+Model

JESIT 82 1–19 ARTICLE IN PRESS

Available online at www.sciencedirect.com

ScienceDirect
Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx

1 Multi-robot path planning in a dynamic environment using improved

2 gravitational search algorithm
3 Q1 P.K. Das a,∗ , H.S. Behera a , P.K. Jena b , B.K. Panigrahi c
Q2 a Dept. of Comp. Sc. and Engineering and Information Technology, VSSUT, Burla, Odisha, India
4
5
b Dept. of Mechanical Engineering, VSSUT, Burla, Odisha, India

6
c Dept. of Electrical Engineering, IIT, Delhi, India

Received 17 August 2015; received in revised form 17 November 2015; accepted 20 December 2015

8 Abstract
9 This paper proposes a new methodology to optimize trajectory of the path for multi-robots using improved gravitational search
10 algorithm (IGSA) in a dynamic environment. GSA is improved based on memory information, social, cognitive factor of PSO
11 (particle swarm optimization) and then, population for next generation is decided by the greedy strategy. A path planning scheme
12 has been developed using IGSA to optimally obtain the succeeding positions of the robots from the existing position. Finally, the
13 analytical and experimental results of the multi-robot path planning have been compared with those obtained by IGSA, GSA and
14 PSO in a similar environment. The simulation and the Khepera environmental results outperform IGSA as compared to GSA and
15 PSO with respect to performance matrix.
16 © 2016 Production and hosting by Elsevier B.V. on behalf of Electronics Research Institute (ERI). This is an open access article
17 under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

18 Keywords: Gravitational search algorithm; Multi-robot path planning; Average total trajectory path deviation; Average uncovered trajectory target
19 distance; Average path length
20

21 1. Introduction

22 Q4
Q5 Gravitational search algorithm (GSA) is effective and efficient using an alternative approach to the multi-robot
23 path planning. Although many algorithms (Tuncer and Yildirim, 2012; Guo and Parker, 2002) have been proposed
24 and proven to be feasible and efficient for robot motion planning and collision avoidance, classic techniques for
25 path planning problem (Konar, 1999; Banerjee et al., 2011) are general methods like Roadmap, Cell Decomposition,
26 Potential Fields, Optical Tweezers and Mathematical Programming. Many authors have proposed multi-robot and the

∗ Corresponding author.
Q3
E-mail addresses: [email protected] (P.K. Das), hsbehera [email protected] (H.S. Behera), [email protected] (P.K. Jena),
[email protected] (B.K. Panigrahi).
Peer review under the responsibility of Electronics Research Institute (ERI).

http://dx.doi.org/10.1016/j.jesit.2015.12.003
2314-7172/© 2016 Production and hosting by Elsevier B.V. on behalf of Electronics Research Institute (ERI). This is an open access article under
the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
2 P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx

27 single robot path planning problems using different classical techniques (Kcymeulcn and Decuyper, 1994; Li et al.,
28 2009), Neural Network (Yu and Kromov, 2001), artificial immune system (Das et al., 2012; Luh and Cheng, 2002) and
29 heuristic optimization algorithms (Das et al., 2010, 2011; Geem et al., 2001; Yang, 2009; Regele and Levi, 2006). High
30 time complexity in large problem spaces and trapping in local optimum are drawbacks for classic techniques and in
31 many meta-heuristic algorithms. These drawbacks cause the classical techniques and inefficient in the various problem
32 spaces. To improve the efficiency of classical methods, probabilistic algorithms like PRM and RRT are proposed
33 for improving the local optimization problem, many evolutionary algorithms like Genetic Algorithms (Tuncer and
34 Yildirim, 2012; Gong and Lincheng, 2001), PSO (Zhang et al., 2013; Masehian and Sedighizadeh, 2010), bee colony
35 optimization (Bhattacharjee et al., 2011) and differential evolution algorithm (Chakraborty et al., 2009) are used in
36 multi-robot path planning problem.
37 The gravitational search algorithm (Verma et al., 2013; Eldos and Qasim, 2013; Chatterjee et al., 2011) is a recent
38 algorithm that has been inspired by the Newtonian’s law of gravity and motion. GSA has undergone a lot of changes to
39 the algorithm itself and has been applied in various applications. At present, there are various variants of GSA (Precup
40 et al., 2012; Rashedi et al., 2010, 2009b; Purcaru et al., 2013) which have been developed to enhance and improve the
41 original version. The algorithm has also been explored in many areas (Sabri et al., 2013; Eldos and Qasim, 2013).
42 For realization multi robot path planning problem with different goal of the respective robots with GSA (Precup
43 et al., 2012; Tuncer and Yildirim, 2012) by the centralized approach, a fitness function is constructed to determine the
44 next position of the robots that lie on optimal trajectories leading toward the respective goals. The fitness function of
45 the GSA (Alba and Dorronsoro, 2005) has two main components: first one is the objective function describing the
46 selection of next position on an optimal trajectory based on velocity, and the second one is the constraint on acceleration
47 representing avoidance of collision with other robots and with static obstacles. The path-planning problem considered
48 here is formulated by a centralized approach, where an iterative algorithm is invoked to determine the next position of
49 all the robots satisfying all the constraints imposed on the multi-objective function. The algorithm is iterated until all
50 the robots reach their destination (goal position).
51 The advantages of GSA are (1) easy to implement with higher computational efficiency; (2) few parameters to
52 adjust, but the disadvantages of this algorithm are as follow (1) if premature convergence occurs, there will not be
53 any recovery for this algorithm; (2) the algorithm loses its ability to explore and then becomes inactive only after
54 Q6 becomes convergence. Due to above difficulties in GSA, further improvements are required for the optimal solution
55 to the complex problem. Here, we consider the improvement of GSA which is based on the communication and
56 memory characteristics of PSO (particle swarm optimization). Therefore, we called it improved gravitational search
57 algorithm.
58 The main objective of this paper is summarized as follows: (i) we study the problem of multi-robot path planning
59 in a clutter environment and formulated the above problem as multi-objective optimization problem with constraints;
60 (ii) a novel method to the solution of an optimal trajectory path generation for multi robot path planning problem using
61 IGSA is proposed in this article; (iii) the proposed algorithm has been applied for multi robot path planning in a clutter
62 and dynamic environment and obtained results are compared to other optimization algorithms like GSA, DE; (iv) the
63 performance of the proposed IGAS, as an optimizing tool in solving multi robot path planning problem, is applied in
64 the simulation as well as Khepera-II environment and result is presented; (v) the performance matrix of the proposed
65 approach is successfully validated in simulation and Khepera-II.
66 In this paper, the implementation of the modified gravitational search technique has been proposed to determine
67 the trajectory path for multiple robots from predefined initial positions to predefine target positions in the environment
68 with an objective to minimize the path length of all the robots. The result shows that the algorithm can improve
69 the solution quality in a reasonable amount of time and also improves the convergence rate. This paper improves
70 the gravitational search algorithm (IGSA) for improving the global path planning problem of the multi-robots by
71 improving the convergence rate. Finally, the efficiency of the IGSA has been proved through the simulation as well
72 as Khepera environment and a result obtained is compared with other evolutionary computing such an GSA and
73 DE.
74 Q7 The rest of the paper is outlined as follows: Section 3 briefly describes the improved gravitational search algorithm.
75 Formulation of the problem for multi-robot path planning has been elaborated in Section 4. Multi-objective optimization
76 problem solving using improved GSA is described in details in Section 5. Section 6 demonstrates the result of path
77 planning for multi-robot through simulation. In Section 7, the experiment has been conducted in Khepera II environment
and finally, the conclusion of the work is presented in Section 8.

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx 3

78 2. Gravitational search algorithm (GSA)

79 Recently, the scientific community has gained the interest on GSA. It is a meta-heuristic optimization algorithm
80 inspired by nature which is based on the Newton’s law of gravity and the law of motion (Rashedi et al., 2009a; Sabri
81 et al., 2013). GSA is grouped under the population based approach and is reported to be more natural. The algorithm
82 has been planned to improve the performance in the exploration and manipulation capabilities of a population based
83 algorithm, based on gravity rules.
84 GSA is based on the two important formulas about Newton gravity laws given by Eqs. (1) and (2). Eq. (1) is the
85 gravitational force equation between the two particles, which is directly proportional to their masses and inversely
86 proportional to the square of the distance between them. But in GSA instead of the square of the distance, only the
87 Q8 distance is used. Eq. (2) is the equation of acceleration of a particle when a force is applied to it (Rashedi et al., 2009a;
88 Sabri et al., 2013).
M 1 M2
89 F =G (1)
R2
F
90 a= (2)
M
91 G is gravitational constant, M1 and M2 are masses and R is the distance between them, F is gravitational force, and a
92 is acceleration. Based on these formulas, the heavier object with more gravity force attracts the other objects as it is
93 seen in Fig. 1.
94 In GSA, each mass (agent) has four characteristics, namely: position, inertial mass, active gravitational mass, and
95 passive gravitational mass. The position of the mass corresponds to a solution of the problem, and the fitness function
96 is used to determine the gravitational and inertial masses (Verma et al., 2013; Sabri et al., 2013). The more precisely
97 masses obey the following laws.
98 Law of gravity: Each particle attracts every other particle and the gravitational force between two particles is directly
99 proportional to the product of their masses and inversely proportional to the distance between them, R. We use here
100 R instead of R2 , because according to our experimental results, R provides better results than R2 in all experimental
101 cases.
102 Law of motion: The current velocity of any mass is equal to the sum of the fraction of its previous velocity and the
103 variation in the velocity. Variation in the velocity or acceleration of any mass is equal to the force acted on the system
104 divided by mass of inertia.

105 2.1. Agents initialization

106 Consider a system with N masses in which position of the ith mass is defined as follows:
 
107 Xi = xi1 , . . .xid , . . ., xin for i = 1, 2, . . .N (3)

108 where xid is the position of ith mass in dth dimension and n is the dimension of the search space.

F2 M2

M1
F1

M3
F3

Fig. 1. The Newton gravitational force representation.

