Arabian Journal for Science and Engineering

https://doi.org/10.1007/s13369-021-05487-w

RESEARCH ARTICLE-MECHANICAL ENGINEERING

GWO‑Potential Field Method for Mobile Robot Path Planning

and Navigation Control
Mohd. Nayab Zafar1 · J. C. Mohanta1 · Anupam Keshari2

Received: 21 September 2020 / Accepted: 18 February 2021

© King Fahd University of Petroleum & Minerals 2021

Abstract
A navigation control methodology is proposed in this paper, which utilizes a hybrid concept of grey wolves optimisation
(GWO) algorithm and an artificial potential field (APF) method for on-time path planning of mobile robot. The proposed
methodology runs in two folds. The first fold defines the focus region (FR), which shows the obstacle-free locations of the
all possible robot movements. However, in the second fold step GWO algorithm searches the shortest path by minimizing the
artificial potential field value of the location generated within FR. The presented methodology was simulated and verified in
various real and simulation environments. The simulation and experimental results reveal that the presented methodology
can not only be capable of finding an optimal or near-optimal robot-path in a complex obstacle-present environment but also
provides an effective path planning strategy on-time basis. The results show that the proposed method has out-performed
to shorten the path length as well as ensured collision-free navigation. At the same time, virtual intermediate targets (VITs)
makes the navigation free from any dead-end situation, even in a cluttered environment.

Keywords  Mobile robot navigation · Grey wolf optimization algorithm · Artificial potential field · Collision avoidance

1 Introduction been an emerging area. Many conventional, as well as AI

techniques, have been utilized to tackle this problem [1–3],
The motion planning of mobile robots has been a lucra- such as fuzzy-logic techniques [4], optimization algo-
tive research issue among the engineers and scientific rithms [7], artificial potential field (APF) techniques [5,
community because of its broad applicability in social, 6], artificial neural networks (ANN) approaches [8, 9] and
manufacturing and non-manufacturing practices. Due to even conglomeration of hybrid techniques. Each method/
the advancements in digital electronics, computer tech- techniques have their advantages and disadvantages com-
nology and their ability to integrate with Artificial Intel- pared to others in certain aspects. The main problems for
ligence (AI) techniques, motion planning became more mobile robot motion planning are computational-complex-
autonomous and programmable. For navigating the mobile ity, adaptability and optimal selection of navigational path,
robot in obstacle-present circumstances, it is necessary to which ensures successful and collision-free navigation by
find a suitable collision-free path from the start location to avoiding the situations like a dead-end. The scientific com-
the target location. During the sequence of robot motion, munities have always been searching for an alternate and
the navigation path is much desirable to be optimal/near- more effective method to tackle these problems.
optimal with time, distance, energy and smoothness. From It is noticeable that the path planning can be considered as
which the minimal length is most desirable criteria. Since an optimization problem (i.e. in terms of shortest distance)
the last four decades, mobile robot motion planning has over certain limitations (i.e. particular environmental-sce-
narios and collision-free condition). Mobile robot path plan-
ning optimization (Problem) can be broadly categorized into
* J. C. Mohanta two groups, i.e. (1) Optimization to obtain the pre-defined
[email protected] shortest path, (2) Optimized path extension on-time basis.
1 The proposed methodology is an example of optimized
Department of Mechanical Engineering, MNNIT-Allahabad,
Prayagraj, India path extension on-time basis. The methodology deals with
2 the optimization instances wherein, on-time path step
National Institute of Industrial Engineering, Mumbai, India

13
Vol.:(0123456789)
 Arabian Journal for Science and Engineering

decision is done by following optimized path extension complex scenario. They demonstrated simulation results
concept. The path extension decision has to be done at each to validate their proposed model. Tsourveloudis et al. [20]
step; thus, the quick decision making capability is one of described the electrostatic potential field-based path planner
the crucial aspects of such methodology. A limited work with double layered fuzzy-inference system and tested their
is noticed for navigation control of mobile robot based on model using “Nomad 200” robot experimentally.
an on-time path planning strategy. However, in majority of Some authors [21, 22] have presented a new scheme
researches, computation time for decision making is further through simulation experiments, based on improved APF
required to be dealt with. Some of the related researches for autonomous navigation of a mobile robot. Huang et al.
with their advantages and limitations are mentioned here for [23] discussed a novel methodology for vision-guided local
path planning of mobile robots. navigational system, based on human locomotion strategy.
Khan et al. [10] discussed the determination of optimal- The target seeking uses the relative heading angles with the
path from a specified initial position of a static/dynamic sce- goal and obstacles, their distances and the angular projection
nario using quadrant-based random particle optimization. of obstacles in order to compute the potential field.
Paniagua et al. [11] presented a multi-objective variable The collision-free navigation of autonomous mobile robots
neighbourhood search algorithm for robot path planning. has been described by Gharajeh et al. [24]. They have com-
Parhi et al. [12] developed a navigation control of multiple bined the advantages of global positioning system and adap-
mobile robots based on petri-potential-fuzzy (PPF) con- tive neuro-fuzzy inference system for finding a collision-free
troller. However, Mohanta et al. [7] proposed the hybrid path. This hybrid technique gives a shorter collision-free
approach capable of finding a near-optimal robot-path for path to reach the goal in comparison with fuzzy and neuro-
multi-robot with the multi-target system in obstacles prone fuzzy-based methods. Ajeil et al. [25] combined the particle
circumstances. swarm optimization and modified frequency bat (PSOMFB)
Tsuji et al. [13] discussed a novel trajectory-generation techniques integrated with local search and obstacle detec-
technique which helps in the control of transient behav- tion features. This technique solves the problem in three steps.
iour (i.e. time-to-target and velocity-profile) using APF for Firstly, the hybrid technique minimizes distance and follows
real-time navigation of mobile robots. Oishi et al. [14] pro- path smoothness criteria. Secondly, novel Local Search (LS)
posed a new view-based localization method SeqSLAM++, algorithm integrated with the hybrid PSOMFB algorithm
which can estimate the robot location by comparing image converted into feasible solutions. The third module features
sequences. The SeqSLAM++ also cope with the changes obstacle detection and avoidance (ODA), within the sensing
in robot’s heading angle and speed through view changes region. Lazarowska [26] discussed discrete artificial poten-
using wide-angle images and a Markov localization scheme. tial field (DAPF) method integrated with a path optimiza-
Wachter et al. [15] demonstrated a video for high-speed, tion algorithm for finding a smooth and shorter path in static/
dynamic, non-holonomic robots based on APF methods. dynamic environment. They utilize the discrete potential field
The video includes the experimental test-bed with a fleet of algorithm for collision-free motion and demonstrated the fea-
inexpensive four-wheel drive (skid-steered) robot named as sibility and effectiveness by comparing with an ant colony
“Dynabots” which can run up to a speed 10 m/s. optimization method in terms of the path length and run time.
Besides the above-mentioned researches, there are some Wahab et al. [27] discussed the comparative study of classical
researches which utilize APF method to regulate the mobile technique and meta-heuristic techniques. This discussion pro-
robot navigation on-time basis. In such type of literature vides a detailed comparison between meta-heuristic techniques
series, Montiel et al. [16] used the APF with bacterial evo- against the set of well-known classical techniques. The total
lutionary algorithm (BEA) to have an enhanced flexible path travel time/distance, number of collisions, energy consump-
planner, while utilizing the advantages of APF method by tion and displacement errors are taken into account to exam-
considerably minimising its drawbacks. ine the feasibility of the techniques. Among these techniques,
Jean et al. [17] have proposed a methodology based on Dijkstra’s algorithm (DA) is considered a benchmark solution
APF method which eliminates the problems of motor schema and constricted particle swarm optimization (CPSO) is found
especially in trapped situation due to existence of local-min- performing better than other meta-heuristic approaches in
ima, and oscillations in the presence of narrow passages. The unknown environments.
similar problem (i.e. inherent oscillation in closely packed In all such studies, it has been found that there are very few/
obstacles and in narrow passages) based on potential field limited research is shown to tackle the problem like main-
methods (PFMs) has been reported by Jing et al. [18].They taining a consistent gap between the robots (in case of mul-
validated their approach through performance comparison tiple robot navigation) and between robot and obstacles. In
with other potential field approaches by altering the param- this paper apart from the criteria like shortest distance, run/
eters. A decentralized traffic control system using APF was computational time, optimal/near-optimal path, collision-free
demonstrated by Masoud [19] for each mobile agent in a motion, target seeking and smoothness, the attention has also

