Automatica 119 (2020) 109086

Contents lists available at ScienceDirect

Automatica
journal homepage: www.elsevier.com/locate/automatica

Prescribed performance distance-based formation control of

Multi-Agent Systems✩
∗
Farhad Mehdifar a , , Charalampos P. Bechlioulis b , Farzad Hashemzadeh c ,
Mahdi Baradarannia c
a
INMA, ICTEAM, UCLouvain, Louvain-la-Neuve, Belgium
b
School of Mechanical Engineering, National Technical University of Athens, Athens, Greece
c
Faculty of Electrical and Computer Engineering, University of Tabriz, Tabriz, Iran

article info a b s t r a c t

Article history: This paper presents a novel control protocol for robust distance-based formation control with
Received 19 July 2019 prescribed performance in which agents are subjected to unknown external disturbances. Connectivity
Received in revised form 20 November 2019 maintenance and collision avoidance among neighboring agents are also handled by the appropriate
Accepted 13 May 2020
design of certain performance bounds that constrain the inter-agent distance errors. As an extension to
Available online xxxx
the proposed scheme, distance-based formation centroid maneuvering is also studied for disturbance-
Keywords: free agents, in which the formation centroid tracks a desired time-varying velocity. The proposed
Distance-based formation control and control laws are decentralized, in the sense that each agent employs local relative information
maneuvering regarding its neighbors to calculate its control signal. Therefore, the control scheme is implementable
Rigid graphs on the agents’ local coordinate frames. Using rigid graph theory, input-to-state stability, and Lyapunov
Prescribed performance control
based analysis, the results are established for minimally and infinitesimally rigid formations in 2-D
Connectivity maintenance
or 3-D space. Furthermore, it is argued that the proposed approach increases formation robustness
Collision avoidance
Unknown external disturbances against shape distortions and can prevent formation convergence to incorrect shapes under the effect
Input-to-state stability of external disturbances, which is likely to happen in conventional distance-based formation control
methods. Finally, extensive simulation studies clarify and verify the proposed approach.
© 2020 Elsevier Ltd. All rights reserved.

1. Introduction
During the past several years, cooperative control of Multi-
Agent Systems (MAS), which deals with achieving a global group
behavior that is beyond each individual capabilities through lo-
cal interactions, has attracted increasing attention due to its
broad applications (Dorri, Kanhere, & Jurdak, 2018). Control prob-
lems in MAS are mainly classified into consensus, formation,
containment, flocking, coverage, and rendezvous (Anderson, Yu,
Hendrickx, et al., 2008; Cao, Yu, Ren, & Chen, 2013; Wang, Zeng,
& Cong, 2016). Particularly, formation control refers to the design
of appropriate control protocols for stabilizing agents’ positions
with respect to each other so that they set up and maintain
a predefined geometrical shape. Based on a recent survey (Oh,
✩ F. Mehdifar is an FRIA/FNRS fellow and his work is also supported by the
concerted research action (ARC) ‘‘RevealFlight’’. The material in this paper was
not presented at any conference. This paper was recommended for publication
in revised form by Associate Editor Michael M. Zavlanos under the direction of
Editor Christos G. Cassandras.
∗ Corresponding author.
E-mail addresses: farhad.mehdifar@uclouvain.be (F. Mehdifar),
chmpechl@mail.ntua.gr (C.P. Bechlioulis), hashemzadeh@tabrizu.ac.ir
(F. Hashemzadeh), mbaradaran@tabrizu.ac.ir (M. Baradarannia).
Park, & Ahn, 2015), the existing approaches can be categorized
into position-, displacement-, and distance-based formation con-
trol schemes, depending on the sensing and controlled variables.
Among them, distance-based formation control is considered to
be an attractive architecture since it imposes less implementation
issues compared to other methods. In distance-based formation
control, agents measure the relative positions with respect to
their neighbors and actively control their inter-agent distances in
order to reach a desired predefined shape. This approach enables
us to design formation control laws in agents’ local coordinate
frames, which neither requires global position measurements
(e.g., using GPS) nor pre-alignment of agents’ local coordinate
frames (e.g., using a compass) (Meng, Anderson, & Hirche, 2016;
Oh et al., 2015). In particular, this formation control approach is
advantageous not only due to lower agents’ costs (since they use
less complex equipment for sensing and local interactions) but
also for operation in GPS-denied environments, where unmanned
multi-agent systems are used for search and rescue operations,
planetary explorations, indoor navigation, and so on (Ramazani,
Selmic, & de Queiroz, 2017).