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
4 P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx

109 2.2. Fitness and best fitness computation

110 worst(t) and best(t) are defined as follows for the minimization case:
111 worst(t) = maxi ∈ p fiti (t), p = 1, 2, . . ., N (4)
112 best(t) = mini ∈ p fiti (t), p = 1, 2, . . ., N (5)

113 2.3. Gravitational constant (G) computation

114 The gravitational constant G is computed at iteration (Sabri et al., 2013).

115 G(t) = Go e(-αt/T ) (6)
116 Here, T is the maximum iteration, t is the current iteration and α  0 is the weight factor, computed as follows.
αmax − αmin
117 α = αmax − ×t (7)
T

118 2.4. Masses of the agents’ calculation

119 Each agent’s mass is calculated after computing current population’s fitness as:
fiti (t) − worst(t)
120 mi (t) = (8)
best(t) − worst(t)
mi (t)
121 Mi (t) = N (9)
j=1 mj (t)

122 where Mi (t) and fiti (t) represent the mass and the fitness value of the agent i at iteration t, respectively.

123 2.5. Velocity and positions of agents

124 The velocity and position of the agents are updated as:
125 Vid (t + 1) = βV di (t) + aid (t) (10)
126 xid (t + 1) = xid (t) + Vid (t + 1) (11)
127 Here, β is the random number, 0 ≤ β ≤ 1 and an acceleration of the ith agents at iteration ‘t’ is computed as,
Fid (t)
128 aid (t) = (12)
M i (t)
129 Fid (t) is the total force acting on ith agent calculated as:

130 Fid (t) = βFijd (t) (13)
j ∈ kbest ,j =
/ i

131 Kbest is the set of first K agents with the best fitness value and biggest mass, which is a function of time, initialized to
132 k0 at the beginning and decreasing with time. Here k0 is set to N (total number of agents) and is decreased linearly to 1.
133 Fijd (t) is computed using the following equation:
  
Maj (t)
134 Fijd (t) = G(t) × Mpi (t) × × Xjd (t) − Xid (t) (14)
disij (t) + ε
135 Here Xi and Xj are the position vector of the ith and jth agent in dth dimension, Fijd (t) is the force acting on agent i
136 from agent j at dth dimension and ith iteration. disij (t) is the Euclidian distance between two agents i and j at iteration
137 t. G(t) is the computed gravitational constant at the same iteration while ε is a small constant. Mpi (t) is the passive

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx 5

138 gravitational mass of the agent i at the instance t. Maj (t) is the active gravitational mass of the agent j at time t, these
139 masses being calculated according to Precup et al. (2012), Rashedi et al. (2010, 2009b) and Purcaru et al. (2013).

140 3. Improvement of the gravitational search algorithm based on PSO and greedy strategy

141 Most of the meta-heuristic searching algorithm find its best solution due to good balance of exploration and exploita-
142 tion (Alba and Dorronsoro, 2005; Liu et al., 2013). The exploration capability of the algorithm provides the connectivity
143 relationship of the search space, which helps to find global optimal solution. The exploitation helps to find the better
144 optimal solutions in the visited domain, which reinforce the convergence capability of local search. So, good meta-
145 heuristic algorithm should improve the exploration ability in the first step and then exploitation ability with increasing
146 of iteration in second step. Therefore, the gravitational search algorithm has been improved to maintain the good
147 balance between exploration and exploitation. In GSA, the moment direction of each agent is based on the total force
148Q9 act by other agents on it and lacking the communication between the agents. Therefore, improvement of the searching
149 ability of GSA based on the memory and social information of PSO and to accelerate the convergence rate, weight
150 value is assigned to inertia mass of every agent in each iteration (Sarafrazi et al., 2011) and then, optimized solution
151 saving strategy is used with reference to DE (Sarafrazi et al., 2011). The PSO updates the velocity using the cognitive
152 and social factor. The velocity and position update equation of PSO are as follow:
153 Vid (t + 1)PSO = wVid + C1 × ϕ1 × (xpbest
d
i
− xid (t)) + C2 × ϕ2 × (xgbest
d
− xid (t)) (15)

154 xid (t + 1)PSO = xid (t + 1) + Vid (t + 1) (16)

155 vdi (t + 1)GSA = βvi (t) + aid (t) (17)
Q10
156 where Eq. (17) is the GSA velocity formulation obtained from Eq. (10). In this paper, GSA is improved by adopting
157 the memory, social and cognitive information of PSO. The velocity updating equation in GSA can be defined as
158 Vid (t + 1)IGSA = βVid (t)GSA + aid (t) + C1 × ϕ1 × (xpbest
d
i
− xid (t)) + C2 × ϕ2 × (xgbest
d
− xid (t)) (18)

159 xid (t + 1)IGSA = xid (t) + Vid (t + 1)IGSA (19)

160 where Eq. (19) is the IGSA velocity formulation, which is formulated and updated using PSO velocity by considering
161 the memory, social and cognitive factor and GSA acceleration. C1 and C2 balance the effectiveness of “law of gravity
162 and memory and social information”. The optimized solution saving strategy is used for deciding the member for next
163 generation t + 1 with reference to differential evolution (DE) (Sarafrazi et al., 2011). The “survival of fittest” strategy
164 is used to decide the member for next generation. Here, the greedy strategy has been devised for deciding better target
165 vector. The population for next generation is decided by comparing the trial vector xid (t + 1) with the target vector
166 xid (t). The selection procedure can be expressed by the following expression.

