Sie sind auf Seite 1von 8

International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 9 (2016) pp 6583-6590

© Research India Publications. http://www.ripublication.com

Decentralized Non-Linear Control of Leader-Follower Formation of


Multiple Autonomous Mobile Robots

Azza El-Sayed Bayoumi Ibrahim


Department of Computers and Systems,
Electronics Research Institute, Giza, Egypt
E-mail: azza@eri.sci.eg

Abstract maintain the desired formation parameters (distance and


Control of the cooperative multi-robot system is one of the orientation) with the leader robot. The main critique of the
most challenging problems. This paper investigates a novel leader-follower approach is that the formation is depended on
implement of integral time varying sliding mode control the leader for achieving the goal and this over-reliance on a
(ITVSMC) on the leader-follower formation of nonholonomic single agent may be undesirable. In spite of these shortages,
mobile robots with bounded control inputs. The objective of the leader-follower architectures are particularly appreciated
the controller is to form up and maintain the follower robots in for their simplicity and scalability [9].
tracking the leader motion. The controller parameters are The formation of a team of autonomous mobile robots with
automatically computed using mathematical formulas. The nonholonomic constraints is considered as a nonlinear system.
annoying chattering problem in the control law is solved using Linearization techniques are used to deal with the
the boundary layer technique. The effectiveness of the nonlinearities. Although the nonlinear formation system can
suggested strategy is verified by simulating the system using be linearized, there exist many problems in practice. Research
MATLAB/SIMULINK software. Simulation results papers try to handle these problems as in [10,15, and 17], so
demonstrate that the designed control strategy success in using a nonlinear control method is preferred. Sliding Mode
getting suitable values of controller parameters resulting in Control (SMC) is a non-linear systematic control method. Its
perfect performance for different system conditions. Also, it is main advantage is the sliding motion exhibits complete
proved that the time-varying sliding surface technique is more robustness to system uncertainties [12, 13]. Some previous
accurate even in the presence of external disturbances than the works utilized SMC to design the followers controllers
conventional time-invariable sliding surface. Therefore, [11,14]. However, this control strategy has two main
ITVSMC technique is a good solution to the problem of drawbacks: the well-known chattering phenomenon and the
leader-followers trajectory tracking in the presence of sensitivity of the system motion to disturbances during the
disturbances. reaching phase. Thus, efforts have been made to minimize or
even remove the chattering and the reaching phase. A time
Keywords: Formation control; two-wheeled mobile robots; varying sliding mode control (TVSMC) method is suggested
Input constraints; time varying sliding mode control. to eliminate the reaching-phase [4]. In this method, the sliding
surface translates in the phase plane with constant slope
without rotating. The sliding surface should be chosen such
Introduction that the error along this surface vanishes.
In recent years, research on coordination control of multiple In this paper, the formation controller is designed based on
mobile robots system has attracted many researchers attention TVSMC method. The fundamental idea is that the time-
and shown its broad application prospect in the military, space varying sliding surface passes through the initial system states
exploration, medical, service industry and other fields [2]. at the start of the motion and then moves towards a
Cooperative/formation control appears because a group of predetermined time-invariant sliding surface with the form of
robots can accomplish a mission more efficiently over a single shifting and/or rotating [5]. Thus, the robustness of the
robot. Differential drive wheeled mobile robots (WMR) are trajectory for the nominal system can be guaranteed from the
usually employed in multi-robots applications, due to their initial time instant [4]. Therefore, TVSMC technique can be a
fast manoeuvring, low cost and simplicity [1]. very powerful solution to the problem of leader-follower
Different strategies have been introduced to control the robots trajectory tracking control in the presence of disturbances.
formation [3], the most popular approaches are: (i) The
behaviour-based approach [6], where each vehicle is
described by several desired behaviours and the final control Problem Formulation
is computed from a weighting of the each behaviour; (ii) The Given a group of mobile robots (WMRs), one robot is selected
virtual structure approach [7], where the entire agents are as a leader and the others are as followers. It is assumed that
treated as a single structure. The desired motion is assigned to the leader robot moves along a predefined collision-free
the virtual structure which trails trajectories for each member trajectory, the vector ( , ) can be measured and passed to
in the formation to follow; (iii) The leader-follower approach each follower. It is required to design a control law such that a
[8]. In this approach, one of the robots is selected as a leader, desired formation for the whole system can be achieved while
while other robots are followers. The leader moves along a tracking the motion of the leader. The objective of the
predesigned collision free trajectory and the followers controller of each follower is to find the values of its

6583
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 9 (2016) pp 6583-6590
© Research India Publications. http://www.ripublication.com

translational, and rotational velocities ( ) such that the


relative position errors to the leader tend to zero.

