mathematics

Article
Faster Implementation of The Dynamic Window Approach
Based on Non-Discrete Path Representation
Ziang Lin * and Ryo Taguchi

Department of Computer Science, Nagoya Institute of Technology, Nagoya 466-8555, Japan;

taguchi.ryo@nitech.ac.jp
* Correspondence: cls14130@nitech.ac.jp

Abstract: The dynamic window approach (DWA) serves as a pivotal collision avoidance strategy
for mobile robots, meticulously guiding a robot to its target while ensuring a safe distance from any
perceivable obstacles in the vicinity. While the DWA has seen various enhancements and applications,
its foundational computational process has predominantly remained constant, consequently resulting
in a heightened level of time complexity. Inspired by the velocity invariance assumption inherent
in the DWA and the utilization of polar coordinate transformations in the model, we introduce a
high-speed version of the DWA.

Keywords: DWA; wheeled robot; autonomous control; local path planning; path representation;
computational efficiency

MSC: 93C85; 93B27; 68W40; 68Q25

1. Introduction
Owing to labor shortages, the utilization of robots across diverse fields has notably
increased. Beyond their traditional applications in automated guided vehicles (AGVs) and
Citation: Lin, Z.; Taguchi, R. Faster robotic arms in factories [1–4], they are also employed in airports and restaurants as guiding
Implementation of The Dynamic and catering robots [5–7]. Given that general environments tend to be more intricate than
Window Approach Based on factory settings, ensuring the safety of robots’ autonomous movements is paramount.
Non-Discrete Path Representation. Autonomous control typically comprises three principal elements: mapping, self-
Mathematics 2023, 11, 4424. position estimation, and path planning [8–11]. Path planning can be subdivided into global
https://doi.org/10.3390/ and local planning. Global path planning delineates the route from a starting point to
math11214424 a destination on a map, with algorithms such as A* [12], D* [13], and RRT* [14] being
Academic Editor: Sanda conventionally utilized for this function. On the other hand, local path planning formulates
Florentina Mihalache collision-free paths that adhere to a global path, necessitating real-time analysis of obstacle
distance and orientation. The dynamic window approach (DWA) [15], artificial potential
Received: 29 September 2023 field (APF) [16], and vector field histogram (VFH) [17] are among the widely adopted
Revised: 23 October 2023
algorithms for local path planning.
Accepted: 24 October 2023
Local path planning algorithms, such as APF or VHF, yield only the current optimal
Published: 25 October 2023
heading direction, sans specific velocity commands, necessitating additional computational
steps. Conversely, the DWA operates as a model predictive control (MPC) method, accept-
ing a robot’s current velocity as input. Mindful of the robot’s performance constraints, the
Copyright: © 2023 by the authors. DWA samples feasible velocity commands and constructs predicted paths based on these.
Licensee MDPI, Basel, Switzerland. These paths are then evaluated to determine the optimal velocity command, allowing the
This article is an open access article DWA’s output to be directly integrated with the robot. Nevertheless, this incurs substantial
distributed under the terms and computational costs within a single control cycle. Despite recent studies concentrating on
conditions of the Creative Commons enhancing or adopting the DWA [18–20], methods for path generation and evaluation have
Attribution (CC BY) license (https:// largely remained static. Conventional DWA methods grapple with processing speed issues
creativecommons.org/licenses/by/ due to the escalating demand for sensor accuracy and responsiveness in robots.
4.0/).

Mathematics 2023, 11, 4424. https://doi.org/10.3390/math11214424 https://www.mdpi.com/journal/mathematics

Mathematics 2023, 11, 4424 2 of 20

Conversely, the conventional DWA presupposes constant velocity, which engenders a

significant discrepancy between the array of generated path reach points and the actual
reachable range of a robot. This constraint hinders the formulation of paths that incorporate
avoidance maneuvers based on the robot’s acceleration and deceleration during the simula-
tion period. To counter this limitation, Lin et al. introduced a variable velocity approach
incorporating an acceleration or jerk model [21], capable of generating paths factoring in
acceleration and deceleration. In this study, an acceleration model was deployed to craft a
variable velocity path. When solely considering the reach points’ range of the generated
paths, the acceleration model exhibited a more extensive reach point range compared to
the jerk model.
In this study, our analysis leads us to conclude that the primary factor contributing to
the elevated computational cost of the conventional DWA resides in the generation of path
points through the iterative kinematic model. Consequently, we have introduced a method
for non-discrete path representation, upon which we have developed a corresponding
approach for collision detection. In our proposed method, we represent paths using circular
arcs and conducting collision detection while utilizing the conventional DWA constant
velocity model for path prediction. Moreover, when employing the acceleration model [21]
for path prediction, we extended this method to approximate paths with multiple circular
arcs and executed collision detection. Experiments demonstrated that both the computation
time and accuracy of the proposed collision detection method surpassed those of the
conventional approach. The remainder of this paper is structured as follows: Section 2
offers a review of the DWA; Section 3 comprehensively details the proposed method; and
Section 4 assesses the efficacy of the proposed method through simulation experiments.

2. Dynamic Window Approach

The DWA is a local path planning algorithm in which the path is planned based on the
kinematic parameters of a robot. Because the object of study is a robot that moves solely in
a plane, x (t), y(t), and θ(t) describe the configuration space of the robot, where x (t) and
y(t) denote the coordinates, and θ(t) represents the heading of the robot. Because wheeled
robots that move in a plane have two degrees of freedom, translational and rotational
velocities, let v(t) and ω(t) denote the translational and rotational velocities of the robot.
The kinematic model is defined by (1)–(3).

Z tp
x t p = x ( t0 ) + v(t) · cos(θ (t))dt, (1)
t0

Z tp
y t p = y ( t0 ) + v(t) · sin(θ (t))dt, (2)
t0

Z tp
θ t p = θ ( t0 ) + ω (t)dt. (3)
t0

For practical applications, Fox et al. presented the discrete forms of (1)–(3) expressed
in (4)–(6). This model assumes that the robot moves a distance of v·∆t along the heading of
t p−1 , and then rotates an angle of ω ·∆t. This method computes the coordinates and heading
of the robot from time t0 to t p by iterating the input velocities. A series of coordinates
corresponding to a series of input velocities is known as a path. This model is widely
employed in related studies of wheeled robots [22–26].

x t p = x t p−1 + v · ∆t · cos θ t p−1 ,

(4)

y t p = y t p−1 + v · ∆t · sin θ t p−1 ,

(5)

θ t p = θ t p−1 + ω · ∆t.



(6)
Mathematics 2023, 11, 4424 3 of 20

Other models predict a path from the input velocity. Equation (7)–(9) represent another
kinematic model that assumes that the robot first rotates an angle of ω ·∆t, and then moves
a distance of v·∆t along the heading of t p . Yang et al. adopt this model in their work and
added a factor of ω · ∆t to reduce the error in path simulation [27].

θ t p = θ t p−1 + ω · ∆t,



(7)

x t p = x t p−1 + v · ∆t · cos θ t p ,





(8)

y t p = y t p−1 + v · ∆t · sin θ t p .





(9)
In this study, we refer to the models in (4)–(6) as tangent models and the models in
(7)–(9) as secant models.
Given the absence of restrictions on velocity variation, both kinematic models yield
a multitude of potential paths. In response to this, Fox et al. introduced a DWA that
designates the current moment as t0 and simulates the robot’s future path over a specified
period by iterating (4)–(6). The conclusion of this simulation is marked as t p , a timespan
also referred to as the simulation period. The DWA posited that velocities remained
constant throughout this simulation period. Consequently, a singular path is generated
for each set of input pairs of translational and rotational velocities. These velocities were
determined based on the robot’s current velocity and acceleration constraints, resulting
in the number of generated paths being equivalent to the product of the translational and
rotational velocities.
Figure 1 depicts a flowchart detailing the general process of robot navigation and
provides an in-depth explanation of the operational mechanics of the conventional DWA.
In robot navigation, the sequence initiates with the task planner setting one or multiple
objectives. Following this, the global path planner is responsible for formulating the global
path. Throughout the execution phase, the robot persistently scans its surrounding milieu,
strategizes local trajectories, and advances, iterating this sequence until the destination
is attained. The symbols 'Y' and 'N' in the flowchart represent 'Yes' and 'No,' respectively.
Within the framework of the conventional DWA, the initial step, Step (A), involves the
sampling of all feasible velocity pairs, grounded on the existing velocity and acceleration
constraints of the robot. For each input velocity pair, the DWA computes the coordinates
of the path points by iterating over (4)–(6) in Step (B), because the computation

time of
Step (B) primarily depends on the number of iterations n p . where t B1 n p denotes the
computational time for Step (B). The relationship between the number of iterations n p , time
interval ∆t, and simulation period t p can be expressed as

t p = n p · ∆t. (10)

In Step (C), the DWA computes the distance between each obstacle and path point.
This step can be efficiently performed using matrix operations to determine the minimum
distance. The computation time for Step
(C) depends on the number of paths and obstacle
points. Therefore, we adopt t B2 n p , no to denote the computation time of Step (C). no
denotes number of obstacles. The maximum value of no depends on the resolution and
range of the sensor that detects obstacles, such as a laser rangefinder. In Steps (D) and (E),
the DWA only computes the heading of the ending point of the path and the distance to
the goal; hence, the computation time is extremely short and can be disregarded. After
evaluating all executable velocity pairs, we output the optimal velocity command in Step
(F). Therefore, the computation time of the conventional DWA, which is the baseline in this
study, can be expressed as




t B n p , no = t B1 n p + t B2 n p , no . (11)
Mathematics 2023, 11, 4424 4 of 20
Mathematics 2023, 11, x FOR PEER REVIEW 4 of 20

Navigation start
(A) Sampling
Task planning
N
exists
Global path planning Y
(B) Path generation:
Sensing
(C) Collision evaluation:
Local path planning (DWA)
(D) Heading evaluation
Move
(E) Togoal evaluation
N
Reach goal
Y (F) Output optimal command
Navigation End

Figure1.1.Flowchart
Figure Flowchartof
ofrobot
robotnavigation
navigationand
andconventional
conventionalDWA.
DWA.

