ŝi + c wi0 eTSi eSi . (13)
2 i=1
2 i=1 i0
N

S̄˙ i = − csij wij (Ŝi − Ŝj ) + csi0 wi0 Ŝi Substituting (12) and (13) into (11) yields
j=1
N N N
N
  γs     γs 
V̇s ≤ − wij ŜiT Ŝi − Ŝj − wi0 ŜiT Ŝi
ξ˙i =Āi ξi − cξij wij (ξˆi − ξˆj ) + wi0 ξˆ (8) 2 i=1 j=1 2 i=1
j=1
 N
N  N
1 s
and + csij wij eTSi eSi + c wi0 eTSi eSi . (14)
2 i=1 i0
ċsij =ιij wij (Ŝi − Ŝj )T (Ŝi − Ŝj ) i=1 j=1

ċξ =ιij wij (ξˆi − ξˆj )T (ξˆi − ξˆj )

ij (9) Furthermore,
N N
where i = 1, · · · , N and j = 0, 1, · · · , N. γs    
Subsequently, the main results of this subsection are − wij ŜiT Ŝi − Ŝj (15)
4 i=1 j=1
presented.
N N N  N
Theorem 1. If Assumptions 1-2 hold, then the designed γs     γS 
observers can estimate the leader’s matrices and state ≤− wij S̄iT S̄i − S̄j + wij eTSi eSi
8 i=1 j=1 2 i=1 j=1
under the ETM − α (6) and ETM − β (7). Moreover,
the MAS does not exhibit the Zeno behavior. and
Proof. The results are proved in three parts, i.e., proving N
limt→∞ S̄ = 0, limt→∞ ξ¯ = 0, and the MAS does not γs 
− wi0 ŜiT Ŝi
exhibit the Zeno behavior. 4 i=1
Part-I: Consider the following Lyapunov function as N N
γs  γs 
N
 N 
 N
(csij − γs ) 2 ≤− wi0 S̄iT S̄i + wi0 eTSi eSi (16)
1 8 i=1 4 i=1
Vs = S̄iT S̄i +
2 i=1 i=1 j=1
8ιij
Then, (14) can be written as the following form by (15),
N
(csi0 − γs )2 (16) and the ETM − α (6),
+ (10)
4ιi0
γs   
i=1 N N
γs T
V̇s ≤ − S̄ (H ⊗ I)S̄ + (1 + μcsij )wij eTSi eSi
where γs is a positive constant satisfying γs ≥ μ2 . Form the 8 2 i=1 j=1
first equation of (8), the derivative of Vs can be obtained N
1   1 
 N
N 
  N
 − wij S̃iT S̃i − S̃j + (1 + μcsi0 ) wi0 eTSi eSi
V̇s = − csij wij S̄i Ŝi − Ŝj −
T
csi0 wi0 S̄iT Ŝi 2 j=1 2
i=1 j=1 i=1 1 
N  N − wi0 (S̃i − S0 )T (S̃i − S0 )
 csij − γs  T   2
+ wij Ŝi − Ŝj Ŝi − Ŝj γs γs
4 ≤ − S̄ T (H ⊗ I)S̄ + N σe−ψt . (17)
i=1 j=1 8 2