Leader-Follower Model
In an inertial reference frame, the configuration of the robot i Controller Design
is given by the position of the centre between its two wheels Based on the TVSMC method, the switching surface starts
(xi,yi) and the angle θi of its heading direction. The heading moving uniformly (i.e. with a constant velocity) in the state
direction is the line perpendicular to the axes of wheel space and then it stops at a time instant > , where is the
rotations, and it is positive in the direction of forward motion. initial time, the details of the method is well described by
The position of the robot is defined by qi = (xi,yi,θi ). The Andrzej Bartoszewicz in [5]. As shown in Fig2. Once the
kinematics of a robot "i" is described by Equation (1) and the surface reaches the origin of the error coordinate frame, it
nonholonomic constraint, where the driving wheels are stops moving and remains fixed and time-invariant.
assumed to roll purely and do not slip is expressed by
Equation (2), [16].

Where the translational velocity vi and the angular velocity ωi


are the input control signals.

Figure 2: Time Varying Sliding Surface phase portrait

A follower tracking errors and the errors rates are introduced


as:
, and
, and (7)
To develop the ITVSMC approach, the integral-type time
varying sliding surface is defined twice times as declared in
Figure 2, for time t ϵ { .

Figure 1: Sketch of leader-follower coordinated


(8)
Where, ci, Ai and Bi are constants. The selection of these
The appropriate modelling of leader-follower formation is constants will be addressed later.
derived in [11]. It is evident from Fig.1 that (xL,yL) represents Since the considered surface stops moving at the time , for
the leader’s position, and (xF,yF) for the follower’s position. any t ≥ it is fixed and can be described as follows:
The angle between leader’s and the x-axis is θL and the
angle between follower’s and the x-axis is θF. The relative
distance between the leader and the follower is represented by
and it is inclination with the x-axis by the angle .
Therefore, the formation parameters can be expressed by Eq.
(3). (9)
, (3) Therefore, two controllers are designed and switching
On the other hand, the relationships between velocities and between them is done at time or when the surfaces reach
angles are obtained by Equations (4), (5), and (6). the origin [4]. In the period of , the sliding surface is

6584
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 9 (2016) pp 6583-6590
© Research India Publications. http://www.ripublication.com

fixed and described by equations (9). The controller is Sliding Surface Parameters Design Subject to Input Signal
designed as in the conventional SMC [12], the variable s is Constraint
differentiated and equated to zero to get the equivalent control The constants As, Bs and Cs are chosen in such a way that the
signal on the sliding surface. representative point of the system at the initial time belongs to
(10) the switching surface and subject to a constraint on the input
signal, such that the maximum admissible values of the input
(11) signals are u1max and u2max. It means that the following
Since the desired formation parameters described by and inequalities should be held.
are chosen to be constants values, , and . and (17)
Then equations (4), (5), and (6) are substituting into the
Where, u1max, u2max are constants, which satisfies the following
resulting formulas, the control signals can be obtained with
conditions:
the knowledge that and . Therefore,
equations (12) and (13) express the final control signal form. and

(12) (18)
Which imply that there exist such strictly positive constants
(13) satisfy the following conditions.
To keep the system trajectory staying perfect on the sliding
surface despite the influences of the disturbances, the (19)
reachability condition is essential. Such a condition is satisfied
by using the switching function sgn (si ), is the switching
gain constant, should be chosen to compensate for the system (20)
disturbances, in which the upper limit of the disturbances is The optimal switching surface parameters are given by the
assumed to be known. The sgn (si ) function is always following formulas, considering are strictly positive
replaced by sat (si / i ) to avoid occurring of fast oscillation, constants, which represent the lower bound of the absolute
where i is a small positive value which expresses the value of control signal multiplier, therefore, the complete
boundary layer thickness, the mathematical description of it derivation is illustrated in [4].
is declared in Equation (14). , ,