3.3.Proposed
ProposedMethodMethod
3.1. Proposed Method for Constant Velocity Model
3.1. Proposed Method for Constant Velocity Model
In the conventional DWA, the robot’s velocity is presumed to remain invariant fol-
In the conventional DWA, the robot’s velocity is presumed to remain invariant fol-
lowing the initial time interval. As a result, throughout the simulation period, the robot’s
lowing the initial time interval. As a result, throughout the simulation period, the robot’s
motion is characterized by constant velocity circular motion, rendering the simulation path
motion is characterized by constant velocity circular motion, rendering the simulation
as a circular arc. The relationship among translational velocity, rotational velocity, and path
path as a circular arc. The relationship among translational velocity, rotational velocity,
radius is articulated in (12) [28–32].
and path radius is articulated in (12) [28–32].
v = ω · rp, (12)
𝑣 𝜔⋅𝑟 , (12)
where v and ω represent the translational and rotational velocities of a robot during the
where 𝑣 and
simulation period,ω represent the translational
respectively, r p denotes the and rotational
radius velocities
of simulation of a robot during the
path.
simulation period, respectively, 𝑟 denotes the radius of simulation path.
Accordingly, we can establish a polar coordinate system with the center of the circular
Accordingly,
arc path we origin.
(x p , y p ) as the can establish a polar coordinate
Subsequently, we can convertsystem thewith the center
coordinates of the circu-
of obstacles in
lar arc path (𝑥 , 𝑦 ) as the origin. Subsequently, we can convert the
the robot coordinate system (xo , yo ) to the polar coordinate system (ro , θo ) usingcoordinates of (13)
obstacles
and
in the
(14). robot
The coordinate
function atan2 system (𝑥 , 𝑦 to
was adopted ) tocalculate
the polar coordinate
the angle after system (𝑟 , 𝜃 transformation.
coordinate ) using (13) and
(14). The function atan2 was adopted q to calculate the angle after coordinate transfor-
mation. ro =

2
xo − x p + yo − y p ,

2
(13)
(13)
𝑟 𝑥 𝑥 𝑦 𝑦
,
θo = atan2 yo − y p , xo − x p . (14)
𝜃
Finally, we can employ the radial atan2 𝑦
coordinates 𝑦of, 𝑥the obstacles
𝑥 . ro and path radius r p(14)
to
compute the distance from the obstacles to the simulation path using (15).
Finally, we can employ the radial coordinates of the obstacles 𝑟 and path radius 𝑟
to compute the distance from the obstacles o −the
d = rto r p simulation
. path using (15). (15)

Figure 2a illustrates this process. (x𝑑o , yo )𝑟 represents

𝑟 . (15)
the coordinates of an obstacle,
and (x p , y p ) denotes the center of the circular arc path in the robot coordinate system. As
Figure 2a illustrates this process. (𝑥 , 𝑦 ) represents the coordinates of an obstacle,
illustrated in Figure 2b, the coordinate of the obstacle is (ro , θo ) after conversion. Next,
and (𝑥 , 𝑦 ) denotes the center of the circular arc path in the robot coordinate system. As
the distances from the circular arc path to all obstacle points are computed. When the
illustrated in Figure 2b, the coordinate of the obstacle is (𝑟 , 𝜃 ) after conversion. Next, the
minimum value of the distance is smaller than the safe distance, the algorithm abandons
distances from the circular arc path to all obstacle points are computed. When the mini-
the simulation path.
mum value of the distance is smaller than the safe distance, the algorithm abandons the
simulation path.
 Mathematics 2023, 11, x FOR PEER REVIEW 5

Mathematics 2023,2023,
Mathematics 11, x11,
FOR4424PEER REVIEW 5 of 20 5 of

(a) In robot coordinate system (b) In polar coordinate system

Figure 2. Examples illustrating how the proposed method applies to conventional DWA.
(a) In robot coordinate system (b) In polar coordinate system
Figure A flowchart of the proposed method ismethod
presented in to Figure 3. In contrast
DWA. to the
Figure2.
2. Examples illustrating
Examples illustrating how
how the the proposed
proposed method appliesapplies conventional
to conventional DWA.
ventional DWA, the proposed method does not require the generation of discrete
points, which enables
AAflowchart
flowchart of
of the it to omit
theproposed
proposed iterative
method
method iscomputations
is presented
presented in in
in Figure time.
3. In3.
Figure TheIn symbols
contrast ʹYʹthe
to theto
contrast andco
conventional
the flowchart
ventional DWA, the
DWA,represent proposed
the proposed ʹYesʹ and method
method does not
ʹNo,ʹ respectively. require
does not require the generation
In Step the(B), of discrete
the coordinates
generation path of thepa
of discrete
points, which enables it to omit iterative computations in time. The symbols 'Y' and 'N'
origins
points, are computed
which enables it toasomit denoted iterative 𝑛 , wherein𝑛time.
by 𝑡computations represents
The symbols the number
ʹYʹ and of ʹN
in the flowchart represent 'Yes' and 'No,' respectively. In Step (B), the coordinates of the
theorigins.
flowchartWhen generating
represent ʹYesʹ paths
and ʹNo,ʹ using the constant
by trespectively.
velocity model, 𝑛 can
In Step (B), the coordinates of the po be viewed
polar origins are computed as denoted P1 ( nc ), where nc represents the number of polar
origins. When generating paths using the constant velocity𝑛model,
constant
origins are of 1. In
computed step (C),
as the
denoted proposed
by 𝑡 𝑛
method, whereperforms polar
represents
nc can transformations
the
be number
viewed as ofonpo a
astacle
origins.
constantpoints;
When hence,
of 1.generating
In thethe
step (C), computation
paths thetime
using method
proposed constantof Step
performs velocity
polar model, 𝑛
(C) depends on can
transformations the on
number
be viewed
all ofa
obstacle
origins 𝑛
points;
andhence, the
obstacle computation
points 𝑛 .time of
Therefore, Step (C)
we depends
utilize
constant of 1. In step (C), the proposed method performs polar transformations on all 𝑡
on 𝑛
the , 𝑛
number to of polar
denote the como
origins
tation
stacle n and
time
c
points; obstacle
of points
Step the
hence, n . Therefore,
(C).computation
The o total computation we utilize
time of Step t (
time
P2 n , n
c of
(C) ) to denote
the proposed
o depends the computation
on themethod
number is of
expre
po
time of Step (C). The total computation time of the proposed method is expressed as (16).
as (16).
origins 𝑛 Because
and obstacle𝑛 ispoints 𝑛 . Therefore,
significantly smallerwe thanutilize𝑛 , we 𝑛 , 𝑛 tothat
𝑡 presume thethe
denote comput
comp
Because nc is significantly smaller than n p , we presume that the computation time of the
timetime
tation
proposedof theofproposed
Step
method method
(C). The
is shorter than total iscomputation
that shorter
of than time
the conventional that DWA.of
of the
the conventional
proposed method DWA.is express
as (16). Because 𝑛 is significantly smaller than 𝑛 , we presume that the computati
t P (ncis 𝑡o ) =𝑛 t,P1𝑛(nc ) +𝑡t P2 (𝑛nc , no )𝑡. 𝑛 , 𝑛 .
, nshorter
time of the proposed method than that of the conventional DWA. (16)

𝑡 𝑛 ,𝑛 𝑡 𝑛 𝑡 𝑛 ,𝑛 . (
DWA Start

(A) DWA Start

Sampling

(A) Sampling exists N

Y
N
(B) exists
Path generation:
Y
(B)(C) Collision
Path generation:
evaluation:

(C)(D)Collision Heading
evaluation:
evaluation

(D)(E) Heading evaluation

Togoal evaluation

(E)(F) Togoaloptimal
Output evaluation
command

(F) Output optimal

DWAcommand
End

Figure
Figure 3. 3. Flowchart
Flowchart of proposed
of proposed
DWA End
method.
method.

Figure 3. Flowchart of proposed method.

 Mathematics 2023, 11, x FOR PEER REVIEW 6

Mathematics 2023, 11, x FOR PEER REVIEW 6 o

Mathematics 2023, 11, 4424 6 of 20

3.2. Proposed Method for the Variable Velocity Model
When the
3.2. Proposed method
Method proposed
for the Variable in [17] isModel
Velocity adopted to generate variable velocity path
circular
3.2. Proposedarc
When the paths
Method become curved
for theproposed
method Variable Velocitywith
in [17]Modelmonotonically
is adopted to generate increasing andvelocity
variable decreasing
paths,cu
tures,
When
circular asarc
illustrated
thepaths
method in Figure
proposed
become in4.[17]
curved The black
is adopted
with line represents
to generate
monotonically a circular
variable velocity
increasing andarc path.
paths, the The red
decreasing cu
blue
circularlines
arc represent
paths become the paths
curved during
with acceleration
monotonically and
increasing deceleration,
and decreasing
tures, as illustrated in Figure 4. The black line represents a circular arc path. The red respectively.
curvatures, The
as illustrated
vature in Figure
ofrepresent
the 4. The blacktoline
path continued represents
increase in thea circular
case arcacceleration
path. The redand and decreased
blue
blue lines the paths during acceleration and of deceleration, respectively. Thei
lines represent the
case ofofdeceleration. paths during acceleration
𝑣 and ω increase and deceleration,
denote thethe robot’s respectively.
initial The curvature
translational and rotationa
vature
of the path the path continued
continued to increase toin the case ofin case of
acceleration andacceleration
decreased inand decreased
the case of in
locities, respectively, and 𝑣 denotes the terminal translational velocity. The curvatu
case
deceleration. v0 and ω0𝑣denote
of deceleration. ω robot’s
and the denoteinitial
the robot’s initial
translational translational
and and rotational
rotational velocities,
the path
respectively, is
locities, respectively,
and vat
p time
denotes 𝑡 ,
theand
terminal at time 𝑡
translational.
and 𝑣 denotes the terminal translational velocity. path
velocity. The curvature of the The iscurvatur
v0 vp
ω0 at
the timeist0 , andatωtime
path 0 𝑡 , tand
at time p. at time 𝑡 .

Figure 4. Simulation path using acceleration model.