13
Arabian Journal for Science and Engineering

been emphasized to focus the issue of maintaining a consist- The inputs to the proposed navigational controller are Left
ent gap between obstacles and other robots. The grey wolf Obstacle Distance (LOD), Front Obstacle Distance (FOD),
optimization (GWO) technique utilized here which ensures Right Obstacle Distance (ROD), Local Heading Angle (LHA)
the on-time navigation and maintains a consistent gap between and Global Heading Angle (GHA). The sign convention of
robot and obstacles as well. heading directions is ‘+ve’, ‘zero’ and ‘−ve’ in relation to the
In this research, grey wolf optimization (GWO) technique reference line, is represented by clockwise, straight and anti-
is used as an optimization tool to search out an on-time move- clockwise orientation, respectively.
ment plan, wherein a newly defined artificial potential field On the basis of LOD, FOD, ROD and heading direction,
paradigm is integrated to implement the target-seeking and navigation controller monitors local heading angle (LHA) of
obstacle avoidance behaviour. The GWO is to extend the on- the mobile robot. The correct selection of the local heading
time path network for ensuring the shortest navigational refer- angle (LHA) is responsible for obstacle avoidance, while speed
ence path. This path planning control finds the shortest path of the robot is uninfluenced. A navigational instance of mobile
by minimizing the distance from the target and the robot’s robot locomotion is shown in Fig. 1, where robot sensors uti-
shortest distraction under the influence of target seeking and lized to detect the immediate obstacles present in left, front
obstacle avoidance behaviour. The newly defined artificial and right directions, respectively.
potential field paradigm ensures the appropriate heading angle At the start of navigation, robot sensors first project the
with respect to target seeking and obstacle avoidance inclina- signals over the obstacles in all three directions (i.e. left, front
tion together. and right).It is worthy to mention that, out of all projected dis-
The structure of this manuscript is as follows: Sect. 2, dis- tances in each direction the most minimal distances are used
cusses the mobile robot path planning issues, sensing systems, for computing the data. The straight-line joining the start and
data collections and processing. The proposed methodology target location is termed as reference line, from where the
is illustrated in Sect. 3, which includes the GWO algorithm, heading angles are measured. During navigation, the robot
newly defined artificial potential field paradigm and GWO- has to pass through multiple obstacles with various shape and
Potential field technique and its implementation for real-time size in order to reach the target, while avoiding the obstacles
path planning. Sect. 4 illustrates the implementation of the as shown in Fig. 1. The origin of mobile robot is considered at
proposed methodology for navigation control of the mobile mid-point of the line joining through robot’s centre of gravity
robot. Sect. 5 demonstrates the carried-out simulations and (CG) and the fore-head axis.
experiments. Six case-studies with various complexity lev-
els of obstacle environments are described, and performance
comparison of GWO-Potential field method is presented in 3 GWO‑Potential Field Method: Proposed
this section. The obtained results with discussions in view of Methodology
benefits and limitations are composed in Sect. 6. Section 7
concludes the research interest, outcomes and future scope. In this control methodology, the advantages of GWO-
Potential field method have been utilized for autonomous
navigation of mobile robot. Herein, optimized influences of
2 Problem Definition: Mobile Robot Path distance (between robot and target) and resultant artificial
Planning potential fields (induced by obstacles and target) are the
basis to decide the robot movement on-time basis. Accord-
The path planning strategy for the robot presented in the study ing to the methodology, the robot is assumed to be a positive
deals with the rear wheel drive mobile robot, which can readily point charge which is attracted by the target (negative point
move in any direction with any arbitrary angle. The mobile charge) and repelled by the obstacles (positive point charge).
robot is mounted with sensing devices which avails the real- The vector sum of the pulling force and the repulsive forces
time information about the presence of obstacles as well as the regulate the path-planning at every instance of the motion
distances of obstacles from the robot. The sensors detect the [28].
obstacles from left, right, and front to their respective distances
from the robot centre (i.e. origin of robot). 3.1 Grey Wolf Optimization Algorithm
The proposed mobile robot decides its movement and
path on-time basis. It explores the movement step which pro- Grey Wolf Optimization algorithm is a population-based
vides the possibility of minimum path. In each exploration, it meta-heuristic approach, (inspired by hunting behaviour
searches a path segment and follows the same within the influ- of grey wolves) proposed by Mirjalili et al. [29]. Recent
ences of the potential field projected by Right Obstacle (RO), practical applications of GWO [30–32] reveal its benefits
Left Obstacle (LO), Front Obstacle (FO) and Local Target at implementation stage like avoidance of local optima,
Point (LTP). quick search capability in small search space and less