https://doi.org/10.1016/j.automatica.2020.109086
0005-1098/© 2020 Elsevier Ltd. All rights reserved.
2 F. Mehdifar, C.P. Bechlioulis, F. Hashemzadeh et al. / Automatica 119 (2020) 109086
An introduction to distance-based formation control for both
undirected and directed inter-agent sensing graphs is found in An-
derson et al. (2008). Some early results in distance-based forma-
tion control are also given in Cai and de Queiroz (2014), Dorfler
and Francis (2010), Krick, Broucke, and Francis (2009), Oh and
Ahn (2011). In these works, formation acquisition (convergence
to a stationary shape) for single integrator agents with undirected
minimally and infinitesimally rigid interaction graphs is consid-
ered. In recent years, various modified controllers for undirected
distance based formation acquisition were also developed consid-
ering event-triggered control (Sun, Liu, Huang, Yu, & Anderson,
2019), quantized distance measurements (Sun, Garcia de Marina,
Anderson, & Cao, 2018), exponential convergence (Sun, Mou, An-
derson, & Cao, 2016), finite-time convergence (Sun, Mou, Deghat,
& Anderson, 2015), optimal control (Babazadeh & Selmic, 2019),
source-seeking (Barogh & Werner, 2017), nonlinear dynamics (Cai
& de Queiroz, 2015), formation scaling (Yang, Sun, Cao, Fang, &
Chen, 2019), formation acquisition with desired orientation (Sun,
Park, Anderson, & Ahn, 2017), and a multi layered version of
distance-based formation acquisition (Ramazani et al., 2017). A
method for dealing with the problem of convergence to incorrect
equilibrium points (undesired shapes) of distance-based forma-
tion acquisition controllers was recently proposed in Anderson,
Sun, Sugie, Azuma, and Sakurama (2017), Liu, de Queiroz, Zhang,
and Khaledyan (2019), introducing an additional control vari-
able (triangular signed areas among agents) for a class of 2-D
shapes. Furthermore, the authors in Mou, Belabbas, Morse, Sun,
and Anderson (2015) analyzed the practical issue of distance
mismatches among neighboring agents for undirected distance-
based formations and a solution was proposed in De Marina, Cao,
and Jayawardhana (2014).
In addition to formation acquisition control, there have been
attempts to solve distance-based formation control problems
with moving shapes (such as formation tracking and formation
maneuvering). In Deghat, Anderson, and Lin (2015), by combin-
ing distance-based formation control with consensus protocols,
agents move in a target formation shape with a common constant
speed. The results are enhanced taking into account the collision
avoidance problem and finite time convergence scheme in Wang
and Guo (2018) and Sun et al. (2015), respectively. Recently,
in Yang, Cao, de Marina, Fang, and Chen (2018), a weighted
centroid formation tracking with distance-based control laws was
introduced, where the formation centroid tracks a predefined
path. However, this methodology requires agents to sense relative
orientations as well. Moreover, Babazadeh and Selmic (2019) de-
veloped a distance-based optimal formation tracking using State
Dependent Riccati Equation with energy constraints, nonetheless,
the design methodology relies on a centralized control approach.
A distance-based formation maneuvering controller is proposed
in Cai and Queiroz (2015) provided that all agents have di-
rect access to the desired time-varying swarm velocity, hence,
it requires pre-alignment of agent’s local coordinate systems.
Later, Khaledyan, Liu, Fernandez-Kim, and de Queiroz (2019),
Mehdifar, Hashemzadeh, Baradarannia, and de Queiroz (2019)
utilized distributed velocity estimators in distance-based forma-
tion maneuvering for single integrator and unicycle agent models,
respectively. In these works, agents estimate the desired group
velocity that is only available to a leader in order to relax the
requirement of direct access to the desired time-varying group
velocity. Nevertheless, these schemes also require inter-agent
relative orientation measurements in order to be applicable in
arbitrary oriented local coordinate systems. In Rozenheck, Zhao,
and Zelazo (2015), a distance-based centroid formation maneu-
vering with a leader is investigated, where the centroid of the
formation tracks a constant (or at best a very slowly varying)
desired reference velocity.
The existence of external disturbances that affect the agents
dynamics is a significant issue of practical interest for MAS ap-
plications. It is noteworthy to mention that in distance-based
formation control problems, none of the aforementioned works
have taken into account external disturbances. Recently, Bae,
Lim, Kang, and Ahn (2018) studied the disturbance attenuation
problem with the LMI approach in distance-based formation con-
trol. However, its results depend on certain LMI feasibility tests,
which are not favorable in practice and may increase complex-
ity in the controller design. Moreover, as a practical problem,
collision avoidance among agents has been addressed partially
in a few of the above mentioned works, such as Babazadeh
and Selmic (2019), Barogh and Werner (2017), Wang and Guo
(2018). In addition, none of these works has addressed con-
nectivity maintenance among neighboring agents, which is also
critical since agents have limited sensing capabilities in practice.
Finally, another crucial issue concerns the transient response of
the MAS. Towards this direction, Prescribed Performance Con-
trol (PPC) (Bechlioulis & Rovithakis, 2008), proposes a simple
and constructive procedure based on which the transient per-
formance of the closed-loop system is predetermined by certain
user defined performance bounds. This method has been also
applied in MAS control (Bechlioulis, Demetriou, & Kyriakopou-
los, 2018; Karayiannidis, Dimarogonas, & Kragic, 2012; Macellari,
Karayiannidis, & Dimarogonas, 2017). Recently, PPC has been
utilized for formation control problems as well, however, these
results are mainly applied to displacement-based formation con-
trol methods (Bechlioulis & Kyriakopoulos, 2014; Bechlioulis &
Rovithakis, 2016; Wang, Xi, Yuan, & Liu, 2018). A recent paper
(Verginis, Nikou, & Dimarogonas, 2019) has also employed PPC for
MAS, however, rather than solving a formation control problem,
by controlling inter-agent relative orientations and distances it
solves a distance- and orientation- based multi-agent coordi-
nation problem (that cannot ensure convergence to a specific
pre-defined shape) where inter-agent interactions are modeled
by undirected tree graphs.
In this paper, we propose a robust distance-based formation
acquisition control protocol with guaranteed transient perfor-
mance, connectivity maintenance and collision avoidance among
neighboring agents. The target formation is assumed to be mini-
mally and infinitesimally rigid in two or three dimensional space.
User defined performance guarantees on the system’s response
are achieved by employing time-varying decreasing performance
bounds on the inter-agent distance errors. We prove that the
proper choice of performance bounds handles the problems of
connectivity maintenance and collision avoidance among neigh-
boring agents as well. More specifically, an error transformation
technique is used to convert the original constrained error system
into a new equivalent unconstrained one, whose stability only en-
sures satisfaction of certain time-varying constraints of the inter-
agent distance errors. Afterwards, we develop a distance-based
formation maneuvering control protocol, based on which the
centroid of the formation tracks a desired time-varying velocity.
It is assumed that the desired centroid velocity is only available
to the leader of the group. The proposed control approach is
independent of a global coordinate system and can be applied
to arbitrarily oriented local coordinate frames. Furthermore, we
prove that the proposed approach naturally reduces potential
distortions induced by external disturbances, thus preventing
convergence to undesired shapes.
The contributions of this work are summarized as follows:1
1 Our previous work (Mehdifar, Hashemzadeh, & Baradarannia, 2017) guaran-
tees only connectivity maintenance and collision avoidance among neighboring
agents by enforcing constant constraints on the distance errors. Hence, it is not
capable of: (i) imposing predefined transient and steady state performance on
the closed-loop system, (ii) dealing with external disturbances while achieving
the desired formation and (iii) solving the formation maneuvering problem.
 F. Mehdifar, C.P. Bechlioulis, F. Hashemzadeh et al. / Automatica 119 (2020) 109086
• To the best of our knowledge, there is no previous work ad-
dressing distance-based formation control with guaranteed
transient and steady state performance as well as connec-
tivity maintenance and collision avoidance.
• The performance of the proposed scheme does not depend
on the upper bound of the external disturbances.
• This paper solves the distance-based formation maneuver-
ing problem with time-varying reference velocity for arbi-
trarily oriented local coordinate frames of agents and with-
out requiring any additional measurements on relative ori-
entations.
2. Preliminaries on graphs and rigidity theory
Consider an undirected graph with l edges and n vertices,
denoted by G ≜ (V , E ) where V = {1, 2, . . . , n} is the set of
vertices and E = {(i, j)|i, j ∈ V , i ̸ = j} is the set of edges. The
neighbor set of vertex i is defined as Ni (E ) = {j ∈ V | (i, j) ∈ E }.
The incidence matrix H = {hij } ∈ Rl×n , relates the edges of G
with its vertices. Assuming arbitrary edge orientation, the entries
of H are defined as hij = +1 when the ith edge sinks at node j,
hij = −1, when the ith edge leaves node j, and hij = 0, otherwise.
For any connected graph it is known that ker(H) = span{1n } (Sun
et al., 2015), where 1n stands for n × 1 vector of ones. Let pi ∈ Rm ,
where m ∈ {2, 3}, denotes a point that is assigned to vertex i ∈ V .
The stacked vector p = col(pi ) ≜ [pT1 , . . . , pTn ]T ∈ Rmn , represents
the realization of G in Rm . The pair F ≜ (G , p) is said to be a
framework of G in Rm . By introducing the matrix H̄ = H ⊗ Im ∈
Rml×mn , the relative position vectors corresponding to the edges
can be constructed as p̃ = H̄p, where p̃ = col(p̃ij ) ∈ Rml with
p̃ij = pi − pj ∈ Rm being the relative position vector defined for
any pair (i, j) ∈ E . Given an arbitrary ordering of edges in E , an
edge function (rigidity function) ΦG : Rmn → Rl associated with
(G , p) is given as:
]T
ΦG (p) = . . . , ∥pi − pj ∥2 , . . . , (i, j) ∈ E .
[
(1)
such that its kth component, i.e., ∥pi − pj ∥2 , relates to the kth edge