Figure 4. Simulation path using acceleration model.
Figure 4. Simulation path using acceleration model.
As depicted in Figure 5, the perpendicular line drawn from the obstacles to the
As depicted in Figure 5, the perpendicular line drawn from the obstacles to the
doesAsnot consistently
depicted run parallel
in Figure to the line extending from thethe obstacles to the orig
path does not consistently run5,parallel
the perpendicular line drawn
to the line extending fromfrom obstacles
the obstacles to the to the p
the polar
does
originnot coordinate
consistently
of the system.
run
polar coordinate Consequently,
parallel
system. toConsequently, thethedistance
the line extending from
fromfrom
distance theobstacles
the the obstacles
obstacles thetoorigi
toto the
cannot
the path
the be calculated
polarcannot using
be calculated
coordinate system. (15).
using In light of this,
(15). In light ofthe
Consequently, the
this, present
the present
distance paper
frompaper introduces
theintroduces a metho
obstaclesato the p
infer
method the
to distance
infer the from
distanceobstacles
from to
obstacles a path,
to a employing
path, employing multiple
multiple
cannot be calculated using (15). In light of this, the present paper introduces polar
polar coordinate
coordinate tran
a metho
transformations.
mations. This
This method method is applicable
is applicable to paths with monotonically increasing or
infer the distance from obstacles to a to paths
path, with monotonically
employing multiple polar increasing or decrea
coordinate trans
decreasing
curvatures. curvatures.
mations. This method is applicable to paths with monotonically increasing or decreas
curvatures.

Figure5.5.Schematic
Figure Schematicillustration of the
illustration application
of the of the
application ofproposed method
the proposed to DWA
method to using
DWAtheusing the
acceleration model.
eration model.
Figure 5. Schematic illustration of the application of the proposed method to DWA using the a
eration model.
Figure 6 presents an example of a path generated by the acceleration model. The side
where Figure
the center6 presents
of curvature an resides
example of a path
is referred generated
to as by the
the inner side, andacceleration
the opposing model.
side The
iswhere
termed the
Figure center
the 6
outer of Initially,
side.
presents curvature resides
we established
an example isareferred
of a path polar to as
by the
coordinate
generated inner
system,
the side, and
designating
acceleration the The
any
model. oppos
side
point is
on termed
the inner the
sideouter
of side.
the path Initially,
as the we
origin, established
and a
subsequentlypolar coordinate
drew
where the center of curvature resides is referred to as the inner side, and the oppos a circle system,
centered desig
at this
ing is origin.
any This
pointthe circle
on outer may
the inner either be separate from, tangential to, or intersecting with
side termed side.side of thewe
Initially, path as the origin,
established a polar and subsequently
coordinate drew
system, ac
desig
the path.
centered at this origin. This circle may either be separate from, tangential to, or inter
ing any point on the inner side of the path as the origin, and subsequently drew a ci
ing withatthe
centered path.
this origin. This circle may either be separate from, tangential to, or inters
ing with the path.
 Mathematics
Mathematics 2023,2023, 11, x FOR PEER REVIEW
11, 4424 7 of720of 20
Mathematics 2023, 11, x FOR PEER REVIEW 7 of 20

(a) Example in which distance can be estimated (b) Example in which distance cannot be estimated
(a) Example in which distance can be estimated (b) Example in which distance cannot be estimated
Figure 6. Generating a circle on the inner side of a curved path.
Figure 6. Generating a circle on the inner side of a curved path.
The line from the obstacles to the center of the circle intersects the circle and path at
The line from the obstacles to the center of the circle intersects the circle and path at
Points C and B, respectively. The foot of the perpendicular line from the obstacles to the
Points C and B, respectively. The foot of the perpendicular line from the obstacles to the
path is calledrespectively.
Point A. Owing to the the lengthperpendicular
of OA being the shortest distance from the
path is called Point A. Owing to the length of OA being the shortest distance from the
obstacles to the path, OA ≤ OB. As illustrated in Figure 6a,shortest
called being the when the circle is separated
obstacles
obstacles to the path,
totangential
the path, to OA
OA ≤ OB. As illustrated in Figure 6a, when the
the circle is
is separated
from or the≤ path,
OB. As OBillustrated
≤ OC; hence, in Figure
OA ≤ OC. 6a, Aswhen
illustrated circlein Figure separated
6b, when
from
from or
or tangential
tangential to
to the
the path,
path, OBOB ≤ ≤
OC; OC; hence,
hence, OA OA
≤ OC.≤ OC.
As As illustrated
illustrated in Figurein Figure
6b, when 6b,
the circle intersects the path, we cannot determine the size relationship between OA and
when
the OC the
circle circle intersects
intersects the path, we cannot determine the size relationship between OA
owing to thethe path, we
location of thecannot determine
obstacles. Thus, thewesize
canrelationship
establish a between
size relationshipOA and be-
and OC owing
OCtween
owing to to location
the the location of of obstacles.
the the obstacles. Thus, Thus,
we we can
can establish
establish a a size
size relationship
relationship be-
between OAOAandandOC OCififthethecircle
circledoesdoesnot notintersect
intersectwith withthethepath.
path.
tween OA and OC
Furthermore, if the circle
the larger does not
the radius intersect with
of the circle, the path.
the smaller
Furthermore,
Furthermore, the
thelarger
larger the
theradius
radius ofofthe
thecircle,
circle, the
thesmaller
smaller OCOC is,OC
and
is,
is,the
and
and
the
the closer
closer
closerit isit to
is
it is
OA. to OA.
To To To
improve improve
thethe the
distance distance
estimation estimation
accuracy, accuracy,
thethe the
largest largest
circle circle
that doesthat does not inter-
to OA. improve distance estimation accuracy, largest circle hthat inotnot
does intersect
inter-
the sect
path the
shouldpath beshould
adopted. be adopted.
Because Because
the path’s the
rangepath’s
of range
curvature of curvature
is v0 v p is
, , the , , the
largest
sect the path should be adopted. Because the path’s range of curvature ω0 ωis 0 , , the
largestthe
radius radius ofisthe v0 circle is .
largest ofradiuscircle ω0 . is
of the circle .
As As illustrated
illustrated in in Figure
Figure 7a,7a,multiple
multiplecirclescircleswere were generated on onthetheinner
innerside. side.AsAs afore-
As illustrated
mentioned,
aforementioned, OA
OA in≤≤Figure
min(OC1,
min(OC1,7a, multiple
OC2).
OC2). This circles were that
implies
implies generated
thatasasmore oncircles
more the inner
circlesareareside. As afore-
generated,
generated, the the
mentioned,
probability
probability OAthat
that ≤ min(OC1,
min(OC1,
min(OC1, OC2,OC2).
OC2, This
. . .)...)
is is implies
close
closeto tothatthat
thatofofasOA
OA more circlesIf,If,
increases.
increases. are
and
and generated,
only
onlyif,if,Pointthe A
Point
probability
A isisthe
thecontactthat min(OC1,
contactpoint
pointofof theOC2,
thecircle...)
circle andis
andclose
path, to
path, that
as as of OA
illustrated
illustratedincreases.
in Figure
in Figure If, and
7b,7b,
Pointsonly
Points if,
A, B, Point
A,and
B, and A
C C
is the
coincide contact
to point
form a of
singlethe circle
point, and
and path,
the as
equalityillustrated
sign in in
the
coincide to form a single point, and the equality sign in the inequality OA ≤ OC holds, Figure 7b,
inequality Points
OA ≤ A,OC B, and
holds, C
coincide
then the the
then to form
proposed amethod
proposed single point,
method can can and
compute thethe
compute equality
true
the true sign
value in OA
of
value the inequality
accurately.
of OA accurately. OA ≤ OC holds,
then the proposed method can compute the true value of OA accurately.

(a) (b)
(a) (b)
Figure
Figure 7. Error
7. Error analysis
analysis of inner
of inner sideside circles.
circles. (a) When
(a) When generating
generating multiple
multiple circles,
circles, as mentioned
as mentioned
Figure 7. Error
earlier, OA analysis
≤ min(OC1, of inner
OC2).side
This circles.
implies (a) When
that as generating
more circles multiple
are circles,
generated, theasprobability
mentionedthat
earlier,
earlier, OA ≤
OA
min(OC1,
min(OC1, OC2).
≤ min(OC1, OC2). This
OC2, ...) approximates
This implies
implies that
that as
OA increases.
as more
more circles
circles are
(b) Only when
generated,
are Point
generated, the probability
the as
A serves probability that
the contactthat
point
min(OC1,
min(OC1, OC2,
OC2, . .
...).) approximates
approximates OA
OA increases.
increases. (b)
(b) Only
Only when
when Point
Point A
A serves
serves as
as the
the contact
contact point
point
between the circle and the path, Points A, B, and C coincide to form a single point. In this scenario,
between
between the
the circle
circle and
andinthe
the path, Points
Points A, B, and C coincide to
to form aa single point. In this
this scenario,
scenario,
the equality sign thepath,
inequality A,
OAB,≤andOCCholds,
coincide
enablingformthe single point.
proposed In
method to accurately
the equality sign in
in the
the inequality
compute the true value of OA.inequality OA
OA ≤≤ OC holds, enabling the proposed method to accurately
compute the true value of OA.
At this point, the boundary on one side of the distance from the obstacle to the path
At this point,
point, thethe boundary
boundary on one
oneside ofofthe
thedistance from the obstacle toto thethe
path is
is determined. Similarly, the on
circle sidebe
can distance
separated from
from, the obstacle
tangential to, or path
intersecting
determined.
is determined. Similarly, the
Similarly, circle can
the circle be separated from, tangential to, or intersecting with the
with the path on the outer side, can be separated
as illustrated from, 8.
in Figure tangential
The line to,fromor the
intersecting
obstacles to
path
with on the outer side, as illustrated inillustrated
Figure 8. The line from the obstacles to the
the center of the
the center of the circle intersects with the circle at Point C. The foot of theobstacles
the path on the outer side, as in Figure 8. The line from to
perpendicular
circle intersects
the line
center of with
the the
circle circle at
intersects Point
with C. The
the foot
circle of
at the perpendicular
Point C. The foot line
of from
the the obstacles
perpendicular
from the obstacles to the path is called Point A, and the perpendicular line intersects
to the
line paththe
from is called
obstacles Point
to A,
theand theisperpendicular A,line
andintersects with the circle
line at Point B.
with
Because
thelength
the
circle at Point B. path
Becausecalled
of OAB.is Because
the shortest
Point of
the length
distance from
OA
the
isthe
theperpendicular
obstacles
shortest
to the
distance
path, OB
intersects
from
≤ OA.
the
As
ob-
with the circle at Point the length of OA is the shortest distance
stacles to the path, OB ≤ OA. As illustrated in Figure 8a, when the circle is separated from from the ob-
stacles to the path, OB ≤ OA. As illustrated in Figure 8a, when the circle is separated from
 Mathematics2023,
Mathematics 2023,11,
11,4424
x FOR PEER REVIEW 88 of
of2020
Mathematics 2023, 11, x FOR PEER REVIEW 8 of 20

or tangential
illustrated to the8a,
in Figure path,
whenOCthe
≤ OB;
circlehence, we canfrom
is separated deduce that OC ≤toOA.
or tangential the As
path, OC ≤ OB;
illustrated in
or tangential to the path, OC ≤ OB; hence, we can deduce that OC ≤ OA. As illustrated in
Figurewe
hence, 8b,can
when the circle
deduce that OC ≤ OA. with
intersects the path, in
As illustrated weFigure
cannot determine
8b, when thethe size
circle relation-
intersects
Figure 8b, when the circle intersects with the path, we cannot determine the size relation-
shipthe
with between OAcannot
path, we and OC owing tothe
determine the location
size of the between
relationship obstacles.OABased
and on
OCtheowingabove dis-
to the
ship between OA and OC owing to the location of the obstacles. Based on the above dis-
location
cussion,ofitthe obstacles.
is evident Based
that on the
we can above discussion,
establish it is evident
a size relationship that we
between OAcanand establish a
OC if the
cussion, it is evident that we can establish a size relationship between OA and OC if the
size relationship
circle between with
does not intersect OA andthe OC
path.if the circle does not intersect with the path.
circle does not intersect with the path.

