IEEE Control Systems Letters paper presented at

2023 American Control Conference (ACC)

San Diego, CA, USA. May 31 - June 2, 2023

Distributed Nash Equilibrium Seeking for Single-Integrator Dynamics

Subject to Disturbances with Unknown Bounds
Xiongnan He and Jie Huang

Abstract— In this paper, we study the problem of Nash disturbances with unknown bounds. Compared with the
equilibrium seeking of N -player games for single integrator dy- existing results, two new features are worth mentioning:
namics subject to bounded disturbances with unknown bounds.
Compared with the existing results, two new features are worth 1) Except for [6], [7], the existing results all assume
mentioning. First, the communication network among players is the communication networks are static, connected, and
jointly strongly connected, which can be disconnected at every undirected graphs while our result applies to jointly
time instant. Second, the class of the disturbances contains any strongly connected graphs, which can be disconnected
bounded time function with the bounds unknown. To achieve at every time instant.
our objective, we have proposed a novel approach by integrating
the distributed estimator, some nonlinear control technique, 2) The class of disturbances in this paper is model-free
and adaptive control technique. Our design is illustrated by and is only required to be bounded with the bounds
the example of a group of velocity-actuated robots in sensor unknown. Thus, the existing results cannot handle such
networks. a class of disturbances.
Index Terms— Nash equilibrium seeking, jointly strongly
connected switching graphs, multi-agent systems, disturbances To achieve our objective, we need to adopt a different
of unknown bounds, adaptive control. approach to handle our disturbances, which integrates the
distributed estimator, some nonlinear control technique, and
I. I NTRODUCTION adaptive control technique. It is noted that a similar class of
The problem of the distributed Nash equilibrium seek- disturbances has been handled for the attitude control prob-
ing over networks has received increasing attention over lem in, for example, [14], [19]. Nevertheless, the problems
the years. Some of the representative publications are, for in [14], [19] do not involve the seeking of the unknown Nash
instance, [1], [3]–[5], [8], [9], [11]–[13], [15]. As a control equilibrium, and are much simpler than the problem studied
system is invariably subject to some external disturbances here. Indeed, the disturbances in [14], [19] can be readily
caused by the delay of communication networks, insensitivity handled by the standard adaptive control technique.
of the sensors, uncertainty of dynamics, and so on. More The rest of the paper is organized as follows. Section
and more references have further considered the problem of II provides preliminaries for game theory and a summary
the distributed Nash equilibrium seeking for players whose of the existing results of [7]. Section III presents the main
dynamics is subject to various disturbances [2], [6], [9], [10], result. Section IV illustrates our approach with a numerical
[18], [20], [21]. Most references assume that the disturbance example. Section V closes this paper with some remarks.
is generated by a known linear autonomous system such Notation The norm ||x|| denotes 2-norm of a vector x
as [2], [6], [9], [10], [20]. Such a system can produce while ||P || is the induced Euclidean norm for a matrix P .
disturbances consisting of exponential functions, sinusoidal Rn is the n-dimensional Euclidean space. Rm×n is the set
functions with known frequencies, and polynomial functions. of all m × n real matrices. For column vectors ai ∈ Rni , i =
This type of disturbance can be handled by using the Lu- · , aTN ]T ∈ Rn with
1, · · · , N , col(a1 , · · · , aN ) = [aT1 , · ·  
enberger observer. Reference [18] considered the continuous a1
and differentiable disturbances with bounded time derivative, PN
n = i=1 ni and diag(a1 , ..., aN ) = 
 .. ∈

.
which may not be produced by a linear autonomous system.
aN
Nevertheless, the approach in [18] cannot fully reject the
Rn×N . ⊗ is the notation of the Kronecker product. 1N is
disturbances. Recently, [6] considered the disturbance in the
the N -dimensional column vector with all elements 1. IN
form of trigonometric polynomials with unknown frequen-
represents an N × N identity matrix.
cies, which can be produced by a linear exosystem with
unknown parameters. II. P RELIMINARIES
In this paper, we will further study the distributed Nash In this section, we introduce the background of non-
equilibrium seeking with disturbance rejection for bounded cooperative game theory and existing results based on [5],
This work was supported in part by the Research Grants Council of the [7], [11], [15].
Hong Kong Special Administration Region under grant No. 14202619, and
in part by National Natural Science Foundation of China under Project A. Game Theory
61973260.
The authors are with the Department of Mechanical and Automation A non-cooperative game denoted by Γ is defined by a
Engineering, The Chinese University of Hong Kong, Hong Kong (e-mail: triplet as follows:
[email protected]; [email protected]. Corresponding author:
Jie Huang.) Γ , {N , fi , Ui }, (1)