of E connecting the ith and jth vertices.
Definition 1 (Oh & Ahn, 2011). A framework F = (G , p) is
rigid if there exists a neighborhood Up of p ∈ Rmn such that
ΦG−1 [ΦG (p)] ∩ Up = ΦH −1
[ΦH (p)] ∩ Up , where H is a complete
graph of n vertices and Φ⋆−1 is a set of positions q ∈ Rmn satisfying
Φ⋆ (p) = Φ⋆ (q) for any graph ⋆.
Definition 1 implies that in a rigid framework, preserving the
length of the graph edges guarantees that all distances among
all vertices of the graph remain unaltered (i.e., the shape is
preserved). The rigidity matrix R : Rmn → Rl×mn of F = (G , p)
∂ Φ (p)
is then defined as R(p) = 12 ∂Gp = 21 ∂ Φ∂Gp̃(p) ∂∂ p̃p = P T H̄, where
P = blockdiag(p̃ij ) for the same ordering of edges as in (1). Note
that, each row of the rigidity matrix R(p) takes the following form:
0T1×m . . . p̃Tij . . . 0T1×m . . . − p̃Tij . . . 0T1×m
[ ]
(2)
Hence, the rigidity matrix depends solely on the relative positions
and can be written as R(p̃ ). It is known that rank[R(p)] ≤ 2n − 3
in R2 , and rank[R(p)] ≤ 3n − 6 in R3 (Asimow & Roth, 1979).
Definition 2 (Anderson et al., 2008). A rigid framework is min-
imally rigid if no single inter-agent distance constraint can be
removed without losing its rigidity. In R2 ( R3 ) a rigid framework
(G, p) is minimally rigid if l = 2n − 3 (l = 3n − 6).
Definition 3 (Hendrickson, 1992). A framework F = (G , p) is
infinitesimally rigid in m-dimensional space if: rank[R(p)] = mn−
m(m + 1)
. the neighboring agents i and j, respectively. Furthermore, define
2
3
If a framework is infinitesimally rigid in R2 (R3 ) and its under-
lying graph has exactly 2n − 3 (3n − 6) edges, then it is called a
minimally and infinitesimally rigid framework. If ΦG (p) = ΦG (q)
holds for two frameworks Fp = (G , p) and Fq = (G , q), they are
said to be equivalent. Also, if ∥pi − pj ∥ = ∥qi − qj ∥, ∀i, j ∈ V , then
the two frameworks are congruent. Let Iso(F ) denotes the set of
all isometric frameworks (rotation, translation, and reflection) of
F . Note that, (1) is invariant under isometric motions of F . Two
infinitesimally rigid frameworks Fp = (G , p) and Fq = (G , q)
are said to be ambiguous, if they are equivalent but not congru-
ent (Anderson et al., 2008). We denote the set of all ambiguities
of an infinitesimally rigid framework F and its isometries by
Amb(F ). It can be reasonably assumed that all frameworks in
Amb(F ) are infinitesimally rigid (Cai & de Queiroz, 2014; Cai &
Queiroz, 2015) (according to Anderson et al. (2008) and Asimow
and Roth (1979, Theorem 3), this assumption holds almost ev-
erywhere; therefore, it is not restrictive). Fig. 1 illustrates the
aforementioned concepts.
Lemma 1 (Cai & Queiroz, 2015). If Fp = (G , p) is infinitesimally
rigid then there exists a small positive constant ∑ϑ such that all
frameworks Fq = (G , q) that satisfy Ψ (Fq , Fp ) ≜ (i,j)∈E (∥qi −qj ∥−
∥pi − pj ∥)2 ≤ ϑ , are also infinitesimally rigid.
Lemma 2 (Sun et al., 2015). If the framework F = (G , p) is
minimally and infinitesimally rigid in m-dimensional space, then the
matrix R(p)R(p)T is positive definite.
3. Problem statement
Consider n interacting agents in an m-dimensional space, with
m ∈ {2, 3}, governed by:
q̇i = ui + δi (t), i = 1, . . . , n (3)
where qi ∈ Rm is the position, ui ∈ Rm is the velocity control
input of agent i with respect to a fixed coordinate frame, and
δi (t) ∈ Rm is an unknown, bounded and piece-wise continuous
external disturbance vector. Let the desired formation be defined
by a minimally and infinitesimally rigid framework F ∗ = (G ∗ , q∗ )
where G ∗ = (V ∗ , E ∗ ), dim(V ∗ ) = n, dim(E ∗ ) = l, and q∗ =
col(q∗i ) ∈ Rmn . Moreover, assume that the actual framework
(actual formation) of the agents, which shares the same graph
with F ∗ , is represented by F (t) = (G ∗ , q(t)), where q(t) =
col(qi (t)) ∈ Rmn . Let the desired distances between agent i and its
neighboring and non-neighboring agents in the rigid framework
be given by dij = ∥q∗i − q∗j ∥ > 0, for i, j ∈ V ∗ , and the relative
positions between neighboring agents as:
q̃ij = qi − qj , (i, j) ∈ E ∗ . (4)
Thus, for each edge of the rigid graph, the distance error is given
by:
eij = ∥q̃ij ∥ − dij , (i, j) ∈ E ∗ , (5)
where ∥q̃ij ∥ is the actual distance between agents i and j. From
(5) it is clear that eij ∈ [−dij , ∞). In what follows, collision avoid-
ance and connectivity maintenance among neighboring agents
are formulated with respect to eij .
Collision Avoidance of Neighboring Agents: The agents should
not collide during their motion towards the desired formation. In
general, to cope with this issue, a circular safety region around
each agent is assumed and the controller is designed to ensure
that the aforementioned safety regions do not overlap during the
operation. In this paper, we assume that all agents have spherical
shapes. Let rsi ∈ R+ , rsj ∈ R+ be the geometrical radius of
4 F. Mehdifar, C.P. Bechlioulis, F. Hashemzadeh et al. / Automatica 119 (2020) 109086
Fig. 1. Graph (a) is a non-rigid (flexible) graph. Graph (b) is a minimally and infinitesimally rigid graph. Graph (c) belongs to the Amb set of graph (b).
Fig. 2. The sensing and safety regions of two neighboring agents.
rsij = (rsi + rsj ) > 0, (i, j) ∈ E ∗ . Obviously, it is necessary that
dij > rsij , otherwise, the formation is not feasible. Thus, in order
to ensure collision avoidance between neighboring agents, it is
required that ∥q̃ij (t)∥ > rsij for all t ≥ 0, which can be restated in
terms of the distance errors as: rsij − dij < eij (t), (i, j) ∈ E ∗ , ∀t ≥ 0,
where rsij − dij is a negative value.
Connectivity Maintenance: In practice, since each agent has
a limited sensing capability it is also necessary for neighboring
agents to remain within their common sensing area during the
operation, otherwise, the whole system might become unstable
or inactive. In this paper, connectivity maintenance is equivalent
to not losing any edge in the actual undirected rigid formation
graph during the operation. Let rci ∈ R+ and rcj ∈ R+ be the
sensing radius of agents i and j, respectively. Furthermore, let us
define rcij = min{rci + rsj , rcj + rsi }, (i, j) ∈ E ∗ . Notice that dij should
be less than rcij so that we seek for a feasible target formation.
Therefore, securing ∥q̃ij (t)∥ < rcij for all t ≥ 0 is sufficient to en-
sure connectivity maintenance, which can be formulated in terms
of the distance errors as: eij (t) < rcij − dij , (i, j) ∈ E ∗ , ∀t ≥ 0. Notice
that, always rci > rsi , ∀i ∈ V ∗ holds. Moreover, collision avoidance
and connectivity maintenance conditions together indicate that
rcij > rsij , (i, j) ∈ E ∗ . Fig. 2 depicts two neighboring agents.
Control Objective: Under the assumption that initially the ac-
tual formation graph is minimally and infinitesimally rigid, which
is common in distance-based formation control literature (Cai &
Queiroz, 2015; Rozenheck et al., 2015; Sun et al., 2016, 2015), the
objective is to design a decentralized robust control protocol such
that:
F (t) → Iso(F ∗ ) as t → ∞ (6a)
−eij (t) <eij (t) < eij (t), (i, j) ∈ E , ∀t ≥ 0,
∗
(6b)
where eij (t) > 0 and eij (t) > 0 are decreasing performance
bounds satisfying limt →∞ eij (t) > 0 and limt →∞ eij (t) > 0. Notice
further that if the performance bounds are selected appropriately
to satisfy eij (0) ≤ rcij − dij and eij (0) ≤ dij − rsij , then connectivity
maintenance and collision avoidance among neighboring agents
will be ensured for all t ≥ 0.
4. Controller design and stability analysis
The following lemma provides a sufficient condition to estab-
lish infinitesimal rigidity of the actual formation F (t) based on
the distance error bounds (6b).
Lemma 3. If F ∗ is infinitesimally rigid and ϑ̄ is a sufficiently small
positive constant
∑ satisfying:
Ψ (F , F ∗ ) ≜ (i,j)∈E ∗ max{|eij (0)|, |eij (0)|} ≤ ϑ̄ , then securing (6b)
guarantees that F is also infinitesimally rigid for all time.
Proof. Using (4) and (5),∑Ψ (Fq , Fp ) in Lemma 1 ∑ can be repre-
sented as: Ψ (F , F ∗ ) = (i,j)∈E ∗ ( ∥ q̃ij ∥ − d ij ) 2
= 2
(i,j)∈E ∗ eij ≤
ϑ , for a small positive constant ϑ . A sufficient condition for
(i,j)∈E ∗ |eij | < ϑ̄ , where ϑ̄ is a
∑
holding this inequality is:
sufficiently small positive constant such that ϑ̄ 2 ≤ ϑ . More-
over, if we select eij (0), eij (0) and guarantee (6b) ∑ for decreasing
performance
∑ functions e ij (t) , e ij (t) we arrive at: (i,j)∈E ∗ |eij (t)| <
(i,j)∈E ∗ max{|eij (0)|, |eij (0)|}, ∀t ≥ 0. Clearly, establishing
Ψ (F , F ∗ ) ≤ ϑ̄ in Lemma 3 is sufficient for the above inequality,
and thus it is sufficient for holding Ψ (F , F ∗ ) ≤ ϑ in Lemma 1 as
well. Finally, Lemma 1 yields that F is infinitesimally rigid, which
concludes the proof.
For an infinitesimally rigid target formation F ∗ , Lemma 3
indicates that if eij (0) and eij (0) are properly selected and (6b)
is satisfied, then the infinitesimal rigidity of the actual formation
graph F (t) is ensured for all time. Note that, the assumption on
infinitesimal rigidity of the actual rigid{ graph at t = 0 can be
demonstrated by e(0) ∈ ΩI where ΩI = e ∈ Rl | Ψ (F , F ∗ ) ≤ ϑ̄ ,
}
and e = col(eij ) ∈ Rl for the same ordering as in the edge
function (1). Moreover, since rigid graphs are connected, the
aforementioned assumption also secures the initial connectivity
specification of neighboring agents.
4.1. Proper Selection of eij (0) and eij (0)
As mentioned earlier, since eij (t) and eij (t) are assumed to be
decreasing, by selecting eij (0) and eij (0) appropriately, one can ob-
tain the distance error bounds which ensure infinitesimal rigidity
of the actual formation as well as connectivity maintenance and
collision avoidance among neighboring agents. Note that, when
e(0) ∈ ΩI and |eij (t)| < |eij (0)| + µ, where µ is a sufficiently
small positive constant, then from Lemma 3 it is reasonable to
set eij (0) = eij (0) = |eij (0)| + µ as the maximum allowable
bounds for eij in order to guarantee infinitesimal rigidity of the
actual formation graph. It is clear that increasing µ reduces the
conservativeness in choosing the bounds of eij that ensure this
property. Notice that since eij ∈ [−dij , ∞), if eij (0) > 0 and
−(|eij (0)| + µ) < −dij then we should take eij (0) = dij . We also
define the formation robustness constants µij and µ with respect
ij
to each edge of the rigid graph where 0 < µij ≤ rcij − dij and
0 < µ ≤ dij − rsij , (i, j) ∈ E ∗ . Adjusting these constants is useful
ij
for improving robustness against formation distortion caused by
external disturbances, which will be discussed later.
 F. Mehdifar, C.P. Bechlioulis, F. Hashemzadeh et al. / Automatica 119 (2020) 109086 5