(a) Example in which distance can be estimated (b) Example in which distance cannot be estimated
(a) Example in which distance can be estimated (b) Example in which distance cannot be estimated
Figure 8. Generating a circle on the outer side of a curved path.
Figure
Figure 8.
8. Generating
Generating aa circle
circle on
on the
the outer
outer side
side of
of aa curved
curved path.
path.

Moreover,the
Moreover, thesmaller
smaller the the radius
radius of of the circle,
thecircle,
circle,the the larger
thelarger
larger OC
OC is and the closer it isisto
Moreover, the smaller the radius of the OC is is
andandthethe closer
closer it isit to
OA.
to OA. To improve the accuracy of distance estimation, the smallest circle that does not in-
OA. To To improve
improve thethe accuracy
accuracy of distance
of distance estimation,
estimation, the the smallest
smallest circlecircle
thatthat
doeshdoesnot notin-i
tersect the path should be employed. Because the path’s range ofof
curvature is ,v v, pthe
intersect
tersect thethe
pathpath should
should be be employed.
employed. Because
Because thethe
path’s path’srangerange curvature
of curvature is is , ω00 ,, ωthe 0
,
smallest radius of the circle is . v p
the smallest
smallest radiusradius
of theof circle
the circle
is is. ω0 .
Asillustrated
As illustrated in Figure
Figure 9a, 9a,multiple
multiplecircles
circleswere
were generated
generated onon thethe
outer side.side.
outer As men- As
As illustrated in Figure 9a, multiple circles were generated on the outer side. As men-
mentioned
tioned above, above,
the the
more more circles
circles generated
generated on the
on the outer outer
side,side,
the the
higherhigher
the the probability
probability that
tioned above, the more circles generated on the outer side, the higher the probability that
that max(OC1,
max(OC1, OC2)OC2) is close
is close to that to of
that
OA. ofIf,
OA.andIf,only
andif,only Point if,APoint
is theAcontact
is the contact
point between point
max(OC1, OC2) is close to that of OA. If, and only if, Point A is the contact point between
between the
the circle circle
and path,andaspath, as illustrated
illustrated in Figurein Figure 9b, Points
9b, Points A, B, A, andB,C and C converge
converge into into
one,one, and
the circle and path, as illustrated in Figure 9b, Points A, B, and C converge into one, and
and
the the equality
equality signsign in inequality
in the the inequality OC OC ≤ OA ≤holds.
OA holds.In this Incase,
this case, the proposed
the proposed method methodcould
the equality sign in the inequality OC ≤ OA holds. In this case, the proposed method could
could accurately
accurately computecompute
the true the OAtruevalue.
OA value. By combining
By combining the distance
the distance estimations
estimations fromfrom both
accurately compute the true OA value. By combining the distance estimations from both
both sides, the range of the distance from the
sides, the range of the distance from the obstacle to the path obstacle to the path can be
be obtained.Because
obtained. Because
sides, the range of the distance from the obstacle to the path can be obtained. Because
multiple
multiplecircles
circlesare
aregenerated
generated on on
thethe inner andand
inner outerouter
sidessidesof theofpath,
the path, 𝑛 is
nc in (16) inno(16)longer
is no
multiple circles are generated on the inner and outer sides of the path, 𝑛 in (16) is no
unity.
longerNonetheless, nc is still𝑛much
unity. Nonetheless, smaller
is still much than
smallern p , than 𝑛 , so acomputation
so a shorter shorter computationtime cantime be
longer
expected unity.
for Nonetheless,
the proposed 𝑛 is stillcompared
method much smaller
to than
that for 𝑛the, so a shorter computation
conventional DWA. time
can be expected for the proposed method compared to that for the conventional DWA.
can be expected for the proposed method compared to that for the conventional DWA.

(a) (b)
(a) (b)
Figure9.9.Error
Figure Erroranalysis
analysisofofouter
outerside circle.
side (a)(a)
circle. WhenWhen generating
generating multiple
multiplecircles, as mentioned
circles, as mentionedearlier,
ear-
Figure 9. Error analysis of outer side circle. (a) When generating multiple circles, as mentioned ear-
lier, max(OC1,
max(OC1, OC2) ≤OC2)
OA. ≤This
OA.implies
This implies that as
that as more more
circles arecircles are generated,
generated, the probability
the probability that max(OC1,that
lier, max(OC1, OC2) ≤ OA. This implies that as more circles are generated, the probability that
max(OC1,
OC2, OC2, …) approximates
. . .) approximates OA increases. OA increases. (b)Point
OnlyAwhen Point A serves as the contact point
max(OC1, OC2, …) approximates OA(b) Only when
increases. (b) Only whenserves as A
Point the contact
serves as point between
the contact the
point
between
circle and the
the circle
path, and theA,
Points path,
B, Points
and C A, B, and
coincide to C coincide
form a to form
single point. aIn
single
this point. In this
scenario, the scenario,
equality
between the circle and the path, Points A, B, and C coincide to form a single point. In this scenario,
the equality sign in the inequality OC ≤ OA holds, enabling the proposed method to accurately
signequality
the in the inequality
sign in the ≤ OA holds,
OCinequality OCenabling the proposed
≤ OA holds, enablingmethod to accurately
the proposed method compute the true
to accurately
compute the true value of OA.
compute
value of OA. the true value of OA.
Mathematics 2023, 11, x FOR PEER REVIEW 9 of 20

3.3. Computations
Mathematics 2023, 11, 4424 9 of 20
According to [17], by replacing the constant translational velocity sequence in the
conventional DWA with a variable translational velocity sequence, a larger range of path
reach points can be achieved. As illustrated in Figure 10, under the conditions of a robot’s
3.3. Computations
initial velocity of (1.0, 0.0), maximum translational velocity of 2.0 , and maximum ac-
According to [17], by replacing the constant translational velocity sequence in the
celeration
conventional m/s
of 1.0DWA with, we compare
a variable the predicted
translational range
velocity of path
sequence, waypoints
a larger range offorpath
the con-
ventional DWA, constant translational acceleration model, and constant
reach points can be achieved. As illustrated in Figure 10, under the conditions translational
of a robot’s jerk
0.8 and 2.0 𝑠 . The acceleration model exhibits a larger range
m
model atvelocity
initial prediction times
of (1.0,
 0.0),
of
 maximum translational velocity of 2.0 s , and maximum
of acceleration
reaching points. m/s2we
of 1.0Here, , we compare
adopt the predicted
the acceleration rangeasofapath
model waypoints
variable velocityformodel,
the as
conventional DWA,
defined in (17) and (18). constant translational acceleration model, and constant translational
jerk model at prediction times of 0.8 and 2.0 [s]. The acceleration model exhibits a larger
range of reaching points. Here, we 𝑎⋅𝑡0
𝑣 adopt 𝑡 𝑡 𝑜𝑟 0model
the acceleration 𝑡 as𝑡 a variable velocity
model, as defined in (17)𝑣and
𝑡 (18). 𝑣 𝑡 𝑡 𝑡 𝑎𝑛𝑑 𝑎 0 , (17)
 0 𝑡 𝑡 𝑡 𝑎𝑛𝑑 𝑎 0
 v0 + a · t 0 ≤ t ≤ tea | or | 0 ≤ t ≤ ted
v(t) = v tea < t ≤ t p | and | a > 0 , (17)
 max 𝜔 𝑡 𝜔 . (18)
0 ted < t ≤ t p | and | a ≤ 0

ω ( t ) = ω0 . (18)

(a) When simulation time is 0.8 s (b) When simulation time is 2.0 s
Figure
Figure 10.10. Comparisonof
Comparison ofpath
path reach
reachpoints’
points’range.
range.

This model assumes that translational acceleration and rotational velocity remain
This model assumes that translational acceleration and rotational velocity remain
constant between time t0 and t p . v0 denotes the initial translational velocity of the robot,
constant between
vmax denotes thetime 𝑡 and
maximum 𝑡 . 𝑣 denotes
translational the ainitial
velocity, translational
represents velocity
the selected of the robot,
acceleration
𝑣 (a ∈denotes the maximum translational velocity, 𝑎 represents the selected
[− amax , amax ]), and t p is the simulation period. For simplicity, we denote vmax −vacceleration
0
as tea
a
∈ a𝑎 as t,ed𝑎.
(𝑎 and v0
), and 𝑡 is the simulation period. For simplicity, we denote as
𝑡 andWith asthe𝑡acceleration
. case as an example, we substitute (17) and (18) into (1)–(3) and
use v0 (t) and θ 0 (t) in place of v0 + a · t and θ (t0 ) + ω0 · t; consequently, (19), (20), (21)
With the acceleration case as an example, we substitute (17) and (18) into (1)–(3) and
are obtained.
use 𝑣 𝑡 and 𝜃 𝑡 in place of 𝑣( ω0𝑎·v0⋅(t𝑡)sinand 𝜃 a𝑡·cos(θ 0 (𝜔
(θ 0 (t))+ t)) ⋅ 𝑡; consequently, (19), (20), (21)
+ c1
are obtained. x path (t) = vmax
ω0 2 , (19)
sin 0
( ( ))
θ t + c
ω0 2
⋅ ⋅
(
−ω0 ·v0 (t)cos(θ 0 (t))+ a·sin(θ 0 (t)) 𝑐
𝑥 (t) =
y path 𝑡 ω0 2
+ c3
, , (20) (19)
− vωmax0
cos sin
( θ 𝜃 𝑡
0 ( t )) + c
4 𝑐

θ path (t) = ω0 · t. (21)

⋅ ⋅
Next, we simplify the𝑦 first interval
𝑡 of (19) as A(t) + c1 , the second 𝑐interval
, as B(t) + c2 , (20)
first interval of (20) as C (t) + c3 , and the second interval
cos 𝜃 𝑡as D ( t ) + c
𝑐 4 . Finally, we compute
the constant of integration using Equations (22)–(25).