13
 Arabian Journal for Science and Engineering

Fig. 1  The instantaneous navi-

gation scenario in the presence
of obstacle and acquired real-
time information

computational burden, etc. The methods can be implemented grown old. The beta wolf must respect the alpha wolf,
both in binary/decimal encoders based on position of the but can command the other lower level wolves. Beta can
grey wolf during hunting mechanism. This method can be advise the alpha at the time of need and maintain dis-
very effective for distance optimization (in small search cipline among the pack. The beta wolf re-inforces the
space) problems, which can be an essential part for optimiza- alpha’s commands among the pack and communicates the
tion of mobile robot path planning problems on-time basis. feedback to alpha. Delta wolves obey the commands of
The GWO algorithm is inspired by the hunting behaviour alphas and betas, but they lead omega wolves. The lowest
and social leadership phenomena which occurs naturally in rank grey wolves are known as omega (which plays the
a grey wolf group. Grey wolves are found living in groups role of scapegoat), responsible for viewing the territorial
at an average size of 5–12 wolves/group. One interesting boundary and warns the group in case of danger. They are
thing that exists in their socially living behaviour is that they the last level wolves in hierarchy to eat the prey.
strictly follow the social dominant hierarchy. In hierarchy, The main activities happen in a grey wolf hunting event
the top-level wolf member is called alpha (α), second level are as follows (Fig. 2).
member is named as beta (β), subsequently third and fourth
level member are called delta (δ) and omega (ω), respec- • Tracking and chasing the prey
tively. The alpha wolf works as the leader and takes full • Encircling and harassing the prey till it stops moving
responsibility for decision making about hunting, sleeping, • Carefully moving towards the prey.
waking time and so on. The alpha wolf is also called the
dominant wolf, since his/her orders should be followed by It is assumed that, the alpha (is best candidate solution)
the pack. It is an important fact that, the alpha wolf is not have the complete knowledge about the potential position
essentially the strongest member of the group but the best in of prey. The best solutions obtained during hunting are
terms of managing the wolves attack on the prey. recorded and communicated to other search agents (beta,
The betas are known as subordinate wolves that helps delta and omegas) to update their positions.
the alpha in decision making or other pack activities dur- The concerned computational scenario (quick decision
ing hunting. The beta might be either male or female, before every single movement) requires an optimization
and he/she is the best candidate to take over the charge technique which can quickly search the best path-step out
of alpha, in case one of the alpha wolves passes away or of all probable steps. In such scenario, the search space is

13
Arabian Journal for Science and Engineering

small and well defined, but computational time ensured to Phase 3: when the searching and encircling phase is
be as quick as possible. In a real-time mobile robot path accomplished, then their next phase is hunting. The loca-
planning problem, GWO technique provides advantages in tion update of wolf pack is mathematically modelled as:
terms of less computational burden, faster rate of conver-
gence in short search space and less computational time 𝜆𝛼 = ||𝜇 ∙ Q𝛼 (s) − Gw (s)||
requirement. Further, GWO converges at a much faster | |
𝜆𝛽 = |𝜇 ∙ Q𝛽 (s) − Gw (s)|
| | (5)
rate than other artificial intelligence techniques [29]. The
𝜆𝛿 = ||𝜇 ∙ Q𝛿 (s) − Gw (s)||
result insures quick decision ability during every single
step movement of the mobile robot.
The mathematical model of GWO is as follows: Gw1 = Q𝛼 (s) − 𝜂1 ∙ 𝜆𝛼
Gw2 = Q𝛽 (s) − 𝜂2 ∙ 𝜆𝛽 (6)
Phase 1: This is basically a searching phase in which the
Gw3 = Q𝛿 (s) − 𝜂3 ∙ 𝜆𝛿
grey wolves strongly follows the social hierarchy during
searching of quarry/prey.
Phase 2: This phase is named as encircling phase, when Gw1 + Gw2 + Gw3
Gw (s + 1) = (7)
the searching process is complete their next step is encir- 3
cling the quarry. The representation of encircling the Phase 4: This final phase decides the pack weather to
quarry is as follows: attack or search for other fitter quarry to fulfil the desired
𝜆 = ||𝜇 ∙ Q(s) − Gw (s)|| (1) hunger of the entire pack.

i. If |η|≥ 1 search another quarry (global search)

Gw (s + 1) = Q(s) − 𝜂 ∙ 𝜆 (2) ii. If |η|< 1 finally attack (optimal solution).

𝜂 = 2xv1 − x (3)
3.2 A Newly Defined Artificial Potential Field
𝜇 = 2v2 (4) Paradigm
where λ = Interspace between wolves and prey;
Q = Quarry/prey location/position; Gw = grey wolves’ The mobile robot navigation using potential field-based
location/position; η & μ = coefficient vectors; x = lin- approach was first introduced by Khatib [28] in the year
early decrease from 2 to 0 with increase in number of 1979. However, many recent papers on mobile robot path
iterations; v = random vectors (i.e. 0, 1); s = number of planning utilize the advances of PF technique for smooth
iterations. navigation and collision free robot motion [19, 34, 35].
The potential field technique is rapidly gaining popular-
ity in navigation and obstacle avoidance applications for
mobile robots because of its elegant mathematical analy-
sis, high efficiency and simplicity. The artificial poten-
tial field (APF) technique is a motion planning algorithm
highly appreciated to solve the path planning problems of
mobile robot from a start location to the target location
through a collision-free trajectory.
In this approach, the robot in the configuration space
is represented as a charge particle under the influence of
an artificial potential field. It is based on the metaphor
that, the target influences an attracting force towards
robot, whereas obstacles repel the robot from them. The
potential-function can be defined over free surface as the
summation of the attractive and repulsive potential, which
decides the steering of motion of the robot. The attractive
potential function is also used for target seeking of mobile
robot.
Fig. 2  Location of Grey Wolf with respect to Prey during a hunting
The APF technique may be a continuous/discrete
incident [29] form and can be best suited for both on-line and off-line

