You are on page 1of 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/330123069

3D Robots Formation cooperative Control Using EFDs

Conference Paper · October 2018


DOI: 10.1109/PAIS.2018.8598535

CITATIONS READS
0 31

5 authors, including:

Allam Ahmed M. Tadjine


National Polytechnic School of Algiers National Polytechnic School of Algiers
9 PUBLICATIONS   7 CITATIONS    239 PUBLICATIONS   1,895 CITATIONS   

SEE PROFILE SEE PROFILE

Elhaouari Kobzili Abdelghani Boucheloukh


National Polytechnic School of Algiers National Polytechnic School of Algiers
9 PUBLICATIONS   5 CITATIONS    8 PUBLICATIONS   11 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Hybrid Fuzzy Sliding Mode Control of a DFIG Integrated into the Network View project

Hybrid vehicle View project

All content following this page was uploaded by Allam Ahmed on 02 February 2019.

The user has requested enhancement of the downloaded file.


3D Robots Formation cooperative Control Using EFDs
ALLAM Ahmed (1), TADJINE Mohamed (2), KOBZILI Elhaouari (4) NEMRA Abdelkrim (3)
BOUCHELOUKH Abdelghani(5)

National Polytechnic School (1),(2),(4),(5) Military Polytechnic School, EMP (3)
El-Harrach Algiers, Algeria Bordj El-Bahri, Algiers, Algeria
ahmedallam900@gmail.com (1), tadjine@yahoo.fr (2) karim_nemra@yahoo.fr (3)
elhaouari.kobzili@g.enp.edu.dz (4), abdelghanibouch@gmail.com(5)

leader position and the geometric structure model. The virtual


Abstract—We study in this work the formation cooperative leader is considered independent of the robots and has a
control of 3D holonomic robots group utilizing parametric and predefined trajectory shared by all the robots. Therefore,
implicit properties of the desired 3D planar curve (pattern there is no feedback to the formation.
shape). This latter is considered as a closed contour and
generated based on Elliptical Fourier Descriptors (EFDs). The The behavioral approaches consider the robot as an
robots control consists of two stages. Firstly, robots move electrical charge subjected to several forces, where those
towards the desired closed contour using its Implicit virtual forces are reflecting desired behaviors (i.e. go to goal,
Polynomial function (IP). The IP is deduced from the EFDs and obstacle avoidance, inter-robots spacing), and are in function
considered as a potential function. The realized formation of the distance to goal position, neighboring robots
shape is considered as a virtual Structure modeled by EFDs, coordinates and the obstacles position. Therefore, the
where a desired relative position is assigned to each robot with resultant total force defines the robots behavior (control
respect to the desired contour frame. The contour center is inputs).
considered as the virtual leader that has a reference path along
to a predefined target. Secondly, a controller is designed to Formation control using elliptical Fourier descriptors
allow robots tracking their reference path and eventually the EFDs was first used for 2D robots formation control [6,7].
time-varying desired configuration. The robots coordinate each The EFDs are used in the field of computer vision and image
to other by introducing virtual forces (linear springs) exercised processing to represent free-form shapes in digital images in
from each robot to its nearest neighbors. The proposed a static fashion. The idea proposed in [6] was to describe a
approach is validated in simulation using different robots desired 2D curve using EFDs, then using both the EFDs and
group sizes showing its feasibility and effectiveness. the implicit function of the desired curve as a potential
Keywords— Robots formation control; Coordination; Elliptic
function to design the robots generation control, while in [7],
Fourier Descriptors EFDs; Virtual structure approach; Implicit the latter work was extended to 2D non-holonomic robots
Polynomial functions. control. In [8], a new formation control based on dynamical
EFDs was used for switching between two 2D robots
I. INTRODUCTION formation shape.
Recently, a lot of attention has been pointed onto the In this paper, we introduce an extension of the approach
cooperative control of multi-robots system from the scientific presented in[8], where we propose a formation cooperative
community (robotic field) [1]. One of the most important part controller for a group of 3D holonomic robots that uses
of formation cooperative control, is to deal with controlling a parametric (based EFDs) and implicit representation of the
group of interacting robots to move while preserving some desired formation shape (planar configuration) combined
given pattern, following a given path and respecting some with virtual structure approach. The designed controller aims
constraints. The cooperative control consider also the in a decentralized way generating 3D planar free-form
switching between two formation shapes or reconfigurations, formation shape, then the robots formation must reach a
the flexibility and the formation preserving during the predefined target and eventually switch gradually to new
navigation task including the obstacles avoidance. time-varying formation shape. During this process, the robots
Different approaches of formation control can be coordinate among themselves by keeping a desired distance
classified according to different criteria. Based on [2], these between each other and ensure an appropriate dispatching in
approaches can be classified into three groups: leader- formation pattern.
follower [3], behavioral [4] and virtual structure [5]. The next section will be reserved for introducing the
In leader follower approach, a group of robots are concept of the EFDs tool and the implicit function of a 3D
considered to follow a predefined leader robot. This method closed parameterized curve. In the third section, we will
presents some drawbacks (i.e. error accumulation form the explain the formation control design, and then we showcase
leader to the followers, robots pattern deformation when some simulation results, conclusion and perspectives.
passing through obstacles). However, in the virtual structure II. PARAMETRIC AND IMPLICIT REPRESENTATION OF 3D
approach, the whole robots system will be considered as a CLOSED CURVE
virtual rigid vehicle, that follows a predefined path. This
latter is calculated based on the knowledge of the virtual A. Elliptical furriers descriptors EFDs
The EFDs were initially introduced in 1982[9], in which
an alternative representation of Fourier descriptors using