𝑡 c1 =𝜔0, ⋅ 𝑡.
lim𝜃A(t) + (22) (21)
t →0+
Mathematics 2023, 11, 4424 10 of 20

lim C (t) + c3 = 0, (23)

t →0+

lim A(t) + c1 = lim B(t) + c2 , (24)

t→tea − t→tea +

lim C (t) + c3 = lim D (t) + c4 . (25)

t→ted − t→ted +

Subsequently, the position of the center must be determined because the radius of the
circle has been determined. First, we generate a pair of circles in contact with the starting
point of the path. The function of the inner circle is defined in (26) and (27), and
 that of the
outer circle is expressed in (28) and (29). The center points of the circles are 0, ωv00 and
 
v
0, ωp0 , respectively.
v0
xic (t) = sin(ω0 · t), (26)
ω0
v0
yic (t) = (1 − cos(ω0 · t)), (27)
ω0
vp
xoc (t) = sin(ω0 · t), (28)
ω0
vp
yoc (t) = (1 − cos(ω0 · t)). (29)
ω0
The other circles can be considered as the translation of this pair of circles, and the
magnitudes  of the translation are computed
  using (30)–(33). Hence,the centers of other
v0 v
circles are 0 − ∆xic (t), ω0 − ∆yic (t) and 0 − ∆xoc (t), ωp0 − ∆yoc (t) .

∆xic (t) = xic (t) − x path (t), (30)

∆yic (t) = yic (t) − y path (t), (31)

∆xoc (t) = xoc (t) − x path (t), (32)

∆yoc (t) = yoc (t) − y path (t). (33)

An example of the generation of two pairs of circles is shown in Figure 11. In this
example, the starting and ending points of the path were adopted as the contact points with
circles. Circles c1 and c3 are inner side, circles c2 and c4 are outer side circles. Equations
(34)–(37) define the distance from obstacle O to the center of each circle.

ric1 = k( xo , yo ) − (∆xic (0), ∆yic (0))k2 , (34)

ric2 = ( xo , yo ) − ∆xic t p , ∆yic t p

2
, (35)

roc1 = k( xo , yo ) − (∆xoc (0), ∆yoc (0))k2 , (36)

roc2 = ( xo , yo ) − ∆xoc t p , ∆yoc t p

2
. (37)
Mathematics 2023, 11, x FOR PEER REVIEW 11 of 20

Mathematics 2023, 11, x FOR PEER REVIEW 11 of 2

Mathematics 2023, 11, 4424 11 of 20

Figure 11. Example generating two pairs of circles.

Figure 11. Example generating two pairs of circles.

The distance 𝑑 from the obstacle to the path is expressed by (38).
Figure 11. Example generating two pairs of circles.
The distance d from the obstacle to the path is expressed by (38).
max 𝑟 ,𝑟 𝑑 min 𝑟 ,𝑟 . (38)
The distance 𝑑 from vp
the obstacle
vp
 to the path
 is expressed by(38).
v0 v
However, max roc1 − steps
additional , roc2 −are required
≤ d ≤ min ic1 −
for rpractical , ric2 − 0 .
applications. (38)
As illustrated in
max 𝑟v 0 v
,𝑟0 𝑑 min 𝑟 v 0 v, 0𝑟 . (38
Figure 12, a polar coordinate system centered on the intersection of the normals of the
However,
starting point additional
(A) and endingsteps are required
point (B) offor
thepractical
path applications.
was created. As illustrated
Two normal in divide
lines
Figure
However,
12, a polar
additional
coordinate
steps
system
are required
centered on
for intersection
the
practical applications.
of the normals
Asofillustrated
the
in
the plane into
Figure point
12, a (A)four
polar regions. The polar angle of the obstacle was then adopted to determine
starting andcoordinate
ending point system centered
(B) of the path was oncreated.
the intersection
Two normal of lines
the normals
divide of th
the four areas
starting point to(A)
which
and it belonged.
ending pointFor
(B)obstacles
of the such
path was 𝑂1 and Two
as created. 𝑂4 in the grey
normal area,
lines the
divid
the plane into four regions. The polar angle of the obstacle was then adopted to determine
distance
thefour
the plane tointo
thetofour
areas path
whichis itthe
regions. minimum Fordistance
The polar
belonged. angle of from
obstacles the
such the start
obstacle
as O1 was
andand end
then
O4 of grey
the path.
in adopted
the For ob-
to determin
area,
stacles
the such
thedistance
four areas as
to the𝑂2 and
path is it
to which 𝑂3
the in the white
minimumFor
belonged. area,
distance the proposed
fromsuch
obstacles as 𝑂1
the start method
andandend 𝑂4determines
of the
in path. the range
For area,
the grey th
of distances,
obstacles such and
as O2the
andmiddle
O3 in value
the of
white the
area,range
the is adopted
proposed as
method the distance.
determines
distance to the path is the minimum distance from the start and end of the path. For ob the range
of distances,
stacles suchand
as the
𝑂2 middle
and 𝑂3value of the
in the range
white is adopted
area, as the distance.
the proposed method determines the rang
of distances, and the middle value of the range is adopted as the distance.

Figure 12. Different methods are employed for obstacles in different areas.
Figure 12. Different methods are employed for obstacles in different areas.
4. Experiment
When generating paths using the constant velocity model within the conventional
4. Experiment
DWA,
Figureall
12.paths are circular
Different methodsarcs. Therefore,for
are employed when applying
obstacles the proposed
in different areas. method, only
When generating
one transformation paths with
is required, using nc the constant
set to 1. Hence,velocity model
the proposed withinnecessitates
method the conventional
only
DWA, a single pathare
all paths
4. Experiment through all obstacle
circular points to compute
arcs. Therefore, distances,the
when applying contrasting
proposed with the
method, only
conventional DWA
one transformation which
When generating
requires
is required,
paths using with the𝑛 set
traversing both obstacle
to 1. Hence,
constant
and
velocity
path
themodelpoints.
proposed This results
method
within
in
necessitates
the conventiona
a noteworthy
only a single advantage
path in computational
through all obstacle time
pointsfor to
thecompute
proposeddistances,
method. contrasting with the
DWA, all paths are
Furthermore, circular arcs.
as illustrated Therefore,
in Figure when applying
2, the proposed the proposed
method reliably method,
calculated the only
conventional
onedistance DWA
transformation which
is the requires
required, traversing
with 𝑛DWA, both
set predicted obstacle
to 1. Hence, and
theare path
proposed points.
method This results
necessitate
true values. In conventional paths represented through
in a
onlynoteworthy
a single advantage
path through in
allcomputational
obstacle points time
to for the
compute proposed
distances, method.
discrete path points, and collision detection is executed by determining the minimum with th contrasting
Furthermore,
distance between
conventional theas
DWA illustrated
sets of path
which and
requiresin obstacle
Figure 2, theboth
points.
traversing proposed
A lower method
number
obstacle andof reliably calculated
pathpoints.
path points can the
This result
true distance values. In the conventional DWA, predicted
in a noteworthy advantage in computational time for the proposed method. paths are represented through
discrete path points,
Furthermore, as and collision
illustrated detection
in Figure is executed
2, the proposedby determining
method reliablythe minimum
calculated th
true distance values. In the conventional DWA, predicted paths are represented through
discrete path points, and collision detection is executed by determining the minimum
 distance between the sets of path and obstacle points. A lower number of path points can
Mathematics 2023, 11, 4424
result in substantial errors in collision detection, thereby highlighting a significant 12 of 20
accu-
racy advantage of the proposed method.
Given that the proposed method demonstrated both swifter computation time and
enhanced accuracyerrors
result in substantial compared to the
in collision conventional
detection, thereby DWA employing
highlighting the constant
a significant accuracyvelocity
model, we restricted our assessment
advantage of the proposed method. to the efficacy of the proposed method when gener-
atingGiven
pathsthat
using
thethe more intricate
proposed acceleration model.
method demonstrated both swifter computation time and
The experimental
enhanced accuracy comparedenvironment and controlDWA
to the conventional program were the
employing constructed utilizing the
constant veloc-
ity model, we restricted our assessment to the efficacy of the proposed
Robot Operating System (ROS) [33]. In these experiments, a two-dimensional simulator method when
generating
named Stagepaths
[34]using
wasthe more intricate
adopted, acceleration
given the model.
robot’s movement was confined to a plane. Due
The experimental environment and control program were constructed utilizing the
to the extensive computation time needed to ascertain the true distance values from the
Robot Operating System (ROS) [33]. In these experiments, a two-dimensional simulator
obstacle points to the path, it was impractical to concurrently compute the accuracy of
named Stage [34] was adopted, given the robot’s movement was confined to a plane. Due
distance calculation
to the extensive and computation
computation time neededtime of the two
to ascertain methods
the true in the
distance navigation
values from the experi-
ment. Consequently, two distinct experiments were conducted. In
obstacle points to the path, it was impractical to concurrently compute the accuracy the initial experiment,
of
the accuracy
distance of distance
calculation calculation
and computation timeofofthe twomethods
the two methods wasnavigation
in the compared, with the robot
experiment.
stationary
Consequently, andtwo
onlydistinct
predicting the path.
experiments were conducted. In the initial experiment, the
accuracy
In theof distance
second calculation
experiment, of the
thetwo methods wastime
computation compared,
for eachwith the robot
method stationary
during navigation
and only predicting
was measured. the path.
In the second experiment, the computation time for each method during navigation
was measured.
4.1. Conditions of Distance Calculation Accuracy Experiment
4.1. Conditions
Figure 13ofillustrates
Distance Calculation Accuracy Experiment
the environment of the first experiment. The initial position of
the robot
Figureis13
setillustrates
to (0, 0),the
and the black line
environment of therepresents the simulation
first experiment. path.
The initial We randomly
position of
the robot is100
generated set to (0, 0),within
points and thea black linearea
circular represents
in fronttheofsimulation
the robotpath. We randomly
to simulate the obstacles.
generated
The radius100 points
of the withinarea
circular a circular
was 5 aream ,inwhich
front of
wastheconsistent
robot to simulate
with the the obstacles.distance
detection
The
of radius
the laserofrangefinder.
the circular area
Thewas 5 [m], which
simulation periodwas 𝑡consistent
was setwith the sdetection
to 2.0 distance
. The translational and
of the laser rangefinder. The simulation period t p was set to 2.0 [s]. The translational and
rotational velocities were set to 1.0 [m/s] and 1.0 rad/s , respectively. The translational
rotational velocities were set to 1.0 [m/s] and 1.0 [rad/s], respectively. The translational
acceleration
acceleration ininthethesimulation
simulation period
period waswas set−to
set to 1.0,−1.0,
−0.5,−0.5, 0.0, 0.5
0.0, 1.0, 0.5 2 m/s
1.0, m/s . .

