You are on page 1of 6

8th IFAC Symposium on Mechatronic Systems

8th IFAC Symposium on Mechatronic Systems


Vienna,
8th IFACAustria, Sept. on
Symposium 4-6, 2019
Vienna, Austria, Sept. 4-6,Mechatronic
2019 Systems
Available
8th IFACAustria,
Vienna, Symposium
Sept. on
4-6,Mechatronic
2019 Systems online at www.sciencedirect.com
Vienna, Austria, Sept. 4-6, 2019
ScienceDirect
IFAC PapersOnLine 52-15 (2019) 127–132
Formation
Formation control
control with
with collision
collision avoidance
avoidance
Formation
for control
first-order with
multi-agentcollision avoidance
systems:
Formation
for control multi-agent
first-order with collision avoidance
systems:
for first-order multi-agent
Experimental resultssystems:
for first-order multi-agent
Experimental systems:
Experimental results results
Experimental results
J. F. Flores-Resendiz ∗∗ J. Meza-Herrera ∗∗
∗∗
J. F. Flores-Resendiz ∗ J. Meza-Herrera ∗∗ ∗∗
E.
E. Aranda-Bricaire
J. F. Flores-Resendiz Aranda-Bricaire J. Meza-Herrera
∗∗
∗ ∗∗
J. F. Flores-Resendiz
E. Aranda-Bricaire J. Meza-Herrera
∗∗
∗∗

∗ Facultad
E. Aranda-Bricaire
Facultad de de Ingenierı́a
Ingenierı́a yy Negocios,
Negocios, Tecate,
Tecate, Universidad
∗ Baja California, Mexico (e-mail:francisco.flores32@uabc.edu.mx)
Universidad Autónoma
Autónoma de de