13
 Arabian Journal for Science and Engineering

navigation. It requires less computation due to its simple Difference Angle (θ): it is the angle between the resultant
mathematical calculations (for computing the function to force direction and instant heading direction.
represent the potential field, and the desired direction for [ ]
the agent at various positions on the potential field). F���R⃗ ⋅ d�⃗
−1
In ideal case, the mobile robot should reach the target
𝜃 = cos (9)
|������ ⃗| ����⃗
|FR | ⋅ |d|
in a straight line by heading exactly towards the target.
However, to avoid collision and to achieve shortest dis- d�⃗ = Instant movement vector (line meeting the current loca-
tance, heading angle has to be adjusted accordingly. For tion and the random point location).
that a decision function is defined as such, the influence The influence of N number of objects can be counted to
of distance and heading angle can be utilized to decide the decide the step movements. In this case (Fig. 3), the only three
instantaneous step movement. By minimizing the objec- closest obstacles are taken into consideration (i.e. N = 3). The
tive function, GWO technique works to decide next step closest obstacles sensed in Right, Left and Front directions,
movement of the mobile robot. The decision function is: and their distances are represented as x1, x2 & x3, respectively.
It is assumed that mobile robot and the obstacles (RO,
Decision function = w1 ⋅ (D) + w2 ⋅ |𝜃| (8)
LO & FO) are alike the positive charges, thus mobile robot
w1 and w2 are the weight factors, which assigns the prior- experience a repulsive force from each of the obstacles.
ity weightage to the associated factors, respectively, i.e. the Moreover, the target is similar to a highly negative charge
distance between robot’s instant position and target posi- particle, which attracts the mobile robot continuously. The
tion, (D), and the angle between the resultant force direction vector summation of repulsive and attractive forces provides
and instant heading direction, (θ). These weights assign the a resultant force direction, which drive the robot towards
importance to the factors for deciding the movement step. In target in a collision free path in each step motion.
the presented case studies, we are giving more importance to
���R⃗ = F
F �⃗1 + F
�⃗2 + F
�⃗3 + F
�⃗target (10)
find a shorter path length. Thus, the weight assigned to the
distance factor (w1) is kept larger than the weight factor asso- �⃗1 = Repulsive Force felt by the mobile robot from the right
F
ciated with heading direction (w2). Wherein w1 = 1, w2 = 0.3.
obstacle (RO);
The weights factors also play a role to balance the contri-
bution of both the factors in deciding the movement step. ���⃗1 =
qMR ⋅ qRO
The weights are mainly affected by the ratio of the distance
F
4𝜋𝜀◦ (x1 )2 (11)
between the start and goal position and focal angle. Some
�⃗2 = Repulsive Force felt by the mobile robot from the left
F
experimental trials are done to decide appropriate weights.
Before every new step movement decision, a number of ran- obstacle (LO);
dom points are generated in the focus region. If the numbers of qMR ⋅ qLO
���⃗2 =
F (12)
randomly generated point increases, the path smoothness will 4𝜋𝜀◦ (x2 )2
also increase and vice-versa. However, at the same time it will
increase the processing time too to decide the robot movement. �⃗3 = Repulsive Force felt by the mobile robot from the front
F
A step movement instance of the mobile robot and representa- obstacle (FO);
tion of potential field forces along with focus region of random qMR ⋅ qFO
point generation are shown in Fig. 3. ���⃗3 =
F (13)
4𝜋𝜀◦ (x3 )2
Focus Region: it is the area in front of the mobile robot (area
within the minimum and maximum focus radius) wherein a The charge density coefficient,
point location is chosen during navigation to move the robot. a1 =
Total distance between start point & target point
.
average step length
Before each step of motion, random points are generated λ = distance & force adjusting coefficient (0 > λ ≥ 0.2).
within the focus region and allow the robot to move through �⃗target = Attraction force felt by the mobile robot towards
F
the selected point. the target
Average step length (S): average distance moved by the
robot in one step (for the case studies, S = 6 cm). �⃗target =
(a1 ⋅ q) ⋅ qMR
Inner radius to define focus region (R1): depends upon sens-
F ; (i ≥ 1) (14)
4𝜋𝜀◦ (|D|)2 (e𝜆i )
ing limit and the robot speed/average step length.
Outer radius to define focus region (R2): depends upon
sensing limit and the robot speed/average step length. i = step number (when Robot is beginning its motion:
Focal Angle (φ): Focal Angle describes the angular extent i = 1)
of the focus region at robot’s centre (Fig. 3). qMR = Charge defined at the mobile robot (positive)
qFO = Charge of the front obstacle (positive)

13
Arabian Journal for Science and Engineering

Fig. 3  A step movement

instance of mobile robot show-
ing the representation of poten-
tial field forces along with the
random point generation region

qLO = Charge of the left-hand obstacle (positive) 3.3 Performance Analysis of Newly Defined

qRO = Charge of the right-hand obstacle (positive). Potential Field Paradigm

We assume that; qMR  =  qFO  =  qLO  =  qRO  =  q  , and The newly defined potential field paradigm is applied for a
qtarget = q ⋅ a1. mobile robot step movement in two different test platforms
The proposed GWO-Potential Field Method demon- (in simulation mode) as shown in Figs. 4 and 5. This is a
strates the combined use of GWO and Artificial Potential separate test of potential field paradigm without the involve-
Field method. The GWO algorithm helps to decide the ment of the GWO algorithm. These tests are carried out
shortest distance with a minimum computational time (due for analysing the potential field variation (i.e. force varia-
to its first convergence rate), while Potential Field methods tion) and the robot’s heading direction deviations during its
help the robot to reach the target (as per Eq. 14) by avoid- motion along the local reference line (LRL). There are some
ing obstacles. In each instantaneous situation, the sensed parameters (a, λ) which are defined through the experimental
distances of the three closest obstacles, instantaneous dis- trials. These parameters directly affect the magnitude and
tance from the obstacle, possible movement region are the direction of the resultant force. Thus, a separate realization/
basis to decide the instantaneous robot motion. demonstration (without GWO involvement) of the resultant
force along with the target distance makes it essential to
realize potential field variations.
The potential field performance analysis was carried
out while robot moving along the Local Reference Line,
Potential field (total force magnitude) variation while its

13
 Arabian Journal for Science and Engineering

movement from instant location to end point, and heading robot moves ahead. Robot updates its location in each step
direction deviation from the local reference line for two movement and memorizes the path during navigation. Same
different situations are shown in Figs. 4 and 5. The results steps are repeated till the robot reaches the endpoint (target).
obtained are quite relevant and show full proof evidence in The implementation of GWO-Potential Field method was
terms of deviation angle and total force magnitude derived carried out using MATLAB®based coding, and the case
from the newly defined potential paradigm. studies for selected environments are demonstrated below
to verify the robustness of the controller.