13. Random
Figure 13.
Figure Randompoints
pointsforfor
simulating obstacles.
simulating obstacles.
The number of circles in the proposed method was set to 1, 2, and 3 pairs; hence, nc
The number of circles in the proposed method was set to 1, 2, and 3 pairs; hence, 𝑛
values were 2, 4, and 6. The generated circle contacted the start, middle, and end points of
values were 2, path.
the simulation 4, andIn6.this
Thestudy,
generated circle contacted
we consider only the case theinstart,
whichmiddle,
a pair of and end ispoints of
circles
the simulation
generated path.tangent
at the same In thispoint;
study, we consider
therefore, when nonly
c is 2,the case
there arein which
three casesaofpair of circles is
tangent
generated
points at theatstarting
the same(s), tangent point;
middle (m), and therefore,
ending points whenof the𝑛 path.
is 2,When
there nare
c is three
4, therecases
are of tan-
three cases of tangent points at the starting and middle (s + m), starting
gent points at the starting (s), middle (m), and ending points of the path. When and ending (s + e), 𝑛 is 4,
and middle and ending (m + e) points of the path. When n = 6, only one
there are three cases of tangent points at the starting cand middle (s + m), starting and tangent point
exists at(s
ending the+ starting, middle, and
e), and middle and ending
ending(s(m + m+ +e)e)points
points ofof the
the path.
path. When 𝑛 = 6, only one
For the conventional DWA, the time interval ∆t is set to 0.10, 0.05, 0.02 [s]; hence, n p
tangent point exists at the starting, middle, and ending (s + m + e) points of the path.
values are 20, 40, and 100, respectively.
For the conventional DWA, the time interval ∆𝑡 is set to 0.10, 0.05, 0.02 [s]; hence, 𝑛
values are 20, 40, and 100, respectively.
Mathematics 2023, 11, 4424 13 of 20

4.2. Results of Distance Calculation Accuracy Experiment

We computed the distance from each obstacle to the path using analytic geometry as
the baseline. The difference between the distance obtained by the proposed method and the
baseline was used as the error, and we adopted the average value of the errors to evaluate
the performance of the distance estimation.
Table 1 presents the average error of the distance estimation in millimeters. First, as
the number of circles increases, the mean of the average error in the distance estimation
for each acceleration decreases. Specifically, when nc = 2, the circle with the tangent point
at the midpoint (m) of the path better represents the trend of the entire path, resulting in
a lower error. Similarly, when nc = 4, the circle with tangent points at the starting and
middle points (s + m) of the path provides a better description of the path trend, resulting
 (s +m). Finally, when nc = 6, for translational accelerations ranging
in a lower error for
from −1.0 to 1.0 m/s2 , (s + m + e) has the smallest estimation error. This matches the
expectation that the greater the number of circles, the smaller the distance estimation error.

Table 1. Average error of distance estimation under proposed method.

Trans nc = 2 nc = 4 nc = 6
acc. s m e s+m s+e m+e s+m+e
−1.0 34 28 181 29 30 29 31
−0.5 102 22 84 16 19 27 13
0.0 0 0 0 0 0 0 0
0.5 98 30 161 33 30 47 30
1.0 55 78 317 46 81 71 42
Mean 58 32 148 25 32 35 23
Std 38.9 25.5 105.6 15.7 27.0 23.6 14.9

Subsequently, the proposed method approximates the variable velocity path with
multiple constant velocity circular paths through vertical observation. Consequently, when
variations in velocities were minimal, the error in distance estimation was reduced. Specif-
ically, when the translational acceleration during the simulation period is 0.0 m/s2 ,


resulting in the robot’s constant velocity circular motion, the proposed method can accu-
rately compute the actual distance between obstacles and paths, particularly when the
translational acceleration during the simulation period is nullified.
Table 2 displays the average error in distance estimation, measured in millimeters.
Examining horizontally, it can be observed that as the number of path points escalates, the
mean of the average error in distance estimation for each acceleration diminishes under
both tangent and secant models. However, a decrease in estimation error corresponds to
an increase in computing cost. The accuracy of distance estimation employing the secant
model consistently surpassed that of the tangent model.

Table 2. Average error of distance estimation under conventional DWA.

Trans np =20 np =40 np =100

acc. Tangent Secant Tangent Secant Tangent Secant
−1.0 55 51 42 40 34 33
−0.5 23 18 12 9 5 4
0.0 32 1 16 1 7 1
0.5 78 39 51 32 36 28
1.0 104 52 69 43 48 38
Mean 59 33 38 25 26 21
Std 29.5 19.9 21.5 17.1 17.1 15.6
Mathematics 2023, 11, x FOR PEER REVIEW 14

Mathematics 2023, 11, 4424 When observed vertically, both the tangent and secant models demonstrate 14 of 20 the c
acteristic that a reduction in translational acceleration corresponds to a decrease in
tance estimation error.
We compared
When the distance
observed vertically, both theestimation
tangent and results
secant of the best-performing
models configuratio
demonstrate the charac-
teristic that a reduction in translational acceleration corresponds to a decrease
the proposed method, that is, (s, s + m, s + m + e), with the tangent and secant mode in distance
estimation error.
the conventional DWA under various 𝑛 conditions. In Figure 14, the horizontal axi
We compared the distance estimation results of the best-performing configuration of
notes
the the number
proposed method,of path
that is, (s, s + m, s𝑛+ m
points in+the conventional
e), with the tangent DWA, themodels
and secant number of circular
in the
𝑛 in the proposed
conventional DWA under method,
various n and the vertical
p conditions. axis 14,
In Figure denotes the average
the horizontal error of dist
axis denotes
estimation.
the number of Thepath error
points bars indicate
n p in the the standard
conventional DWA, thedeviation
number of caused by nthe
circular arcs c in accelera
the proposed
change. Themethod, and the shows
comparison vertical axis
thatdenotes the average
the proposed error ofachieves
method distance estimation.
a similar level o
The errorinbars
curacy indicateestimation
distance the standard as deviation
the secant caused by the
model, acceleration
which performs change.
betterThe
in the con
comparison shows that the proposed method achieves a similar level of accuracy in distance
tional DWA.
estimation as the secant model, which performs better in the conventional DWA.

Figure Comparison
Figure14.14. of distance
Comparison estimation
of distance error. error.
estimation
4.3. Conditions of Navigation Experiment
4.3. The
Conditions
second of Navigation
experiment wasExperiment
conducted to compare the performances of the two
methods in terms of the computational
The second experiment was conducted cost of navigation.
to compare Table
the3performances
lists the parameters
of the two m
of
ods in terms of the computational cost of navigation. Table 3 listsenvironment
the PC and other libraries, and Figure 15 illustrates the experimental the parameters in of th
the simulator; the environment shown in (b) is narrower than (a). The maximum velocity
and other libraries, and Figure 15 illustrates the experimental environment in the sim
of the robot was set to 2.0 [m/s], maximum acceleration to 1.0 [m/s2 ], the robot’s initial
tor; the to
position environment
(−18, 0), andshown in (b)
the target is narrower
to (18, than (a).
0). The control The maximum
frequency velocity
was established at of the r
was set to 2.0 [m/s], maximum acceleration to 1.0 [m/s
10 Hz. In this configuration, we utilized an automatic time padding technique. Should the positio
], the robot’s initial
(−18, 0), and
cumulative the target
duration consumed to (18,
by all0).procedures
The controlwithinfrequency
a loop fallwas below established at 10 Hz. In
100 milliseconds,
the system automatically allocates the idle time to uphold
configuration, we utilized an automatic time padding technique. Should the a consistent control frequency of cumul
10 Hz. The range of the laser rangefinder was set to 5 [m], and
duration consumed by all procedures within a loop fall below 100 milliseconds, the the resolution was set to
1, 2, or 3 [nums/◦ ]. For the environment illustrated in Figure 15, the average numbers of
tem automatically allocates the idle time to uphold a consistent control frequency of 10
obstacle points detected per frame during autonomous navigation no is approximately 130,
Theand
260, range
390.of Wethe laser rangefinder
controlled the value ofwas no byset to 5 [m],
adjusting theandlaserthe resolution
rangefinder was set to 1, 2
resolution;
therefore, no is the result and is for reference only. The simulation period was set to of obs
[nums/°]. For the environment illustrated in Figure 15, the average numbers
2points
[s], anddetected
the time perintervalframe ∆t of
during autonomous
the conventional DWA navigation
was set to 1.00,𝑛 is0.50,
approximately
0.20, 0.10, 130,
and 390. We controlled the value of 𝑛 by adjusting the laser rangefinder resolu
0.05, 0.02 [s], according to (7); n p values are 2, 4, 10, 20, 40, and 100, respectively. The
number
therefore,of circles
𝑛 isinthe theresult
proposed
and method was set toonly.
is for reference 1, 2, The
and 3simulation
pairs; hence, nc values
period was set to
were 2, 4, and 6. Both the conventional DWA and proposed method divide the selectable
and the time interval Δ𝑡 of the conventional DWA was set to 1.00, 0.50, 0.20, 0.10,
translational acceleration and rotational velocity ranges into five equal parts, generating
0.02
25 [s], according
simulation to (7); 𝑛 values
paths simultaneously. Ten are 2, 4, 10,
automatic 20, 40, and
navigations 100,
were respectively.
performed in theThe num
of circles environment
simulation in the proposed method
and the averagewas set to 1, 2,
computation and
time of 3the
pairs;
DWAhence, 𝑛 values were
was determined
in
andmilliseconds.
6. Both the Since all navigation
conventional DWA experiments in both environments
and proposed method divide werethe
successful
selectable tra
only for n > 20, using the results obtained with n = 20 and
tional acceleration and rotational velocity ranges into five equal parts, generating
p p n c = 2 as the baseline, we 25
attempted to determine whether there was a significant change in the calculation time
ulation paths simultaneously. Ten automatic navigations were performed in the sim
when n p and nc were varied. We adopted * to indicate p-values less than 0.05 and ** to
tion
indicateenvironment
p-values less and than the
0.01.average computation time of the DWA was determined in
liseconds. Since all navigation experiments in both environments were successful onl
𝑛 20 , using the results obtained with 𝑛 20 and 𝑛 2 as the baseline, w
tempted to determine whether there was a significant change in the calculation time w
𝑛 and 𝑛 were varied. We adopted * to indicate p-values less than 0.05 and ** to ind
p-values less than 0.01.
Mathematics 2023, 11, 4424 15 of 20
Mathematics 2023, 11, x FOR PEER REVIEW 15 of 20

Table 3. Parameters of experiment environment.