We propose Algorithm 1 for selecting the allowable upper and Remark 1. Note that, by using (7) one can write ηij = eij (eij +2dij )
lower bounds of eij in (6b) (eij (0) and eij (0)) to ensure infinitesimal and −η = −eij (−eij + 2dij ). Since eij (t) > 0 and 0 < eij (t) <
ij
rigidity, connectivity maintenance and collision avoidance among dij , then we can obtain the following
neighboring agents. Algorithm 1 is applied in a distributed fashion
√ √ relations: eij (t) = −dij +
d2ij + bij ρij (t) and eij (t) = dij − d2ij − bij ρij (t) between perfor-
and each agent calculates its corresponding edges’ maximum
allowable distance error bounds. Agents should sense the initial mance bounds of ηij (t) and eij (t). Therefore, satisfaction of (8),
distance errors eij (0) with their neighbors and use the sensing and fulfills the condition in (6b) with the aforementioned decreasing
the geometric radius. Note that, when the sensing and geometric prescribed performance bounds.
radii are not identical for all agents, then agent i may receive
scalars rcj and rsj through communication with its neighbors at 4.3. Transformed error system
t = 0.
Algorithm 1: Distributed selection of eij (0) and eij (0), j ∈ Ni (E ∗ ) The distance error dynamics corresponding to each edge of the
for i-th agent at t = 0. rigid framework can be obtained invoking (3) and (5) as ėij =
Input: eij (0), dij , rci , rcj , rsi , rsj , µ q̃Tij (ui − uj + δi − δj )/(eij + dij ), (i, j) ∈ E ∗ . Using ėij and (7), the time
1: rsij = (rsi + rsj ) and rcij = min{rci , rcj } derivative of ηij is given by:
2: Choose µij and µ such that 0 < µij ≤ rcij − dij and
ij η̇ij = 2q̃Tij (ui − uj + δi − δj ), (i, j) ∈ E ∗ . (10)
0 < µ ≤ dij − rsij
ij
3: if (eij (0) ≥ 0) then Based on the structure of rigidity matrix in (2), (10) can be written
4: eij (0) = min{|eij (0)|+µ , rcij − dij } in compact form as:
5: eij (0) = min{|eij (0)|+µ , µ }
ij η̇ = 2R(q)(u + δ ), (11)
6: else if (eij (0) < 0) then
7: eij (0) = min{|eij (0)|+µ , µij } where u = col(ui ) ∈ Rmn , δ = col(δi ) ∈ Rmn , and η = col(ηij ) ∈
8: eij (0) = min{|eij (0)|+µ , dij − rsij } Rl , (i, j) ∈ E ∗ for the same ordering as in (1). The problem of
9: end if designing a controller that meets error constraints in (8) can be
10: return eij (0), eij (0) simply transformed into establishing the boundedness of certain
error signals via the prescribed performance control methodol-
ogy (Bechlioulis & Rovithakis, 2008). Specifically, to handle the
4.2. Prescribed performance bounds time-varying constraints in (8), an error transformation technique
is used to convert the original dynamical error system (10) with
Define the squared distance error ηij = ∥q̃ij ∥2 − d2ij , (i, j) ∈ E ∗ , constraints (8) into a new equivalent unconstrained one, whose
then it follows from (5) that: stability ensures satisfaction of the restrictions given in (8). First,
consider the following modulated squared distance errors:
ηij = eij ∥q̃ij ∥ + dij .
( )
(7)
ηij (t)
Since eij ∈ [−dij , ∞), from (7), it is easy to see ηij = 0 if η̂ij (t) ≜ , (i, j) ∈ E ∗ . (12)
and only if eij = 0. Therefore, instead of using eij directly, we ρij (t)
will use the squared distance error to impose specific prescribed To transform the constrained error system (in the sense of (8))
performance bounds for all times. Similar to the results developed into an equivalent unconstrained one, we introduce the following
in Bechlioulis and Rovithakis (2008), for each edge of the rigid error transformation corresponding to each edge in the rigid
framework, first a smooth decreasing performance function ρij (t) : framework:
[0, ∞) → R+ , (i, j) ∈ E ∗ with limt →∞ ρij (t) = ρij∞ > 0 is chosen. ( )
In this paper, we adopt the following exponential performance 1 bij η̂ij + bij bij
σij = Tij (η̂ij ) = ln , (i, j) ∈ E ∗, (13)
function: ρij (t) = (ρij0 − ρij∞ ) exp(−aij t) + ρij∞ , (i, j) ∈ E ∗ , where 2 bij bij − bij η̂ij
ρij0 > ρij∞ and aij > 0. The objective of guaranteeing transient
performance can be achieved if the following condition is always where σij is the transformed error corresponding to ηij and Tij (.) :
satisfied (−bij , bij ) → (−∞, +∞) is a smooth, strictly increasing bijec-
tive mapping satisfying Tij (0) = 0. It is clear that ηij → 0
− ηij (t) < ηij (t) < ηij (t), (i, j) ∈ E ∗ , (8) if and only if σij → 0. Finally, notice that maintaining the
boundedness of σij ∈ R, enforces −bij < η̂ij (t) < bij or
where ηij (t) = bij ρij (t) and η (t) = bij ρij (t), in which bij , bij > 0
ij equivalently (8). Now, taking the time derivative of σij , yields
are positive scalars. It is clear that ηij (0) = bij ρij (0) and η (0) = σ̇ij = (dTij /dη̂ij )η̂˙ ij = 0.5 [1/(η̂ij + bij ) − 1/(η̂ij − bij )] × [(η̇ij /ρij ) −
ij
bij ρij (0) are the maximum overshoot and the maximum under- (ηij ρ̇ij /ρij2 )] = 0.5 ξij (η̇ij − η̂ij ρ̇ij ), where
shoot (i.e. negative overshoot) of ηij (t), respectively. Furthermore,
the decreasing rate of ρij (t) introduces a lower bound on the
[ ]
1 1 1
convergence speed of ηij (t). Moreover, based on (7), eij (0) and ξij ≜ − , (i, j) ∈ E ∗ . (14)
ρij η̂ij + bij η̂ij − bij
eij (0) can be also related to ηij (0) = bij ρij0 and η (0) = bij ρij0
ij
as: Then σ̇ij is given in compact form as:
1
eij (0) + 2dij eij (0) = bij ρij0 ,
2
(9a) σ̇ = ξ (η̇ − ρ̇ η̂) (15)
2
2dij eij (0) − e2ij (0) = bij ρij0 . (9b)
where σ = col(σij ) ∈ Rl , ξ = diag(ξij ) ∈ Rl×l , ρ̇ = diag(ρ̇ij ) ∈
Consequently, if eij (0) and eij (0) are given from Algorithm 1, then Rl×l and η̂ = col(η̂ij ) ∈ Rl , for (i, j) ∈ E ∗ and the same
one can always assume ρij0 = 1 and find bij and bij to encapsulate ordering as in (1). In the next subsection σ is employed in the
the requirements of infinitesimal rigidity, connectivity mainte- design of a distance-based formation controller that stabilizes
nance and collision avoidance between neighboring agents within the transformed error dynamics (15), thus satisfying the error
the performance bounds defined in (8). transformation in (8) for all time.
6 F. Mehdifar, C.P. Bechlioulis, F. Hashemzadeh et al. / Automatica 119 (2020) 109086