2024 36th Chinese Control and Decision Conference (CCDC) 4089

where we have used the fact that γs ≥ μ2 . Set A0  < o < Then, according to ETM −β (7), (22) satisfies the following
γs
4 λmin (H). Then, (17) satisfies
inequality
o γs N
V̇s ≤ − Vs + N σe−ψt (18) γξ 1
2 2 V̇ξ ≤ − o1 ξ¯T ξ¯ + N σ̄e−ψ̄t + (Āi − A0 )x0 2 .
2 2 i=1
which implies that limt→∞ S̄ = 0.
Part-II: Consider the following Lyapunov function as Similar to Part-I, one obtains limt→∞ ξ¯ = 0.
N N N ξ Part-III: The contradictory method is used to prove that
1  ¯T ¯   (cij − γξ )2 the MAS does not exhibit Zeno behavior. We first prove that
Vξ = ξi ξi +
2 i=1 i=1 j=1
8ιij the MAS will not exhibit Zeno behavior under the ETM −α
N
(6). Suppose Zeno behavior occurs in the j th subsystem,
 (cξi0 − γξ )2 namely, limt→∞ tiα+1 = T < ∞. Then,
+ (19)
i=1
4ιi0 1
(1 + μcsi0 )wi0 eTSi (tiα+1 )eSi (tiα+1 )
where γξ is a enough large positive constant to be given. 2
Form the second equation of (8), the derivative of Vξ satisfies N
i
+ (1 + μcsij )wij eTSi (tiα+1 )eSi (tiα+1 ) ≥ σe−ψtα+1 .
N
  N
  j=1
V̇ξ = ξ¯iT Āi ξi − A0 x0 − cξij wij (ξˆi − ξˆj ) + wi0 ξˆi
i
i=1 j=1 which implies that 0 ≥ σeψtα+1 > 0. It is contradictory with
N  N
cξij − γξ  T   limt→∞ tiα+1 = T < ∞. Hence, the MAS will not exhibit
+ wij ξˆi − ξˆj ξˆi − ξˆj Zeno behavior under the ETM − α (6) for any limited time.
i=1 j=1
4 Moreover, it can also be proved that Zeno behavior will not
N occur under the ETM − β (7) for any limited time, which
cξi0 − γξ
+ wi0 ξˆiT ξˆi . (20) is the same as the proof of ETM − α. That is the end of
i=1
2 proof.
Further, consider the following inequality by Young’s
inequality 3.2. Design the controller and stability analysis
N 
   0ni ×n0
ξ¯iT Āi ξi − A0 x0 Set Γi = vec
−C0o×n0
, Ωi = ÂTi ⊗
i=1
I ni 0ni ×mi Aini ×ni Bini ×mi
N
   − I n0 ⊗ .
0o×ni 0o×mi Cio×ni 0o×mi
= ξ¯iT (Āi − A0 )(ξ¯i + x0 ) + A0 ξ¯i Then, we give the following lemma to solve the solution
i=1

pair Xi (t), Ui (t) regulator equation (3) online.
N N
1 1 Lemma 1. [4] If Assumption 3 holds, then ηi exponentially
≤ξ¯T ξ¯ + (Āi − A0 )x0 2 + (Āi − A0 )ξ¯i 2 
2 i=1 2 i=1 Xi (t)
converges to vec , and it exponential con-
Ui (t)
+ ξ¯T (I ⊗ A0 )ξ¯ (21)
vergence rate is at least faster than S̄i . The dynamics of
Set θ = max{i=1,··· ,N } {Āi − A0 2 }, similar to (17), ηi can be modeled by η̇i = − ΩTi (Ωi ηi −Γi ) with > 0
the following inequality using (21) can be obtained, being positive constant.
1  γξ  γξ Set x̄i = xi − Xi (t)ξi , i = 1, · · · , N . Then, design the
V̇ξ ≤ ξ¯T I ⊗ (A0 + AT0 + θI + 2I) − H ⊗ I ξ¯ + controller for ith follower based on observers (4) as follows:
2 4 2
N  N N
1   ui = K̂xi xi + K̂ξi ξi (23)
× (1 + μ̄cξij )wij eTξi eξi − wij ξ˜iT ξi − ξj
2 j=1
i=1 j=1 where K̂xi = col(K̂x1i , · · · , K̂xmi i ), K̂ξi =
1  1  1 mi j 1×ni j 1×n0
+ (1 + μ̄cξi0 ) wi0 eTξi eξi − wi0 (ξ˜i − x0 )T (ξ˜i − x0 ) col(K̂ξi , · · · , K̂ξi ), K̂xi ∈ R , K̂ξi ∈ R for
2 2 j = 1, · · · , mi , K̂xj i and K̂ξji are updated by the following
N
1 equations
+ (Āi − A0 )x0 2 . (22)
2 i=1 ˙j ˙
K̂xi = −x̄Ti Fi bji x̄Ti , K̂ξj = −x̄Ti Fi bji ξiT (24)
From the proof of Part-I, Āi − A0 exponentially tends to
where bi ∈ Rmi is the j th column for matrix Bi , and Fi is
zero with rate 2o . Since A0  < o < γ4s λmin (H), one can
N positive definite matrix to be given.
obtain that 12 i=1 (Āi − A0 )x0 2 exponentially tends to Then, according to Assumption 3 and controller (23),
zero. one obtains
Parameter γξ is selected to satisfy γξ ≥ μ̄2 and A0 +
T γ
A0 + θI + 2I − 4ξ λH < −o1 I and o1 is positive constant. x̄˙ i = (Ai + Bi Kxi )x̄i + Bi K̃xi x̃i + Bi K̃ξi ξi + χi (25)

