Advances in Engineering Software 79 (2015) 4756

Contents lists available at ScienceDirect

Advances in Engineering Software

journal homepage: www.elsevier.com/locate/advengsoft

Efcient collision-free path-planning of multiple mobile robots system

using efcient articial bee colony algorithm
Jun-Hao Liang, Ching-Hung Lee
Department of Mechanical Engineering, National Chung Hsing University, Taichung 402, Taiwan, ROC

a r t i c l e i n f o a b s t r a c t

Article history: This paper aims to propose a novel design approach for on-line path planning of the multiple mobile
Received 16 July 2014 robots system with free collision. Based on the articial bee colony (ABC) algorithm, we propose an
Received in revised form 4 September 2014 efcient articial bee colony (EABC) algorithm for solving the on-line path planning of multiple mobile
Accepted 21 September 2014
robots by choosing the proper objective function for target, obstacles, and robots collision avoidance.
Available online 14 October 2014
The proposed EABC algorithm enhances the performance by using elite individuals for preserving good
evolution, the solution sharing provides a proper direction for searching, the instant update strategy
Keywords:
provides the newest information of solution. By the proposed approach, the next positions of each robot
Multiple mobile robots
Path planning
are designed. Thus, the mobiles robots can travel to the designed targets without collision. Finally,
Articial bee colony algorithm simulation results of illustration examples are introduced to show the effectiveness and performance
Instant update of the proposed approach.
Optimization 2014 Elsevier Ltd. All rights reserved.
Collision avoidance

1. Introduction
Many researches are conducted for multiple mobile robots sys-
tem since multiple robots can improve the working capability and
performance. The task of the path planning is to guide the mobile
robots toward the goal without collision with obstacles and other
robots. Design of a fast and efcient method is one of the important
problems for multiple mobile robots. Recently, there have been
many interesting research work in the literature for path planning
of mobile robots [1,3,416,2632,35]. Many literatures indicated
swarm intelligence techniques are among the methods that have
recently received considerable interest in the area of mobile robot
path planning [6,7,9,14,27]. As above, path planning is an impor-
tant issue for multiple mobile robots system navigation, especially
on-line path planning approach. In this paper, we consider the on-
line path planning problem for multiple mobile robots system,
which is the task of determining an optimal path from the initial
position to the target while avoiding collision between robots
and obstacles. Based on the articial bee colony (ABC) algorithm,
we propose an efcient method for solving the on-line path plan-
ning by choosing the proper objective function for target, obstacles,
and robots collision avoidance.
The ABC algorithm is a new population-based algorithm which
has the advantages of nding global optimization solution, being
Corresponding author.
E-mail address: chleenchu@dragon.nchu.edu.tw (C.-H. Lee).
simple and exible, and using very few control parameters
[2,1721]. The ABC algorithm has been applied to many real-world
applications, e.g., function optimization, real-parameter optimiza-
tion, digital lter design, clustering, neural network training
[2,1721,34]. Although the ABC algorithm benets to the afore-
mentioned advantages, it has something disadvantages of accuracy
of the optimal value and low convergence speed. Herein, we pro-
pose the efcient articial bee colony (EABC) algorithm to improve
the performance of optimization. The proposed EABC algorithm
enhances the performance by using elite individuals for preserving
good evolution, the solution sharing provides a proper direction for
searching, the instant update strategy provides the newest infor-
mation of solution.
In this paper, the on-line path planning method for multiple
mobile robots system without collision with obstacles and other
robots is developed using the proposed EABC algorithm. This task
can be transferred as an optimization problem with hybrid
objective function, i.e., the proposed EABC with hybridization of
objective functions for distances between goal, other robots, and
obstacles are adopted to on-line plan the path of each mobile
robot. This method results optimal paths of mobiles robots toward
to the goal and without collision in travel period. Simulation
results of illustration examples are shown to demonstrate the
effectiveness and performance of the proposed approach.
The rest of the paper is as follows: Section 2 introduces problem
formulation for path planning of multiple mobile robots. The on-
line path planning design using EABC is introduced in Section 3.

http://dx.doi.org/10.1016/j.advengsoft.2014.09.006
0965-9978/ 2014 Elsevier Ltd. All rights reserved.
48 J.-H. Liang, C.-H. Lee / Advances in Engineering Software 79 (2015) 4756
Simulation results of illustration examples and discussion are pre-
sented in Section 4. Finally, conclusion is given.
2. On-line path planning strategy design of multiple mobile
robots
Multiple mobile robots usually operate in the same environ-
ment for many situations. Thus, path planning problem becomes
an important issue for multiple mobile robots system
[3,7,11,14,27,29,30]. The off-line method is adopted usually for
the case in which the environment is exactly known, and the on-
line path planning is used for unknown environment. In this paper,
the on-line collision-free path planning strategy for multiple
mobile robots system is developed by using the proposed EABC
algorithm with hybrid objective functions.
2.1. Problem description
At rst, we introduce the environment of on-line path planning
strategy for multiple mobile robots system shown in Fig. 1. The
design goal is to determine the corresponding optimal path for
obstacle
obstacle
obstacle
Robot 1
obstacle
Robot 2
Robot 3
obstacle
Robot P
Robot 4
Fig. 1. Environment illustration of on-line path planning strategy for multiple
mobile robots system.
Fig. 2. The illustration of EABC-based approach for multiple mobile robots system.
each mobile robot such that the mobile robots move from the cor-
responding initial positions to the target Goal without any colli-
sions with obstacles and other robots. Fig. 2 shows the illustration
of EABC-based approach for multiple mobile robots system.
Detailed description will be introduced as follows.
The proposed EABC algorithm is used to choose the optimal
local position for next position in the neighborhood of each mobile
robot. Herein, each mobile robots state in the current time instant
is dened by (xci, yci, hci), i = 1, 2, . . ., p, where p denotes number of
robots. In this paper, the optimization problem for next position
is evaluated by three values: objective value for Goal, objective
value for obstacles avoidance, and objective value for robots avoid-
ance. That is, the next planned positions (xri, yri) for ith robot is
designed, where the subscript r denotes the planned position by
EABC, and it also play the role of reference trajectory for tracking
control of robots in the current time instant. In addition, the deriv-
 