Facultad
Baja de Ingenierı́a
California, Mexico y Negocios, Tecate, Universidad
(e-mail:francisco.flores32@uabc.edu.mx) Autónoma de
Facultad
∗∗ de
∗∗ Electrical Ingenierı́a
Engineeringy Negocios, Tecate,
Department, Universidad Autónoma de
Baja
Baja∗∗
California,
Electrical
California, Engineering
Mexico Department, Mechatronics
Mechatronics Section,
Mexico (e-mail:francisco.flores32@uabc.edu.mx)
(e-mail:francisco.flores32@uabc.edu.mx) Section,
CINVESTAV,
Electrical Mexico
CINVESTAV, EngineeringCity, Mexico
Mexico (e-mail:jmezah01@gmail.com,
City, Department, Mechatronics Section,
∗∗
Electrical Mexico
CINVESTAV, Engineering
Mexico City, Department,
Mexico
earanda@cinvestav.mx)
(e-mail:jmezah01@gmail.com,
Mechatronics Section,
(e-mail:jmezah01@gmail.com,
CINVESTAV, Mexicoearanda@cinvestav.mx)
City, Mexico (e-mail:jmezah01@gmail.com,
earanda@cinvestav.mx)
earanda@cinvestav.mx)
Abstract: This
Abstract: This paper
paper deals
deals with
with the the general
general formation
formation controlcontrol problem
problem for for first
first order
order multi-
multi-
agent systems.
Abstract:
agent systems. This We
We investigate
paper deals
investigate the
with
the behaviour
the general
behaviour of
of an
an arbitrary
formation
arbitrary number
control
number of
problem
of mobile
mobile for agents
first
agents moving
order
moving on aa
multi-
on
Abstract:
plane
agent with
systems. This We paper
constrained deals velocity
input
investigate with
the thewhen
general
behaviour they
of formation
an are required
arbitrary control
to
number problem
reach
of a foragents
specific
mobile first order
spatial
moving multi-
pattern.
on a
plane
agent with constrained
systems. input velocity
We investigate when they
the behaviour of anare required
arbitrary to reach
number a specific
of mobile spatial
agents movingpattern.
on a
A
A general
plane with
general communication
constrained input graph among
velocity when agents
they is
are considered,
required to relaxing
reach a conditions
specific to
spatial the only
pattern.
plane
A with communication
requirement
general constrained
of containing
communication input graph
a directed
graph
among
velocity
among when agents
spanning
agents
is considered,
theytree.
are required
The
is considered, torelaxing
solution reachto
relaxing athisconditions
specific
problem
conditions
to the
spatial
is
to
only
pattern.
provided
the only
requirement
A general two of containing
communication a directed
graph among spanning
agents tree. The
is considered, solution to
relaxing this problem
conditions is
to provided
the only
attending
requirement
attending two of issues.
containingFirst, a the
directedasymptotic
spanning convergence
tree. The to the
solution desired
to this formation
problem ispattern
provided is
requirement
ensured.
attending Then
two of issues.
containing
a
issues.
First,a directed
non-collision
First,
the asymptotic
the analysis spanning
asymptotic is
convergence
tree. The
presented
convergence
to the desired
solution
regarding
to the the tomost
desired this formation
problem
general
formation ispattern
provided
geometrical
pattern
is
is
ensured.
attending Then
two can a non-collision
issues. theanalysis is presented regarding the most formation
general geometrical
ensured.
scenario which
scenario
Then
which can leadFirst,
a non-collision
lead to
asymptotic
analysis
to aa collision
collision among
among
convergence
is presented
agents.
agents. This
This
to the
regarding
later
later issue
issue
desired
the ismost
is solved
solved
general
by using
by using pattern
geometrical
repulsive
repulsive
is
ensured.
vector
scenario Then
fields
which with a unstable
can non-collision
lead to focus
a analysis
behaviour.
collision among is Numerical
presented
agents. regarding
simulations
This later the
issueas ismost
well as
solved general
experimental
by usinggeometrical
results
repulsive
vector fields
scenario which with unstable focus behaviour. Numerical simulations as well as experimental results
are provided
vector
are provided
fields incan
with
in orderlead
to to
unstable
order to
a collision
illustrate
focus
illustrate theamong
behaviour.
the
agents.ofThis
effectiveness
Numerical
effectiveness of the later
proposed
simulations
the proposed
issue is solved
control
as control
well by using repulsive
law.
as experimental
law. results
vector fields with
are provided unstable
in order focus behaviour.
to illustrate Numerical
the effectiveness of simulations
the proposed as control
well as experimental
law. results
© 2019,
are IFAC (International
provided in order to Federationthe
illustrate of Automatic
effectiveness Control)
of theHosting
proposed by Elsevier
control Ltd.
law. All rights reserved.
Keywords: Mobile
Keywords: Mobile robots,
robots, Discontinuous
Discontinuous control, control, Formation
Formation control,
control, Collision
Collision avoidance,
avoidance,
First-order
Keywords:
First-order systems.
Mobile
systems. robots, Discontinuous control, Formation control, Collision avoidance,
Keywords: Mobile
First-order systems. robots, Discontinuous control, Formation control, Collision avoidance,
First-order systems.
1. INTRODUCTION
1. INTRODUCTION Martinez and
Martinez and Aranda-Bricaire
Aranda-Bricaire (2013), (2013), Flores-Resendiz
Flores-Resendiz and and
1. INTRODUCTION Aranda-Bricaire
Martinez and
Aranda-Bricaire (2014),
Aranda-Bricaire Flores-Resendiz
(2014), Flores-Resendiz (2013), et al. (2015)).
Flores-Resendiz
et al. (2015)).and As
and
As
1. INTRODUCTION Martinez
a result,
Aranda-Bricaireandundesired
no Aranda-Bricaire
(2014), equilibrium (2013),
Flores-Resendiz Flores-Resendiz
points etappear
al. and
(2015)). the
As
In the
the last
last decade,
decade, multi-agent
multi-agent systems systems (MAS) (MAS) has has been a result, no undesired
been Aranda-Bricaire equilibrium
(2014),equilibrium
Flores-Resendiz pointsetappear and the
al. (2015)). As
In agents
aagents avoid
result,avoid collisions
no undesired
collisions by