Remark 2 (PPC Design Philosophy).
When −η (0) < ηij (0) < ηij (0), based on the properties of the
ij
error transformation (13), prescribed performance in the sense of
(8) is achieved, if σij (t), (i, j) ∈ E ∗ is kept bounded. Moreover, it
is important to note that the specific bounds of σij (t) (no matter
how large they are, which is the key property of the adopted error
transformation) do not affect the achieved transient and steady
state performance on ηij (t), which is solely determined by (8) and
edges, whenever connectivity maintenance is ensured for F (t).
Thus, since F ∗ is minimally rigid, F (t) is also minimally rigid for
all t ∈ [0, τmax ). Consequently, it can be deduced that F (t) is
minimally and infinitesimally rigid for all t ∈ [0, τmax ). Further-
more, owing to (18), the error signals σij in (13) are well defined.
Therefore, consider the potential function: Vij = 0.5 kij σij2 , (i, j) ∈
E ∗ , and define the overall candidate Lyapunov function: V =
(i,j)∈E ∗ Vij (σij ) = 0.5 σ K σ . Employing (11) and (15), we ar-
T
∑

thus by selection of the performance functions ρij (t) as well as bij rive at: V̇ = σ K ξ Ru + σ T K ξ Rδ − 0.5 σ T K ξ ρ̇ η̂. By Young’s
T

and bij , (i, j) ∈ E ∗ . inequality we get: σ T K ξ Rd ≤ 0.5 (σ T K ξ RRT ξ K σ ) + 0.5 ∥δ∥2 .