978-1-5386-4238-2/18/$31.00 ©2018 IEEE


elliptical description of the curves was presented. Thus, the B. Implicit polynomial function (IP) of a closed 3D curve
EFDs model uses planar free-form curves as a sum of 3 Curves can also be represented using implicit
ellipses. EFDs are rotation, scale and translation invariant. polynomials function IP as ( , , ) = 0.
Most of the application areas of EFDs are related to biology,
and anatomy [10,11]. The EFDs can be used to represent and The IP function of a 3D planar curve is used to define an
to model any 2D closed curve [6],[9]. Similarly to [6], we algebraic distance between any point in space and the curve,
propose to model an arbitrary 3 curve using EFDs as: and can be used to find whether a point is inside or outside
the planar curve. This IP function can be found by
( )= + ∑ ( cos( ) + sin ( )) implicitization of the EFDs. In this study, the IP function of a
( ) = + ∑ ( cos( ) + sin ( )) (1) 2D closed planar curve is obtained using the method detailed
( ) = + ∑ ( cos( ) + sin ( )) in [10], similarly, the description of a 3D parameterized
closed curve with an IP function is given by,
In the equation above, k is an index and n is a positive
number representing the harmonics number used to model ( , , )=∑ = 0. (4)
the closed curve. Where ( ), ( ) and ( ), the
coordinates of the points on the curve are written as functions Where are coefficients, = 2 is the degree of the
of a normalized parameter defined as: polynomial function, as detailed in [10].
Any 3D planar curve defined by a set of 3D points with
= ×2 . (2) coordinates , and could be modeled by calculating its
Clearly, ∈ [0, 2 ] for ∈ [0, ], where the length of EFDs [7,9],
the 3D closed curve is . Note that (1) is periodic with period = ∑ ( ) ; = ∑ ( ) ; = ∑ ( ).
.
Where , and are curve center coordinates, and the = ∑ cos ( ) ; = ∑ sin ( ).
vectors A, B, C, D, E, and D of dimension are the 3D curve
descriptors. The accuracy or the level of detail in the = ∑ cos ( ) ; = ∑ sin ( ). (5)
characterization of the representation increases with the
harmonics number used to describe the 3D curve. On the = ∑ cos ( ) ; = ∑ sin ( ).
other side, with more harmonics the computation time Where M is the number of the sample points taken from
increases. the 3D curve to be modeled. Based on the calculated EFDs,
It’s important to note that the curve shape generated by we can then calculate its IP function (4).
(1) can be either a planar curve or a non-planar curve. By In Fig. 2, based on 20 coordinate points that describe
using an arbitrary EFDs values, we usually obtain a non- geometrically a 3D planar curve in blue color, the EFDs are
planar curve. To get a free-form 3 planar curve, at first we estimated using (5) with a selected number of harmonics of
need describe our desired curve in 2 frame using 2 = 2. The estimated EFDs are = 61.9, = 66.3, =
version of EFDs [6], then based on a predefined matrix 20.1, = [7.2 − 0.1], =[−19.3 3.1], = [17.2 0.4],
transformation , we describe this latter on the 3 frame (3). =[1.8 0.1], = [1.1 0] and = [0.4 0.1].