by staying points
staying at aa distance
at distance
appear greater
and the
greater
In
an the
area
an the last
of decade,
intensive
area of decade, multi-agent
research
intensivemulti-agent
research because systems
because of(MAS)
its wide
of its wide has been
range
range a result,
than
agents or no
equal
avoid undesired
to a equilibrium
predefined
collisions by bound.
staying points
at In
a appear
distance and
Flores-Resendiz the
greater
In
an arealast
of applications,
applications, such research
of intensive as surveillance, systems
surveillance,
because of(MAS)
seeking and
its and
widehas been agents
rescue,
range than oravoid equal collisions
to a predefined by staying bound. In
at In Flores-Resendiz
a distance greater
of such as seeking rescue, et al.
than
et al. or
(2015),
equalthe
(2015), the particular
to particular
a predefined casebound.
case of communication
of communication
Flores-Resendizgraphs
graphs
an
etc.,
of area
(Kim of and
applications, intensive
such research
Bang (2016),
as because
Lin
surveillance, et al.
al. of(2004),
seeking its and
wide range than or equal to a predefined bound. In Flores-Resendiz
Marshall
rescue,
etc., (Kim and Bang (2016), Lin et (2004), Marshall composed
et al.
composed (2015),exclusively
the
exclusively of
particular cyclic
of cyclic case paths
of
paths was considered.
communication
was considered. graphs
of applications,
et al.
al.
etc., (2004),
(Kim and such as
Mousavi
Bang etsurveillance,
al. (2010)).
(2016), (2010)).
Lin et al. seeking
Problems
(2004), and rescue, et
asMarshall
motion al. (2015), the particular casepaths
of communication graphs
et (2004), Mousavi et al. Problems as motion composed exclusively of cyclic was considered.
etc.,
et al.
without(Kim
(2004), and
collisionsBang
Mousavi have (2016),
et al. Lin
(2010)).
attracted etto al. (2004),
Problems
much asMarshall
motion
attention in In this
composed
In this paper,the
exclusively
paper,the problem
problem of is
cyclic
is solved
paths
solved in
was
in itsconsidered.
its full generality;
full generality;
without
et collisions
al. (2004), haveet attracted to much attention in namely, no assumption on the type of communication
order
without
order to be Mousavi
to collisions
be applied
applied have in
al.
real(2010)).
in attracted
real to Problems
scenarios.
scenarios. much Both,
Both,
ascollisions
motion
attention
collisionsin In this paper,the
namely, no assumption problemonis the solvedtypein of
its communication
full generality;
without
among
order collisions
agents
to be as well
applied have
asin attracted
with
real to much
environmental
scenarios. attention
obstacles
Both, in In
have
collisions graphthis ispaper,the
namely,
graph is no no
made,
made,
problem
assumption
other
other than on
than is the
solvedtype
containing
containing
in of
itsa communication
full generality;
spanning
a communication
spanning tree
tree
amongto
order agents
be as well asinwith
applied real environmental
scenarios. obstacles
Both, have namely,
collisions (Lafferriere assumption
et al. (2005), on
Ren theand type
Beardof (2008)). When
been
among regarded
agents (Ahmadi
as well as Barogh
with et al.
environmental (2015), Kostic
obstacles et al.
have graph is
(Lafferriere made,
et al. other
(2005), thanRen containing
and Beard a spanning
(2008)). tree
When
been regarded (Ahmadi Barogh et al. (2015), Kostic et al. graph is made, other in than
among agents
been regarded
(2010), Phan et as well(2017)).
(Ahmadi
et al. as with
Barogh environmental
et al. (2015), obstacles
Kostic et have the agents
al. (Lafferriere
the agents areal.not
et
are not(2005),
in Rencontaining
risk
risk ofand
of Bearda (2008)).
collision,
collision,
spanning
they
they move
move
tree
Whenon
on
(2010),
been Phan
regarded al. (2017)).Barogh et al. (2015), Kostic et al. (Lafferriere et al.not(2005), Renbounded
and Beard (2008)). When
(2010), Phan et(Ahmadi
al. (2017)). the plane
agents
the agents
plane at at
are some known
in risk of velocity
collision, they by
move using
on
In aa first
(2010),
In first approach,
Phan collisions can
et al. (2017)).
approach, collisions can be be predicted,
predicted, but but notnot the aresome
plane functions
saturation
the at not while
some
functions
known
in
knownrisk bounded
of collision,
bounded
in Flores-Resendiz
velocity
they
velocity
Flores-Resendiz
by
move
by
et al.
al.
using
on
using
(2015)
In a first
avoided, approach,
from initial collisions
conditions can be predicted,
(Hernandez-Martinez but not
and saturation
the plane at some while
known in bounded velocity et by (2015)
using
avoided, from initial conditions (Hernandez-Martinez and the agents
saturation
the agents move move
functionsat some
at somewhile constant
in
constant velocity
Flores-Resendiz because
velocity because et al. the use
(2015)
the use
In a firstfrom
avoided, approach,
Aranda-Bricaire (2011),
initial collisions
conditions can
Hernandez-Martinezbe predicted,
(Hernandez-Martinez but and
and Bricaire
Bricairenot saturation functions whilevectorin Flores-Resendiz et to
al.the
(2015)
Aranda-Bricaire (2011), Hernandez-Martinez and of normalized
the
of normalized
agents move attractive
at some
attractive constant
vector fields.
fields. Similarly
velocity because
Similarly to Flores-
use
Flores-
avoided,
(2012)). If from
If the
Aranda-Bricaire initial
the initial
initialconditions
conditions
(2011),conditions (Hernandez-Martinez
can be
Hernandez-Martinez be known,
known, and
or even
and Bricaireeven the agentsetmove at somethe constant velocity because
(2012)). can or Resendiz
of normalized
Resendiz al.attractive
et al. (2015),
(2015), vector
the collision
fields.
collision avoidance
Similarly
avoidance to the
problem use
Flores-
problem is
is
Aranda-Bricaire
selected,
(2012)). this
If the (2011),conditions
approach
initial Hernandez-Martinez
could be enough.
can be The
known,and
most Bricaire
orknown
even of normalized
Resendiz et al.attractive
(2015), vector
the fields.
collision Similarly
avoidance to Flores-