979-8-3503-2806-6/$31.00 ©2023 AACC 515

Authorized licensed use limited to: University of Virginia Libraries. Downloaded on April 08,2024 at 21:34:57 UTC from IEEE Xplore. Restrictions apply.
where N is the set of N players. For each player i ∈ a directed path of G(t) from node i1 to node ik at time t, and
N , the strategy
PN of player i is denoted by xi ∈ Ui ⊆ in this case, we say node i1 can reach node ik at time t. The
Rni . Let i=1 ni = n and U = U1 × U2 × · · · × graph G(t) is said to be static if G(t) = G(0) for all t > 0.
UN ⊆ Rn , which is called the strategy space. Then, fi : A static graph is denoted by G = (V, E). A static graph is
U → R is the payoff function for player i. Let x = said to be connected if there is a node that can reach every
col(x1 , x2 , · · · , xN ) ∈ Rn , which is called the strategy other node, and is said to be strongly connected if there is
vector, x−i , (x1 , x2 , · · · , xi−1 , xi+1 , · · · , xN ), and U−i = a directed path between any two nodes. An edge (i, j) is
U1 × · · · × Ui−1 × Ui+1 · · · × UN . Then, the goal of each called an undirected edge of E if (i, j) ∈ E ⇔ (j, i) ∈ E. G
player i is to minimize its payoff function fi (xi , x−i ) over is called an undirected graph if every edge in E is undirected.
xi ∈ Ui , that is, The weight adjacency matrix of a graph G is a nonnegative
matrix A = [aij ] ∈ RN ×N where, for i, j = 1, 2, ..., N ,
minimize fi (xi , x−i ) subject to ∀xi ∈ Ui . (2)
i 6= j, aij > 0 if there is an edge from node j to node i and
∗
A strategy vector x = (x∗i , x∗−i )
∈ U is said to be a Nash aij = 0 if otherwise. Since, for i = 1, 2, ..., N , there are no
equilibrium point (NE) if it is such that such edgesPNas (i, i), we have aiiP= 0. For i = 1, 2, ..., N , let
N
din
i = j=1 a ji and dout
i = j=1 aij , which are called
fi (x∗i , x∗−i ) ≤ fi (xi , x∗−i ), ∀i ∈ N , ∀xi ∈ Ui . (3) the in-degree and out-degree of node i, respectively. Let
Thus, any unilateral change of its own strategy increases its D = diag (dout out out
1 , d2 , · · · , dN ), which is called the degree
payoff value. matrix of G. The matrix L = D − A is called Laplacian of
Two standard assumptions frequently used in G corresponding to A.
the distributed Nash equilibrium seeking of the A time function σ : [0, ∞) 7→ P = {1, 2, . . . , n0 } with
game (1) are n0 being some positive integer is said to be a piecewise
 given next. For this purpose, let
constant switching signal if there exists a sequence {tj : j =