ing angles are calculated by hri tan1 yxriri y ci
xci
. Subsequently, the
tracking control of each mobile robot can be solved by the results
of [2225], i.e., tracking controller generates suitable control
action/signals (ddi and dhi) such that all mobile robots move to
the next planned states (xri, yri, hri), i = 1, 2, . . ., p.
Goal 2.2. Collision-free path planning by local navigation
Herein, the on-line path planning of multiple mobile robots is
regarded as an optimization problem for searching local optimal
position step by step. The optimization considers the distances
between goal, obstacles, and other mobiles for designing the next
position and collisions. The swarm intelligence techniques have
recently received considerable interest in the area of mobile robot
path planning. Herein, we use the proposed EABC algorithm to nd
the optimal path for next position. The next target position is
searched in the neighborhood with a pre-selected radius r of the
current position shown virtual mobile robots in Fig. 3. From
Fig. 3, each robot uses its own EABC algorithm to nd the optimal
solution for next position. The current position of mobile robot is
denoted by (xc, yc) and the virtual mobile robot position (xi, yi),
i = 1, 2, . . ., PS, where PS is the chosen number of virtual robots, i.e.,
population number. It presents the individuals for minimizing
the objective function, where d denotes the changing distance, dx
denotes the changing distance in horizontal, dy denotes the
 J.-H. Liang, C.-H. Lee / Advances in Engineering Software 79 (2015) 4756 49

Goal

Local Navigation region for mobile robot i

Initial position
Fig. 3. Illustration of local navigation region (ve virtual mobile robots for searching).

changing distance in vertical, Dh denotes the changing angle. Then, function g1() implies to minimize the distance between virtual
the mobile robots nd its optimal path from some initial locations mobile robot and Goal. The optimal individual is chosen to be the
to target Goal and avoid collision with obstacles and other next position for ith robot in current location (xc, yc) for local navi-
robots. As shown in Fig. 3, the population size (number of virtual gation area. That is, g1() has the ability of deriving the mobile robot
robots) is ve and the fth robot performs the smallest objective to achieve the Goal step by step.
value (the optimal solution in the current instant), i.e., the optimal (2) Objective function for obstacles avoidance
solution of next position. Therefore, the EABC algorithm for on-line
path planning with lower and upper bounds is formulated in the The objective value for obstacles avoidance provides distance
form of information between robots and obstacles dened as
(  
Min gX i 0; X i  Obsj  > dobs ; j 1; 2; . . . ; q
1 g j2 X i  
subject to X i 2 S; S fX i xi ; yi jjjX i  X c jj 6 rg; 1
d 1
; X i  Obsj  6 dobs ; j 1; 2; . . . ; q
kX i Obsj k obs

where r is the radius of searching area, |||| denotes the Euclid- 5

ean distance operator, and g(Xi) is the objective function that is min-
imized. Note that the radius r can be selected according to the where Xi denotes the position of the ith individual (or ith virtual
mobile robots maximum speed. Subsequently, we generate popula- robot), Obsj denotes the center of jth obstacle, dobs denotes security
tion in the searching area dened as: value between robots and obstacles. Herein, dobs can be chosen by
the radius of obstacles and add some security value, q denotes num-
xi xc /  r  cosh ber of obstacles. Herein, our purpose is to minimize the objective
2
yi yc /  r  sinh function g2(). In order to avoid collision with obstacles, the shortest
distance between mobile robots and obstacles should be larger than
where h is random value between [0, 2p], / is a random value
the security value dobs. When the virtual mobile robot Xi is close to
between [0, 1]. The hybrid objective function introduced as below.
the obstacle j, g2() results a nonzero value, otherwise zero objective
value is obtained. Finally, the objective value of avoiding the obsta-
2.3. Hybrid objective function P
cles is g 2 X i qj1 g j2 X i for local navigation.
In this phase, we introduce the used hybrid objective function (3) Objective function for robots avoidance
for EABC algorithm. The objective function depends on three rules:
reach to Goal, avoid collisions with obstacles, and avoid collisions Similar to g2() the objective value for robots avoidance provides
with other robots. As the results of literature [9,14,27], the objec- the distance information about distance between other robots is
tive function for optimizing the robots path is dened as dened as
(  
gX i c1 g 1 X i c2 g 2 X i c3 g 3 X i 3 0; X i  X cj  > drobot ; j 1; 2; . . . ; p
g j3 X i  
P3 1
d 1
; X i  X cj  6 drobot ; j 1; 2; . . . ; p
where ci denotes positive weighting values satisfying c 1 for
i1 i kX i X cj k robot