problem is
selected, this approach could be enough. The most known attended
attended et by
by al.using
using repulsive
repulsive vector fields
vector avoidance in
fields in such such a way
a way
(2012)).
selected, If
approach, the
this initialin conditions
approach
consists could
the use canRepulsive
be enough.
use of be The
known,most orknown
even Resendiz
Potential (2015), theacollision problem is
approach, consists in could
the of Repulsive Potential that every
attended
that everybyagent
agent
using regards
repulsive
regards scaled
a scaled unstable
vector unstable focus
fields in focussuch vector
avector
way
selected,
Functions
approach, this approach
(RPF)
consists in incombination
the be enough.
use with
of The most
Attractive
Repulsive known attended
Poten-
Potential by using repulsive vector fields in such a way
Functions (RPF) in combination with Attractive Poten- field
that centred
every
field centred at
agent the position
regards
at the position a of the
scaled
of rest of
unstable
the rest them.
of them. focus vector
approach,
tial Functions
Functions
Functions consists
(RPF) in in
(APF) the use of
(Dimarogonas
combination with Repulsive
and Potential
Kyriakopoulos
Attractive Poten- that every agent
tial (APF) (Dimarogonas and Kyriakopoulos field centred at theregards
position a ofscaled unstable
the rest of them. focus vector
Functions
(2006),
tial Do
Functions(RPF)
(2006)). in
(APF) combination
This approach
(Dimarogonas with
could Attractive
lead
and to high Poten-
Kyriakopoulos mag- This
field
This paper
centred
paper is
isatorganized
the position
organized as
as follows.
of the
follows. In
rest
In Section
of them.
Section II we
II we in-
in-
(2006),
tial Do (2006)).
Functions (APF) This(Dimarogonas
approach could lead
and to high mag- troduce some preliminary definitions and present
Kyriakopoulos an im-
(2006),
nitude Do (2006)).
control This
signals approach
while the could
distance lead to
between high mag-
agents This
troduce paper
some is organized
preliminary as follows.
definitions InandSection
present II we
an in-
im-
nitude control signals while the distance between agents
(2006), Do (2006)).
could become
nitude
could become
control This
arbitrarily
signals
arbitrarily
approach
whilesmall.
small. In
the In could
orderlead
distance
order to
to
to high
take
between
take into ac- This
mag-
agents
into ac-
portant
troduce
portant
paper
Lemma
some
Lemma
is preliminary
organized
which is
which is as follows.
useful
useful to
Inand
to prove
definitions prove Section
asymptotic
present
asymptotic
II an
wecon-
in-
im-
con-
nitude control
count become
could the physicalsignals
physical while
dimensions
arbitrarily the
small. of distance
of agents
In agents
order toandbetween
and agents
guarantee
takeguarantee troduce
portant
vergence. some
Lemma
In preliminary
which
Section is
III, we definitions
useful
into ac- vergence. In Section III, we state the problem and pro- to
state the and
prove present
asymptotic
problem andan im-
con-
pro-
count
could the
become dimensions
arbitrarily small. of In agents
order to takeguarantee
into ac- portant Lemma
pose aa solution
solution whichIII,
regarding is both,
useful to prove
convergence asymptotic
to the
theand con-
desired
collision-free
count the tracking,
physical in Mastellone
dimensions et al. (2008),
and Mondal vergence.
pose In Section
regarding we state
both, the problem
convergence to pro-
desired
collision-free
count tracking,
the physical in Mastellone
dimensions of et al. (2008),
agents and Mondal
guarantee vergence.
formation In
and Section
collision III, avoidance
we state the problem
problems. and
Numericalpro-
et al. (2017),
collision-free Rodrı́guez-Seda
tracking, in et
Mastellone al. (2014),
et al. a modification
(2008), Mondal pose a
formation solution regarding
and regarding both,
collision avoidance convergence
problems. to the desired
Numerical
et al. (2017),
collision-free Rodrı́guez-Seda
tracking, et
in Mastellone al. (2014), a
et al. (2008),modification
Mondal formationpose a solution
simulation and and real
real both,
time experiments
experiments convergence to
were carried
carriedthe desired
out and
and
to al.
et this(2017),
last approach
approach has
Rodrı́guez-Seda beenetproposed.
proposed.
al. (2014), a modification simulation collision
time avoidance problems.
were Numerical
out
to
et this last has been formation and collision avoidance problems. Numerical
to al.
this(2017), Rodrı́guez-Seda
last approach has beenetproposed.
al. (2014), a modification simulation the results
the results andare presented
are presented
real in Section IV.
time experiments
in Section IV. Finally,
were carried
Finally, in out
in Section
and
Section
Finally,
to this last
Finally, theapproach
the design of
design ofhas thebeen
the repulsive
proposed.
repulsive component as
component as been
been the simulation
V wewe and
summarize
results real
some
are presented timeconclusions
experiments
in Sectionabout were
about
IV. carried
this
Finally, in out
work. and
Section
modified
Finally, by
the using
design unstable
of the focus
repulsivestructures,
component (Hernandez-
as been V
the summarize
results some
are presented conclusions
in Sectionabout this
IV. Finally, work.
in Section
modified by using unstable focus structures, (Hernandez- V we summarize some conclusions this work.
Finally,
modifiedthe designunstable
by using of the repulsive component
focus structures, as been V we summarize some conclusions about this work.
(Hernandez-
modified by using unstable focus structures, (Hernandez-
2405-8963 © 2019, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Copyright © 2019 IFAC 408
Copyright
Peer review©under
2019 responsibility
IFAC 408Control.
of International Federation of Automatic
Copyright © 2019 IFAC 408
10.1016/j.ifacol.2019.11.662
Copyright © 2019 IFAC 408
2019 IFAC MECHATRONICS
128
Vienna, Austria, Sept. 4-6, 2019 J.F. Flores-Resendiz et al. / IFAC PapersOnLine 52-15 (2019) 127–132