F (x) = col ∂f1 (x 1 ,x−1 ) ∂f2 (x2 ,x−2 )
∂x1 , ∂x2 , · · · , ∂fN (xN ,x−N )
∂xN ,
0, 1, 2, . . .} satisfying t0 = 0 and tj+1 − tj ≥ τ for some
which is called the pseudogradient of f .
positive constant τ such that, for all t ∈ [tj , tj+1 ), σ(t) = p
Assumption 1: For all i ∈ N , i) Ui is nonempty, closed
for some p ∈ P. P is called the switching index set, tj is
and convex; ii) the payoff function fi (xi , x−i ) is convex and
called the switching instant, and τ is called the dwell time.
continuously differentiable in xi for every fixed x−i ∈ U−i ;
A graph Gs = (Vs , Es ) is a subgraph of G = (V, E) if
iii) F is strongly monotone on U , i.e., for some µ > 0,
Vs ⊆ N and Es ⊆ E ∩ (Vs , Vs ). Given a set of r graphs
(x − x0 )T (F (x) − F (x0 )) ≥ µ||x − x0 ||2 , ∀x, x0 ∈ U .
{Gi =S(V, Ei ), i = 1, 2, · · · , r}, the graph G = (V, E) where
Remark 1: The following system r
E = S i=1 Ei is called the union of graphs Gi , denoted by
r
ẋi = −∇i fi (xi , x−i ), i ∈ N (4) G = i=1 Gi .
Given a piecewise constant switching signal σ(t) and a
where ∇i fi (xi , x−i ) = ∂fi (x i ,x−i )
∂xi , is called gradient dy- set of n0 static graphs Gi = (V, Ei ), i = 1, . . . , n0 , one
namics [4], [5], which can be put into the following compact can define a time-varying graph Gσ(t) = (V, Eσ(t) ), which is
form: called a switching graph. A switching graph Gσ(t) is said to
ẋ = −F (x). (5) be every time connected (every time strongly connected), if
for t ≥ 0, the graph Gσ(t) is connected (strongly connected).
By Theorem 3 of [11], under Assumption 1, there exists a
unique NE x∗ ∈ U which satisfies C. Imperfect Information Case
In this section, we consider games whose actions xi are
F (x∗ ) = 0n . (6) governed by the following single integrator dynamics:
Moreover, by Lemma 2 of [5], with Ui = Rni , the unique ẋi = ui + di , i ∈ {1, 2, · · · , N } (7)
NE x∗ of (5) is globally exponentially stable.
Assumption 2: The gradient dynamics ∇i fi (xi , x−i ) is where xi , ui ∈ Rni , and di represents some external distur-
Lipschitz continuous, i.e., for all i ∈ N , ||∇i fi (x) − bance.
∇i fi (x0 )|| ≤ θi ||x − x0 ||, ∀x, x0 ∈ U, for some constant The gradient dynamics (4) can be viewed as the closed-
θi > 0. loop system resulting from the composition of the single-
In what follows, like in [5], we focus on obtaining the integrator dynamics (7) with di = 0ni , and the feedback
global result. Thus, we let Ui = Rni , ∀i = 1, · · · , N . control ui = −∇i fi (xi , x−i ). The control law ui =
−∇i fi (xi , x−i ) assumes every player can observe the ac-
B. Graph Theory tions of all other players. Such a scenario is called perfect
A time-varying graph is denoted by G(t) = (V, E(t)), information. In what follows, we assume each player can
where V = {1, . . . , N } is the node set , and E(t) ⊆ V × V only observe the actions of those players who are his/her
is the edge set. For i, j = 1, 2, ..., N and i 6= j, if there is an neighbors. Such a scenario is called imperfect information.
edge from node j to node i at time t, then (i, j) ∈ E(t). In To describe the information exchange of the players of
this case, we say j is a neighbor of i at time t. A subset of the game (1), let a switching graph Gσ(t) = (V, Eσ(t) ) with
E(t) of the form {(i1 , i2 ), (i2 , i3 ), ..., (ik−1 , ik )} is said to be V = {1, . . . , N }, where σ(t) is the switching signal with