Table 3. Parameters of experiment environment.
System Configuration Details
System Configuration
OS Details
Ubuntu 18.04.5 LTS
OS Ubuntu 18.04.5 LTS
ROS Melodic 1.14.9
ROS Melodic 1.14.9
Stage
Stage 4.3.04.3.0
CPU CPU IntelIntel
CoreCore i7-8700
i7-8700 3.20Ghz×12
3.20Ghz×12
Numpy
Numpy 1.16.4
1.16.4
Scipy
Scipy 1.3.01.3.0

(a) Environment 1 (b) Environment 2

Figure15.
Figure 15. Experiment
Experiment environment.
environment.

4.4.
4.4. Results
Results of
of Navigation
Navigation Experiment
Comparison
Comparison of the the travel
traveldistance
distanceandandtime
timeisisshown
shown in in Tables
Tables 4 and
4 and 5. the
5. In In exper-
the ex-
perimental
imental setup,setup,
thethe straight-line
straight-line distance
distance between
between the the initial
initial position
position andand
the the target
target is
is 36
36 [m].
[m]. TheThe results
results inin Table4 4indicate
Table indicatea asignificant
significantdifference
differenceinintravel
traveldistance
distanceand and other
other
results for n𝑛p < 10.
results for 10. This
This is
is attributed
attributed toto the
the sparse
sparse distribution of path points,
points, causing
causing
the robot to lose its way
the robot to lose its way during autonomous navigation. Since Environment 2 is narrower
Environment 2 is narrower
than
than Environment
Environment 1, examining the results results inin Table
Table 6, when 𝑛n p is less than 20, successful
6,when successful
navigation
navigation becomes
becomes challenging. Hence, for
challenging. Hence, for more
more complex
complex environments,
environments, robots
robots require
require
more
morepath
path points
points for
for successful
successful navigation.
navigation.
When there are enough path points to guarantee successful navigation, one might
Table 4. Travel
observe distance
that there is and time of conventional
no substantial disparityDWA andnavigation
in the proposed method
travel in Environment
distance 1.
and time
between the conventional DWA and our proposed method. This similarity arises because
Laser Rangefinder the experiments 1 were conducted at an identical 2 control frequency. In real-world
3 robotic
Resolution [nums/◦ ]
operations, each control loop encompasses not just local path planning, but also tasks of
Number of Detected environmental perception and specific ntask processing, such as real-time
no ≈ 130 o ≈ 260 no ≈ 390image pro-
Obstacle Points per Frame cessing. Hence, curtailing the time spent on local path planning affords additional time
for other
Travelcrucial operations.
Travel Travel Travel Travel Travel
Distance [m] Time [s] Distance [m] Time [s] Distance [m] Time [s]
Table 4. Travel distance and time of conventional DWA and proposed method in Environment 1.
np = 2 90.6 79.9 90.4 79.9 89.4 79.9
Laser Rangefinder
n p Resolution
=4 44.8
1 26.7 44.4
2 26.5 43.0
3 25.5
[nums/°]
Conventional n p = 10 39.2 23.4 39.4 23.1 39.0 23.5
Number of Detected Obstacle
DWA n = 20 39.0 𝒏𝒐 𝟏𝟑𝟎
23.6 39.2 𝒏𝒐 𝟐𝟔𝟎
23.5 39.2 𝒏𝒐 𝟑𝟗𝟎 23.6
Points per pFrame
n p = 40 39.0 Travel 25.0
Travel Time 39.0
Travel 25.2 Time
Travel 39.0
Travel 25.0Time
Travel
n p = 100 Distance [m] 24.8 [s]
39.0 Distance
39.0 [m] 25.0[s] Distance
39.6 [m] [s]
23.6
𝑛 2 90.6 79.9 90.4 79.9 89.4 79.9
nc = 2 38.4 21.5 38.4 21.6 38.4 21.5
Proposed
𝑛 4 44.8 26.7 44.4 26.5 43.0 25.5
Conventional nc = 4 38.4 21.6 38.0 21.3 38.1 21.4
method 𝑛 10 39.2 23.4 39.4 23.1 39.0 23.5
DWA nc =𝑛6 20 39.1 39.0 25.1 23.6 39.039.2 24.9
23.5 39.1
39.2 25.0
23.6
𝑛 40 39.0 25.0 39.0 25.2 39.0 25.0
Mathematics 2023, 11, 4424 16 of 20

Table 5. Travel distance and time of conventional DWA and proposed method in Environment 2.

Laser Rangefinder
1 2 3
Resolution [nums/◦ ]
Number of Detected
no ≈ 130 no ≈ 260 no ≈ 390
Obstacle Points per Frame
Travel Travel Travel Travel Travel Travel
Distance [m] Time [s] Distance [m] Time [s] Distance [m] Time [s]
np = 2 86.0 79.9 85.2 79.9 87.6 79.9
np = 4 66.2 52.7 63.4 46.8 61.2 42.9
Conventional n p = 10 60.8 42.5 61.2 41.6 61.2 41.5
DWA n p = 20 53.4 30.7 53.6 31.4 53.8 30.9
n p = 40 53.6 32.1 53.2 32.5 53.6 32.2
n p = 100 55.8 31.7 56.0 31.8 55.6 31.5
nc = 2 51.6 30.5 54.3 31.6 52.6 31.1
Proposed nc = 4 53.1 31.4 54.4 32.2 53.6 31.7
method
nc = 6 53.6 31.7 54.1 32.0 54.3 32.1

Table 6. Computing time of conventional DWA and proposed method in Environment 1.

Laser Rangefinder Resolution

1 2 3
[nums/◦ ]
Number of Detected Obstacle
no ≈ 130 no ≈ 260 no ≈ 390
Points per Frame
Mean p Mean p Mean p
np = 2 16.1 - 20.6 - 24.0 -
np = 4 17.8 - 21.3 - 23.8 -
Conventional n p = 10 25.4 - 30.9 - 32.9 -
DWA n p = 20 34.6 - 36.1 - 39.1 -
n p = 40 37.8 * 38.7 * 39.3
n p = 100 47.3 ** 47.5 ** 48.8 **
nc = 2 24.7 - 25.0 - 30.1 -
Proposed nc = 4 25.1 28.7 31.5
method
nc = 6 29.4 * 30.2 34.7
* to indicate p-values less than 0.05 and ** to indicate p-values less than 0.01.

When there are enough path points to guarantee successful navigation, one might observe
that there is no substantial disparity in the navigation travel distance and time between the
conventional DWA and our proposed method. This similarity arises because the experiments
were conducted at an identical control frequency. In real-world robotic operations, each control
loop encompasses not just local path planning, but also tasks of environmental perception
and specific task processing, such as real-time image processing. Hence, curtailing the time
spent on local path planning affords additional time for other crucial operations.
A comparison of the computing time per frame is shown in Tables 6 and 7. According
to Equations (7) and (8), computing cost escalates with sensor resolution. Examining the
results in Tables 6 and 7, an anticipated increase in computation time is observed with the
rise in the number of detected obstacle points for both methods. Nonetheless, in the case of
the conventional DWA method, there exists a notable augmentation in computation time as
the number of path points n p increases. The generation of paths in the conventional DWA
necessitates iterative computations, which are inherently time consuming. Consequently,
we may infer that the computation time of the conventional DWA experiences a substantial
Mathematics 2023, 11, 4424 17 of 20

surge with an increased number of iterations. While it is feasible to diminish computation

Mathematics 2023, 11, x FOR PEER REVIEW 17 of 20
time by enlarging time intervals, this adjustment could potentially incur a higher risk of
navigation failure due to discrete path points.

Table 7.
Table 7. Computing
Computing time
time of
of conventional
conventional DWA
DWA and
and proposed
proposed method
method in
in Environment
Environment2.
2.