2. PRELIMINARIES where ni is defined as above and cji = [hji , vji ]T ∈ C,


∀j ∈ Ni define the desired geometrical distribution of
A formation graph G = {V, E, C} can be used to describe agents. Moreover, inside a region around, the agent Ri
the communication among a group of agents. It consists of can sense and determine any other agent’s position. This
a set of vertices V = {R1 , . . . , Rn }, corresponding to each region is assumed to be a circle of radius d. The subset of
agent and a set of edges E = {(Rj Ri ) ∈ V × V, i �= j}, agents inside the detection region of robot Ri is denoted
which provide information about the information exchange as Mi (t) = {Rj ∈ N | �zi (t) − zj (t)� ≤ d}. The goal is
between a pair of agents. If (Rj Ri ) ∈ E, the agent Ri to design distributed control laws ui = α(zi , zi∗ , Mi ∪ Ni ),
receives information from Rj , but not necessarily vice i = 1, . . . , n such that:
versa. Finally, the set C = {cji ∈ R2 | (Rj Ri ) ∈ E, i �= j}
of constant vectors represents the relative desired position i) The agents reach a desired formation, that is,
of agent Ri with respect to Rj . For the edge (Rj Ri ), Rj lim (zi (t) − zi∗ (t)) = 0, i = 1, . . . , n, and
t→∞
is called the parent node and Ri is the child node and it ii) the agents avoid collisions in finite-time and, after
is said Rj is neighbour of Ri . The set of all neighbours of that, they remain at some distance greater than or
Ri is denoted as Ni . A directed tree is a directed graph in equal to a predefined bound d from each other, i.e.,
which every node has exactly one parent, except for one �zi (t) − zj (t)� ≥ d, ∀t ≥ t1 ≥ 0, i �= j.
single node called the root. A directed spanning tree of a
directed graph G is a directed tree involving every node A solution to this problem can be found attending the
in G. Associated to a formation graph G, the Laplacian two objectives separately. The convergence issue is treated
matrix is given by as in Flores-Resendiz and Aranda-Bricaire (2014) and
Flores-Resendiz et al. (2015), with a relaxation in the
L (G) = ∆ − Ad , (1) conditions imposed to the communication graph. The only
where ∆ is the in-degree matrix defined as ∆ = requirement is to contain a directed spanning tree. Once
diag{n1 , . . . , nn }, where ni ∈ R is the cardinality of Ni , this objective is ensured, a complementary control law to
i = 1, . . . , n, and Ad = [aij ] ∈ Rn×n is the adjacency avoid collisions is designed. The proposed control law to
matrix of G defined as  reach the desired formation is given by
1, if (Rj Ri ) ∈ E
aij = (2) γi = −µφ(z̃i ), i = 1, . . . , n, (5)
0, otherwise.
As mentioned before, we consider physical constraints in where z̃i = zi − zi∗
is the position error, µ > 0 ∈ R and
the input signals. φ(·) is a saturation function. Without loss of generality,
we assume φ(·) is parametrised by unity gain, this implies
Definition 1. Let φ : R → R be a saturation function, then that agents can move at a maximum velocity µ.
it is indexed by a parameter r > 0 in such a way that
Theorem 3. Consider a group of n first-order agents, de-
−r ≤ φ(x) ≤ r, ∀x ∈ R, scribed by (3) along with the desired positions (4) and
the control law (5). Consider also that the communication
xφ(x) > 0, ∀x �= 0. graph contains a directed spanning tree and �zi (0) −
zj (0)� ≥ d, ∀i �= j. Then, in the closed-loop system
Now, we recall a useful Lemma. (3)-(5) the agents converge asymptotically to the desired
Lemma 2. Consider the dynamical system ẋ = Ax, where formation.
T
x = [x1 , . . . , xn ] ∈ Rn and the matrix A ∈ Rn×n is
Hurwitz. Then, the saturated system ẋ = Aφ(x) with Proof. See the detailed proof in Flores-Resendiz and
T
φ(x) = [φ(x1 ), . . . , φ(xn )] ∈ Rn , is globally asymptoti- Aranda-Bricaire (2019).
cally stable.
In order to avoid collisions among agents while they reach
Proof. See the detailed proof in Flores-Resendiz and the desired configuration a complementary reactive control
Aranda-Bricaire (2019). law is needed. This new component was presented firstly
in Flores-Resendiz et al. (2015). A novelty in our current
3. CONTROL DESIGN approach is to find design parameters that minimize the
control input signal. This issue is solved by using repul-
Consider a group of N mobile agents on a plane. Each sive vector fields, based on the relative distance between
of this robots are numbered by R1 , . . . , Rn . The posi- any pair of agents. The relative position coordinates are
tion coordinates of agent Ri are specified by zi (t) = defined as
[xi (t), yi (t)]T ∈ R2 , i = 1, . . . , n. Every robot is modelled
as a single integrator, pij = xj − xi , (6)
żi = ui , i = 1, . . . , n, (3)
qij = yj − yi , (7)
T
where ui = [ui1 , ui2 ] ∈ R2 are the velocities along the X
and Y axes. We assume that robot Ri can determine the for i, j = 1, . . . , n, i �= j. The repulsive vector fields are
position of a specific set of neighbours, say Ni ⊂ N , at such that for robot Ri there exists an unstable counter-
any time instant. This set is defined by the corresponding clockwise focus, centred at the position of any other agent
communication graph G. The desired position zi∗ of robot in risk of collision. The general expression of the repulsive
Ri is defined according to its set of neighbours as vector fields is
n  
1   pij − qij
zi∗ = (zj + cji ), (4) βi = −ε δij , (8)
ni pij + qij
j∈Ni j=1,j�=i

409
2019 IFAC MECHATRONICS

Vienna, Austria, Sept. 4-6, 2019 J.F. Flores-Resendiz et al. / IFAC PapersOnLine 52-15 (2019) 127–132 129