Substituting (16) into V̇ and employing the aforementioned in-
4.4. Main result equality and Lemma 2, yields: V̇ ≤ −0.5 (σ T K ξ RRT ξ K σ ) +
0.5 ∥δ∥2 − 0.5 σ T K ξ ρ̇ η̂ ≤ −0.5 λmin (RRT )σ T K ξ ξ K σ + 0.5 ∥δ∥2 −
Theorem 1. Consider n agents with dynamics (3) in an m- 0.5 σ T K ξ ρ̇ η̂. Let ϵ be an arbitrarily small positive constant satis-
dimensional space (m ∈ {2, 3}) with the actual formation F (t) = fying λmin (RRT ) > ϵ 2 /2. Invoking Young’s inequality on −(1/2)
(G ∗ , q(t)). Let the initial conditions be such that e(0) ∈ ΩI , which σ T K ξ ρ̇ η̂ and exploiting the diagonality of K , ξ , ρ matrices we
implies infinitesimal rigidity of the actual rigid graph at t = 0. If the obtain:
desired formation F ∗ is minimally and infinitesimally rigid and also [
λmin (RRT ) ϵ2
]
∥δ∥2 1
eij (0), eij (0) are selected according to Algorithm 1, then the following V̇ ≤− − σ TK ξ ξ K σ + + η̂T ρ̇ 2 η̂
control law: 2 4 2 4ϵ 2
λmin (RRT ) ϵ 2 ∥δ∥2
[ ]
u = −RT ξ K σ , (16) ≤− − λmin (ξ 2 )λmin (K 2 )∥σ ∥2 +
2 4 2
where K = diag(kij ) > 0, R is the rigidity matrix, σ = col(σij ), 1
and ξ = diag(ξij ) with ξij , σij defined in (13) and (14), respectively, + λmax (ρ̇ )∥η̂∥ ,
2 2
∀t ∈ [0, τmax ), (19)
4ϵ 2
ensures prescribed performance in the sense of (8) and leads to con-
nectivity maintenance and collision avoidance among neighboring where ∥δ∥2 , λmax (ρ̇ 2 ), and ∥η̂∥2 are bounded. Let λ ≜ [λmin (RRT )/2−
agents. ϵ 2 /4]λmin (ξ 2 )λmin (K 2 ) > 0. Moreover, by (18), we get: ∥η̂∥2 =
2
(i,j)∈E ∗ η̂ij < (i,j)∈E ∗ max{bij , bij } = γ . Then from (19) we
2 2
∑ ∑
Proof. The proof of Theorem 1 proceeds in three phases. First, arrive at: V̇ ≤ −λ∥σ ∥ + (1/2)∥δ∥2 + (γ /4ϵ 2 )λmax (ρ̇ 2 ), ∀t ∈
2