516
Authorized licensed use limited to: University of Virginia Libraries. Downloaded on April 08,2024 at 21:34:57 UTC from IEEE Xplore. Restrictions apply.
switching instants {ti : i = 0, 1, 2, . . . } ⊂ [0, +∞), and III. M AIN R ESULT
Eσ(t) ⊆ V × V for all t ≥ 0. For any t ≥ 0, Eσ(t) contains
In this section, we will deal with the problem of Nash
an edge (j, i) if and only if the player i is able to observe
equilibrium seeking over jointly strongly connected networks
the action xj of player j at time t. We make the following
with the disturbances satisfying the following assumption.
assumption on graph Gσ(t) .
Assumption 4: For i ∈ V, the external disturbances di (t)
Assumption 3: There exists a subsequence {ik } of {i :
are bounded over t ∈ [0, ∞) with the bounds unknown.
i = 0, 1, 2, . . . } with tik+1 − tik < ν for some positive
S ti Remark 3: Assumption 4 means di (t) can be any time
number ν such that the union graph tik+1 Gσ(t) is strongly function such that, for some unknown positive constant Di ,
k
connected. supt≥0 ||di (t)||∞ ≤ Di , where || · ||∞ denotes the infinity
Remark 2: For convenience, we say a switching graph norm of a vector. In contrast, in [6], [9], [10], [20], [21],
Gσ(t) satisfying Assumption 3 is jointly strongly connected. the disturbance di is assumed to be generated by a linear
Under Assumption 3, the graph can be disconnected at autonomous system as follows:
every time instant. The Nash equilibrium seeking for games (
over jointly strongly connected communication networks was v̇i = Si vi , vi (0) ∈ Rqi
studied in [6] and [7]. Di : (11)
di = Di vi
Next, we summarize the existing results without distur-
bances from [7], i.e., di being zero. In this case, let yi = where vi ∈ Rqi , di ∈ Rni , Si ∈ Rqi ×qi and Di ∈ Rni ×qi .
col(yi1 , yi2 , · · · , yiN ) ∈ Rn denote the estimate of player i As a result, di consists of exponential functions, polynomial
on the actions of all players. Then the distributed control law functions, and trigonometric polynomials. On the other hand,
in [7] is as follows: ∀i, j ∈ V, in [18], the disturbance is required to be continuous and
ui = −δ k̄i ∇i fi (yi ) (8a) differentiable, and the derivative of the disturbance needs to
N
! be bounded.
The class of the disturbances considered in the existing
X
ẏij = − aik (t)(yij − ykj ) + aij (t)(yij − xj ) (8b)
k=1 literature is somehow narrow. Under Assumption 4, our result
can handle some disturbances that cannot be handled by the
where k̄i is a fixed positive number, δ is some positive
existing literature.
number for modulating the gradient dynamics (8a), and
To handle such a class of disturbance, we modify the
aij (t) is the element of the weighted adjacency matrix of
control law (9) to the following form: ∀i, j ∈ V,
the graph Gσ(t) . The variable yij ∈ Rnj is the estimate of
player i on the action xj of the player j. The control law (8) ui = −δ k̄i ∇i fi (zi ) − sgn(xi − x̂i )D̂i − κ(xi − x̂i ) (12a)
can be viewed as an extension of the control law in [15] to ˙
the jointly strongly connected networks. Under the control D̂i = (sgn(xi − x̂i ))T (xi − x̂i ) (12b)
law (8), the closed-loop system is as follows: ∀i, j ∈ V, ˙x̂i = −δ k̄i ∇i fi (zi ) (12c)
N
!
ẋi = −δ k̄i ∇i fi (yi ) (9a) X
N
! żij = − aik (t)(zij − zkj ) + aij (t)(zij − x̂j ) .
X
k=1
ẏij = − aik (t)(yij − ykj ) + aij (t)(yij − xj ) . (9b)
(12d)
k=1
Let Lσ(t) denote the Laplacian matrix of the switching where κ is some positive constant, zi =
graph Gσ(t) , and col(zi1 , zi2 , · · · , ziN ) ∈ Rn and, for any scalar α,

Bσ(t) = diag (a11 (t)In1 , · · · , a1N (t)InN , a21 (t)In1 , · · · , 1
 if α > 0,
a2N (t)InN , · · · , aN 1 (t)In1 , · · · , aN N (t)InN ). sgn(α) = 0 if α = 0, (13)

−1 if α < 0.