Evaluating in the dynamics of the relative position vari-


ables
σab � ��
− µ [pab , qab ] (φ (z̃b ) − φ (z̃a )) + 2ε p2ab + qab
2

V̇ = .
2
(15)
Now, the first term inside the parenthesis depends on the
desired position of agents, then it is bounded by the worst
case. Regarding that functions φ(z̃i ) are parameterized
by unity gain and the trajectories are inside σab where
p2ab + qab
2
≤ d2 , we have
1
V̇ ≤ σab −2µd + 2εd2 .
� �
Fig. 1. P-Q Plane (16)
2
where ε > 0 and the functions δij depend on the distance Regarding σab < 0, it can be written σab = −|σab |
between Ri and Rj , in the following way
� V̇ ≤ −d|σab | (−µ + εd) . (17)
1, if �zi − zj � ≤ d, Thus, the parameter ε should be selected in such a way
δij = (9)
0, if �zi − zj � > d, that ε > 2 µd . Assume ε = 2k µd with k > 1, then
where d is the minimum allowed distance between any pair
of agents. If the sensed area is the same for all agents, it is V̇ ≤ −dµ(2k − 1)|σab |, (18)
clear that δij = δji , i �= j. In Figure ??, an auxiliary plane which can be written as
1
Pij − Qij is shown. Outside the circumference of radius d, V̇ < −c|σab | < −cV 2 , (19)
only attractive vector fields exists (δij = 0) while in the
for some c > 0 and |σab | > 0. The last expression,
inner region the repulsive vector fields appear (δij = 1).
according to Bhat and Bernstein (2000), implies not only
Then, the composite control law for each agent is
that the surface σab is attractive from inside but also that
ui = γi + βi , i = 1, . . . , n, (10) the surface is reached in finite-time. Once the conflict be-
where γi and βi are given by (5) and (8), respectively. tween the agents has been solved, the direct application of
Theorem 4. Consider the system (3) and the control law Theorem 1 ensures the agents reach the desired formation
(10) along with definitions (4)-(9). Suppose at the time asymptotically. This concludes the proof.
instant t0 there exists a pair of agents in risk of collision, Remark 5. The generalization of the previous Theorem for
i. e. �zi (t0 ) − zj (t0 )� < d, and ε > 2 µd . Then, in the closed- any number of agents could be obtained by adding agents
loop system (3)-(10) the mobile robots avoid the collision in risk of collision with an arbitrary agent Rr .
in finite-time, t1 > t0 end they reach their desired position
asymptotically while staying at a distance greater than or 4. SIMULATION AND EXPERIMENTAL RESULTS
equal to d for all t ≥ t1 ≥ t0 .
In this Section, we present the results of a numerical
Sketch of proof For the sake of space, here we start
simulation as well as its experimental validation. The
by proving that no collision occur for the most simple
control law previously presented needs to be modified for
case which consist in two agents in risk of collision,
the case of this unicycle-type mobile robots. The kinematic
while the rest of them are not involved. Under this same
model for each robot Ri , according to Figure 2, is given
approach, more complex interactions among agents could
by
be treated. Suppose the agents Ra and Rb get closer
than the minimum allowed distance, that is, their relative
position trajectory gets into the circumference of radius d,
where �[pab , qab ]� ≤ d. Then, the relative position is inside
the region
σab = p2ab + qab2
− d2 ≤ 0 (11)
and it is of interest because in the boundary of this region
the control law becomes discontinuous. Once the repulsive
vector fields appear, we need to verify that the surface
σab = 0 is attractive from the inner region which means
the conflict is solved. The relative position dynamics in the
closed-loop system (3)-(10) is Fig. 2. Kinematic model of the uniclycle-type mobile

ṗab
� �
pab − qab
� robot.
= −µ (φ (z̃b ) − φ (z̃a )) + 2ε . (12)
q̇ab pab + qab   �
ẋi cos θi 0 � �

Therefore, in order to prove convergence to the discontinu-  ẏi  = sin θi 0 vi
, i = 1, . . . , n (20)
ous surface σab , we take the Lyapunov function candidate wi
θ̇i 0 1
1 2 where vi is the longitudinal velocity of the middle point of
V = σab , (13)
4 wheels axis of the i-th robot, wi its angular velocity and
which evaluated along the relative position trajectories θi the orientation with respect to the X axis. According
becomes � � to Brockett (1983), mobile agents modelled by (20) cannot
1 1 ṗab be stabilized by any continuous and time-invariant control
V̇ = σab σ̇ab = σab [ pab qab ] . (14)
2 2 q̇ab law . Then, to avoid singularities in the control law, we

410
2019 IFAC MECHATRONICS
130
Vienna, Austria, Sept. 4-6, 2019 J.F. Flores-Resendiz et al. / IFAC PapersOnLine 52-15 (2019) 127–132

study the kinematics of a point αi off the wheels axis,


Hernandez-Martinez and Aranda-Bricaire (2013). The co-
ordinates of point αi are given by
   
αxi xi + ℓ cos θi
αi = = , (21)
αyi yi + ℓ sin θi
and its dynamics is given by
   
α̇xi vi
α̇i = = Ai (θi ) (22)
α̇yi wi Fig. 3. AmigoBot Robots.
where  
cos θi −ℓ sin θi • A positioning system. The position and orientation
Ai (θi ) = (23)
sin θi ℓ cos θi of each robot is measured through a vision system
is the so-called decoupling matrix for each robot Ri . The composed of 12 cameras Flex 13 manufactured by
decoupling matrix is non-singular since det (Ai (θi )) = ℓ �= Natural Point (Figure 4) which are located at a height
T
0. By defining auxiliary control inputs ui = [uix , uiy ] we of 4 meters. These cameras have a resolution of 1280×
can establish a strategy for controlling the position of the 1024 pixels with at a frequency of 120 frames per
point αi by second. To detect an object, it must have a minimum
 
vi

uix
 of 3 markers and at least 3 cameras must locate the
−1
= Ai (θi ) (24) object within their range of vision.
wi uiy
where A−1 i (θi ) is the inverse of the decoupling matrix. The
closed-loop system (22)-(24) becomes
α̇i = ui (25)
The desired position of robot Ri , related to the coordinates
αi , is given by
1 
α∗i = (αj + cji ) (26)
ni Fig. 4. Cameras Flex 13.
j∈Ni
Then, a formation control with collision avoidance, similar • One Intel core i3-based computer. The software Mo-
to the presented previously is written as tive calculates the position of the centroid of the
ui = γ˜i + β˜i , i = 1, 2, . . . , n (27) geometric figure formed by the markers and its orien-
˜ tation. The control law is calculated in Visual C++
where γ˜i and βi are defined in a similar way than in (5)
using Aria libraries which are also used to communi-
and (8), respectively, but in terms of αi . Thus, the system
cate with the robots. The protocol VRPN is used to
is reduced to the case presented in previous Sections. It
share information between Motive and Visual C++.
is important to remark that (24) steers the coordinates
Finally, the velocities of each wheel are sent through
of the points αi to a desired position while the angles θi
Wi-Fi to the robots.
remain uncontrolled.