we show that all η̂ij (t) remain within (−bij , bij ) for a specific
time interval [0, τmax ) (i.e., the existence and uniqueness of a
maximal solution). Next, we prove that the proposed control
scheme guarantees, for all [0, τmax ): (a) the boundedness of all
closed loop signals as well as (b) that η̂ij (t) remains strictly in
a compact subset of (−bij , bij ), which by contradiction leads to
τmax = ∞ (i.e., forward completeness) in the last phase, thus
completing the proof.
In particular, the compact form of the modulated errors in (12)
is given by η̂ = ρ (t)−1 η where ρ (t)−1 = diag(ρij (t)−1 ) ∈ Rl×l .
Differentiating η̂ as well as employing (11) and (16), yields:
η̂˙ = hη̂ (t , η̂) = ρ (t)−1 (η̇ − ρ̇ (t)η̂)
= ρ (t)−1 (−2RRT ξ K σ (η̂) + 2Rδ − ρ̇ (t)η̂).
Let us also define the open set Ωη̂ = (i,j)∈E ∗ (−bij , bij ).
∏ bij (exp(2σ ) − 1)
Phase I. Algorithm 1 and (9) define bij and bij , which guarantee
that the set Ωη̂ is nonempty and open. Moreover, Algorithm 1,
ensures that η̂(0) ∈ Ωη̂ . Additionally, hη̂ is continuous on t and
locally Lipschitz on η̂ over the set Ωη̂ . Therefore, the hypotheses
of Theorem 54 in Sontag (1998, p. 476) hold and the existence and
uniqueness of a maximal solution η̂(t) of (17) for a time interval
[0, τmax ) such that η̂(t) ∈ Ωη̂ , ∀t ∈ [0, τmax ) is guaranteed.
Equivalently, we infer that:
η̂ij (t) ∈ (−bij , bij ), (i, j) ∈ E ∗ , ∀t ∈ [0, τmax ),
from which we obtain that ηij (t) are absolutely bounded as in (8)
for all t ∈ [0, τmax ). Accordingly, from Remark 1, it can be deduced
that −eij (t) < eij (t) < eij (t), (i, j) ∈ E ∗ , ∀t ∈ [0, τmax ). Recall
that eij (t) and eij (t) are decreasing functions of time. As a result,
it ensures: (a) connectivity maintenance and collision avoidance
of neighboring agents in the actual formation graph F (t) as well
as (b) infinitesimal rigidity of F (t) (due to e(0) ∈ ΩI ), for all
t ∈ [0, τmax ).
Phase II. From Phase I we know that infinitesimal rigidity
and connectivity of the neighboring agents is guaranteed for all
t ∈ [0, τmax ). Note that, F ∗ and F (t) have the same number of
[0, τmax ). Assume that φ = col(δ, λmax (ρ̇ 2 )) ∈ Rn+1 . Using Lemma
11.3 in Chen and Huang (2015, p. 343) yields: V̇ ≤ −λ∥σ ∥2 +
β (∥φ∥), ∀t ∈ [0, τmax ), where β is a K∞ -function. This inequality
satisfies the conditions of Theorem 2.7 in Chen and Huang (2015,
p. 33) (note that, all K∞ -functions are indeed class K functions).
Therefore, σ is Input-to-State Stable (ISS) with respect to input φ
for all t ∈ [0, τmax ). Due to ISS and the boundedness of φ for all
t ≥ 0, there exists an ultimate bound σ > 0 independent of τmax
such that ∥σ ∥ ≤ σ . Based on this result it is clear that (16) is also
bounded for all t ∈ [0, τmax ). Now using σ and taking the inverse
logarithmic function in (13), leads to:
bij (1 − exp(−2σ ))
−bij < − bij = η̂ ≤ η̂ij (t)
(17) bij exp(−2σ ) + bij ij
(20)
≤ η̂ij = bij < bij ,
bij exp(2σ ) + bij
for all t ∈ [0, τmax ) and (i, j) ∈ E ∗ .
Phase III. Up to this point, what remains to be shown is that
τmax can be extended to ∞. Towards this direction, ∏ notice by (20)
that η̂(t) ∈ Ωη̂′ , ∀t ∈ [0, τmax ) where Ωη̂′ = (i,j)∈E ∗ [η̂ ij , η̂ ij ] is
a nonempty compact subset of Ωη̂ . Hence, assuming τmax ≤ ∞
and since Ωη̂′ ⊂ Ωη̂ , Proposition C.3.6 in Sontag (1998, p. 481)
dictates the existence of a time instant t ′ ∈ [0, τmax ) such that
(18) η̂(t ′ ) ∈
/ Ωη̂′ , which is a clear contradiction. Therefore, τmax = ∞.
Thus, all closed loop signals remain bounded and moreover η̂(t) ∈
Ωη̂′ ⊂ Ωη̂ , ∀t ≥ 0. Finally, multiplying (20) by ρij (t) results in:
−bij ρij (t) < η̂ij ρij (t) ≤ ηij (t) ≤ η̂ij ρij (t) < bij ρij (t), for (i, j) ∈ E ∗
and t ≥ 0, which further guarantees (6b) for all t ≥ 0.
Remark 3. In the absence of external disturbances (i.e. δi =
0, i = 1, . . . , n) or even when δi (t) are vanishing with time
(i.e. δi (t) → 0 as t → ∞), the proposed controller guarantees ex-
act zero convergence of the errors with prescribed performance.
Note that, since ρij (t) is smooth and limt →∞ ρij (t) = ρij∞ > 0
then limt →∞ ρ̇ij = 0 and so λmax (ρ̇ 2 ) converges to zero as t → ∞.
 F. Mehdifar, C.P. Bechlioulis, F. Hashemzadeh et al. / Automatica 119 (2020) 109086
Accordingly, whenever δ is zero or vanishing with time we have
φ → 0 as t → ∞. Therefore, based on the ISS property for
vanishing input φ , we have σ (t) → 0, which implies η(t) → 0,
and hence e(t) → 0 as t → ∞.
Remark 4. Note that (6a) is met when ∥qi (t) − qj (t)∥ → dij
as t → ∞, for i, j ∈ V ∗ . In general, eij (t) → 0, as t →
∞, (i, j) ∈ E ∗ is a necessary condition for (6a) and it is equiv-
alent to (6a) only when Amb(F ∗ ) is an empty set (i.e., when
F ∗ is a complete graph, also known as global rigidity). Recall
that Amb(F ∗ ) is the set of undesired shapes with correct dis-
tance constraints between neighboring agents. Notice that, since
F ∗ is assumed to be minimally rigid, Amb(F ∗ ) is not neces-
sarily an empty set. Hence, e(t) → 0 implies either F (t) →
Iso(F ∗ {
) or F (t) → Amb(F ∗ ). Now consider the} following set
ΩF = e ∈ Rl | dist(q, Iso(F ∗ )) < dist(q, Amb(F ∗ )) . Since stabil-
ity of e = 0 is ensured in Theorem 1, in order to guarantee (6a),
the initial formation shape should be closer to Iso(F ∗ ). Thus, con-
verging to the correct formation can be achieved by considering
e(0) ∈ ΩI ∩ ΩF in Theorem 1. This condition for convergence to
the right shape is an inherent issue in distance-based formation
control problems and is a common requirement among various
related works such as Cai and Queiroz (2015), Krick et al. (2009),
Ramazani et al. (2017), Sun et al. (2016, 2015).
Remark 5. Notice that in the conventional approaches (e.g., Cai
and de Queiroz (2014, 2015), Cai and Queiroz (2015), Krick et al.
(2009), Sun et al. (2016, 2015)) when external disturbances affect
the agents’ dynamics as in (3), the condition e(0) ∈ ΩI ∩ ΩF ,
discussed in Remark 4, is not sufficient to secure convergence
to the correct shape (formation), since unpredictable transient
behavior in agents’ motion may lead the actual formation closer
to an undesired shape (F (t) → Amb(F ∗ )). On the other hand,
in our approach the proposed control scheme prevents such phe-
nomena since it establishes a predetermined transient response.
Moreover, according to Algorithm 1, it is clear that one can limit
the initial bounds of eij (i.e. eij (0) and eij (0)) to further increase
the robustness against formation distortions as much as possible
by decreasing µ , µij (formation robustness constants) and/or µ.
ij
Note that, decreasing µ results in more conservativeness for
choosing the proper bounds that ensure infinitesimally rigidity
of the actual formation graph.
Remark 6. The proposed control protocol (16) is in a sim-
ilar form with the conceptual schemes of Cai and de Queiroz
(2014), Cai and Queiroz (2015), Krick et al. (2009), Sun et al.
(2016, 2015), but it further encapsulates guaranteed transient and
steady state performance with increased robustness against any
bounded external disturbances and ensures connectivity mainte-
nance and collision avoidance among neighboring agents. Notice
that, The control
∑ law in (16) can be expressed for each agent
as: ui = − j∈N (E ∗ ) kij gij (ηij ) q̃ij with gij (ηij ) = ξij σij , which
i
is decentralized since each agent is required to sense only the
relative position with respect to its neighbors via on-board sen-
sors. Hence, the control law is independent of a global coordinate
frame and does not require agents’ local coordinate systems to
be aligned (this can be proved rigorously similarly to Sun et al.
(2015)).
4.5. Extension to distance-based formation centroid maneuvering
In this case, a single agent will be considered as the leader
that is injected with an external reference velocity command,
while the rest should follow the leader and maintain the for-
mation shape such that the centroid of the formation moves
with a desired time-varying reference velocity. Thus, consider
7
n disturbance-free single integrator agents in an m-dimensional
space modeled by:
q̇i = ui , i = 1, . . . , n (21)
Let vd (t) ∈ R be the bounded time-varying desired velocity
m
of the formation’s centroid, which is known only to ∑the leader.
n
The formation’s centroid is given by: qc (t) = (1/n) i=1 qi (t) =
(1/n)(1Tn ⊗ Im )q(t). Now, the problem is to design a proper control
protocol that ensures (6) as well as q̇c (t) = vd (t). Let the proper
formation maneuvering controller be designed as:
um = −RT ξ K σ + nML vd (t), (22)
where ML ∈ R mn×m
denotes the pinning matrix to the leader
that has direct access to the desired centroid velocity vd (i.e., if
agent i is the leader, then the ith block of ML is Im ). Invoking
the boundedness of nML vd (t) instantly leads us to infer that δ (t)
in Theorem 1 is equivalent to nML vd (t) in this case. Hence, the
achieved results in Theorem 1 (satisfaction of (6b)) may be easily
extended as follows (detailed analysis are available in Mehdifar,
Bechlioulis, Hashemzadeh, and Baradarannia (2019)).
Corollary 1. For the agents’ dynamics (21), the control law (22)
guarantees q̇c (t) = vd (t) and (6), which solves the prescribed
performance distance-based formation centroid maneuvering con-
trol problem with connectivity maintenance and collision avoidance
among neighboring agents.
Remark 7. The control law (22) can be expressed component
wise for each agent exactly the same as in Remark 6 except
for the leader that has an extra term nvd (t). The structure of
the leader’s controller indicates that in addition to the desired
centroid velocity, the leader should also know the total number of
agents in the formation (for the rigorous analysis of this reasoning
see Rozenheck et al. (2015)). Moreover, since the leader is the
only agent that has access to the desired centroid velocity, vd is
indeed available w.r.t. to the leader’s local coordinate system. Due
to these facts, all the points stated in Remark 6 hold for (22) as
well. Therefore, the control law (22) is decentralized and also it
is independent of a global coordinate frame.
5. Simulation results
5.1. Formation acquisition and robustness to distortions and unde-
sired shapes 2
Consider a group of five agents modeled by (3) in a two dimen-
sional space. Assume that the desired formation is a pentagon de-
fined by a minimally and infinitesimally rigid graph with edge set
E ∗ = {(1, 2), (1, 3), (1, 4), (1, 5), (2, 3), (3, 4), (4, 5)}. The desired
distances between neighboring agents (desired edges lengths) in
the
√ rigid framework are assumed to be √ d12 = d23 = d34 = d45 =
2(1 − cos(2π/5)) and d13 = d14 = 2(1 + cos(π/5)). The ini-
tial positions of the agents are given by q1 (0) = [−0.8049, 0.6951],
q2 (0) = [−1.3941, −0.1340], q3 (0) = [−0.4940, −0.7153],
q4 (0) = [1.5028, 0.1060], q5 (0) = [1.8808, 1.2388] indicating
that F (0) is infinitesimally rigid and e(0) ∈ ΩI ∩ ΩF (see
Remark 4). Moreover, in the simulations we assumed that all
agents have the same geometrical and sensing radius, i.e., rsi =
0.2 and rci = 5 for i = 1, . . . , 4. In Algorithm 1, we selected
µ = 0.12, and µij = µij = 0.3, ∀(i, j) ∈ E ∗ . The parameters
in the performance function ρij (t) are considered as aij = 1,
ρij0 = 1, ρij∞ = 0.03 for all (i, j) ∈ E ∗ . Moreover, the controller
2 Extended simulation results (including 3D formations, etc.) are available
in Mehdifar et al. (2019). Moreover, an illustrative simulation video is available
at: https://youtu.be/eIxCKLcVnM8
8 F. Mehdifar, C.P. Bechlioulis, F. Hashemzadeh et al. / Automatica 119 (2020) 109086

gains are set to kij = 0.3, (i, j) ∈ E ∗ . Without loss of generality,

consider that only the second and third agents are subject to
external disturbances. To this end, assume δ1 (t) = δ4 (t) = δ5 (t) =
[0, 0], δ2 (t) = [0.6 sin(1.2π t) − 0.3 sin(0.6π t), 0.5 sin(π t)], and
δ3 (t) = [0.3 sin(0.6π t) − 0.6 sin(1.2π t), −0.5 sin(π t)]. Recall that
δ (t) = col(δi (t)). Now, we provide a comparative simulation
study to show that the proposed formation control law (16) can
effectively prevent formation distortions that lead to undesired
shapes when external disturbances affect the agents dynamics
(see Remark 5). First, notice that in the absence of external
disturbances, the conventional distance-based formation control
law u = −kRT η, (k > 0) as proposed in Cai and de Queiroz
(2014), may establish the desired formation. Alternatively, in the
presence of external disturbances, in order to provide reliable
comparisons, we introduce a decentralized and robustly modified
distance-based formation control law:
diag(D̂)kRT η
u = −kRT η − diag(D̂) tanh( )
ε (23)
˙ = W |kRT η| − θ D̂
D̂

where k, ε, θ are positive scalars and W = diag(wij ) ∈ Rl×l with

wij > 0, ∀(i, j) ∈ E ∗ , Moreover, |.| is the element-wise absolute
value operator for vectors. Notice that, D̂ = col(D̂i ) is an estimate
of the upper bounds of the agents’ external disturbances. One can
prove that the modified conventional control law (23) is capable
of stabilizing inter agent distance errors in a sufficiently small
neighborhood of e = 0 by utilizing the uniform ultimate bound- Fig. 3. Comparison of results using control protocols: (a) (23), and (b) (16) when
edness notion. In the simulations, it is assumed that k = 0.3 (the δ (t) is considered as the external disturbance of the agents. The dashed lines
same as kij ), ε = 0.01, θ = 0.01, and wij = 1.5 for (i, j) ∈ E ∗ . In represent the prescribed performance bounds.