d
xi (t), if fit(xid (t)) < fit(xid (t + 1))
xi (t + 1) =
d
167
xid (t + 1), otherwise (20)

168 4. Problem formulation for multi robot path planning

169 The problem formulation for multi-robot path planning is to determine the next position of the robot from their
170 existing positions in its workspace by avoiding the collision with other robots and obstacles (which are static in nature)
171 in its path to reach at the goal. Multi-robot path planning problem is formulated by considering the set of principles
172 using the following assumptions by a uniform treatment.
173 Assumptions

174 a For each robot the current position (recent position) and goal position (target position) is known in a given reference
175 coordinate system.
176 b The robot can select any action in a given time from a fixed set of actions for its motions.
177 c Each robot is performing its action in steps until all robots reached in their respective target positions.

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
6 P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx

Y
y i (t + 1)
vi (t )
( x i (t ), y i (t )) θi
y (t )
i

x (t ) xi (t + 1) X
i

Fig. 2. Representation of next position from current position of the i-th robot.

178 The following principles have been taken care for satisfying the given assumptions.

179 1. For determining the next position from its current position, the robot tries to align its heading direction toward the
180 goal.
181 2. The alignment may cause a collision with the robots/obstacles (which are static in nature) in the environment, hence,
182 the robot has to turn its heading direction left or right by a prescribed angle to determine its next position.
183 3. If a robot can align itself with a goal without collision, then, it will move to that determine the position.
184 4. If rotating the heading direction left or right requires the same angle of rotation of the robot about the z-axis, if it is
185 tied then, broken randomly.

186 Consider the initial position of the ith robot at a time t is (xi (t), yi (t)), the next position of the same robot at a time
goal goal
187 (t + δt) is (xi (t + δt), yi (t + δt)), vi (t) is the velocity of the robot Ri and (xi , yi ) is the target or goal position of
188 the robot Ri .
189 So, the expression for the next position (xi (t + δt), yi (t + δt)) can be derived from Fig. 2 as follows
190 xi (t + δt) = xi (t) + vi (t)cos θ i δt (21)
191 yi (t + δt) = yi (t) + vi (t)sin θ i δt (22)
192 When δt = 1, Eqs. (21) and (22) are reduced to
193 xi (t + 1) = xi (t) + vi (t)cos θi (23)
194 yi (t + 1) = yi (t) + vi (t)sin θi (24)
195 Consider initially, the robot Ri is placed in the location at (xi (t), yi (t)). We want to find the next position of the
196 robot (xi (t + 1), yi (t + 1)), such that the line joining between {(xi (t), yi (t)), (xi (t + 1), yi (t + 1))} and {(xi (t + 1),
goal goal
197 yi (t + 1)), (xi , yi )} should not touch the obstacle in the world map is represented in Fig. 3 and minimizes the
198 total path length from current position to a goal position without touching the obstacle by forming constraint. Then

Y
( x i (t + δt ), y i (t + δt ))

Obstacle

( x i (t ), y i (t ))

Fig. 3. Selection of next position (xi (t + δt), yi (t + δt)) from the current position (xi (t), yi (t)) to avoid collision with obstacles.

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx 7

199 objective function fit1 that determines the length of the trajectory for n number of robots,
n  
 goal 2 goal 2
200 fit1 = ((xi (t) − xi (t + 1)) + (yi (t) − yi (t + 1) ) + ((xi (t + 1) − xi ) + (yi (t + 1) − yi ) )
2 2

i=1
201 (25)
202 By putting the value xi next and yinext from Eqs. (21) and (24) into Eq. (25), we obtain,

 
 n
2 2
203 fit1 = vi (t) + (xi (t) + vi (t)cos θi − xigoal ) + (yi (t) + vi (t)sin θi − yigoal ) (26)
i=1
204 The multi-robot path-planning is now represented as an optimization problem by minimizing the objective function
205 in Eq. (26) with considering the penalty function as the constraints in the objective function. Minimizing the objective
206 function in Eq. (26) shows that the robot will follow the shortest distance from the initial point to target point. The
207 constraints here are two types of penalty. The first penalty is to avoid collision between teammates (any two mobile
208 robots), whereas the second penalty is to avoid collision of a robot with a static obstacle. By combining these two
209 penalties a linear fuzzy function is developed for evaluating the obstacle present in the path. So, the objective function
210 formed based on the fuzzy function is denoted as fitj .
⎧
⎪
⎪ 1 d(Oj ) ≤ d(Oj )min
⎪
⎪  
⎨ d(Oj ) − d(Oj )min
211 fitj = exp −α d(Oj )min ≺ d(Oj ) ≺ d(Oj )max (27)
⎪
⎪ d(O j )max − d(Oj )min
⎪
⎪
⎩
0 d(Oj ) ≥ d(Oj )max
212 where α is a positive constant, d(Oj ) be the distance between mobile robot and obstacles, d(Oj )max is maximum
213 distance and d(Oj )min is the minimum distance with respect to the obstacle Oj . The path is safe and collision free
214 path, when d(Oj ) ≥ d(Oj )max and path is unsafe if, d(Oj ) ≥ d(Oj )min .
215 Thus, mathematically, the optimization problem for obstacles can be formulated as follows:
216 fit2 = maxj=1,2,...N (fitj ) (28)
217 Thus, the optimization problem can be represented as follows:
λ
218 fit = fit1 + (29)
fit2
219 Here, λ is positive constant. The above optimization problem is to minimize the Euclidean distance between the current
220 position and their goal position which is presented by the objective function fit1 and the second objective function is a
221 constraint to find the safe path.