4090 2024 36th Chinese Control and Decision Conference (CCDC)

where K̃xi = K̂xi − Kxi , K̃ξi = K̂ξi − Ui with Kxi = Carrying out coordinate transformation xi1 = yi , xi2 =
−BiT Fi , and χi exponentially converges to zero from The- ẏi , one can obtain that system (28) can be convert-
orem 1 and Lemma 1. ed to the dynamics as described in (2) with Ai =
The main results of this subsection are presented. [0 1; −gi , −mi ], Bi = [0; 1], Ci = [1 0]. The dy-
Theorem 2. If Assumptions 1-3 hold, Fi is the solution of namics of the harmonic oscillator is given in (1) with
the following inequality: A0 = [0 1; −1 0] and C0 = [1 0].

ATi Fi + Fi Ai + Fi Fi − 2Fi Bi BiT Fi + Pi < 0 (26)

with Pi is the given positive definite matrix. Then,
the leader-follower output consensus problem is solved The initial state is randomly selected from the interval
under the designed controller (23). (0, 1), and the simulation results of Theorems 1-2 are shown
in Figs. 1-4.
Proof. Consider the following Lyapunov function as
N
 mi
N 
  j j T j j T
6

Vu = x̄Ti Fi x̄i + K̃1i (K̃1i ) + K̃2i (K̃2i ) .

5
i=1 i=1 j=1

Then, calculate the derivative of Vu along (26) 4

N
 N

3

V̇u = x̄Ti (Fi Ai + ATi Fi + 2Fi Bi Kxi )x̄i + x̄Ti Fi χi 2

i=1 i=1
1
N
≤ x̄Ti (Fi Ai + ATi Fi − 2Fi Bi BiT Fi + Fi Fi )x̄i 0
0 5 10 15 20 25 30 35 40

i=1 Time t(s)

N
 Figure 1. The norm of S̄i under Theorem 1.
+ χTi χi
i=1
N
 N

≤− x̄Ti Pi x̄i + χi χi (27)
i=1 i=1
1.5 4

where the Young’s inequality is used to obtain the first 3

ti +1 -t i (sec)

ti +1 -t i (sec)
1

inequality. Since χi exponentially converges to zero, similar 0.5

2

to Theorem 1, limt→∞ x̄i = 0, i = 1, · · · , N can be

1

0 0

obtained. 0 10 20
Agent 1
30 40 0 10 20
Agent 2
30 40

Then, according to (3), the following equation is feasible 8 4

  6 3
ti +1 -t i (sec)

yi − y0 =Ci x̄i + Xi (t)ξi − C0 x0 4 ti +1 -t i (sec) 2

2 1

=Ci x̄i + C0 ξi − C0 x0 0 0
0 10 20 30 40 0 10 20 30 40

=Ci x̄i + C0 (ξi − x0 ) Agent 3 Agent 4

which implies that limt→∞ yi − y0 = 0 by Theorem 1. That Figure 2. The inter-event times under ETM − α (6).
is the end of proof.

4. Example
70

The following example is introduced to illustrate the 60

effectiveness of main results, which is borrowed from [1]. 50

Example 1. This example consider a MAS composed of four 40

mass-damper-spring follower systems and one leader

30
which is used to generate the tracking signal by the
harmonic oscillator. The matrix H of the communication 20

graph is given by H = [3 − 1 0 − 1; −1 2 − 1 0; 0 − 10

1 2 − 1; −1 0 − 1 2]. The mass-damper-spring system 0

is described by 0 5 10 15 20
Time t(s)
25 30 35 40

yi + mi yi + gi yi = ui , i = 1, · · · , 4. (28) Figure 3. The norm of ξ̄i under Theorem 1.