4 Implementation of Proposed
Methodology for Mobile Robot 5 Case Studies
Navigation Control
5.1 Case Studies Comparing the Simulation
The implementation of GWO-Potential Field method is and Experimental Runs
briefly described in this section. The flow chart shown in
Fig. 6 depicts the various steps of the algorithm which are To validate the efficacies of the proposed control method-
sequentially followed by GWO-Potential Field controller to ology, MATLAB® simulations and real-time experiments
incorporate the navigation. The step to step movement-path are conducted by considering the different obstacle prone
is generated on-time basis, and navigation is done accord- environments using prototype model. The path followed by
ingly. The path concerning minimum objective value for the robot during real-time experiments was marked on the
each movement-step is searched by GWO technique. The floor by means of a pen attached on the rear side of the
start point and end point for the chosen environment, vir- robot frame, and traced path length was measured in ‘cm’.
tual intermediate target point (VITP) locations and their The path length in simulation environment was measured
sequence, sensing radius, focus radius and the parameters by counting the pixel along the path of the robot. The path
related to the potential field method are pre-stored in the of the robot was developed by the proposed methodology
database. in simulation and real-time experiments for three different
Virtual intermediate target generation is an optional facil- cases of obstacle prone environments which are shown in
ity, which depends on the user’s interest and is preferably Fig. 7a1–c2. The straight-line path, experimental path and
defined when the robot path is long & complicated. VITs simulation path lengths were measured for the environments
encompass high significance to simplify the robot path concerned. The maximum heading angles and time taken to
planning. It is generated between Start Point and End Point. reach the target were also measured for the experiments, and
When VIT(s) are defined, it is assumed that long & compli- results are shown in Table 1.
cated path will be travelled into several pieces. In the first Simulation model and experimental platform used
piece of movement, the robot travels between Start Point for three dissimilar environments (C-shaped continuous
and first VIT, and then it travels between successive VITs obstacle, S-shaped continuous obstacle and rectangular
sequentially. The final piece of motion is performed between plate type arranged obstacle) are defined as follows: navi-
the last VIT and Endpoint. If VIT is not defined, then robot gational space = 400 × 400 pixels; Global Heading Angle
target is to move for End point directly. Defining the Virtual (GHA) = zero degree (robot’s first heading angle along ref-
intermediate target points is an optional feature only. How- erence line); robot dimension = 8 × 6 cm rectangular shape;
ever, many times users are informed with some crucial point Robot’s reference line is passed through the CG (Centre
locations in an unfamiliar environment even. In such cases, of Gravity) of the robot on which the fore-head sensor is
VIT concept can be utilized, and the concerned advantages mounted. Robot’s average step length = two pixels in simula-
can be taken. In the programming structure, the VIT points tion scenario; 1 pixel = 3.00 × 3.00 ­cm2 area; highest robot
can be stored along with the start location and end location speed = 10 cm/s; real-time experiment area = 120 × 120 ­cm2;
coordinates. Simulation area (Extended C-shaped continuous obstacle,
At any instant, once the robot identified the FO, LO, RO S-shaped continuous obstacle and rectangular plate type
and its distances and recognized the current focus region arranged obstacle) = [400  pixels, 400  pixels]; Number
(free from obstacles), then there are three sequential steps to of randomly generated location nodes = 20. The robot is
plan the path, i.e. (i) defining the random points within the mounted with three IR sensors; one at the front, one on left
focus region (probable search area); (ii) calculation of deci- side and one on right side to get the information about the
sion function values related to the random points and (iii) obstacles in all three directions. The range of the sensors is
searching a locational point which returns minimum deci- about 40 cm.
sion function value. Random point offering the minimum
decision function value is the location through which the

13
Arabian Journal for Science and Engineering

Fig. 4  Potential field perfor-

mance analysis while robot
in moving along the Local
Reference Line; a Test plat-
form-1 layout, b Potential field
variation while movement from
instant location to end point,
c heading direction deviation
from the local reference line

13
 Arabian Journal for Science and Engineering

Fig. 5  Potential field perfor-

mance analysis while robot in
moving along the Local Refer-
ence Line; a Test platform-2
layout, b Potential field (total
force magnitude) variation
while its movement from instant
location to end point, c Heading
direction deviation from the
local reference line

13
Arabian Journal for Science and Engineering

Fig. 6  Flow chart for GWO-

potential field method imple- Data Base
mentation Start position, target position or
VITs with their sequence,
sensing radiuses (R1 and R2), and
robot speed.

Start

Trace the focus region

(Region between R1 and R2, as shown in Fig. 3)

Allocate the random points in focus region (analogous to

initialize the grey wolf’s sample vectors)

Measure the decision function (potential field-based decision function; Eq. 8) value
associated with random points (analogous to the calculationof fitness)

Run the GWO Algorithm to minimize the

objective function and decide the move

Updates/store the navigation step

Robot Movement Command

Yes If |Actual Target-

Immediate
Target|>0?

No

Stop

5.2 Case Studies Showing the Proposed with different navigational parameters (i.e. maximum turn-
Methodology Performances with other Models ing angles, average turn angle, path length and the distance
in some Complex Simulation Scenarios (minimum) between obstacle & path). The performance
results are compared in Table 2. The GWO-Potential Field
Furthermore, some experimental scenarios are taken from controller’s results are graphically represented in Map 8
the work presented by Han and Seo [33] and proposed meth- ­(a2, ­b2, ­c2, ­d2, ­e2, ­f2) for Environment-1 to Environment-6,
odology is implemented to demonstrate the similar situa- respectively.
tions. The efficacy of GWO-Potential Field controller is In case study-4 (Fig. 8a1), it can be noticed that, the robot
compared with the control methodology presented by Han makes many sharp turns during its movement while con-
and Seo [33]. The performances of these two approaches are trolling through Han and Seo [33] methodology. During
evaluated on the basis of average turn angle to decide next the course of navigation, robot takes intermediate change
step-definition, path length and the consistent gap with the in heading angle near the corners. These sudden changes
obstacles. The six complex scenarios (Environments: 1–6) of robot step-movement are in reality a typical job for any
have been selected from Han and Seo [33] results for the mobile robot controller in actual practice. With the use
comparison and are shown in Fig. 8a1–f2.The similar sce- of GWO-Potential Field method, it is experimented that,
narios were created in simulation environment and utilized the generated-path is quite smooth at the corner and the