Fig. 1. Constructing 3D parametrized planar curve.

( ) ( )
( ) ( )
= × . (3)
̃( ) 0
1 1 Fig. 2. EFDs estimation based on a set of points coordinates.
Where ( ), ( ) and ̃ ( ) are the 3D planar curve
The estimated EFDs are used to reconstruct the 3D curve
points coordinates. The parameters of this latter 3D curve ,
in red color using (1). The IP function of the modeled curve
, , , , , , and are deduced from (3) by
above is too long to be carried in the paper.
identification with (1).
The EFDs is a very interesting mathematical tool to form III. FORMATION CONTROL STRATEGY
with high flexibility and to model any virtual planar 3D In this work, we consider the robot as a 3D mass point
curve, this latter could be considered as a geometrical particle to rend the formation control problem more
configuration for a group of robots. simplified. This latter assumption allows eliminating the
complexity related to the robot dynamics and its non-
holonomic constraints (i.e. UAVs) from the analysis, and
enables studying the efficiency and the feasibility of the The first component _ of (9) will be formulated based
proposed formation control approach. We first present the on the framework of [6] by using the IP of the desired
robot kinematic model and then we detail the formation formation shape (desired planar curve). This first component
control strategy. control input will ensures only the robot convergence to a
A. Robot Kinematic model virtual 3D curve which is parallel to the desired 3D curve as
it is shown in Fig. 3.
We opted for a 3D holonomic robot that is modeled as
point particle. The kinematic model for the robot in the
formation is given:
= ,.
,
= . (7)
= , .
Where , and are the velocities of the robot
in the x, y and z axis, respectively, with respect to the world
,
coordinate frame, and =[ , , , , ] is its
corresponding control input.
B. Formation control strategy
We assume that a 3D holonomics robots group moves in
a 3D space, where each robot position is given by
= [ ] . Initially, the robots are randomly
positioned in the defined space. Fig. 3. Sketch of the robots formation generation control.