Then control law (9) can be written in the following compact
form:
where D̂i ∈ R is the estimate of Di .
ẋ = −δ k̄H(y) (10a) Remark 4: One specific challenge of our problem is that
ẏ = −(Lσ(t) ⊗ In + Bσ(t) )y + Bσ(t) (1N ⊗ x) (10b) the Nash equilibrium x∗ is unknown. If the Nash equilibrium
x∗ were known, one may have considered a simplified
where k̄ = diag{k̄i Ini }, y = col(y1 , · · · , yN ) and H(y) =
control law as follows:
col(∇1 f1 (y1 ), ∇2 f2 (y2 ), · · · , ∇N fN (yN )). It can be veri-
fied that Assumption 2 impliesp||H(y) − H(y 0 )|| ≤ θ||y − ui = −δ k̄i ∇i fi (zi ) − sgn(xi − x∗i )D̂i (14a)
y 0 ||, ∀y, y 0 ∈ RN n where θ = θ12 + · · · + θN 2 .
˙
The following theorem is a rephrasing of Theorem 2 in D̂i = (sgn(xi − x∗i ))T (xi − x∗i ) (14b)
[7]. N
!
X
Theorem 1: Under Assumptions 1 to 3, there exists δ ∗ > żij = − aik (t)(zij − zkj ) + aij (t)(zij − x∗i ) .
0 such that, for 0 < δ < δ ∗ , the equilibrium (x∗ , 1N ⊗ x∗ ) k=1
of systems (10a) - (10b) is globally exponentially stable. (14c)

517
Authorized licensed use limited to: University of Virginia Libraries. Downloaded on April 08,2024 at 21:34:57 UTC from IEEE Xplore. Restrictions apply.
 ˙
To overcome the difficulty of the unknown x∗ , we have to D̂ = (Sgn(x̃))T x̃. (20b)
introduce (12c) to estimate x∗ , thus leading to the control
law (12). Let D̃ = D̂ − D with D = col(D1 , · · · , DN ). Choose the
Substituting (12) into (7) gives the following closed-loop Lyapunov function candidate of (20a)-(20b) as follows:
system: ∀i, j ∈ V,
1 1
V1 (x̃, D̃) = x̃T x̃ + D̃T D̃. (21)
ẋi = −δ k̄i ∇i fi (zi ) − sgn(xi − x̂i )D̂i − κ(xi − x̂i ) + di 2 2

(15a) Differentiating (21) along the trajectories of (19a)-(19b)

˙ gives
D̂i = (sgn(xi − x̂i ))T (xi − x̂i ) (15b)
˙
x̂˙ i = −δ k̄i ∇i fi (zi ) (15c) V̇1 = x̃T x̃˙ + D̃T D̂
= x̃T (−Sgn(x̃)D̂ − κx̃ + d) + D̃T (Sgn(x̃))T x̃
N
!
X
żij = − aik (t)(zij − zkj ) + aij (t)(zij − x̂j ) .
k=1
= −κ||x̃||2 + x̃T Sgn(x̃)(D̃ − D̂) + x̃T d
(15d) = −κ||x̃||2 − x̃T Sgn(x̃)D + x̃T d (22)
N
Let x̂ = col(x̂1 , x̂2 , · · · , x̂N ) ∈ Rn , d = X
col(d1 , · · · , dN ), z = col(z1 , · · · , zN ) ∈ RN n , H(z) = ≤ −κ||x̃||2 − (||xi || − xi )Di
i=1
col(∇1 f1 (z1 ), ∇2 f2 (z2 ), · · · , ∇N fN (zN )), Sgn(x − x̂) =
diag(sgn(x1 − x̂1 ), · · · , sgn(xN − x̂N )) ∈ Rn×N , and D̂ = ≤ −κ||x̃||2 .
col(D̂1 , · · · , D̂N ) ∈ RN ×1 . Then the compact form of (15)
is as follows: Thus, x̃ and D̃ are both
R t bounded, and hence D̂ is bounded.
Next, define W1 (t) = 0 κ||x̃(r)||2 dr. If one could show
ẋ = −δ k̄H(z) − Sgn(x − x̂)D̂ − κ(x − x̂) + d (16a)
˙ lim Ẇ1 (t) = lim κ||x̃(t)||2 = 0, (23)
D̂ = (Sgn(x − x̂))T (x − x̂) (16b) t→∞ t→∞