user design. 6
(1) Objective function for goal
where Xi denotes the position of the ith individual (or ith virtual
The objective value for Goal provides the distance to Goal robot), Xcj denotes the jth actual robot position. drobot denotes the
security value between robots. Function g j3 X i illustrates the
g 1 X i kX i  Goalk 4
avoidance of collision between virtual robot Xi and jth actual robot
where Goal denotes goal position, Xi denotes the position of the ith Xcj. In order to avoid collision with robots, the shortest distance
individual (or ith virtual robot). Herein, minimize the objective between mobile robots and other robots should be greater than
50 J.-H. Liang, C.-H. Lee / Advances in Engineering Software 79 (2015) 4756

the selected security value. The corresponding objective value of

Start
avoidance robots collision for virtual robot Xi is obtained by
P
g 3 X i pj1 g j3 X i .
Initialization
As above, the corresponding objective value of virtual robot i is
calculate by Eq. (3) and the position of virtual robot i (xri, yri) having
smallest objective value is selected to be the next target position Evaluation & Ranking
for ith mobile robot.
Onlooker Bees Phase

Onlooker bees with

Remarks:

solution sharing
(a) The proposed approach can be applied for dynamic environ- Selection

i=i+1
ment (moving obstacles avoidance) if the obstacles are
regards as the other robots for ith virtual robot, i.e., Obsj = Xcj. No i=Po
On the other hand, the virtual robot Xi considers other robots
as moving obstacles, where the obstacle position is con- Yes
stantly changing.
Employed Bees Phase

solution sharing
(b) The optimization of objective function g1 can be replaced by

Employed bees with

g 1 X i kX i  Goali k when the targets of mobile robots are
different, where Goali denotes the target of ith mobile robot. Selection

i=i+1
(c) Other optimization algorithm can be adopted for the pro-
posed on-line path planning strategy by using the hybrid No i=Ps
objective function. Herein, we introduce the EABC algorithm
to implement. Yes

Scout Bees
3. Efcient articial bee colony (EABC) algorithm

The articial bee colony (ABC) algorithm was proposed for real- Population Adjustment
parameter optimization which simulates the foraging behavior of
No
bee colony [2,1821]. It consists of three kinds of bees, employed
Stop Criterion
bees, onlooker bees, and scout bees. This paper introduces the
modications of ABC to improve the performance of optimization, Yes
includes elite, solution sharing, instant update, and population 0
End
adjustment strategies. The comparison of performance between
traditional ABC and EABC has been introduced in [27]. As the above Fig. 4. Flowchart of the proposed EABC algorithm.
description, we here adopt EABC algorithm with hybrid objective
function to treat the optimization for on-line path planning of mul-
tiple mobile robots. The EABC algorithm for optimization problem (2) Evaluation and ranking
with bounds is in the form of (1), where the parameter dimension
is two and each individual (or bee) Xi represents the ith solution This phase evaluates the corresponding performance of each
agent. In addition, the owchart of the EABC algorithm is shown individual calculates, i.e., objective function values. According to
in Fig. 4. There are four major operations for minimizing the objec- g(Xi), the tness value is dened as
tive function, i.e., onlooker bees with solution sharing strategy,
employed bees with solution sharing strategy, scout bees, and pop- 1
f X i ; i 1; 2; . . . ; PS : 7
ulation adjustment. Herein, we adopt the instant update and solu- 1 gX i
tion sharing strategies to enhance the performance of the EABC In addition, the following probability value is dened for onlooker
algorithm. The corresponding pseudo-code of the proposed EABC bees operation
algorithm is given below:
f X i
PV X i PP ; i 1; 2; . . . ; PS : 8
Initialize population
i1 f X i
S

Do
Evaluation of population Subsequently, all individuals are ranked and indexed. Finally, the
Ranking of population (classify as onlooker bees and employed individual having least objective value is dened as the current best
bees) individual Xbest, i.e., Xbest = {Xi| min g(Xi), Xi 2 S}.
Produce new solutions by onlooker bees with solution sharing (3) Onlooker bees with solution sharing
Produce new solutions by employed bees with solution sharing
Produce new random solution by scout bees (the abandoned Herein, we choose the half better population as onlooker bees,
solution is replaced) i.e., the population size of onlooker bees PO is PS/2 if PS is even or
Update the population size by population adjustment (PS + 1)/2 if PS is odd, i = 1, 2, . . ., PO. To enhance the convergence,
Until stop criterion is met the solution sharing strategy is adopted in this operation, that is,
the information of the current best individual Xbest is provided to
More details are introduced as follows:
be a part of searching direction, i.e., the trial individual is