The proposed robots formation control strategy consists The IP functions are adapted to find if a point is on a
of two stages. The first is to form a desired 3D planar curve or not. In the other side, the IP functions could be used
formation shape (geometric configuration), once the first to measure the distance between a point and a closed curve
stage is achieved, the second stage consists of controlling the [6]. The position error function between the robot and the
robots formation to move to predefined point (target) while 3D virtual planar curve is given by the algebraic distance to
keeping the achieved formation and eventually switching to a the curve using the implicit equation of the desired formation
new desired formation shape. During the two stages, robots shape as,
are expected to avoid collision between each other and to
keep certain coordination in the formation. Then, the input = ( , , ). (10)
control for the robot will be obtained by the sum of Based on (10), if the robot is approximately on the curve,
the control components designed for, the formation the error is zero, while it is negative when the robot is inside
generation or deployment control, coordination control and the curve and positive when outside. The control is designed
target reaching and formation switching control respectively. to force this error to decrease exponentially based on
, , , Lyaponov theory, i.e.
, + + _
, , , , =− . (11)
= = + + _ . (8)
, , , , Where is a positive number. Using (12) into the
+ + _ equation above yields
1) Formation generation control
We can define a 3D desired configuration for a group of ( , , )=− ( , , ). (12)
3D robots by many ways depending on the task to be Using chain rule of differentiation and substituting (7) in
realized. It could be the perimeter of a friend UAV or an (12), we find,
enemy target to be encircled or supervised or the boundary of
a considered geographical area. This boundary (desired 3D ,
_
curve) can be defined based on an extracted and processed ,
images provided by 3D robots onboard camera. The desired ( ) _ =− ( , , ). (13)
3D formation shape (3D curve) is in practice a sequence of ,
_
points, we use (5) to compute its EFDs and the method
detailed in[10] to compute its IP function. Based on Laponov theory, the optimal parallel generation
formation control _ for the robot can be determined
The generation formation control is composed of two using pseudo inverse, namely
components, parallel input _ and normal input _
respectively. ,
_ ( , , )
, , , ,
_ _ _ =− ( , , ) ( , , ) . (14)
, , , ∇ , ,
= = _ + _ . (9) ,
_
( , , )
, , ,
_ _
Where , and are the partial derivatives of Where is an adaptable spring constant. The desired
( , , ) with respect to x, y and z, ∇ = . coordination stiffness could be obtained by tuning .
Therefore, may take relatively grand values if the robots
The second generation control component _ is are far away from the desired configuration, while it takes
formulated so to converge the robot normally to the plane small values when the robots are reaching the desired
containing the desired 3D planar curve, see Fig. 3. In doing configuration.
so, the control input needs to be directed to the opposite We have p and q the indices for the robots that are the
direction of the surface’s normal unitary vector , while its
magnitude will be a function of the normal distance between nearest two neighbors of robot. and are the
the robot and the plane containing the desired curve. current distances of the robot from the and robots
respectively. ( ) and ( ) represent the
,
_ x, y and z position coordinates of the robots p and q with
, respect to the world coordinate frame.
_ = − 2 . (15)
, 3) Target reaching and switching control
_
Once the robots have achieved the desired formation
shape using (9) during the first stage, the control input for the
In (15), R is the rotation matrix from the coordinates
second stage will enable the robots formation reaching a
frame associated to the plane containing the 3D desired curve predefined point or target and eventually switching gradually
to the world coordinates frame, and is a positive number. to a new formation shape at any moment to deal with changes
We consider as the normal distance between the in the environment (i.e. facing an obstacle) or to respond to
robot and the plane containing the desired 3D curve. some mission requirements. This will be done by using
The control parameters and could be tuned in sort formation’s parametric representation (EFDs vectors).
satisfying non-holonomic constraints and/or to optimize a Based on the achieved robots position in the 3D curve, a
certain criteria (i.e. minimizing consumed energy), which is relative angle is assigned to each one,
not within the scope of this work. Thus, and since we opted
for a 3D holonome robots, we can take arbitrary positive ( )2
= (17)
values of the latter parameters to simulate and to check the
feasibility of the proposed formation controller.
Where ∈ [1, ], a robot index and M is the number of
2) Coordination control the points defining the desired 3D curve.
The robots are supposed to coordinate each to other in the
navigation task. Therefore, the robots must avoid collision
among themselves and ensuring a given desired spacing
between each robot and its neighbors, which enables also
having a certain distribution of the robots group in the
achieved formation shape. The coordination controller is
modeled as the sum of virtual forces of linear springs
exercised from each robot to its two nearest neighbors as in
Fig. 5. Robots assignement process.
Fig. 4[6].
Now, for a time-varying desired curve, we can use the
parametric representation EFDs to describe the robot’s
desired position,

( )= ( )+∑ ( ( ) cos( )+ ( )sin ( ))

( )= ( )+∑ ( ( ) cos( )+ ( )sin ( ))