x̂˙ = −δ k̄H(z) (16c) then, one can conclude limt→∞ κ||x̃(t)|| = 0, which implies
ż = −(Lσ(t) ⊗ In + Bσ(t) )z + Bσ(t) (1N ⊗ x̂). (16d) (17a).
Integrating both sides of (22) gives
Now, we are ready to present our main result.
Theorem 2: Under Assumptions 1 to 4, and the control Z t
law (16), there exists δ ∗ > 0 such that, for 0 < δ < δ ∗ , for W1 (t) ≤ −V̇1 (r)dr = V1 (0) − V1 (t) ≤ V1 (0). (24)
0
any initial condition, the solution of the closed-loop system
(16) is bounded, and Since Ẇ1 (t) = κ||x̃||2 ≥ 0, limt→∞ W1 (t) exists.
Since x̃ and D̃ are both bounded, by Theorem 1, x̂ will
lim x(t) = x∗ (17a)
t→∞ converge to x∗ exponentially, which implies that x̂(t) is
lim x̂(t) = x∗ (17b) bounded. By the definition of x̃, x is also bounded. By (19a),
t→∞
x̃˙ is bounded. Thus, Ẅ1 (t) = 2κx̃T x̃˙ is bounded. Applying
lim z(t) = 1N ⊗ x∗ . (17c)
t→∞ Barbalat’s lemma completes the proof of this theorem.
Proof: Let x̃ = x − x̂. Then, Remark 5: Compared with [6], two points are worth
x̃˙ = ẋ − x̂˙ = −Sgn(x̃)D̂ − κx̃ + d. (18) noting. First, [6] considers a class of disturbances which is a
trigonometric polynomial. Thus, the class of the disturbances
Thus, the closed-loop system (16) is transformed to the considered in this paper strictly contains the class of the
following: disturbances in [6] as a subclass. Second, [6] applied the
x̃˙ = −Sgn(x̃)D̂ − κx̃ + d (19a) internal model method to deal with the disturbances. In
contrast, since the disturbances here are model-free, an
˙
D̂ = (Sgn(x̃))T x̃ (19b) adaptive learning control technique is used to estimate the
x̂˙ = −δ k̄H(z) (19c) bounds of the disturbances.
Remark 6: If the bound of the disturbance is known, then
ż = −(Lσ(t) ⊗ In + Bσ(t) )z + Bσ(t) (1N ⊗ x̂). (19d)
there is no need to estimate Di . In this case, the control law
By Theorem 1, there exists δ ∗ > 0 such that, for 0 < (12) can be simplified to the following form:
δ < δ ∗ , (x∗ , 1N ⊗ x∗ ) is the globally exponentially stable
equilibrium of the subsystems (19c) and (19d). Thus, (17b) ui = −δ k̄i ∇i fi (zi ) − sgn(xi − x̂i )Di − κ(xi − x̂i ) (25a)
and (17c) hold. We only need to show that (17a) holds and D̂ x̂˙ i = −δ k̄i ∇i fi (zi ) (25b)
is bounded. For this purpose, consider the following system: X N
!
żij = − aik (t)(zij − zkj ) + aij (t)(zij − x̂j ) .
k=1
x̃˙ = −Sgn(x̃)D̂ − κx̃ + d (20a) (25c)
518
Authorized licensed use limited to: University of Virginia Libraries. Downloaded on April 08,2024 at 21:34:57 UTC from IEEE Xplore. Restrictions apply.
 IV. A N UMERICAL E XAMPLE

In this section, we will use an example of five velocity-

actuated robots in sensor networks to demonstrate the effec-
tiveness of our approach.
As in [9], the motion equations of the mobile robots are
described by the following single integrator system:

ẋi = ui + di (26)

where xi = col(xix , xiy ) ∈ R2 , representing the abscissa

and ordinate of the ith agent’s position, ui , di ∈ R2 with di
taking the following form:

   
d1 D1 sin(t)cos(2t) Fig. 1: The communication graphs Gσ(t) .
d2  
   D2 (−1)d0.25te 

d=d3
 = (D3 + 1 )(−1)d0.25te  ⊗ 12 (27)
t+1
  
d4   D4 sin(b0.2tc)

 By Theorem 2, we can obtain an adaptive control law of
d5 D5 (t − btc) the form (12) with the design parameters as follows:

δ = 0.01, κ̂ = 0.1, k̄i = 1, i = 1, 2, 3, 4, 5.