(1) Initialization V i X i /1 X k  X i /2 X best  X i 9

In the EABC algorithm, each individual X represents a solution where /1 and /2 are random vector between [1, 1] and [0, 1],
agent and initial individual is randomly chosen from the circle respectively, and k is selected by roulette wheel approach according
region (local navigation region in Fig. 3). to PV(Xi), k i, to enhance the evolution.
 J.-H. Liang, C.-H. Lee / Advances in Engineering Software 79 (2015) 4756 51

(4) Employed bees with solution sharing probability of nding the global optimum, but it also spends much
time in each generation [12]. Therefore, the population size (denoted
The remained individuals with lower tness value are set to be PS) is often choice according to users experience. Herein, a method
employed bees, i.e., the population size of onlooker bees PE is PS/2 if to adjust the population size according the current evolution status
PS is even or (PS  1)/2 if PS is odd, i = (PO + 1), (PO + 2), . . ., PS. The is introduced. The population size will be increased or decreased
different between employed bees and onlooker bees is the differ- according to following conditions, individual combination by sim-
ential vector composed of randomly selected individual Xk and ilarity measurement, and evaluation of evolution. We set minima
each individual Xi. Herein, we also adopt solution sharing strategy population size to be two and a maximum population size is also
to enhance the convergence. The corresponding update law is predened to avoid unlimited increase of population. Details
pseudo-code of population adjustment is introduced as follows.
V i X i /1 X k  X i /2 X best  X i 10
Pseudo-code of population adjustment
where /1 and /2 are random vector between [1, 1] and [0, 1], Begin
respectively, and k 2 {1, 2, . . ., PS} and k i. Note that the selection Remove redundant individuals by similarity examination (Eq.
of Xk has the random property and the trial vector may have better (14))
evolution when the onlooker bee is selected. Update PS
(5) Instant update embedded selection If cd > ci and PS > 2 % good evolution
PS = PS  (cd  ci);
Note that the onlooker bees and employed phases operate If PS < 2
selection instantly when the new trial solution is generated, as PS = 2;
shown in Fig. 4. Selection is a general used technique for present- End
ing the better offspring, the operation is Remove the abandoned individuals

Vi if gV i 6 gX i elseif cd < ci and PS < PS % false evolution
Xi 11
Xi otherwise PS = PS + 1;
where Vi denotes the trial individual in evolution process. The cur- Create new individual randomly
rent best individual Xbest is also updated in each selection operation, End
i.e.,
 (a) Individual combination by similarity examination
Xi if gX i 6 gX best
X best 12
X best otherwise The EABC algorithm removes the redundant individuals which
Herein, we update the each individual and best individual instanta- have similar objective values and locations. It can improve the ef-
neously, not until all individuals have been operated. After a new ciency and reduce computation effort. Since the similar individuals
individual been produced, each individual Xi and the current best may converge to the similar locations, we remain the better indi-
individuals Xbest would be updated instantly. Therefore, we can pro- vidual among the similar individuals. The worse individual is
vide the newest information for the next individual in searching removed if the following similarity examination satises
process. The used selection and instant update method is intro-  
gX i  gX k 
duced as follows. The termination condition is all individuals have   < dm and kX i  X k k < r m 14
 gX k 
been done. This modication will enhance the performance of
convergence. where dm is the threshold, rm is the distance between individual, Xi
and individual Xk. Fig. 5 introduces an illustrated example (in one
dimension) of removing the redundant individuals when individu-
Pseudo-code of instant update embedded selection for
als are identied as having the similar objective values and location.
onlooker phase
We can observe that the individuals X1 and X2 have the similar
For i = 1:PO
objective values, but the distance kX 1  X 2 k is not less than rm.
Produce new solution by
Although the individuals X3 and X4 have the similar location, they
V i X i /1 X k  X i /2 X best  X i
If g(Vi) 6 g(Xi)
Xi = Vi; g(Xi) = g(Vi); 1
If g(Vi) 6 g(Xbest) 0.8
Xbest = Xi;
End 0.6
End
0.4

0.2
(6) Scout bees
0
The scout bees carry out a random search to discover new food
-0.2
source. We withdraw the worst individual and generate the new
individual as scout bee randomly, i.e., -0.4
X new LB /2  U B  LB 13 -0.6
where /3 is random value between [0, 1]. -0.8
(7) Population adjustment
-1
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
For solving the optimization problem, there is no general
method to decide how many individuals are suitable. Having more Fig. 5. Illustrated example of how to remove the redundant individuals by
individuals can extend the searching space and increase the similarity examination (two dimension case).
52 J.-H. Liang, C.-H. Lee / Advances in Engineering Software 79 (2015) 4756