222 5. Multi-objective optimization problem solving using IGSA

223 In this section, multi-robot path planning algorithm has been proposed using IGSA. The proposed IGSA algorithm
224 is used to evaluate the next positions of every robot by presuming the current positions of robots and speeds as the
225 parameter for optimizing the given multi-objective function. It determines the optimized path from each state to the
226 goal state in both dynamic and static environments and the robot measures its distance to obstacles with the help of
227 equipping sensors.
228 The agents are moving in the search space with the help of the gravity is considered in the proposed IGSA based
path planning. The outline of the proposed algorithm is discussed below:

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
8 P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx

229 Procedure IGSA (xcurr i , ycurr i , pos-vector)

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx 9

230 Pseudo code for path planning

231 6. Computer simulation

232 The multi-robot path-planning algorithm is implemented in a simulated environment. The simulation is conducted
233 in a C environment on a Pentium processor and the experiment was performed with 14 robots of circular shape. The
234 radius of each robot is considered as 6 pixels. Before initiating the experiment for multi-robot path planning, each robot
235 starting and goal points are predefined. The experiments were performed with seven differently shaped obstacles and
236 with equal velocities for all the robots in a given run of the program; however, the velocities were adjusted over different
237 runs of the same program. One of our experimental world-maps with an initial configuration of the world-map with 7
238 obstacles and the starting and the goal positions of 12 circular soft-bots are shown in Fig. 4. The intermediate steps of
239 movement of the robots are shown in Figs. 5 and 6. The final stage of world maps, where all the robots reached in their
240 predefine goal respectively is shown in Fig. 7. The simulation is also conducted in the environment as presented in
241 Fig. 4 for same number of robots by GSA and DE. The optimal trajectory of the path has been presented in Figs. 8 and 9
242 for GSA and DE respectively.
243 The experiment has been conducted using a central version of the algorithm using the fitness Function (29) for
244 deciding the next position of the robot. In our experiment, parameters have been described in Table 1 for simulation
245 and Khepera II environment.

246 6.1. Average total trajectory path distance (ATTPD)

247 Consider a trajectory path from the predefine starting point Sk to the goal point Gk generated by the program for
248 robot Rk in the jth run is TPkj . If TPk1 , TPk2 ,. . .., TPkj are the trajectory paths generated over jth runs, then the average
j
249 total trajectory path traversed (ATTPT) by a robot Rk is given by r=1 TPir /j and the average trajectory path distance

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
+Model
JESIT 82 1–19 ARTICLE IN PRESS
10 P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx

Fig. 4. Initial world map with 7 obstacles and 5 robots.

Fig. 5. Intermediate state of the world map during execution using IGSA for 5 robots and 7 obstacles after 9 steps.

Fig. 6. Intermediate stage of the world map during execution of IGSA for 5 robots and 7 obstacles after 17 steps.

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
+Model
JESIT 82 1–19 ARTICLE IN PRESS
P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx 11

Fig. 7. Five robots reached in their respective pre-defined goal.

Fig. 8. All robots reached in their respective pre-defined goal in 29 steps by GSA.

Fig. 9. All robots reached in their respective pre-defined goal in 30 steps by PSO.

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
12 P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx

Table 1
Parameter used in the simulation and Khepera.
Parameters Values

G0 100
αmin 0.2
αmax 0.4
λ 100
C1 0.5
C2 0.5
T (Maxiter) 100
W 0.72
β 0.5
N 50

250 for this robot is evaluated by measuring the difference between ATPT and the ideal shortest path between Sk to Gk . If
251
the ideal trajectory path for robot Rk obtained geometrically is TPk−real , then the average path distance is given by

j
Pir
252 TPk−real − .
j
r=1
253
Therefore, for n robots in the workspace the average total path distance (ATPD) is
 
n j
Pir
254 TPk−real −
j
i=1 r=1

255 6.2. Average uncovered trajectory target distance (AUTTD)