13
 Arabian Journal for Science and Engineering

Fig. 7  a1–c2 Case studies in

both simulation and experimen-
tal environments showing the
mobile robot movements in dif-
ferent obstacle prone scenarios

consistency in gap between the robot path and boundary was methodology encourage developing a real-time navigation
maintained during the navigation. It is further observed that, controller.
the on-time path network for ensuring the shortest navigation In case study-5 (environment-2), a C-type obstacle envi-
path-strategy employed in the robot controller can confirm ronment is selected in which start position is inside the
the smooth movement in any type of cluttered map. Fur- C-type obstacle and the targeted position at the bottom end
thermore, the leading advantages found from the proposed corner of the environment layout. The robot moves straight
towards the small opening passage of C-type obstacle, then

13
Arabian Journal for Science and Engineering

Table 1  Comparison of results obtained from MATLAB® Simulation and real-time experiments

Sr. No Arena type Time taken (s) Straight line Experimental Path-length in Path-length in Maximum
path-length path length Matlab simulation Matlab simulation heading angle
(cm) (cm) (cm) (Pixel) (°)

Case study-1 Extended C-shaped 11 70.3 94.5 82.6 413 38

continuous obsta-
cle (Fig. 7a1, ­a2)
Case study-2 Rectangular 27 131 234.8 222 1073 51
shaped obstacles
(Fig. 7b1, ­b2)
Case study-3 S-shaped con- 26 130 233 220 1062 47
tinuous obstacles
(Fig. 7c1, ­(c2)

by taking desired turning it reaches to the targeted posi- obstacle (wall) during navigation. However, GWO-Potential
tion. The result presented by Han and Seo [33] methodol- Field-based navigational path keeps a consistent gap with
ogy shows that many times a generated path is passing very the boundary of the obstacles and is free from sharp turning-
close to the obstacle wall. Such situation may lead to the movements (Fig. 8e2).
collision of robot with the obstacle boundary. However, by A different situation is considered in environment 6 (case
use of GWO-Potential field approach the robot can success- study-9), where the oval-shape obstacles are arranged in zig-
fully reach the target while maintaining the consistent gap zag way such that the robot is not able to sense the existence
with the obstacle wall. It can further be noticed that, the of target. In this simulation, robot has to move by changing
robot takes approximately straight path trajectory except the heading direction frequently in order to reach the target.
the corner and followed a smooth path without hitting the In comparison, it can be seen that the path traced by GWO-
obstacles walls. Potential Field method is smoother and maintains a proper
In case study-6, four parallel rectangular slabs are placed gap with the obstacle boundary. The generated paths from
in zig-zag fashion, in which robot starts its journey from both the methodology are shown in (Fig. 8f1–f2).
bottom right corner (coordinates 1400, 200), progressed Using both the approaches, robot can reach the target
upwards (right) for short span of time. After that robot takes ensuring the near optimal/minimum path length, but the
a left turn adjacent to wall end while maintaining a desired major difficulties noticed using Han and Seo’s methodology
distance from obstacle, then takes a smooth right and left is that, the robot path collides with the wall in many places
turn to reach the target smoothly. throughout the navigation. Further in some situations, the
In environment 4 (case study-7), two layers of rectan- paths generated using the Han and Seo’s methodologies are
gular-shaped slab obstacles are placed in typical fashion as not straight, even though it is free from nearby obstacles.
such the target is not directly visible from the robot’s start While with the GWO-Potential Field approach the robot
position. The robot has to start the navigation from bottom can successfully reach the target by maintaining the con-
right corner to the target location in the chosen obstacle sce- sistent gap with the obstacles using on-time path network,
narios as given in Fig. 8. The results show that robot hit with for ensuring the shortest navigational path strategy. It can
the boundary wall and taken more deviation angle to reach further be noticed that the robot takes approximately straight
the target. With the current methodology, robot follows the path except the corner and followed a smooth motion with-
wall following rule during locomotion and navigate in an out hitting the obstacles walls. It can further be noticed that,
approximately straight zig-zag path till the end of the first the time required to reach the target in individual environ-
slab and again follows the same pattern to reach the target ment for the proposed methodology is quite comparable with
without colliding with wall (obstacle). the average run time as mentioned in Han and Seo [33]
The case study-8 (Environment 5) presents a fairly dif- Furthermore, the presented GWO-Potential Field method-
ficult scenario, i.e. corridor navigation in which robot ology considers the utilization of virtual intermediate target
frequently trapped into local-minima due to existence of (VIT) points, which makes easy to deploy a successful path
dead-cycle in which the target is not visible directly. In this plan for any lengthy and complex obstacle prone situations.
experiment, the obtained results through both the approaches
are compared figuratively (Fig. 8e1, ­e2). It can be noticed
that, the path generated by [33] is very close to the obsta-
cles and sometimes it touches the wall (Fig. 8e1) in many
places. The robot is unable to retain a consistent gap with the

13
 Arabian Journal for Science and Engineering

Fig. 8  a1–f2 Comparative path obtained from Han and Seo [33] and using proposed GWO-Potential Field methodology

13
Arabian Journal for Science and Engineering

Fig. 8  (continued)

13
 Arabian Journal for Science and Engineering

6 Results and Discussion

Time taken (s)

The developed methodology has been verified rigorously

26.77

25.96

99.12

64.27

72.65

38.85
over simulations and experiments, which presents efficacy
of the method to solve the navigational problems in an opti-
mised way.
between obstacle & the

This section reports the comparison between Han and Seo

[33] and GWO-Potential controller, and the performances
Minimum distance

of both the controller have been highlighted for the six case
path (Pixels)

studies (case study-4 to case study-9). The results obtained

from the simulations experiment (from both the controller)
are given in Table 2. The performance of both the methodol-
16

11

10

11

9
ogy was evaluated on the basis of three criteria (i.e. average
turning angle, path-length and minimum distance between
(Pixels)
GWO-potential field methodology

length

obstacle & the path).