2024 36th Chinese Control and Decision Conference (CCDC) 4091

 3 10
[2] Y. Qian, L. Liu and G. Feng, ”Output Consensus of Heterogeneous
Linear Multi-Agent Systems with Adaptive Event-Triggered Control,”

ti +1 -t i (sec)

ti +1 -t i (sec)
2
5
in IEEE Transactions on Automatic Control, vol. 64, no. 6, pp. 2606-
1 2613, June 2019.
0
0 10 20 30 40
0
0 10 20 30 40
[3] J. Zhang, H. Zhang, B. Zhou, X. Xie, ”Distributed optimal consensus of
Agent 1 Agent 2 multiagent systems with Markovian switching topologies: synchronous
2
2
and asynchronous communications,” Science China Information Sci-
1.5
ences, pp. 1-10, 2023, doi:10.1007/s11432-023-3843-7.
ti +1 -t i (sec)

ti +1 -t i (sec)
1.5
1
1

0.5
[4] H. Cai, F. Lewis, G. Hu, and J. Huang, “The adaptive distributed
0.5

0 0
observer approach to the cooperative output regulation of linear multi-
0 10 20
Agent 3
30 40 0 10 20
Agent 4
30 40
agent systems,” Automatica, vol. 75, pp. 299–305, 2017.
[5] J. Zhang, D. Yang, H. Zhang, Y. Wang, and B. Zhou, “Dynamic event-
Figure 4. The inter-event times under ETM − β (7). based tracking control of boiler turbine systems with guaranteed perfor-
mance,” IEEE Transactions on Automation Science and Engineering,
pp. 1-11, 2023, doi:10.1109/TASE.2023.3294187.
8
[6] P. Tabuada, ”Event-Triggered Real-Time Scheduling of Stabilizing
6 Control Tasks,” in IEEE Transactions on Automatic Control, vol. 52,
no. 9, pp. 1680-1685, Sept. 2007.
4

[7] D. V. Dimarogonas, E. Frazzoli and K. H. Johansson, ”Distributed

2
Event-Triggered Control for Multi-Agent Systems,” in IEEE Transac-
0
tions on Automatic Control, vol. 57, no. 5, pp. 1291-1297, May 2012.

-2
[8] L. Ding, Q. Han, X. Ge and X. Zhang, ”An Overview of Recent
Advances in Event-Triggered Consensus of Multiagent Systems,” in
-4 IEEE Transactions on Cybernetics, vol. 48, no. 4, pp. 1110-1123, April
2018.
-6
0 5 10 15 20 25 30 35 40
Time t(s) [9] H. Zhang, G. Feng, H. Yan and Q. Chen, ”Observer-Based Output
Feedback Event-Triggered Control for Consensus of Multi-Agent Sys-
Figure 5. The output error yi − y0 , i = 1, · · · , 4 under Theorem 2. tems,” in IEEE Transactions on Industrial Electronics, vol. 61, no. 9,
pp. 4885-4894, Sept. 2014.
Figs. 1-4 imply that the observers (4) can estimate the [10] W. Hu and L. Liu, ”Cooperative Output Regulation of Heterogeneous
leader’s matrices and state and the MAS does not exhibit Linear Multi-Agent Systems by Event-Triggered Control,” in IEEE
the Zeno behavior under the ETM − α (6) and ETM − β Transactions on Cybernetics, vol. 47, no. 1, pp. 105-116, Jan. 2017.
(7). Fig. 5 implies that the leader-follower output consensus [11] J. Huang, Nonlinear Output Regulation: Theory and Applications.
problem of MAS (1) and (2) is effectively solved under the Society for Industrial and Applied Mathematics, 2004.
designed controller (23).

5. Conclusion
The leader-follower output consensus problem for het-
erogeneous MASs has been solved under the designed ob-
servers and controller. Theoretical analysis and simulation
results together show the effectiveness of our methods.
Notice that this conference paper studies the ideal situation.
In fact, there may exist DoS attacks and actuator faults in
the system. Therefore, in the future, we will plan to consider
the leader-follower consensus of MASs with actuator faults
and DoS attacks.

Acknowledgments
This work was supported by the Joint Funds of the Na-
tional Natural Science Foundation of China [grant number
U22B20115]; and the National Key Research and Develop-
ment Program of China [grant number 2022YFB4100802].

References
[1] J. Zhang, H. Zhang, Y. Liang, and W. Song, “Adaptive bipartite output
tracking consensus in switching networks of heterogeneous linear
multiagent systems based on edge events,” IEEE Trans. Neural. Netw.
Learn. Syst., vol. 34, no. 1, pp. 79-89, Jan. 2023

4092 2024 36th Chinese Control and Decision Conference (CCDC)