where Di , i = 1, 2, · · · , 5, are some unknown finite positive
constants. Thus, the disturbance d satisfies Assumption 4. Simulation is performed with the following randomly
Fig. 2 shows the profiles of various disturbances. chosen initial conditions:
The payoff function of each player is as follows: 
−0.1
  
0.2
 
0.3
x̂1 (0) = , x̂2 (0) = , x̂3 (0) = ,
−0.1 0.2 0.3
X    
fi (xi , xi−i ) = xTi xi + xTi ri + ||xi − xj ||2 (28) x̂4 (0) =
−1.2
, x̂5 (0) =
0.1
;
j∈V −1.2 0.1
     
1 −1 1
z1j = , z2j = , z3j = ,
where r1 = col(2, −2), r2 = col(−2, −2), r3 = 1 −1 1
   
col(−4, 2), r4 = col(2, −4), r5 = col(3, 3), and −1 1.5
z4j = , z5j = ;
V = {1, 2, 3, 4, 5}. Since these payoff functions −1 1.5
     
are all quadratic, it can be easily verified that 0.71 0.45 0.527
they all satisfy Assumptions 1 and 2. Letting x1 (0) = , x2 (0) = , x3 (0) = ,
1.12 0.57 0.474
∂fi (xi ,xi−i )
= 0 gives the only Nash equilibrium
   
∂xi −0.5 0.69
x∗ = col((− 14 , 12
5 1
), ( 12 5
, 12 ), ( 41 , 12
1
), (− 14 , 12
7
), (− 13 , 0)). x4 (0) = , x5 (0) = ;
0.009 0.064
 
The communication network is shown in Fig. 1 which is 1
dictated by the following switching signal σ(t): D̂i (0) = , i, j = 1, 2, 3, 4, 5.
2

 To avoid the chattering phenomenon caused by the discon-

1, if 5k ≤ t < 5k + 1

 tinuous function sgn, we replace sgn(·) with the continuous
2, if 5k + 1 ≤ t < 5k + 2


 saturation function sat(·) defined as follows:
σ(t) = 3, if 5k + 2 ≤ t < 5k + 3 (29) 

1 if xi ≥ 0.01,
4, if 5k + 3 ≤ t < 5k + 4

 


 sat(xi ) = 100xi if |xi | < 0.01,
5, if 5k + 4 ≤ t < 5k + 5


−1 if xi ≤ −0.01.


with k = 0, 1, 2, · · · . The communication network is periodic Fig. 3 shows the players’ positions of the closed-loop
and switches five times in every period with the period being system under the adaptive control law (12). We can see that
5 seconds. Clearly, the communication network is jointly the positions of all robots converge to the Nash equilibrium
strongly connected even though the graph is disconnected with negligible errors although the communication topology
for all time. Thus, Assumption 3 is satisfied. is not connected at every time instant.

519
Authorized licensed use limited to: University of Virginia Libraries. Downloaded on April 08,2024 at 21:34:57 UTC from IEEE Xplore. Restrictions apply.
 V. C ONCLUSIONS [19] X. Zhang, “Semitensor product based adaptive control for attitude
tracking of spacecraft with unknown external disturbances,” Journal of
In this paper, we have studied the distributed Nash equilib- Control Theory and Applications, Vol. 10, No. 3, pp. 292-296, 2012.
[20] Y. Zhang, X. Wang and H. Ji, “Distributed Nash equilibrium seeking in
rium seeking for single integrator dynamic games subject to aggregative game with disturbance rejection”, IEEE 15th International
the class of the bounded disturbances with unknown bounds. Conference on Control and Automation (ICCA), pp. 1091-1095, 2019.
Such a class of disturbances cannot be handled by the [21] Y. Zhang, S. Liang, X. Wang and H. Ji, “Distributed Nash equilibrium
seeking for aggregative games with nonlinear dynamics under external
approaches in the existing literature. We have achieved our disturbances”, IEEE Transactions on Cybernetics, Vol. 50, No. 12, pp.
objective by developing a novel approach by integrating the 4876-4885, 2020.
distributed estimator, some nonlinear control technique, and
adaptive control technique. The effectiveness of our approach
is illustrated by the example of velocity-actuated robots in
sensor networks.
The extension of the results of this paper to high-order
integrator-type dynamic agents over switching networks is
underway.