4.1 Experimental setup 4.2 Experiment results

The experiments were developed by using an experimental The experiment consists of four agents R1 , ..., R4 and the
platform composed of the following elements: goal is to reach the square formation shown in Figure 5
where √the relative position√
vectors are defined as c31 =
• Four differential-drive mobile robots, model AmigoBot
manufactured by MobileRobots Inc (Figure 3). Each
1
ℓ[ 2 , − 2 ] , c13 = ℓ[− 2 , 2 ] , c12 = ℓ[− 21 , 2√
3 T 1 3 T 1
3
], c32 =

one is furnished on their top with infrared markers ℓ[0, − √13 ]T , c43 = ℓ[ 21 , 23 ]T and c24 = ℓ[− 21 , − 2√
1 T
] . In
3
which form a geometric pattern such that the cen- this case, we take ℓ = 1.5 m, the velocity bound when no
troid of this figure coincides with the middle point collision risk exists is µ = 0.5 m/s, the minimum allowed
of the wheels’ axis of each robot for identification. distance is d = 0.65 m and ε = 2.5. The initial conditions
The parameters of AmigoBot robots are: wheel radius for the agents are [x10 , y10 ]T = [−1, 1]T , [x20 , y20 ]T =
r = 0.06 meters, length of wheels axis is L = 0.28 me- [1, −1]T , [x30 , y30 ]T = [−1, −1]T and [x40 , y40 ]T = [1, 1]T .
ters, distance to the front point of the robot ℓ = 0.15 Figure 5 illustrate the desired formation for this experi-
meters, 8 sonar sensors to avoid collisions or locate ment as well as the communication links among agents.
obstacles and the maximum longitudinal velocity of 1 The results of numerical simulation are shown in Figures
m/s. They feature wireless serial ethernet for remote 6 and 7 while the experimental results are depicted in
operations, two position encoders and built-in veloc- Figures 8 and 9. The trajectories followed by agents are
ity controllers. The workspace measures are 3.6 × 4.8 shown in Figures 6 and 8 while Figures 7 and 9 show the
meters. The linear and angular velocities vi and wi distances between any pair of agents. As can be seen,
obtained from the control law developed in this paper numerical simulation and real-time experiment exhibit
are transformed into linear velocities of the right and some differences. The most relevant reason to explain this
left wheel vri , vℓi through the isomorphism: differences is that the theoretical results are valid for first-
    
vri 1 2 L vi order systems, while the real robots are modelled by sec-
= , i = 1, ..., n. (28)
vℓi 2 2 −L wi ond order differential equations. Despite these differences,

411
2019 IFAC MECHATRONICS

Vienna, Austria, Sept. 4-6, 2019 J.F. Flores-Resendiz et al. / IFAC PapersOnLine 52-15 (2019) 127–132 131

the multi-agent system converge to the desired formation


1.5
and the agents exhibit a behaviour closely related to the
simulation. The minimum distance between any pair of 1
robots is always around the predefined bound. Finally, in
0.5

R1

Y[m]
0 R2
R3
R4
Initial position
−0.5 Final position

−1

−1.5
−1.5 −1 −0.5 0 0.5 1 1.5
X [m]

Fig. 5. Desired formation.


Fig. 8. Real-time experiment: Trajectories of the four
agents at the plane.
1.5

3
1
2.5

αi − αj  [m]
0.5 2

1.5
R1
Y[m]

0 R2
R3 1
R4
−0.5 Initial position
0.5
Final position

0
−1 0 10 20 30 40 50
time [s]

−1.5
−1.5 −1 −0.5 0
X [m]
0.5 1 1.5 Fig. 9. Real-time experiment: Distances among any pair of
agents.
Fig. 6. Simulation result: Trajectories of the four agents at solve the issue but also it is done in finite-time. Simulation
the plane. and real time experiment were performed to verify the
effectiveness of the proposed algorithm. As future work,
the design of formation control strategies with no collisions
3
for second order agents is considered.
2.5

REFERENCES
zi − zj  [m]

1.5
Ahmadi Barogh, S., Rosero, E., and Werner, H. (2015).
Formation control of non-holonomic agents with colli-
1 sion avoidance. Proceedings of the American Control
0.5
Conference, 2015-July, 757–762. doi:10.1109/ACC.2015.
0 5 10 15
time [s]
20 25 30
7170825.
Bhat, S. and Bernstein, D. (2000). Finite-time stability
Fig. 7. Simulation result: Distances among any pair of of continuous autonomous systems. SIAM Journal on
agents. Control and Optimization, 38(3), 751–766. doi:10.1137/
S0363012997321358. URL https://doi.org/10.1137/
Figure 10 are shown the positions of robots at some specific S0363012997321358.
time instants in order to illustrate the convergence to the Brockett, R.W. (1983). Asymptotic stability and feedback
desired formation. stabilization. In R.S. Millman, R.W. Brockett, and
H.J. Sussmann (eds.), Differential Geometric Control
5. CONCLUSIONS Theory, 181 – 191. Birkhauser, Boston, Massachusetts,
USA.
A general solution for the formation control problem Dimarogonas, D.V. and Kyriakopoulos, K.J. (2006).
without collisions has been presented. A control law is Distributed cooperative control and collision avoid-
designed for each of the objectives, formation control and ance for multiple kinematic agents. Proceed-
collision avoidance. A saturated error-based control law is ings of the 45th IEEE Conference on Decision
used to achieve the desired formation, while a distance- and Control, 721–726. doi:10.1109/CDC.2006.376884.
based complement is designed for each agent to avoid URL http://ieeexplore.ieee.org/lpdocs/epic03/
collision with another agents. The convergence to the wrapper.htm?arnumber=4177320.
desired formation agents is proved to be asymptotically Do, K.D. (2006). Formation control of mobile agents using
stable. On the other hand, when two or more agents are in local potential functions. In 2006 American Control
risk of collision, the algorithm proposed is able not only to Conference, 6 pp.–. doi:10.1109/ACC.2006.1656537.