do not satisfy the other condition. In the case of the particles X5 and
X6, they have the similar objective values and locations which sat-
isfy the conditions of Individual Combination. In other words, X5
should be removed.
(b) Individual variation by evaluation of evolution status
In this step, we evaluate the evolution status of current gener-
ation to decide to increase or decrease the population size. As
above introduction, the maximum and minimum population sizes
should be set. Herein, we choose selection of onlooker bees for the
basis of population adjustment. Note that all counter indexes have
PO
zero initial values. Herein, cd Pi1 cdi denotes the update num-
bers (nd the better solutions) of parameters for each generation
PP O
and ci i1 cii memories the number of false searching (cannot
nd the better solutions). Therefore, the proposed population
adjustment technique can be introduced as:
(i) If the update number cd is greater than the number of false
searching ci (cd P ci), then the population can be decreased
and the worse individuals with poor performance in current
generation are removed.
(ii) On the other hand, the false searching dominates the evolu-
tion (ci P cd), and then an additional individual should be
generated randomly. As above description, the maximum
and minimum population sizes are set to preserve the evolu-
tion computational efciency.
4. Simulation results of illustration examples
In this section, several simulation results are introduced to
illustrate the performance of our approach. At rst, the test func-
tions of [33] given in Table 1 are presented to show the perfor-
mance and efciency of the EABC algorithm. And then, examples
for on-line path planning and control of multiple mobile robots
system are presented to illustrate the performance and effective-
ness of proposed approach. For local navigation of next step, the
following EABC algorithm parameters are chosen, Initial popula-
tion size: 10; maximum population size: 20; minimum population
size: 2; total generation number G: 20; dm: 0.1; rm: 0.0292m; the
searching radius r: 0.146 m.
4.1. Optimization of test functions
The simulation results of the testing functions is shown in
Table 1. In order to verify the performance of EABC algorithm,
the comparison results with other algorithms are obtained. We
choose traditional ABC algorithm and two multiple-agents algo-
rithms for comparison. The population size of the traditional ABC
algorithm and two multiple-agents algorithms is set as 20. The
PSO c1, c2 is set as 0.7. The GA crossover rate is 0.5, mutation rate
is 0.1. The comparison results for ten individuals are shown in
Table 2. Table 2 shows the comparison results between EABC,
ABC, PSO, and GA. We can observe that the EABC algorithm has bet-
ter performance of accuracy than traditional ABC for all test func-
tions. On the other hand, the computer time (computational effort)
is greater than the ABC algorithm since several strategies are
adopted. However, due to the population manager the computer
time is not proportional to the uses of additional strategies. Com-
pare with PSO and GA algorithms, we can also nd that EABC algo-
rithm has better result than PSO and GA algorithm, especially for
high dimensional test functions optimization. According to the
simulation results of testing functions, the EABC algorithm has
the ability to nd the global optimum.
4.2. On-line path planning using EABC algorithm
We here introduce the EABC algorithm with hybrid objective
function for on-line path planning of multiple robots. Four illustra-
tion examples are introduced to show the effectiveness and perfor-
mance of our approach.
4.2.1. Example 1: Illustration of objective function g1
The rst task is to plan the path from initial position (0, 0) to
Goal (2, 2). This illustrates the effectiveness of g1 and EABC algo-
rithm, therefore we select (c1, c2, c3) = (1, 0, 0). The stop condition
is objective value is less than 0.1. Fig. 6 shows the planned trajec-
tory and the corresponding objective value of local searching for
next step, respectively. From Fig. 6(a), we can observe that the
mobile robot can move to Goal straightly and the corresponding
objective value is decreased during the mobile robot moving to

Table 1
Test functions [33].

Function Decision space Optimal solution

P
Sphere f 1 x D 2 [100, 100]D 0
i1 xi
P QD
Schwefel 2.22 f 2 x D x [10, 10]D 0
i1 i j 
j i1 jxi j
Schwefel 1.2 PD Pi 2 [100, 100]D 0
f 3 x i1 j1 xj

Schwefel 2.21 f 4 x maxfjxi j; 1 6 i 6 Dg [100, 100]D 0

P  
Rosenbrock f 5 x D1 2 2 [30, 30]D 0
i1 100 xi1  xi xi  1 
PD
Step f 6 x i1 bxi 0:5c 2 [100, 100]D 0
Quartic P 4 [1.28, 1.28]D 0
f 7 x D i1 ixi random0; 1
P
Rastrigin f 8 x D 2 [5.12, 5.12]D 0
i1 xi  cos2pxi 10

q
Ackley P [32, 32]D 0
f 9 x 20 exp 0:2 D1 D i1 xi
2

 P 
 exp D1 D i1 cos2pxi 20 exp1
PD 2 QD  
Griewank 1
f 10 x 4000 xi [600, 600]D 0
i1 xi  i1 cos 1
p
i
n PD1
Generalized penalized f 11 x 0:1 sin 3px1 i1 xi  12 1 sin2 3pxi1 
2 [5, 5]D 0
P
xD  12 1 sin2 2pxD g D i1 uxi ; 5; 100; 4
8 m
< kxi  a ; xi > a
where uxi ; a; k; m 0; a 6 xi 6 a
:
kxi  am ; xi < a
Schwefel
1 [65.536, 65.536]D 0.998004
P
f 12 x 5001
25 j1 P2
1
6
j x aij
i1 i
 J.-H. Liang, C.-H. Lee / Advances in Engineering Software 79 (2015) 4756 53