R EFERENCES
[1] T. Basar and G. Olsder, Dynamic Noncooperative Game Theory, 2nd
ed. Philadelphia, PA, USA: SIAM, 1999.
[2] X. Cai, F. Xiao, B. Wei, M. Yu, and F. Fang, “Nash equilibrium seeking
for general linear systems with disturbance rejection,” 2021, Online.
[3] S. Flam, “Equilibrium, evolutionary stability and gradient dynamics”,
Int. Game Theory Rev., Vol. 4, No. 4, pp. 357-370, 2002.
[4] F. Francisco and J. Pang, Finite-dimensional variational inequalities
and complementarity problems, Springer, 2003.
[5] D. Gadjov and L. Pavel, “A passivity-based approach to Nash equi-
librium seeking over networks”, IEEE Transactions on Automatic
Control, Vol. 64, No. 3, pp. 1077−1092, 2019.
Fig. 2: Profiles of the various disturbances.
[6] X. He and J. Huang, “Distributed Nash equilibrium seeking with
dynamics subject to disturbance of unknown frequencies over jointly
strongly connected switching networks”, IEEE Transactions on Auto-
matic Control, conditionally accepted.
[7] X. He and J. Huang, “Distributed Nash equilibrium seeking over
strongly connected switching networks”, Neurocomputing, DOI:
https://doi.org/10.1016/j.neucom.2023.02.064.
[8] S. Li and T. Basar, “Distributed algorithms for the computation of
non-cooperative equilibria”, Automatica, Vol. 23, No. 4, pp. 523-533,
1987.
[9] A. R. Romano and L. Pavel, “Dynamic Nash equilibrium seeking
for high-order integrators in networks”, European Control Conference
(ECC) pp. 1029-1035, 2019.
[10] A. R. Romano and L. Pavel, “Dynamic NE seeking for multi-integrator
networked agents with disturbance rejection”, IEEE Transactions on
Control of Network Systems, pp. 129-139, 2020.
[11] G. Scutari, F. Facchinei, J.-S. Pang, and D. P. Pallomar, “Real and
complex monotone communication games”, IEEE Transactions on
Information Theory, Vol. 60, No. 7, pp. 400-409, 2014.
[12] J. Shamma and G. Arslan, “Distributed convergence to Nash equilibria
with local utility measurements”, IEEE Conference on Decision and
Control, pp. 1538-1543, 2004.
[13] K. Suad and S. Grammatico, “An integral Nash equilibrium control
scheme for a class of multi-agent linear systems”, IFAC PapersOnLine,
pp. 5375-5380, 2020.
[14] T. Wang and J. Huang, “Leader following consensus of multiple Fig. 3: The trajectories of the players’ positions xi (t), i ∈
spacecraft systems with disturbance rejection over switching networks
by adaptive learning control”, International Journal of Robust and
{1, 2, 3, 4, 5} under control law (12).
Nonlinear Control, Vol. 32, No. 5, pp. 3001-3020. 2022.
[15] M. Ye and G. Hu, “Distributed Nash equilibrium seeking by a
consensus based approach”, IEEE Transactions on Automatic Control,
Vol. 62, No. 9, pp. 4811-4818, 2017.
[16] M. Ye and G. Hu, “Distributed Nash equilibrium seeking in multiagent
games under switching communication topologies”, IEEE Transac-
tions on Cybernetics, Vol. 48, No. 11, pp. 3208-3217, 2018.
[17] M. Ye, G. Hu and F. L. Lewis, “Nash equilibrium seeking for N-
coalition noncooperative games”, Automatica, Vol. 95, pp. 266-272,
2018.
[18] M. Ye, “Distributed Nash equilibrium seeking for games in systems
with bounded control inputs”, IEEE Transactions on Automatic Con-
trol , Vol. 66, No. 8, pp. 3833-3839, 2021.

520
Authorized licensed use limited to: University of Virginia Libraries. Downloaded on April 08,2024 at 21:34:57 UTC from IEEE Xplore. Restrictions apply.