1071

1045

3965

2571

2906

1554
Path

Out of all six experiments, it can be observed that, the

path-length obtained by GWO-Potential controller in com-
turn angle
Average

parison with only Han and Seo [33] controller, there is a

5.62
7.9

0.9

13.2

2.1

16.2

net reduction of about 5% in path length in environment

(°)

4 and 5. However, in environment 1, 2, 3 and 6, there is a

Table 2  Performance comparison (methodology presented by Han and Seo Vs proposed GWO-potential field methodology)

Maximum
turn angle

slight increase in path length. There is not much variation

in number of step movements, when the sharp edges get
64.80
28.2

transformed to smooth curves. Further, there is no appre-

(°)

94

78

89

42

ciable difference in average turning angle in Han and Seo

[33] and GWO-Potential Field controller. However, the pro-
between obstacle & the

posed GWO-Potential Field controller enables the robot for

Minimum distance

maintaining the constituent gap (i.e. > 6 pixels in all the

chosen environments) with the obstacles and helps to reach
path (Pixels)

the target in an optimal/near-optimal path in all the cho-

methodology presented by Han and Seo [33]

sen environments. The Simulation and experimental results

justify that the method is not only capable of finding an
2

optimal/near-optimal robot path in cluttered environments

(Pixels)
length

but also provides an effective path planning strategy on-time

1045

1018

3911

2732

3062

1500
Path

basis. The results show that, the proposed method has out-
performed to shorten the path length as well as ensured a
turn angle
Average

collision-free navigation, while virtual intermediate targets

8.16

20.63
13.5

2.5

7.5

(VIT) makes the navigation ease and totally free from any
(°)

dead-end situation even in complex environments.

Maximum
turn angle

31.5

7 Conclusions and Future Scope

(°)

56

80

93

99

43
Case study-7 Environment-4 (Fig. 8d1,
Case study-5 Environment-2 (Fig. 8b1,
Case study-4 Environment-1 (Fig. 8a1,

Case study-6 Environment-3 (Fig. 8c1,

Case study-8 Environment-5 (Fig. 8e1,

The research proposes a novel control methodology for

Case study-9 Environment-6 (Fig. 8f1,
Environmental scenario/

on-time path planning and navigation of mobile robot. The

methodology utilizes a hybrid concept of grey wolves opti-
misation (GWO) algorithm and a newly defined artificial
potential field (APF) paradigm. The methodology not only
enables the quick decision-making ability (due to very fast
­b2)

­d2)
­a2)

­c2)

­e2)
map

­f2)

convergence rate of GWO) during each step movement,

but also ensures collision free optimized path in obsta-
cle-prone scenarios. It enables minimum computational
time and an optimized control over the decision of each
Sl. No