(21)
(14)
Similarly, for the period , the second control law is
designed, but using the sliding surfaces described by equation
(8). Using the same procedure, the control signals can be
obtained as:
(22)
(15) Where and are the initial values of the linear and
the angular separation errors which depend on the initial
position of the follower and the leader vehicles. The switching
(16) surface stops moving at the time instants, , computed
According to Equations (15) and (16), the control signals
by Equations (21) and (22).
values are dependent on the values of the parameters of the
sliding surfaces, As, Bs and Cs and, ks. Figure 3 illustrates the
block diagram of the overall controlled system of the leader-
Simulation Results
follower approach.
The effectiveness of the proposed algorithm can be verified by
a numerical simulation using Matlab/Simulink software. The
reference desired trajectory is given by y = g(x) =
sin(0.5x)+0.5x+1, and x( ) = . Four robots are used to form
an example of analysis, one robot act as a leader and three act
as followers. For a case study, the initial conditions of the
leader robot (xL(0), yL(0), θL(0)) = (3,2,0), and for a follower
( , ) = (1.5,0.7,2 ) as in follower 3 in Table
[1]. The desired separation parameters are chosen as
and . The parameters of the designed controller
are computed according to the proposed criteria such that
, and and recorded in
Figure 3: The controlled system block diagram of one table [1]. After several running of the simulator, it is found
follower that the suitable value of the bounded layer thickness is equal

6585
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 9 (2016) pp 6583-6590
© Research India Publications. http://www.ripublication.com

to 0.2 of the corresponding feedback gains ( . Figure


(4-a) shows that the follower robot can accurately track the
leader trajectory with very low steady state errors, the
corresponding error signals are depicted in Figure (4-b) and
(4-c). There is no reaching phases and the sliding mode is
starting from the initial time as shown in Fig. 5. The control
signals are drawn in Figure 6.

Figure 5: The behaviours of the sliding surfaces s1 & s2 based


on ITVSMC

Figure 4: (a) The Leader-Follower trajectories based on


ITVSMC. (b) The linear separation error signals; (c) The
angular separation error signals

Figure 6: The control signals u1 & u2 based on ITVSMC

The proposed algorithm is also tested in controlling a group of


robots. Two different cases are suggested.

6586
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 9 (2016) pp 6583-6590
© Research India Publications. http://www.ripublication.com

In the first case, three followers are put in different initial performances are drawn in Fig.(7-b), As shown in the figure
positions, and it is required from each follower to track the the three followers track the leader correctly. The designed
leader with the same separation parameters. Simulation results controller for each follower is able to get precise values of its
are recorded in table [1], and the system responses are parameters, which verify the powerful of the proposed
depicted in Figure (7-a). Referring to the figure, it is evident method.
that all followers track the leader trajectory precisely, and the
proposed methodology success in determining the appropriate Table 2: System parameters with no disturbances for different
controllers’ parameters without trial and error method. Note desired positions, (Case2).
that the follower which starts far from the leader takes a
longer time to reach the desired situation. Case 2 Follower 1 Follower 2 Follower 3

Table 1: System parameters with no disturbances for different


followers initial positions, (Case1). Desired Parameters 3 4 5
0.5 0.6 0.9
Case 1 Follower 1 Follower 2 Follower 3 parameters of A1 -2 -3 -4
Controller (u1) B1 0.51 0.57 0.45
C1 0.71 0.63 0.47
Initial conditions 5 3.1 1.5 parameters of A2 0.2 0.1 -0.2
0.9 0.51 0.7 Controller (u2) B2 -0.14 -0.16 0.13
parameters of A1 2 0.1 -1.5 C2 1.19 1.97 1.14
Controller (u1) B1 0.6 -0.25 0.45 Switching gains K1 0.301 0.01 0.5
C1 0.78 2.7 0.77 K2 0.051 0.001 0.4
parameters of A2 0.4 0.01 0.2 Steady-state errors e1s.s -4.63e-9 -1.8e-7 -4.2e-5
Controller (u2) B2 -0.14 -0.15 -0.12 e2s.s 1.1e-15 0 -5.6e-10
C2 0.84 5.4 1.09
Switching gains K1 0.301 0.01 0.5
K2 0.051 0.001 0.4
Steady-state e1s.s 5.7e-10 0 -5e-11
errors e2s.s 4.2e-12 0 2.27e-16

Figure (7-b): Multi-followers behaviours for different desired


relative position.