256 Given a goal position Gk and the current position Ck of a robot on a 2-dimensional workspace, where Gk of Ck are
257 2-dimensional vectors, the uncovered trajectory distance for the robot k is Gk − Ck , where . denotes Euclidean
n
258 norm. For n robots, uncovered trajectory target distance (UTTD) is UTTD = Gk − Ck . For k runs of the program,
i=1
259 we evaluate the average of UTTDs, and call it the average uncovered trajectory target distance (AUTTD). Fig. 16 shows
260 that by decreasing the velocity, AUTTD takes longer time to converge and gradually terminated with iteration. Again,
261 it is noted that larger the velocity of the robot, the faster falloff in the AUTTD. Fig. 17 shows that, larger the number
262 of robots, slower the convergence rate. Slower the convergence causes the delay in falloff in AUTTD.
263 The performance analysis was undertaken in the simulation environment and the ATPT was plotted for n robots,
264 called average total trajectory path traveled (ATTPT) by varying no. of robots 1–5 presented in Fig. 18 and generate
265 paths using three algorithms, including real-coded DE, GSA and IGSA. It is noteworthy from Fig. 18 that IGSA possess
266 least ATTPT in comparison to the algorithms irrespective of the number of robots.
267 The performance analysis has been performed in terms of AUTTD over the number of steps in Fig. 19. It provides
Q11
268 graphs between AUTTD vs. no. of stages required during the planning of path using three algorithms with number
269 of obstacles = 7 and no. of robots = 5. It is apparent from Fig. 19 that AUTTD returns the smallest value for IGSA
270 irrespective of the number of planning steps.
271 The performance of the result has been analyzed by plotting the average total trajectory path distance (ATTPD)
272 with the number of robots as variable in Fig. 20. This path is generated by three different evolutionary algorithms
273 such as GSA, DE, IGSA. Fig. 20 shows the result of ATTPD computation, when the number of robots varies between
274 1–5. Here, we observed that IGSA performs better than the other two algorithms as ATTPD is smallest for IGSA in
275 comparison to other two algorithms irrespective of the number of robots.
276 Now, the performance analysis was undertaken by comparing the running time over the maximum number of
277 iterations using three algorithms. Fig. 21 provides the time required for robots to reach in their respective goal position
278 by three different algorithms and it shows that IGSA takes less time for robots to reach in destination.

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx 13

Table 2
Description of obstacles presents in Fig. 4.
Robot number No. of step required to goal in IGSA No. of step required to goal in GSA No. of step required to goal in DE

1 17 19 23
2 21 25 29
3 15 23 27
4 26 29 30
5 12 14 17

Table 3
Q12 Comparison of number of steps taken, ATTPT and ATTPD of different algorithms for different no. of robots.

No. of robots Algorithms (steps taken) ATTPT (in inch.) ATTPD (in inch.)

IGSA DE GSA IGSA DE GSA IGSA DE GSA

2 12 16 18 35.7 36.5 38.4 3.7 4.7 5.7

3 15 18 20 37.8 38.6 40.4 4.9 5.6 6.6
4 19 21 24 39.7 40.5 42.6 6.8 7.3 7.9
5 21 24 26 41.3 44.6 45.7 7.6 8.4 9.3

279 Finally, the performance of the simulation result is analyzed in the terms of the number turn, by which we can
280 able to minimize the energy consumption. The number of turn required for three different algorithms for number of
281 robots = 6 is demonstrate in Fig. 22. It shows that IGSA takes less number of turn than other two algorithms and energy
282 consumption to reach in the designation is less than the other two algorithms. The simulation is only presented for five
283 numbers of robots but number of turn is less for irrespective of the robot in the planning scheme of the algorithm.
284 The experiment is conducted in the environment shown in Fig. 4 by the three algorithms for same fitness function
285 in Eq. (29) with same parameter for 30 iteration, the best fitness value for three algorithms is presented in Fig. 23. The
286 fitness value of the robots presented in Fig. 23 indicates that there is no conflict in the next position calculation by the
287 robots, it shows that the best fitness value obtain for IGSA after 26 iteration is 3.638, but that achieved by GSA and
288 DE after 29 and 30 is 4.105 and 4.711 respectively. This presents that IGSA is better than GSA and DE in the terms
289 of avoiding problem at local optima and faster convergence rate.
290 Number of optimal steps reqiured for different robots, number from 1 to 5 of the simulation result for different
291 algorithm is presented in Table 3. Table 3 shows that the number of optimal steps required for IGSA is less than the
292 other algorithm such as GSA and DE. The total number of optimal steps required for IGSA, GSA and DE is 26, 29
293 and 30 respectively.
294 The result of the experiments performed is summarized in Table 2 in the terms of three performance metrics,
295 namely, (1) total no. of steps required to reach in the goal, (2) ATTPT and (3) ATTPD have been used here to determine
296 the relative merits of IGSA over the other algorithms for different robots. Table 1 confirms that the remaining two
297 algorithms perform well with respect to all three metrics for different robots.

298 7. Experiment on Khepera II robot

299 Khepera II (Fig. 10) is a miniature robot (diameter of 8 cm) equipped with 8 built-in infrared range and light sensors,
300 and 2 relatively accurate encoders for the two motors. The range sensors are positioned at fixed angles and have limited
301 range detection capabilities. The sensors are numbered clockwise from the leftmost sensor 0 to sensor 7 and its internal
302 structure (Fig. 12). Sensor values are numerical ranging from 0 (for distance > 5 cm) to 1023 (approximately 2 cm).
303 The on board microprocessor has a flash memory size of 256KB, and the CPU of 8 MHz. Khepera can be used on
304 a desk, connected to a workstation through a wired serial link. This configuration allows an optional experimental
305 configuration with everything at hand: the robot, the environment and the host computer. The Khepera II network and
306 its accessories are presented in Fig. 11 for the conduct of experiments.
307 The initial world map for conducting the experiment in the Khepera II is presented in Fig. 13 to 8 obstacles of
308 different shape and predefine initial state and goal is marked on the map, where different meta heuristic algorithm

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
14 P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx

Fig. 10. The Khepera II robot.