Table 2 obstacles. One robot is initialized at position (0, 0) and nds the
The comparison result of different algorithms for D = 10 in 1000 generation. path to Goal (3, 3) and avoids the obstacles in travel period. There-
Test functions EABC ABC PSO GA fore, the weighting values are set as (c1, c2, c3) = (0.5, 0.5, 0). The
f1 Mean 6.94  10 16
5.62  10 6
5.43  10 6
713.49 stop condition is the same as above example. Fig. 7 introduces
Run time (s) 0.85 0.94 0.24 0.53 the simulation results, Fig. 7(a) shows the planned trajectory from
f2 Mean 1.49  109 2.26  105 2.31 5.72 (0, 0) to (3, 3) without collision with obstacles, the circle denotes
Run time (s) 0.95 0.94 0.26 0.55 the planned mobile robot. The concentric circles show the obstacle
f3 Mean 3.01  1015 7.65  105 1.67  106 3.65  103
Run time (s) 0.81 0.92 0.25 0.54
avoidance areas, the inner circle denotes the obstacles, and the
f4 Mean 5.73  106 1.14 1.16 35.04 outer circle denotes the security area of obstacles. Herein, the secu-
Run time (s) 0.93 0.97 0.27 0.55 rity radiuses of obstacles are 0.25 m, 3 m, 0.35 m. We can observe
f5 Mean 4.22  103 5.21 48.11 9.75  104 that the mobile robot can move to Goal successfully without colli-
Run time (s) 0.69 0.90 0.24 0.53
sions with obstacles. The corresponding objective value is shown
f6 Mean 1.17  1017 3.28  1011 3.22  109 457.42
Run time (s) 0.9 0.90 0.24 0.53 in Fig. 7(b), we nd that while the mobile robot avoids collision
f7 Results 0.33 0.62 0.36 0.42 with obstacles, there are some variant of objective value (see
Run time (s) 1.0 1.9 0.30 0.60 Fig. 7(b)), and the objective value is minimized when the mobile
f8 Mean 1.84  106 0.59 23.48 21.96 robot moving to Goal.
Run time (s) 0.80 0.92 0.25 0.54
f9 Mean 1.25  108 1.60  103 0.91 9.38
Run time (s) 0.89 0.94 0.27 0.56 4.2.3. Example 3: Illustration of the hybridization of objective functions
f10 Mean 3.96  102 0.13 0.25 5.81
g1() and g3()
Run time (s) 1.48 0.96 0.28 0.56
f11 Mean 4.18  1015 5.34  1010 5.50  103 0.21 Herein, two robots from initial positions (0, 0), (2, 2) moving to
Run time (s) 1.64 1.41 0.48 0.77 the corresponding Goals (2, 2), (0, 0) is introduced to examine the
f12 Mean 0.998 0.998 3.26 22.86 objective function respect to robot avoidance. To test the ability of
Run time (s) 4.04 2.78 1.12 1.40 moving to Goal and robots avoidance, the weighting values are set
as (c1, c2, c3) = (0.5, 0, 0.5). The stop condition is the same as above.
Simulation results are shown in Fig. 8. Fig. 8(a) is the planned tra-
Goal, shown in Fig. 6(b). This illustrates the ability of the EABC jectories of two robots by EABC algorithm with hybridization of
algorithm with g1(). g1() and g3(). We can observe that two mobile robots can move
to Goals and avoid collisions with other robot. Fig. 8(b) introduces
4.2.2. Example 2: Illustration of the hybridization of objective functions the distance between two mobile robots. We can observe that the
g1 and g2 distance value between two mobile robots is larger than the secu-
This experiment is introduced to illustrate the effectiveness of rity of robots in travel period (in this case, the security of robots is
objective functions g1() and g2(), i.e., reach to Goal and avoiding setting as 0.25 m).

2.5
(a)
3 (b)
2 2.5
Objective value

Goal
1.5 2
y (m)

1 1.5

0.5 1

0 0.5
Initial
position
-0.5 0
-0.5 0 0.5 1 1.5 2 2.5 0 2 4 6 8 10 12 14 16 18 20
x (m) Time Step

Fig. 6. Simulation results of Example 1, (a) planned trajectory using EABC algorithm and (b) the corresponding objective value.

3 2.5
(a) (b)
Goal
2.5
Obstacles 2
2
Objective value

obstacle
1.5 1.5 avoidance
y (m)

region

1
1
0.5
0.5
0
Initial
position
-0.5 0
-0.5 0 0.5 1 1.5 2 2.5 3 0 5 10 15 20 25 30
x (m) Time step

Fig. 7. Simulation results of Example 2, (a) the planned trajectory using EABC algorithm; (b) the corresponding objective value variation.
54 J.-H. Liang, C.-H. Lee / Advances in Engineering Software 79 (2015) 4756

2.5
(a) 3 (b)
Goal
2
2.5
Initial
position
1.5 Robot 1
2

Distance
y (m)

1 1.5

0.5 Initial Robot 2 1

position

0 0.5
Goal

-0.5 0
-0.5 0 0.5 1 1.5 2 2.5 0 2 4 6 8 10 12 14 16 18 20
x (m) Time step

Fig. 8. Simulation results of Example 3, (a) the planned trajectory using EABC algorithm and (b) the distance between two mobile robots.