Comparing TVSMC with Conventional SMC methods


The SMC controller is also simulated, and the same previous
Figure (7-a): Multi-followers behaviours for different initials case study is used to test it. The follower trajectory designed
positions. by the ITVSMC and the SMC is compared with the leader
trajectory and drawn in Fig. (8-a). In the case of the SMC
method the follower tracks the leader with steady state errors
For the second case, the induced controllers are also examined equal to e1s.s = 0.0315, and e2s.s = 9.82e-8. But for TVSMC,
by another situation of the multiple followers. All followers the error is avoided. The sliding surfaces are registered and
started from the same position and asked to follow the leader illustrated in Figure 9, which shows that the sliding surfaces
with different separation parameters. All results and based on SMC take a longer time to reach the sliding mode
parameters of this case are demonstrated in table2. The system than the TVSMC.

6587
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 9 (2016) pp 6583-6590
© Research India Publications. http://www.ripublication.com

To demonstrate the robustness properties of the ITVSM from


the beginning, Both controllers are tested by the same case
study, but the system is suffered from external additive
disturbances and the results are drawn in Figure (8-b). It is
evident that a bad vibration appears on SMC trajectory in a
period which makes the system out of control over this period,
but the controlled system under TVSMC can withstand the
disturbances more than the SMC.
Moreover, the ITVSMC algorithm calculates the controller
parameters by itself. On the other hand, the SMC parameters
are adjusted manually by trial and error method which is a
tedious method resulting in wasting the efforts and consuming
the time.
Finally, the designed controller based on the ITVSMC is
faster, more accurate, more robust than the one designed
based on the SMC.

Figure 8: (a) Comparison bet. SMC & ITVSMC with no-


disturbances. (b) The comparison with system disturbances

6588
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 9 (2016) pp 6583-6590
© Research India Publications. http://www.ripublication.com

Systems 13(1), 41-47, DOI: 10.7906/indecs.13.1.6.