412
2019 IFAC MECHATRONICS
132
Vienna, Austria, Sept. 4-6, 2019 J.F. Flores-Resendiz et al. / IFAC PapersOnLine 52-15 (2019) 127–132

Robotic Systems, 9(4), 100. doi:10.5772/50722. URL


https://doi.org/10.5772/50722.
Jiang, Z.P., Lefeber, E., and Nijmeijer, H. (2001). Sat-
urated stabilization and tracking of a nonholonomic
mobile robot. Systems & Control Letters, 42(5),
327 – 332. doi:https://doi.org/10.1016/S0167-6911(00)
00104-3. URL http://www.sciencedirect.com/
science/article/pii/S0167691100001043.
Kim, Y. and Bang, H. (2016). Decentralized control
of multiple unmanned aircraft for target tracking and
obstacle avoidance. 2016 International Conference on
Unmanned Aircraft Systems, ICUAS 2016, 327–331.
doi:10.1109/ICUAS.2016.7502547.
Kostic, D., Adinandra, S., Caarls, J., and Nijmeijer, H.
(2010). Collision-free motion coordination of unicy-
cle multi-agent systems. In Proceedings of the 2010
American Control Conference (ACC 2010), 30 June 30
- 2 July 2010, Baltimore, Maryland, USA, 3186–3191.
Institute of Electrical and Electronics Engineers (IEEE),
United States. doi:10.1109/ACC.2010.5530852.
Lafferriere, G., Williams, A., JohnS.Caughman, I.S.C.,
and Veerman, J.J.P. (2005). Decentralized control of
vehicle formations. Systems & Control Letters, 54, 899–
910.
Lin, Z., Broucke, M., and Francis, B. (2004). Local con-
trol strategies for groups of mobile autonomous agents.
IEEE Transactions on Automatic Control, 49(4), 622–
629.
Marshall, J.A., Broucke, M.E., and Francis, B.A. (2004).
Formations of Vehicles in Cyclic Pursuit. IEEE Trans-
actions on Automatic Control, 49(11), 1963–1974.
Mastellone, S., Stipanović, D.M., Graunke, C.R.,
Intlekofer, K.A., and Spong, M.W. (2008). Formation
Fig. 10. Real-time experiment: Agents’ positions in the control and collision avoidance for multi-agent non-
plane at different times. holonomic systems: Theory and experiments. The
International Journal of Robotics Research, 27(1),
Flores-Resendiz, J.F. and Aranda-Bricaire, E. (2014). 107–126. doi:10.1177/0278364907084441. URL
Cyclic pursuit formation control without collisions in https://doi.org/10.1177/0278364907084441.
multi-agent systems using discontinuous vector fields. Mondal, A., Behera, L., Sahoo, S.R., and Shukla, A.
Congreso Latinoamericano de Control Automático, (2017). A novel multi-agent formation control law with
Cancún, México. collision avoidance. IEEE/CAA Journal of Automatica
Flores-Resendiz, J.F. and Aranda-Bricaire, E. (2019). Sinica, 4(3), 558–568. doi:10.1109/JAS.2017.7510565.
A general solution to the formation control problem Mousavi, M.S.R., Khaghani, M., and Vossoughi, G. (2010).
without collisions for first order multi-agent systems. Collision avoidance with obstacles in flocking for multi
ROBOTICA, Cambridge University Press. agent systems. 2010 International Conference on In-
Flores-Resendiz, J.F., Aranda-Bricaire, E., Gonzalez- dustrial Electronics, Control and Robotics, IECR 2010,
Sierra, J., and Santiaguillo-Salinas, J. (2015). Finite- 1–5. doi:10.1109/IECR.2010.5720122.
time formation control without collisions for multiagent Phan, D., Yang, J., Grosu, R., Smolka, S.A., and Stoller,
systems with communication graphs composed of cyclic S.D. (2017). Collision avoidance for mobile robots with
paths. Mathematical Problems in Engineering, 17 pp. limited sensing and limited information about moving
doi:10.1155/2015/948086. obstacles. Formal Methods in System Design, 51(1), 62–
Hernandez-Martinez, E.G. and Aranda-Bricaire, E. 86. doi:10.1007/s10703-016-0265-4.
(2011). Convergence and collision avoidance in forma- Ren, W. and Beard, R. (2008). Distributed consensus
tion control: A survey of the artificial potential functions in multi-vehicle cooperative control: theory and applica-
approach. In Multi-agent Systems- Modeling, Control, tions. Communications and Control engineering series.
Programming, Simulations and Applications. INTECH, Springer-Verlag, London.
Austria, 105–126. Rodrı́guez-Seda, E.J., Tang, C., Spong, M.W., and Sti-
Hernandez-Martinez, E.G. and Aranda-Bricaire, E. panović, D.M. (2014). Trajectory tracking with collision
(2013). Collision avoidance in formation control using avoidance for nonholonomic vehicles with acceleration
discontinuous vector fields. 9th IFAC Symposium on constraints and limited sensing. The International Jour-
Nonlinear Control Systems. Toulouse, France. nal of Robotics Research, 33(12), 1569–1592. doi:10.
Hernandez-Martinez, E. and Bricaire, E.A. (2012). Non- 1177/0278364914537130. URL https://doi.org/10.
collision conditions in multi-agent virtual leader-based 1177/0278364914537130.
formation control. International Journal of Advanced

413

You might also like