( )= ( )+∑ ( ( )cos( )+ ( )sin ( ))
(18)
The robots formation will be considered as virtual
Fig. 4. Modeling of coordination control. structure described by its EFDs, and the virtual leader of the
The spring is proposed to have a normal length which is robots formation will be considered as the center of the actual
equal to the desired distance . The produced force by the (achieved) formation shape [ ] and has a predefined
spring is linearly proportional to the difference between the path along to target, while the formation structure (3D
actual distance and the desired distance between the robots. desired curve) is time-varying along the leader path by using
The spring produces a force directed from from the robot the EFDs. The controller for this stage will be designed in
to its neighbor as below: order to keep the robots on the desired time-varying curve by
,
tracking the formation trajectory as shown in Fig.6.
− −
,
= − − + ( − ) − . (16)
,
− −
correctly and optimally achieving the desired curve 3D curve
achieving by three robots using (9).

Fig. 6. Robots formation Control sketch.

The error enabling the robots tracking the desired 3D


curve while moving along a predefined path of the leader is
defined as the difference between the desired position and the
actual position of the robot as,

( )

_ = ( ) − (19)

( )
The controller will be formulated in sort to force the error
to decrease exponentially,

_ =− _ . (20) Robots initial positions

Where a positive constant, (20) can be written,

,

( ) _
∗ ,
_ = ( ) − _ (21) Fig. 7. 3D curve achieving by three robots using (9).

( ) ,
_ Fig. 8 shows the evolution of the different control inputs
By substituting (21) into (20), the control for stage-2 is acting on the robots and their corresponding IP function
given, variation when achieving the desired formation (first stage).
We see that the control input (9) is acting on robots in sort of
, decreasing their corresponding IP function, which confirms