13
Arabian Journal for Science and Engineering

step movement with the real-time distance’s information 12. Parhi, D.R.; Mohanta, J.C.: Navigational control of several mobile
of LHO, RHO, FHO and the Virtual Intermediate Target robotic agents using Petri-potential-fuzzy hybrid controller. Appl.
Soft Comput. 11(4), 3546–3557 (2011). https:​ //doi.org/10.1016/j.
Point/actual target point. The utilized Virtual Intermediate asoc.2011.01.027
Targets Point(s) concept reduces the chances of any dead- 13. Tsuji, T.; Tanaka, Y.; Pietro, G.M.; Sanguineti, V.; Kaneko, M.:
end situation happening in complex and long navigation Bio-mimetic trajectory generation of robots via artificial potential
environments. field with time base generator. IEEE Trans. Syst. Man Cybern.
Part C Appl. Rev. 32, 426–439 (2002). https​://doi.org/10.1109/
The results justify that the methodology can be an effec- TSMCC​.2002.80727​3
tive alternative to build an on-time basis collision free navi- 14. Oishi, S.; Inoue, Y.; Miura, J.; Tanaka, S.: SeqSLAM++: view-
gation control system. In future, the methodology can be based robot localization and navigation. Robotics Auton. Syst.
extended for applications like dynamic obstacles scenario 112, 13–21 (2019). https​://doi.org/10.1016/j.robot​.2018.10.014
15. Wachter, L.; Murphy, J.; Ray, L.: Potential function control for
as well as multiple mobile robot path planning. Moreover, multiple high-speed nonholonomic robots. In: IEEE International
the methodology can be extended to implement in practi- Conference on Robotics and Automation 2008, pp. 1781–2. https​
cal applications, i.e. autonomous flying robot navigation ://doi.org/10.1109/ROBOT​.2008.45434​64
for investigation of power transmission lines, autonomous 16. Montiel, O.; Rosas, U.O.; Sepulveda, R.: Path planning for
mobile robots using bacterial potential field for avoiding static
mobile robot navigation in surveillance purposes, etc. and dynamic obstacles. Expert Syst. Appl. 42, 5177–5191 (2015).
https​://doi.org/10.1016/j.eswa.2015.02.033
17. Jean, B.M.; Huang, X.; Wang, M.: Fuzzy motion planning among
dynamic obstacles using artificial potential fields for robot
manipulators. Robotics Auton. Syst. 32, 61–72 (2000). https​://
References doi.org/10.1016/S0921​-8890(00)00073​-7
18. Jing, R.; Kenneth, M.A.; Rajni, V.P.: Modified Newton’s method
1. Henrich, D.; Wurll, C.; Worn, H.: On-line path planning by heu- applied to potential field-based navigation for mobile robots. IEEE
ristic hierarchical search. In: IECON’98, Proceedings of the 24th Trans. Robotics 22, 384–391 (2006). https​://doi.org/10.1109/
Annual Conf. of the IEEE Industrial Electronics Society, New TRO.2006.87066​8
York, NY, USA, pp. 2239–2244 (1998). https​://doi.org/10.1109/ 19. Masoud, A.A.: Decentralized self-organizing potential field-based
IECON​.1998.72406​9 control for individually motivated mobile agents in a cluttered
2. When, H.W.; Belanger, P.R.: Ultrasound-based robot position esti- environment: a vector-harmonic potential field approach. IEEE
mation. IEEE Trans. Robot. Autom. 13(5), 682–692 (1997). https​ Trans. Syst. Man Cybern. Part A Syst. Hum. 37(3), 372–390
://doi.org/10.1109/70.63123​0 (2007). https​://doi.org/10.1109/TSMCA​.2007.89348​3
3. Xiao, J.; Michalewicz, Z.; Zhang, L.; Trojanowski, K.: 20. Tsourveloudis, N.C.; Kimon, P.V.; Hebert, T.: Autonomous vehi-
Adaptive evolutionary planner/navigator for mobile robots. cle navigation utilizing electrostatic potential fields and fuzzy
IEEE Trans. Evol. Comput. 1(1), 18–28 (1997). https​://doi. logic. IEEE Trans. Robotics Autom. 17(4), 490–497 (2001). https​
org/10.1109/4235.58588​9 ://doi.org/10.1109/70.95476​1
4. Song, K.T.; Tai, I.C.: Fuzzy navigation of a mobile robot. In: 21. Cosio, F.A.; Castai, E.; Padilla, M.A.: Autonomous robot naviga-
IEEE/RSI International Conference on Intelligent Robots and tion using adaptive potential fields. Math. Comput. Model. 40,
Systems, July 7–10, Raleigh, USA, pp. 621–627 (1992). https​:// 1141–1156 (2004). https​://doi.org/10.1016/j.mcm.2004.05.001
doi.org/10.1109/IROS.1992.58740​5 22. Cai, T.; Shi, E.; He, C.; Guo, J.: Study of the new method for
5. Cosio, F.A.; Castaneda, M.A.P.: Autonomous robot navigation improving artificial potential field in mobile robot obstacle
using adaptive potential fields. Math. Comput. Model. 40(9–10), avoidance. In: IEEE International Conference on Automation
1141–1156 (2004). https​://doi.org/10.1016/j.mcm.2004.05.001 and Logistics, pp. 282–286 (2007). https​://doi.org/10.1109/
6. Rimon, E.; Doditschek, D.E.: Exact robot navigation using arti- ICAL.2007.43385​72
ficial potential fields. IEEE Trans. Robotics Autom. 8, 501–518 23. Huang, W.H.; Fajen, B.R.; Jonathan, R.F.; William, H.W.: Vis-
(1992). https​://doi.org/10.1109/70.16377​7 ual navigation and obstacle avoidance using a steering potential
7. Mohanta, J.C.; Parhi, D.R.; Patel, S.K.: Path planning strategy function. Robotics Auton. Syst. 54, 288–299 (2006). https​://doi.
for autonomous mobile robot navigation using Petri-GA optimi- org/10.1016/j.robot​.2005.11.004
zation. Comput. Electr. Eng. 37, 1058–1070 (2011). https​://doi. 24. Ajeil, F.A.; Ibraheem, I.K.; Sahib, M.A.; Humaidi, A.J.: Multi-
org/10.1016/j.compe​lecen​g.2011.07.007 objective path planning of an autonomous mobile robot using
8. Muniz, F.; Zalama, E.; Gaudiano, P.; Coronado, J.L.: Neural con- hybrid PSO-MFB optimization algorithm. Appl. Soft Comput. J.
troller for a mobile robot in a non-stationary environment. In: Pro- 89, 106076 (2020). https​://doi.org/10.1016/j.asoc.2020.10607​6
ceedings of 2nd IFAC Conf. on Intelligent Autonomous Vehicles, 25. Gharajeh, M.S.; Jond, H.B.: Hybrid global positioning system-
Helsinki, Finland, pp. 279–284 (1995) adaptive neuro-fuzzy inference system based autonomous mobile
9. Yang, S.X.; Meng, M.: An efficient neural network approach robot navigation. Robotics Auton. Syst. 134, 103669 (2020). https​
to dynamic robot motion planning. Neural Netw. 13, 143–148 ://doi.org/10.1016/j.robot​.2020.10366​9
(2000). https​://doi.org/10.1016/S0893​-6080(99)00103​-3 26. Lazarowska, A.: Discrete artificial potential field approach to
10. Khan, H; Agha, I.S.H.S.; Khan, L.A.: Optimal path planning of a mobile robot path planning. IFAC-Papers On Line 52–8, 277–282
mobile robot using quadrant based random particle optimization (2019). https​://doi.org/10.1016/j.ifaco​l.2019.08.083
method. In: Student Research Paper Conference, vol. 1, no. 2, pp. 27. Wahab, M.N.A.; Meziani, S.N.; Atyabi, A.: A comparative review
6–11 (2014). Corpus ID: 15359338 on mobile robot path planning: classical or meta-heuristic meth-
11. Paniagua, A.H.; Rodriquez, M.A.V.; Ferruz, J.: Applying the ods? Annu. Rev. Control (2020). https​://doi.org/10.1016/j.arcon​
MOVNS algorithm to solve the path planning problem in mobile trol.2020.10.001
robotics. Expert Syst. Appl. 58(1), 20–35 (2016). https​://doi.
org/10.1016/j.eswa.2016.03.035

13
 Arabian Journal for Science and Engineering

28. Khatib, O.: Real-time obstacle avoidance for manipulators and 32. Komaki, G.M.; Kayvanfar, V.: Grey wolf optimizer algorithm
mobile robots. Int. J. Robotics Res. 5(1), 90–98 (1986). https​:// for the two-stage assembly flow-shop scheduling problem with
doi.org/10.1177/02783​64986​00500​106 release time. J. Comput. Sci. 8, 109–120 (2015). https​://doi.
29. Mirjalili, S.; Mirjalili, S.M.; Lewis, A.: Grey wolf optimizer. Adv. org/10.1016/j.jocs.2015.03.011
Eng. Softw. 69, 46–61 (2014). https​://doi.org/10.1016/j.adven​ 33. Han, J.; Seo, Y.: Mobile robot path planning with surrounding
gsoft​.2013.12.007 point set and path improvement. Appl. Soft Comput. 57, 35–47
30. Sulaiman, M.H.; Mustaffa, Z.; Mohamed, M.R.; Aliman, O.: (2017). https​://doi.org/10.1016/j.asoc.2017.03.035
Using the gray wolf optimizer for solving optimal reactive power 34. Fan, X.; Guo, Y.; Liu, H.; Wei, B.; Lyu, W.: Improved artificial
dispatch problem. Appl. Soft Comput. 32, 286–292 (2016). https​ potential field method applied for AUV path planning. In: Math-
://doi.org/10.1016/j.asoc.2015.03.041 ematical Problems in Engineering-Hindawi (2020). https​://doi.
31. Song, X.; Tang, L.; Zhao, S.T.; Zhang, X.Q.; Li, L.; Huang, J.Q.; org/10.1155/2020/65231​58
Cai, W.: Grey wolf optimizer for parameter estimation in surface 35. Oscar, M.; Sepúlveda, R.; Orozco-Rosas, U.: Optimal path plan-
waves. Soil Dyn. Earthq. Eng. 75, 147–157 (2015). https​://doi. ning generation for mobile robots using parallel evolutionary arti-
org/10.1016/j.soild​yn.2015.04.004 ficial potential field. J. Intell. Robotic Syst. 79(2), 237–257 (2015)

13