Fig. 11. Khepera network and its accessories.

Fig. 12. Position of sensors and internal structure of Khepera II.

309 is applied. Fig. 14 shows the intermediate moment of the robot in the trajectory path toward the goal by respective
310 robot using IGSA. IGSA is implemented in the Khepera-II robot with considering two robots and compared with a
311 different evolutionary computing algorithm is demonstrated in Fig. 15. It shows better convergence in comparing to
312 the other meta-heuristic algorithm presented in Fig. 15. Finally, different meta-heuristic algorithms have been applied
313 in Khepera environment and results of the trajectory path have been presented in Fig. 15.
314

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx 15

Fig. 13. Khepera environment setup for multi-robot path planning.

Fig. 14. Snapshot of intermediate stage of the multi-robot path planning using IGSA in Khepera environment.

Fig. 15. Optimal path representation of different algorithm for multi-robot path planning in Khepera environment is represented by different color
code.

315

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
+Model
JESIT 82 1–19 ARTICLE IN PRESS
16 P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx

Fig. 16. Average uncovered trajectory distance vs. number of stages with variable velocity for fixed number of obstacles = 7.

Fig. 17. Average uncovered trajectory distance vs. number of stages with variable number of robots for fixed number of obstacles = 7 (constant).

Fig. 18. Average total trajectory path traversed vs. number of robots with variable number of obstacles for fixed velocity.

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx 17

Fig. 19. Average uncovered trajectory target distance vs. number of steps in different algorithms.

Fig. 20. Average total trajectory path deviation vs. no. of robots algorithm with fixed no. of obstacles = 7.

316
Fig. 21. Run time vs. maximum iteration for different algorithms.

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
18 P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx

Fig. 22. Number of turn vs. number of robots in three different algorithms.

Fig. 23. Fitness value of IGSA, GSA and DE for fitness function in Eq. (29).

317 8. Conclusion and future works

318 An improved gravitational search algorithm was proposed for trajectory path planning of multi-robots in order to find
319 collision free smoothness optimal path from predefine start position to end position for each robot in the environment.
320 The obtained results from the experimental work perform better compared with the proposed algorithm. Comparing
321 the performances among different techniques have been carried out. From the simulation and Khepera-II environment,
322 it is observed that the IGSA technique is best over other technique for navigation of multi-mobile robot. However, in
323 this paper, both the environment and obstacles are static relative to the robots; where as other robots are dynamic for
324 priority robots. In future, work will be carried out using dynamic obstacles other than robots such as running vehicle,
325 animals and on board camera during multi-robot path planning.

326 References

327 Alba, E., Dorronsoro, B., 2005. The exploration/exploitation tradeoff in dynamic cellular genetic algorithms. IEEE Trans. Evol. Comput. 9 (2),
328 126–142 (IEEE).

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003
 +Model
JESIT 82 1–19 ARTICLE IN PRESS
P.K. Das et al. / Journal of Electrical Systems and Information Technology xxx (2016) xxx–xxx 19