what follows, the performance of both control laws (16) and (23)
are compared in two cases of disturbance level.
Case (I) In this simulation, the external disturbance vector
δ (t) is applied to the agents’ dynamics for both control schemes.
Figs. 3(a) and 3(b) show the agents’ trajectories towards the
desired formation after 20 s using (23) and the proposed control
protocol (16), respectively. Apparently, in this case both con-
trollers were able to ensure F (t) → Iso(F ∗ ) in the presence of
external disturbances. However, as Fig. 3(a) depicts, (23) cannot
guarantee the errors to remain within the Prescribed performance
Bounds (PPB) and as a result it is not as robust as (16) with
respect to formation distortions. This can be clearly understood
by noticing the evolution of e23 and ∥e∥ in Fig. 3(a).
Case (II) The external disturbance was increased up to 4δ (t).
As it is illustrated in Fig. 4, the control protocol (23) is not able
to ensure convergence to the desired shape, whereas it is still
able to drive the distance errors to a small neighborhood of
zero (see ∥e∥ evolution). As stated in Remark 5, this stems from
the fact that (23) is not able to guarantee predefined quality of
the transient response. Indeed, during the transient phase the
actual formation distorts massively, forcing it to get closer to
one of the shapes lying inside Amb(F ∗ ) before the controller
dominates the effects of external disturbances. On the other
hand, (16) not only ensures convergence to the correct shape but
also shows an outstanding desirable performance, dealing with
the high amplitude external disturbances that try to distort the
formation shape all the time. Notice that, such performance is
achieved without modifying the controller’s gains and is purely
prescribed by the performance functions in the controller design
procedure. Certainly, by increasing the control gains in (23), one Fig. 4. Comparison of results using control protocols: (a) (23), and (b) (16) when
may be able to improve the transient and steady state behavior of 4δ (t) is considered as the external disturbance of the agents.
the closed-loop system in order to avoid/reduce risk of F (t) →
Amb(F ∗ ). However, this approach inevitably leads to a tedious
trial and error procedure to achieve satisfactory performance. On and collision avoidance among neighboring agents even when
the contrary, in the proposed scheme the control gains selec-
agents are subject to high external disturbances, which cannot
tion is significantly simplified. It is also noteworthy to mention
that the control law (16) also ensures connectivity maintenance be guaranteed by a conventional robust control scheme.
 F. Mehdifar, C.P. Bechlioulis, F. Hashemzadeh et al. / Automatica 119 (2020) 109086
Fig. 5. Agents trajectories towards the target shape as well as tracking the
desired centroid velocity.
5.2. Formation maneuvering
Consider a group of five agents with single integrator dy-
namics (21) and the same desired formation as in the previous
subsection. We assume that agent 5 is the leader of the group
and thus has access to the desired velocity for the formation
centroid and knows the total number of agents in the group. Let
the desired velocity be vd = [sin(0.5t), cos(0.5t)] resulting in a
circular motion. The initial positions of the agents are considered
to be q1 (0) = [−0.3639, 0.6361], q2 (0) = [−1.7126, −0.4526],
q3 (0) = [0.4919, 0.2706], q4 (0) = [2.0789, −0.0179], q5 (0) =
[0.9100, 0.2679]. Moreover, for the formation maneuvering con-
trol law in (22) it is assumed that the controller gains are set
to kij = 0.2, (i, j) ∈ E ∗ . All other parameters (for the per-
formance bounds and Algorithm 1) are considered the same as
in the previous subsection. Fig. 5 shows the agents’ trajectories
towards the desired shape while the formation centroid tracks
the velocity vd (t) (the asteroid mark ∗ represents the centroid of
the formation). Fig. 6 shows the evolution of inter-agent distance
errors for each edge as well as their norm. From the evolution
of ∥e∥ it is clear that the agents can maneuver cooperatively
with the given time varying velocity of the formation centroid
without affecting the formation errors (formation maintenance).
Moreover, since the distance errors are kept within the PPB,
the proposed controller (22) is capable of ensuring connectivity
maintenance and collision avoidance of neighboring agents.
6. Conclusion
In this paper, we proposed a novel decentralized robust distance-
based formation control law with guaranteed performance for
single integrator agents affected by unknown external distur-
bances. Moreover, by imposing proper predefined performance
bounds in the design, we solved the problems of connectivity
maintenance and collision avoidance among neighboring agents.
Then, the results were extended to solve distance-based for-
mation maneuvering for nominal single integrator agents, in
which the centroid of the formation tracks a predefined time-
varying velocity. It was assumed that only one agent, i.e., the
leader of the group, had access to the desired centroid velocity.
Moreover, we showed that the designed controllers are indepen-
dent of a global coordinate system. Furthermore, we assumed
9
Fig. 6. Inter-agent distance errors of the pentagon formation for (i, j) ∈ E ∗ as
well as ∥e∥ using the formation maneuvering controller (22).
that the desired formation is defined by a minimally and in-
finitesimally rigid graph in a 2 or 3 dimensional space. We also
highlighted that by using the proposed control scheme, which en-
sures predetermined transient response, formation convergence
to undesired shapes are prevented despite the presence of exter-
nal disturbances. Future research efforts will be devoted towards
considering directed interactions among agents and extending
the results for agents with higher order and nonlinear dynamics.
Another interesting topic for future research would be to check
whether the proposed approach can be used directly or modified
accordingly for solving the problem of distance-based formation
control with neighboring agents’ distance mismatch as addressed
in De Marina et al. (2014), Mou et al. (2015).
References
Anderson, B. D., Sun, Z., Sugie, T., Azuma, S.-i., & Sakurama, K. (2017). Formation
shape control with distance and area constraints. IFAC Journal of Systems and
Control, 1, 2–12.
Anderson, B. D., Yu, C., Hendrickx, J. M., et al. (2008). Rigid graph control
architectures for autonomous formations. IEEE Control Systems, 28(6).
Asimow, L., & Roth, B. (1979). The rigidity of graphs, ii. Journal of Mathematical
Analysis and Applications, 68(1), 171–190.
Babazadeh, R., & Selmic, R. (2019). Distance-based multi-agent formation control
with energy constraints using sdre. IEEE Transactions on Aerospace and
Electronic Systems.
Bae, Y.-B., Lim, Y.-H., Kang, S.-M., & Ahn, H.-S. (2018). Disturbance attenuation
in distance-based formation control: A linear matrix inequality approach. In
2018 IEEE conference on control technology and applications (pp. 1609–1614).
IEEE.
Barogh, S. A., & Werner, H. (2017). Cooperative source seeking with distance-
based formation control and single-integrator agents. IFAC-PapersOnLine,
50(1), 7911–7916.
Bechlioulis, Charalampos P, Demetriou, Michael A, & Kyriakopoulos, Kostas J
(2018). A distributed control and parameter estimation protocol with
prescribed performance for homogeneous lagrangian multi-agent systems.
Autonomous Robots, 42(8), 1525–1541.
Bechlioulis, Charalampos P, & Kyriakopoulos, Kostas J (2014). Robust model-free
formation control with prescribed performance and connectivity mainte-
nance for nonlinear multi-agent systems. In 53rd IEEE Conference on Decision
and Control (pp. 4509–4514). IEEE.
Bechlioulis, C. P., & Rovithakis, G. A. (2008). Robust adaptive control of feedback
linearizable mimo nonlinear systems with prescribed performance. IEEE
Transactions on Automatic Control, 53(9), 2090–2099.
Bechlioulis, C. P., & Rovithakis, G. A. (2016). Decentralized robust synchronization
of unknown high order nonlinear multi-agent systems with prescribed
transient and steady state performance. IEEE Transactions on Automatic
Control, 62(1), 123–134.
Cai, X., & de Queiroz, M. (2014). Rigidity-based stabilization of multi-agent
formations. Journal of Dynamic Systems, Measurement, and Control, 136(1),
014502.
10 F. Mehdifar, C.P. Bechlioulis, F. Hashemzadeh et al. / Automatica 119 (2020) 109086
Cai, X., & de Queiroz, M. (2015). Adaptive rigidity-based formation control for
multirobotic vehicles with dynamics. IEEE Transactions on Control Systems
Technology, 23(1), 389–396.
Cai, X., & Queiroz, M. (2015). Formation maneuvering and target interception
for multi-agent systems via rigid graphs. Asian Journal of Control, 17(4),
1174–1186.
Cao, Y., Yu, W., Ren, W., & Chen, G. (2013). An overview of recent progress in the
study of distributed multi-agent coordination. IEEE Transactions on Industrial
Informatics, 9(1), 427–438.
Chen, Z., & Huang, J. (2015). Stabilization and regulation of nonlinear systems.
Springer.
De Marina, H. G., Cao, M., & Jayawardhana, B. (2014). Controlling rigid formations
of mobile agents under inconsistent measurements. IEEE Transactions on
Robotics, 31(1), 31–39.
Deghat, M., Anderson, B. D., & Lin, Z. (2015). Combined flocking and
distance-based shape control of multi-agent formations. IEEE Transactions on
Automatic Control, 61(7), 1824–1837.
Dorfler, F., & Francis, B. (2010). Geometric analysis of the formation problem
for autonomous robots. IEEE Transactions on Automatic Control, 55(10),
2379–2384.
Dorri, A., Kanhere, S. S., & Jurdak, R. (2018). Multi-agent systems: A survey. IEEE
Access, 6, 28573–28593.
Hendrickson, B. (1992). Conditions for unique graph realizations. SIAM Journal
on Computing, 21(1), 65–84.
Karayiannidis, Yiannis, Dimarogonas, Dimos V, & Kragic, Danica (2012). Multi-
agent average consensus control with prescribed performance guarantees.
In Decision and Control (CDC), 2012 IEEE 51st Annual Conference on (pp.
2219–2225). IEEE.
Khaledyan, M., Liu, T., Fernandez-Kim, V., & de Queiroz, M. (2019). Flocking and
target interception control for formations of nonholonomic kinematic agents.
IEEE Transactions on Control Systems Technology.
Krick, L., Broucke, M. E., & Francis, B. A. (2009). Stabilisation of infinitesimally
rigid formations of multi-robot networks. International Journal of Control,
82(3), 423–439.
Liu, Tairan, de Queiroz, Marcio, Zhang, Pengpeng, & Khaledyan, Milad (2019).
Further results on the distance and area control of planar formations.
International Journal of Control, 1–17.
Macellari, Luca, Karayiannidis, Yiannis, & Dimarogonas, Dimos V (2017). Multi-
agent second order average consensus with prescribed transient behavior.
IEEE Transactions on Automatic Control, 62(10), 5282–5288.
Mehdifar, F., Bechlioulis, C. P., Hashemzadeh, F., & Baradarannia, M. (2019).
Prescribed performance distance-based formation control of multi-agent
systems (extended version). arXiv:1911.07266.
Mehdifar, F., Hashemzadeh, F., & Baradarannia, M. (2017). On distance-based
formation control: A nonlinear mapping approach and its applications. In
2017 IEEE 4th international conference on knowledge-based engineering and
innovation (pp. 0741–0746). IEEE.
Mehdifar, F., Hashemzadeh, F., Baradarannia, M., & de Queiroz, M. (2019). Finite-
time rigidity-based formation maneuvering of multiagent systems using
distributed finite-time velocity estimators. IEEE Transactions on Cybernetics,
49(12), 4473–4484.
Meng, Z., Anderson, B. D., & Hirche, S. (2016). Formation control with
mismatched compasses. Automatica, 69, 232–241.
Mou, S., Belabbas, M.-A., Morse, A. S., Sun, Z., & Anderson, B. D. (2015).
Undirected rigid formations are problematic. IEEE Transactions on Automatic
Control, 61(10), 2821–2836.
Oh, K.-K., & Ahn, H.-S. (2011). Formation control of mobile agents based on
inter-agent distance dynamics. Automatica, 47(10), 2306–2312.
Oh, K.-K., Park, M.-C., & Ahn, H.-S. (2015). A survey of multi-agent formation
control. Automatica, 53, 424–440.
Ramazani, S., Selmic, R., & de Queiroz, M. (2017). Rigidity-based multiagent
layered formation control. IEEE Transactions on Cybernetics.
Rozenheck, O., Zhao, S., & Zelazo, D. (2015). A proportional-integral controller for
distance-based formation tracking. In Control conference (ECC), 2015 European
(pp. 1693–1698). IEEE.
Sontag, E. D. (1998). Mathematical control theory: Deterministic finite dimensional
systems. New York, Inc.: Springer-Verlag.
Sun, Z., Liu, Q., Huang, N., Yu, C., & Anderson, B. D. O. (2019). Cooperative
event-based rigid formation control. IEEE Transactions on Systems, Man, and
Cybernetics: Systems, 1–13.
Sun, Z., Garcia de Marina, H., Anderson, B. D., & Cao, M. (2018). Quantiza-
tion effects and convergence properties of rigid formation control systems
with quantized distance measurements. International Journal of Robust and
Nonlinear Control, 28(16), 4865–4884.
Sun, Z., Mou, S., Anderson, B. D., & Cao, M. (2016). Exponential stability for
formation control systems with generalized controllers: A unified approach.
Systems & Control Letters, 93, 50–57.
Sun, Z., Mou, S., Deghat, M., & Anderson, B. (2015). Finite time dis-
tributed distance-constrained shape stabilization and flocking control for
d-dimensional undirected rigid formations. International Journal of Robust and
Nonlinear Control.
Sun, Z., Park, M.-C., Anderson, B. D., & Ahn, H.-S. (2017). Distributed stabilization
control of rigid formations with prescribed orientation. Automatica, 78,
250–257.
Verginis, Christos K, Nikou, Alexandros, & Dimarogonas, Dimos V (2019). Ro-
bust formation control in se (3) for tree-graph structures with prescribed
transient and steady state performance. Automatica, 103, 538–548.
Wang, L., & Guo, Q. (2018). Distance-based formation stabilization and flock-
ing control for distributed multi-agent systems. In 2018 IEEE international
conference on mechatronics and automation (pp. 1580–1585). IEEE.
Wang, L., Xi, J., Yuan, M., & Liu, G. (2018). Guaranteed-performance time-
varying formation control for swarm systems subjected to communication
constraints. IEEE Access, 6, 45384–45393.
Wang, X., Zeng, Z., & Cong, Y. (2016). Multi-agent distributed coordination
control: Developments and directions via graph viewpoint. Neurocomputing,
199, 204–218.
Yang, Q., Cao, M., de Marina, H. G., Fang, H., & Chen, J. (2018). Distributed
formation tracking using local coordinate systems. Systems & Control Letters,
111, 70–78.
Yang, Q., Sun, Z., Cao, M., Fang, H., & Chen, J. (2019). Stress-matrix-based
formation scaling control. Automatica, 101, 120–127.
Farhad Mehdifar was born in Tabriz, Iran, in 1992.
He received the B.Sc. and M.Sc. degrees in electrical
engineering (control systems) from the University of
Tabriz, Tabriz, Iran, in 2015 and 2018, respectively. He
is currently pursuing his Ph.D. at UCLouvain, Belgium.
His research interests include cooperative control of
multi-agent systems, formation control, nonlinear con-
trol theory, autonomous robots, networked control, and
data driven control/optimization.
Charalampos P. Bechlioulis was born in Arta, Greece,
in 1983. He is currently a postdoctoral researcher in the
Control Systems Laboratory at the School of Mechanical
Engineering of the National Technical University of
Athens. He received a diploma in electrical and com-
puter engineering in 2006 (first in his class), a bachelor
of science in mathematics in 2011 (second in his class)
and a Ph.D. in electrical and computer engineering
in 2011, all from the Aristotle University of Thes-
saloniki, Thessaloniki, Greece. His research interests
include nonlinear control with prescribed performance,
system identification, control of robotic vehicles, multi-agent systems and object
grasping. He has authored more than 80 papers in scientific journals and
conference proceedings and 3 book chapters.
Farzad Hashemzadeh was born in Maku, Iran, in 1981.
He received the B.Sc. degree in Biomedical Engineering
from the Amirkabir University of Technology, Tehran,
Iran, in 2003, the M.Sc. degree in Control Engineer-
ing from the University of Tehran, Tehran, in 2006,
and the Ph.D. degree in Control Engineering from the
University of Tabriz, Tabriz, Iran in 2012. In 2012,
he joined the Department of Electrical and Computer
Engineering, University of Tabriz and he is currently
an Associate professor. His current research interests
include teleoperation, network control and multi-agent
systems.
Mahdi Baradarannia was born in Tabriz, Iran, in 1982.
He received his B.Sc. and M.Sc. degrees in Electrical
Engineering from the Faculty of Electrical and Com-
puter Engineering, University of Tabriz, Iran, in 2005
and 2007, respectively, where he also received the
Ph.D. degree in Electrical Engineering in 2012. He is
now an Associate Professor in the Faculty of Electrical
and Computer Engineering at University of Tabriz. His
research interests currently involve the analysis and
control of nonlinear and optimal systems and their
applications in robotics, multiagent and cooperative
systems and active structures.