2.5
(a) Initial
Initial
position
position
2
Robot 3 Obstacles

1.5 Robot 2
Robot 4
y (m)

Initial Goal
position
0.5

0 Robot 1 Robot 5

Initial Initial
position position
-0.5
-0.5 0 0.5 1 1.5 2 2.5
x (m)
(b)
Objective value

1
Robot1
0.5
0
0 1 2 3 4 5 6 7 8 9
Time step
Objective value

0.4
Robot 2
0.2
0
0 1 2 3 4 5 6 7 8 9
Time step
Objective value

1
Robot 3
0.5
0
0 1 2 3 4 5 6 7 8 9
Time step
Objective value

1
Robot 4
0.5
0
0 1 2 3 4 5 6 7 8 9
Time step
Objective value

1
Robot 5
0.5
0
0 1 2 3 4 5 6 7 8 9
Time step

Fig. 9. Simulation results of Example 4, (a) the planned trajectories using EABC algorithm; (b) The corresponding objective value variation.
 J.-H. Liang, C.-H. Lee / Advances in Engineering Software 79 (2015) 4756 55

Table 3 supported in part by the National Science Council Taiwan, R.O.C.,

Comparisons of ability between our approach and other literatures. under contracts NSC-102-2221-E-005-095-MY2, NSC-102-2221-
On-line Optimization Multiple Valid for E-005-061-MY3 and NSC-102-2218-E-005-012.
approach approach robots dynamic
(shortest path) environment
Literature [1] No No No No
References
Literature [4] No Yes Yes No
Literature [8] Yes Yes No No [1] Abiyev R, Ibrahim D, Erin B. Navigation of mobile robots in the presence of
Literature [26] No Yes No No obstacles. Adv Eng Softw 2010;41:117986.
[2] Akay B, Karaboga D. A efcient articial bee colony algorithm for real-
Literature [28] No Yes No No
parameter optimization. Inf Sci 2012;192(1):12042.
Literature [30] No Yes Yes No
[3] Arai T, Pagello E, Parker LE. Editorial: advances in multiple mobile robots
Literature [31] No No Yes No systems. IEEE Trans Robot Autom 2002;18(5):65561.
Literature [35] No No No Yes [4] Bhattacharjee P, Rakshit P, Goswami I, Konar A, Nagar AK. Multi-robot path-
Our approach Yes Yes Yes Yes planning using articial bee colony optimization algorithm. In: 2011 Third
world congress on nature and biologically inspired computing (NaBIC), 1921
October; 2011. p. 21924.
[5] Borenstein J, Koren Y. Real-time obstacle avoidance for fast mobile robots. IEEE
Trans Syst Man Cybern 1989;19(5):117987.
4.2.4. Example 4: Illustration of the hybridization of objective functions [6] Cai Z, Peng Z. Cooperative coevolutionary adaptive genetic algorithm in path
g1(), g2(), and g3() planning of cooperative multi-mobile robot systems. J Intell Rob Syst
2002;33(1):6171.
As above, we separate the objective functions to illustrate the [7] Cao YU, Fukunaga AS, Kahng AB. Cooperative mobile robotics: antecedents and
effectiveness of objective functions. In this example, we combine directions. Auton Robots 1997;4:727.
the objective functions to examine the ability of local searching [8] Chia SH, Su KL, Guo JH, Chung CY. Ant colony system based mobile robot path
planning. In: 2010 Fourth international conference on genetic and
for next position, consists of moving to Goal, obstacles avoidance, evolutionary computing (ICGEC), 1315 December; 2010. p. 21013.
and robots avoidance. Five robots from initialized at positions (0, [9] Chen X, Li Y. Smooth path planning of a mobile robot using stochastic particle
0), (0, 1), (0, 2), (2, 2), (2, 0) and nd the path to Goal (1, 1), there- swarm optimization. In: Int conf on mechatronics and automation; 2006. p.
17227.
fore, the weighting values are set as (c1, c2, c3) = (0.4, 0.3, 0.3). The
[10] Dudek G, Jenkin MRM, Milios E, Wilkes D. A taxonomy for multi-agent
stop condition is both the objective values are less than 0.1. The robotics. Auton Robots 1996;3(4):37597.
simulation results are shown in Fig. 9. Fig. 9(a) shows the planned [11] Fahimi F, Nataraj C, Ashrauon H. Real-time obstacle avoidance for multiple
mobile robots. Robotica 2008;27:18998.
trajectories of ve robots by EABC algorithm with hybridization of
[12] Hao B, Dai X. The collision-free motion of robot with fuzzy neural network. In:
g1(), g2(), and g3(). In the travel period of robots 1, 4, and 5, they The 2nd int conf on industrial and information systems; 2010. p. 2019222.
also avoid the obstacles and then move to Goal. We can also [13] Hwang YK, Ahuja N. Path planning using a potential eld representation. IEEE
observe that robot 2 stop at Goal and other four robots stop around Int Conf Robot Autom 1988;1:6489.
[14] de Jong WJ. Development of an experimental research platform with mobile
the Goal due to the security of robots. Herein, the security of robots robots including an application of formation control. Master of Science in
is setting as 0.22 m. The corresponding hybrid objective values of Systems and Control at Delft University of Technology; 2008.
ve robots are shown in Fig. 9(b), we can observe that all objective [15] Kambhampati S, Davis L. Multiresolution path planning for mobile robots. IEEE
Trans Robot Autom 1986;2(3):13545.