_ ( ) the effectiveness of the control input based on Lyaponov
, ∗
_ = 3 _ + ( ) (22) theory. We can remark, that the coordination control
, ∗
( ) component is nil since the robots are too far each to other in
_ the achieved formation.
IV. SIMULATION RESULTS
The proposed approach was simulated in matlab, where
the simulation results could be obtained with any desired
EFDs and any number of robots. We validated the
implemented approach with two simulation scenarios. In the
first simulation, three robots are considered to check the
feasibility and efficiency of the proposed formation
generation control (stage-1). In the second simulation, eight
robots are employed to see the performances of the controller
in formation generation, coordination and target reaching
with switching to new formation shape.
The parameters used in the first simulations, d = 25 ,
= 1.4 and λ =1.1. The adaptable parameter used is,
2, ( , , ) <
= (23) Fig. 8. Control inputs acting on the three robots.
6, ( , , ) ≥
In the second simulation, the proposed formation control
Where = 15, is a limit algebraic distance. approach is applied to a group of eight robots. The trajectory
of the virtual leader is a straight line linking the initial robots
In Fig. 7, we can see a scenario when three robots placed
formation center to the target. The EFDs the of the first 3D
randomly in the space, where the desired curve is represented
desired curves in the second simulation is,
using EFDs with 2 harmonics and an IP function with a
degree of 4. The robots under the control input (9) are
= 85, = 85, = 40, = [16.2 − 0.89],
=[4.2 0.54], = [−2.76 − 3.3], =[9.9 − 2.1],
= [0 0] and = [0 0].
The IP function of the first desired 3D curve which has
been used in (9) is,
( , , )
= 10 − 5 ∗ 10 − 0.3 + 2 ∗ 10
+ 0.14 + 32.1 − 3.6 ∗ 10 − 0.04 Fig. 10. Control inputs acting on the three robots selected.
− 10.6 − 1.41 ∗ 10 + 5.2 ∗ 10 − 7.5 ∗ 10
+ 3.86 − 4.4 + 3073. V. CONCLUSION
The above equation doesn’t contain terms, because the We have presented a formation cooperative controller of
desired 3D planar curve (the red half-circle) is parallel to the 3D holonomic robots to form arbitrary desired 3D shapes
plane. (planar curves) and reaching a predefined target while
tracking time-varying configuration using EFDs and the
The virtual leader path and the desired time-varying implicit function of the formation shape. The proposed
EFDs to be tracked by the robots during the stage-2 is given, method introduces flexibility on the choice of desired shapes.
The obtained simulations results demonstrate the feasibility
( ) = 30 + , ( ) = 30 + , ( ) = 50 + . and effectiveness of the formation control.
( )= + /9, ( ) = + cos (2 /40), ( ) = , As perspectives, we think to extend the presented
( )= + , ( )= + 0.1 , and ( ) = + 4 . approach to surface based robots formation shape (i.e. sphere,
The parameters used in the second simulations are d = ellipsoid). Possible extension and implementation of this
5 , time step = 0.05 , = 0.8, λ =0.2 and λ =0.1. work with a group of Quad-copters is under scope taking in
consideration the robots dynamics and providing the whole
From Fig. 9, we see in the first stage, starting from theirs formation stability.
initial positions, the robots have achieved with success the
first desired formation shape (the red half-sphere) while VI. REFERENCE
keeping a good coordination using (9) and (16), then in the
second stage, the robots moved following the virtual leader to [1] R. Murray, “Recent Research in Cooperative Control of Multivehicle
Systems,” Journal of Dynamic Systems, Measurement, and Control,
the target, while switching and tracking gradually the time- Vol. 129, pp. 571-583, 2007.
varying desired formation shape (the blue triangle is the final [2] S. Garrido, L. Moreno, and P. U. Lima, “Robot formation motion
desired shape) using (22). The shown figures reveal the planning using fast marching”, Robot. Auton. Syst., vol. 59, no. 9, Sep
feasibility and the efficiency of the proposed control method 2011.
whatever the planar 3D desired formation shape chosen. [3] Zhao, Yudong, et al. "Leader-follower formation control for multiple
mobile robots by a designed sliding mode controller based on
kinematic control method." Society of Instrument and Control
Engineers of Japan (SICE), 56th Annual Conference of the. IEEE,
2017.
[4] LEE, G and CHWA, D. "Decentralized behavior-based formation
control of multiple robots considering obstacle avoidance". Intelligent
Service Robotics, vol. 11, no 1, p. 127-138, 2018.
[5] L. Chen, B.L. Ma, "A nonlinear formation control of wheeled mobile
robots with virtual structure approach", Proceedings of Chinese
Control Conference, pp. 1080-1085, 2015.
[6] Y. H. Esin, M. Unel, M. Yildiz, “Formation Control of Multiple
Robots Using Parametric and Implicit Representations,” Proceedings
of the ICIC’08, Lecture Notes in Computer Science, Springer-Verlag,
Berlin, pp. 558-565, 2008.
[7] Y. H. Esin M. Unel, M. Yildiz, “Formation control of non-holonomic
Robots initial positions
mobile robots using implicit polynomials and elliptic Fourier
Descriptors,” Turkish Journal of Electrical Engineering and Computer
Sciences 18: 765–780, 2010.
[8] E. Sanem and U. Mustafa. Planar formation control of swarm robots
using dynamical elliptic Fourier descriptors. Transactions of the
Institute of Measurement and Control, vol. 37, no 5, p. 661-671, 2015.
Fig. 9. Sketch of eight robots formation control. [9] F.P. Kuhl, C.R. Giardina, “Elliptic Fourier Features of a Closed
Contour,” Computer Graphics and Image Processing, vol. 18, pp.236-
Fig. 10 shows the evolutions of the different control 258, 1982.
inputs acting on three robots chosen arbitrary from the eight [10] H. Yalcin, M. Unel, W. Wolovich, “Implicitization of Parametric
robots group. We can notice some chattering of the Curves by Matrix Annihilation,” International Journal of Computer
coordination control input during stage-2, because the fact Vision vol. 54, no.1/2/3, pp.105-115, 2003.
that robots could be close each other when tracking the [11] O. Soldea, M. Unel and A. Ercil, “Recursive computation of moments
of 2D objects represented by elliptic Fourier descriptors”. Pattern
desired 3D configuration. Recognition Letters 31: 1428–1436, 2010.

View publication stats

You might also like