2015.
[2] Wang Guanghua, Li Deyi, Gan Wenyan, Jia Peng,
"Study on Formation Control of Multi-Robot
Systems", Third International Conference on
Intelligent System Design and Engineering
Applications, 2013.
[3] A. Farinelli, L. Locchi,and D. Nardi, “Multirobot
systems: A classification focused on coordination,”
IEEE Transactions on Systems, Man, and
Cybernetics, Part B: Cybernetics, vol. 34, no. 5, pp.
2015-2028,October 2004.
[4] Andrzej Bartoszewicz, Aleksandra Nowacka-
Leverton, "Time-Varying Sliding Modes for Second
and Third Order Systems.",Springer-Verlag Berlin
Heidelberg. DOI 10.1007/978-3-540-88063-9, 2009.
[5] Q. JIN Yong, D. LIU Xiang, & H. ChaoZhen,
“Time-varying sliding mode control for a class of
uncertain MIMO nonlinear system subject to control
Figure 9: The behaviors of the sliding surfaces s1 & s2 based input constraint.”, January 2010 Vol. 53 No. 1: 89–
on the SMC & ITVSMC 100, 2010.
[6] Mu Xiaomin; Du Yang; Liu Xing; Wu Sentang,
2009, "Behavior based formation control of multi-
Conclusion missiles", Control and Decision Conference, 2009.
This paper introduces a framework for controlling a group of CCDC '09. Chinese.Pages: 5019 5023, DOI:
autonomous two-wheeled mobile robots based on leader- 10.1109/CCDC.2009.5194937.
follower approach control in the presence of environmental [7] Maithripala, D.H.S.; Berg, J.M.; Maithripala,
disturbances and control input constraint. A decentralized D.H.A.; Jayasuriya, S., "A geometric virtual structure
cooperative control for each follower has been implemented approach to decentralized formation control",
employing a time-varying sliding mode nonlinear controller. American Control Conference (ACC), P: 5736-5741,
Numerical simulations of the designed control strategy are DOI: 10.1109/ACC.2014.6859451, 2014.
done. The suggested approach is tested using different status [8] Mohamed A. Kamel, Youmin Zhang, “Decentralized
of the system. The results show that the follower robots can Leader-Follower Formation Control with Obstacle
accurately track the leader trajectory with very low steady Avoidance of Multiple Unicycle Mobile Robots”,
state errors even there exist an additive bounded external Proceeding of the IEEE 28th Canadian Conference on
disturbances. Electrical and Computer Engineering Halifax,
A comparison between the TVSMC and the conventional Canada, May 3-6, 2015.
SMC is carried out. It is proven that the time-varying sliding [9] Wang Guanghua, Li Deyi, Gan Wenyan, Jia Peng
surface can get faster error convergence and less steady state “Study on Formation Control of Multi-Robot
error than the time-invariable sliding surface. Also, it is Systems”, Third International Conference on
noticed that the system behaviors based on the TVSMC are Intelligent System Design and Engineering
more resistance to the external disturbances than the SMC. Applications, 2013.
Moreover, the time of converging to the equilibrium point is [10] Erfu Yang, Dongbing Gu, and Huosheng Hu
lower than the conventional sliding mode control method. In a “Nonsingular formation control of cooperative
case of SMC simulation, the parameters are changed to mobile robots via feedback linearization”, Intelligent
choose the best ones after a process of trial and error. Robots and Systems, 2005. (IROS 2005). 2005
However, this is not valid in the case of an actual system. But IEEE/RSJ International Conference on Aug.,
the ITVSMC algorithm develops accurate calculations for the Page(s): 826-831, DOI:10.1109/IROS.2005.1545537,
controller parameters. A linear relation between the feedback 2005.
parameter and the thickness of the boundary layer is suggested [11] Zhipeng Wang, Qinghe Wu, 2014 "Tracking and
to avoid occurring of the chattering. obstacle avoidance control of a second-order system
All these aspects make this approach highly attractive in with SMC", 2014 Sixth International Conference on
diverse application domains of mobile vehicles. Intelligent Human-Machine Systems and Cybernetics
978-1-4799-4955-7/14 © 2014 IEEE. DOI
10.1109/IHMSC.2014.47.
References [12] S. Spurgeon, C. Edwards “Sliding Mode Control:
Theory And Applications “,August 27, Reference-
[1] Birol Kocaturk, " Motion Control of Wheeled Mobile 237 Pages. ISBN 9780748406012, 1998.
Robots", Interdisciplinary Description of Complex [13] Parvat B. J, Jadhav V. K, Lokhande N. N, “Design
and Implementation of Sliding Mode Controller for

6589
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 9 (2016) pp 6583-6590
© Research India Publications. http://www.ripublication.com

Level Control “,IOSR Journal of Electronics and


Communication Engineering (IOSR-JECE) ISSN :
2278-2834, ISBN : 2278-8735, PP : 51-54.
www.iosrjournals.org.
[14] Guo Jinrong; Qian Dianwei, "Design and
performance analysis of integral sliding mode for
multi-robot formation control", Society of Instrument
and Control Engineers of Japan (SICE), 2015 54th
Annual Conference, pages 457-462, DOI:
10.1109/SICE.2015.7285347, 2015.
[15] M. A. Kamel; Youmin Zhang, "Linear model
predictive control via feedback linearization for
formation control of multiple wheeled mobile
robots", Information and Automation, IEEE
International Conference, pages: 1283-1288, DOI:
10.1109/ICInfA.2015.7279484, 2015.
[16] Salem, F.A.,"Kinematics and dynamic models and
control for differential drive mobile robots". Int.
J.Current Eng.Technol.,3: 253-263.
http://inpressco.com/wp-content/uploads/Paper6253-
263.pdf, 2013.
[17] Aveek K. Das, et al, “A Vision-Based Formation
Control Framework”, IEEE Trans. on Robotics and
Automation, Vol. 18, NO. 5, pp. 813-825, 2002.

6590

Das könnte Ihnen auch gefallen