values are minimized when the mobile robots moving around the [16] Khatib O. Real-time obstacle avoidance for manipulators and mobile robots.
Goal. For this illustration example, we planned the ve mobile In: IEEE Int conf on robotics and automation, St. Louis; 1985. p. 500505.
robots trajectories successfully and it is feasible of local searching [17] Kang F, Li J, Xu Q. Structural inverse analysis by hybrid simplex articial bee
colony algorithms. Comput Struct 2009;87(1314):86170.
for next position by EABC algorithm. [18] Karaboga D, Basturk B. Articial bee colony (ABC) optimization algorithm for
Finally, to illustrate ability and effectiveness of our approach for solving constrained optimization problems. Found Fuzzy Logic Soft Comput
on-line path planning of multiple robots, a simple comparison is 2007;4529:78998.
[19] Karaboga D, Basturk B. On the performance of articial bee colony (ABC)
introduced in Table 3. Table 3 shows that the proposed approach
algorithm. Appl Soft Comput 2008;8(1):68797.
is valid for dynamic environment and multiple mobile robots. It [20] Karaboga D, Akay B. A efcient articial bee colony (ABC) algorithm for
also has the ability of on-line performance and nding the shortest constrained optimization problems. Appl Soft Comput 2011;11(3):302131.
[21] Karaboga N. A new design method based on articial bee colony algorithm for
path for each mobile robot.
digital IIR lters. J Franklin Inst 2009;346(4):32848.
[22] Lee TC, Song KT, Lee CH, Teng CC. Tracking control of unicycle-modeled mobile
robots using a saturation feedback controller. In: IEEE Trans on control
5. Conclusion systems and technology, vol. 9(2); 2001. p. 30518.
[23] Lee CH, Chang FK, Lee YC. An improved electromagnetism-like algorithm for
In this paper, we have proposed an on-line collision-free path recurrent neural fuzzy controller design. Int J Fuzzy Syst 2010;12(4):28090.
[24] Lee CH, Teng CC. Control of a nonlinear dynamic system via adaptive
planning strategy of multiple mobile robots system using the pro-
saturation PID control scheme. Int J Fuzzy Syst 2002;4(4):9227.
posed EABC algorithm. The EABC algorithm enhances the perfor- [25] Lee CH, Chiu MH. Recurrent neuro-fuzzy control design for tracking of mobile
mance by using elite individuals for preserving good evolution, robots via hybrid algorithm. Expert Syst Appl July 2009;36(5):89939.
the solution sharing provides a proper direction for searching, [26] Lei L, Shiru Q. Path planning for unmanned air vehicles using an improved
articial bee colony algorithm. In: China control conference, Hefei, 2527 July;
the instant update strategy provides the newest information of 2012. p. 246891.
solution. It contains path planning for next position. First, we sep- [27] Liang JH. On-line path planning strategy design of multiple mobile robots
arate the objective functions to show the ability of each objective system using modied articial bee colony algorithm. Mater thesis,
Department of Electrical Engineering, Yuan Ze University, Taiwan; 2012.
function. The EABC algorithm here is to nd the optimal trajectory [28] Liu G, Li T, Peng Y, Hou X. The ant algorithm for solving robot path planning
to the goal for each mobile robot. Finally, we combine path plan- problem. In: Third international conference on information technology and
ning for next position. Several illustrated examples have been applications, 2005, ICITA 2005, vol. 2, 47 July; 2005. p. 25, 27.
[29] Lozano-Perez T, Wesley MA. An algorithm for planning collision-free paths
shown that it is feasible in on-line path planning of multiple among polyhedral obstacles. Commun ACM 1979;22(10):56070.
mobile robots. They also show the performance and effectiveness [30] Saffari MH, Mahjoob MJ. Bee colony algorithm for real-time optimal path
of our approach. planning of mobile robots. In: Fifth international conf on soft computing,
computing with words and perceptions in system analysis, decision and
control. 24 September; 2009. p. 1, 4.
Acknowledgement [31] Sahoo RR, Rakshit P, Haidar MT. Navigation path planning of multi-robot using
honey bee mating optimization algorithm (HBMO). Int J Comput Appl
2011;27(11):18.
The authors would like to thank anonymous reviewers for their [32] Yang SX, Luo C. A neural network approach to complete coverage path
insightful comments and valuable suggestions. This work was planning. IEEE Trans Syst Man Cybern B Cybern 2004;34(1):71825.
56 J.-H. Liang, C.-H. Lee / Advances in Engineering Software 79 (2015) 4756

[33] Yao X, Liu Y, Lin G. Evolutionary programming made faster. IEEE Trans Evol [35] Zhu Y, Zhang T, Song J, li X. A new hybrid navigation algorithm for mobile
Comput 1999;3(2):82. robots in environments with incomplete knowledge. Knowl-Based Syst
[34] Zhang C, Quyang D, Ning J. An articial bee colony approach for clustering. 2012;27:30213.
Expert Syst Appl 2010;37(7):47617.