329 Banerjee, G., Chowdhury, S., Losert, W., Gupta, S.K., 2011. Realtime path planning for coordinated transport of multiple particles using optical
330 tweezers. IEEE Trans. Autom. Sci. Eng. 9 (October), 669–678 (IEEE).
331 Bhattacharjee, P., Rakshit, P., Goswami, I., Konar, A., Nagar, A.K., 2011. Multi-robot path-planning using artificial bee colony optimization algorithm.
332 Proceedings of IEEE Congress on Nature and Biologically Inspired Computing, IEEE, 219–224.
333 Chakraborty, J., Konar, A., Jain, L.C., Chakraborty, U.K., 2009. Cooperative multi-robot path planning using differential evolution. J. Intell. Fuzzy
334 Syst. 20 (1–2), 13–27 (IOS Press, Amsterdam).
335 Chatterjee, Mahanti, G.K., Mahapatra, P.R.S., 2011. Generation of phase-only pencil-beam pair from concentric ring array antenna using gravitational
336 search algorithm. Proceedings of International Conference on Communications and Signal Processing, IEEE, 384–388.
337 Das, P.K., Konar, A., Laishram, R., 2010. Path planning of mobile robot in unknown environment. Int. J. Comput. Commun. Technol. 1 (2–4), 26–31
338 (Inderscience).
339 Das, S., Mukhopadhyay, A., Roy, A., Abraham, A., Panigrahi, B.K., 2011. Exploratory power of the harmony search algorithm analysis and
340 improvements for global numerical optimization. IEEE Trans. Syst. Man Cybern. B 41 (1), 89–106 (IEEE).
341 Das, P.K., Pradhan, S.K., Patro, S.N., Balabantaray, B.K., 2012. Artificial immune system based path planning of mobile robot. Soft Comput. Tech.
342 Vis. Sci. 395, 195–207 (Springer).
343 Eldos, T., Qasim, R.A., 2013. On the performance of the gravitational search algorithm. Int. J. Adv. Comput. Sci. Appl. 4 (8), 74–78 (The Science
344 and Information Organization).
345 Geem, Z.W., Kim, J.H., Loganathan, G.V., 2001. A new heuristic optimization algorithm: harmony search. Simulation 76 (2), 60–68 (Sage).
346 Gong, C., Lincheng, S., 2001. Genetic path planning algorithm under complex environment. Robot 23 (1), 40–43 (Hindawi).
347 Guo, Y., Parker, L.E., 2002. A distributed and optimal motion planning approach for multiple mobile robots. Proceeding of IEEE International
348 Conference on Robotics and Automation (ICRA’02), IEEE vol. 3, 2612–2619.
349 Kcymeulcn, D., Decuyper, J., 1994. The fluid dynamics applied to mobile robot motion: the stream field method. Proceedings of IEEE International
350 Conference on Robotics and Automation, IEEE, 378–385.
351 Konar, 1999. Artificial Intelligence and Soft Computing: Behavioral and Cognitive Modeling of the Human Brain, 1st edition. CRC Press.
352 Li, C., Bodkin, B., Lancaster, J., 2009. Programming Khepera II robot for autonomous navigation and exploration using the hybrid architecture.
353 Proceedings of 47th Annual South East Regional Conference, ACM, p. 31.
354 Liu, S. Hsi, Mernik, M., Hrnclc, D., Crepinsek, M., 2013. A parameter control method of evolutionary algorithms using exploration and exploitation
355 measures with a practical application for fitting Sovova’s mass transfer model. Appl. Soft Comput. 13 (9), 3792–3805 (Elsevier).
356 Luh, G.C., Cheng, W.-C., 2002. Behavior-based intelligent mobile robot using immunized reinforcement adaptive learning mechanism. Adv. Eng.
357 Inform. 16 (2), 85–98 (Elsevier).
358 Masehian, E., Sedighizadeh, D., 2010. Multi-objective PSO- and NPSO based algorithms for robot path planning. Adv. Electr. Comput. Eng. 10
359 (4), 69–76 (Stefan cel Mare University of Suceava, Faculty of Electrical Engineering and Computer Science).
360 Precup, R.E., David, R.C., Petriu, E.M., Preitl, S., Radac, M.B., 2012. Novel adaptive gravitational search algorithm for fuzzy controlled servo
361 systems. IEEE Trans. Ind. Inform. 8, 791–800 (IEEE).
362 Purcaru, C., Precup, R.-E., Iercan, D., Fedorovici, L.-O., David, R.-C., Dragan, F., 2013. Optimal robot path planning using gravitational search
363 algorithm. Int. J. Artif. Intell. 10, 1–20 (CESER).
364 Rashedi, E., Nezamabadi-pour, H., Saryazdi, S., 2009a. GSA: a gravitational search algorithm. Inf. Sci. 179 (13), 2232–2248 (Elsevier).
365 Rashedi, E., Nezamabadi-pour, H., Saryazdi, S., 2009b. GSA: a gravitational search algorithm. Inf. Sci. 179, 2232–2248 (Elsevier).
366 Rashedi, E., Nezamabadi-pour, H., Saryazdi, S., 2010. BGSA: binary gravitational search algorithm. Nat. Comput. 9, 727–745 (Springer).
367 Regele, R., Levi, P., 2006. Cooperative multi-robot path planning by heuristic priority adjustment. Proceeding of IEEE/RSJ International Conference
368 on Robotics System, IEEE, 5954–5959.
369 Sabri, N.M., Puteh, M., Mahmood, M.R., 2013. A review of gravitational search algorithm. Int. J. Adv. Soft Comput. 5 (3), 1–39 (SCRG).
370 Sarafrazi, S., Nezamabadi-pour, H., Saryazdi, S., 2011. Disruption: a new operator in gravitational search algorithm. Sci. Iran. 18 (3), 539–548
371 (Elsevier).
372 Tuncer, A., Yildirim, M., 2012. Dynamic path planning of mobile robots with improved genetic algorithm. Comput. Electr. Eng. 38 (6), 1564–1572
373 (Elsevier).
374 Verma, O.P., Sharma, R., Kumar, M., Agrawal, N., 2013. An optimal edge detection using gravitational search algorithm. Lect. Notes Softw. Eng.
375 1 (2), 148–152 (Springer).
376 Yang, X.-S., 2009. Harmony search as a metaheuristic algorithm. In: in Geem, Z.W. (Ed.), Music-inspired Harmony Search Algorithm: Theory and
377 Applications, Studies in Computational Intelligence, vol. 191. Springer, Berlin, pp. 1–14.
378 Yu, J., Kromov, V., 2001. A rapid path planning algorithm of neural network. Robot 23 (3), 201–205 (Hindawi).
379 Zhang, Y., Gong, D.-W., Zhang, J.-H., 2013. Robot path planning in uncertain environment using multi-objective particle swarm optimization.
380 Neurocomputing 103, 172–185 (Elsevier).

Please cite this article in press as: Das, P.K., et al., Multi-robot path planning in a dynamic environment using improved
gravitational search algorithm. J. Electr. Syst. Inform. Technol. (2016), http://dx.doi.org/10.1016/j.jesit.2015.12.003