Laser Rangefinder Resolution

Laser Rangefinder Resolution 1 2 3
[nums/°]
◦] 1 2 3
[nums/
Number of Detected Obstacle
n𝒏 𝟏𝟑𝟎 n𝒏o𝒐≈ 260
𝟐𝟔𝟎 𝒏n𝒐 ≈ 390
𝟑𝟗𝟎
Number of Detected Obstacle
Pointsper
perFrame
Frame o𝒐≈ 130 o
Points
Mean
Mean p p Mean
Mean pp Mean
Mean p
p
𝑛 2 21.0 25.6 29.7
np = 2 21.0 25.6 29.7
𝑛 4 21.9 32.7 35.5
np = 4 21.9 32.7 35.5
Conventional 𝑛 10 31.0 36.0 37.6
Conventional n p = 10 31.0 36.0 37.6
DWA 𝑛 20 36.7 38.3 38.5
DWA n p𝑛= 2040 36.7 38.3 38.5
37.8 40.1 40.2
n𝑛p = 40100 37.8
46.8 ** 40.1
49.1 ** 40.2
50.0 **
np =𝑛 1002 46.8
28.1 ** 49.1
28.7 ** 50.0
34.5 **
Proposed
nc𝑛= 2 4 28.8
28.1 32.9
28.7 36.1
34.5
method
Proposed nc𝑛= 4 6 33.7
28.8 * 34.6
32.9 39.8
36.1 *
method
nc = 6 * to indicate
33.7p-values less than
* 0.05 and **34.6
to indicate p-values less than 0.01.
39.8 *
* to indicate p-values less than 0.05 and ** to indicate p-values less than 0.01.
Contrastingly, the computation time of the proposed method does not exhibit a sig-
nificant alteration as
Contrastingly, thethe number oftime
computation circles 𝑛 proposed
of the increasesmethod
and remains
does notconsistently lower
exhibit a signifi-
than that of the conventional DWA method. The proposed method solely
cant alteration as the number of circles nc increases and remains consistently lower than necessitates sub-
stituting the velocity into the formula to compute the path equation
that of the conventional DWA method. The proposed method solely necessitates substitut- and ascertain the lo-
cation
ing theof the tangent
velocity point
into the of a finite
formula numberthe
to compute of path
arcs, equation
a processandwithascertain
negligible thecomputa-
location
tional
of overhead.
the tangent Thus,
point of an increment
a finite number in of
thearcs,
number of arcs
a process within
with the proposed
negligible method
computational
does not significantly impinge upon the computation time.
overhead. Thus, an increment in the number of arcs within the proposed method does not
As illustrated
significantly impingein upon
Figurethe 16,computation
based on the given values of 𝑛 or 𝑛 , we calculate the
time.
average computation
As illustrated time for
in Figure 16,different
based onvalues of 𝑛 values
the given , representing the
of n p or n relationship
c , we calculate thebe-
tween 𝑛computation
average or 𝑛 and timethe computation time. InofFigure
for different values 16, the horizontal
no , representing axis denotes
the relationship the
between
nnumber of path
p or nc and points 𝑛 in the
the computation time.conventional
In Figure 16, DWA and the number
the horizontal of circular
axis denotes arcs 𝑛
the number
of
in path points n pmethod,
the proposed in the conventional
and the vertical DWA axisand the number
denotes of circular
the computation nc inbars
arcsError
time. the
indicate deviations caused by the change in 𝑛 . Because the value of 𝑛 increases, there
proposed method, and the vertical axis denotes the computation time. Error bars indicate
deviations
was an overall caused by the
upward change
shift no . Because
in theinresults. In boththe value of no 1increases,
Environment there was2,an
and Environment it
overall upward
is evident shiftconventional
that the in the results.method
In bothconsistently
Environmentexhibits
1 and Environment 2, it is evident
longer computation times
that
than the
the conventional
proposed method, method andconsistently exhibits longer
there is a substantial increasecomputation
in computation timestimethan 𝑛
asthe
increases. method, and there is a substantial increase in computation time as n p increases.
proposed

(a) Environment 1 (b) Environment 2

16. Comparison of computation time.
Figure 16.

4.5. Overall Analysis

Given that the range of randomly generated obstacle points in the first experiment
aligned with that of the laser rangefinder in the second experiment, it is noteworthy that
the accuracy of the proposed method was intrinsically linked to acceleration. To capture
Mathematics 2023, 11, 4424 18 of 20

4.5. Overall
Mathematics 2023, 11, x FOR PEER REVIEW Analysis 18 of 20
Given that the range of randomly generated obstacle points in the first experiment
aligned with that of the laser rangefinder in the second experiment, it is noteworthy that
the accuracy of the proposed method was intrinsically linked to acceleration. To capture
the maximum generated error, the parameters for the first experiment were meticulously
the maximum generated error, the parameters for the first experiment were meticulously
chosen to encompass a broader range of velocity variations throughout the simulation
chosen to encompass a broader range of velocity variations throughout the simulation
period. Hence, we posit that the distance estimation error yielded in the second experi-
period. Hence, we posit that the distance estimation error yielded in the second experiment
ment did not surpass that of the first. By amalgamating the distance estimation accuracy
did not surpass that of the first. By amalgamating the distance estimation accuracy from the
from the first experiment with the computation time from the second, we constructed
first experiment with the computation time from the second, we constructed cost–accuracy
cost–accuracy (COQ) curves for both the conventional DWA and the proposed method.
(COQ) curves for both the conventional DWA and the proposed method. These curves,
These curves,
eschewing eschewing
specific parameterspecific parameter
settings, illuminatesettings, illuminate
the trade-off the trade-off
between computation between
time
computation time and distance estimation accuracy. As depicted
and distance estimation accuracy. As depicted in Figure 17, with calculation time in Figure 17, with calcu-
on
lation
the time on axis
horizontal the horizontal
and distanceaxisestimation
and distance estimation
accuracy on theaccuracy
vertical,on
thethe vertical,ofthe
closeness a
closeness curve
method’s of a method’s curve
to its origin to its origin
signifies signifies
exemplary exemplary In
performance. performance. In both 1Envi-
both Environment and
ronment 1 and
Environment 2, aEnvironment 2, a discernible
discernible negative correlationnegative
between correlation
computation between
time computation
and distance
time and distance estimation accuracy is observable, corroborating
estimation accuracy is observable, corroborating our hypothesis. Moreover, the curveour hypothesis. More-
of
over, the curve of the proposed method, compared to that of the
the proposed method, compared to that of the conventional DWA, is both closer to theconventional DWA, is
both closer to the origin and possesses a steeper gradient. This proximity
origin and possesses a steeper gradient. This proximity to the origin implies higher dis- to the origin
implies
tance higher distance
estimation accuracy estimation accuracy computational
within equivalent within equivalent time,computational
while a steeper time, while
gradient
a steeper gradient denotes more substantial enhancements in accuracy
denotes more substantial enhancements in accuracy per unit of computational time. Thus, per unit of compu-
tational
the time. Thus,
proposed method thedemonstrates
proposed method demonstrates
superior performance superior performance
relative relative to
to the conventional
the conventional
DWA method. DWA method.

(a) Environment 1 (b) Environment 2

17. COQ graph.
Figure 17.

The
The DWA
DWAalgorithm,
algorithm,a aprevalent
prevalentlocal path
local planning
path planningtechnique, is extensively
technique, utilized
is extensively uti-
in robotics navigation and autonomous driving research. Our analysis of references
lized in robotics navigation and autonomous driving research. Our analysis of references [35–40]
indicates that boththat
[35–40] indicates research-oriented mobile robot
both research-oriented platforms
mobile and commercial
robot platforms robots gener-
and commercial ro-
2

ally
botsoperate
generallyat maximum linear velocities
operate at maximum linearunder 2 [m/s]
velocities and
under accelerations
2 [m/s] below 1 m/s
and accelerations below,
placing
1 m/s them within
, placing low-velocity
them environments.
within low-velocity In these settings,
environments. In these our proposed
settings, method
our proposed
showcases benefits in terms of distance calculation accuracy and computational
method showcases benefits in terms of distance calculation accuracy and computational time when
juxtaposed
time when juxtaposed with conventional approaches. Nonetheless, it is imperativethe
with conventional approaches. Nonetheless, it is imperative to acknowledge to
confines of ourthe
acknowledge investigation.
confines of our Forinvestigation.
example, the For
efficacy of ourthe
example, proposed
efficacy method remains
of our proposed
unverified for diverse
method remains robot configurations,
unverified such
for diverse robot as omnidirectional
configurations, suchrobots or drones. More-
as omnidirectional ro-
over, the employment of multiple arcs to approximate non-circular paths
bots or drones. Moreover, the employment of multiple arcs to approximate non-circular in our method
may induce substantial curvature discrepancies in non-circular trajectories, particularly in
paths in our method may induce substantial curvature discrepancies in non-circular tra-
high-acceleration contexts like autonomous driving. These variations might influence the
jectories, particularly in high-acceleration contexts like autonomous driving. These varia-
distance calculation precision of our approach. Expanded research and experimentation,
tions might influence the distance calculation precision of our approach. Expanded re-
especially in high-acceleration situations and with various robot architectures, are essential
search and experimentation, especially in high-acceleration situations and with various
to thoroughly ascertain the effectiveness and versatility of our proposed method across
robot architectures, are essential to thoroughly ascertain the effectiveness and versatility
different real-world scenarios.
of our proposed method across different real-world scenarios.

5. Conclusions
The aim of this study was to diminish the processing time associated with the DWA.
By reducing the processing time of the DWA, the control frequency of a robot can be in-
creased, thus making the robot more flexible and responsive to unexpected situations. In
Mathematics 2023, 11, 4424 19 of 20

5. Conclusions
The aim of this study was to diminish the processing time associated with the DWA. By
reducing the processing time of the DWA, the control frequency of a robot can be increased,
thus making the robot more flexible and responsive to unexpected situations. In pursuit of
this objective, we introduced a method centered on generating circles and executing polar
coordinate transformations. When utilizing the conventional DWA constant velocity model
for path prediction, a novel method was proposed that represents paths with circular arcs
and facilitates collision detection. Furthermore, when applying the acceleration model for
path predictions, we extended this method to approximate paths with multiple circular
arcs and conduct collision detection. Our proposed method employs a representation
that forgoes path discretization, eliminating the need for iterating velocity pairs in the
time dimension. Consequently, the computation of distance from obstacles to the path is
streamlined. We also ascertained that our method could accurately determine the interval
wherein the distance is situated. Simulation experiments substantiated the efficacy of the
proposed approach.
In this discourse, we solely contrast the processing times between the conventional
DWA and the proposed method using a wheeled differential drive robot navigation task.
We posit that path modeling is a fundamental operation capable of reducing the time
required for collision detection computations. Consequently, we aspire to extend the
applicability of our proposed method to other prevalent robots, including those with
car-like or omnidirectional drives.

Author Contributions: Methodology, Z.L. and R.T.; Validation, Z.L.; Writing—original draft, Z.L.;
Writing—review & editing, R.T.; Supervision, R.T. All authors have read and agreed to the published
version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: Research data will be uploaded to our github page: github.com/arialn/.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Zhao, X.F.; Liu, H.Z.; Lin, S.X.; Chen, Y.K. Design and implementation of a multiple AGV scheduling algorithm for a job-shop.
Int. J. Simul. Model. 2020, 19, 134–145. [CrossRef]
2. Reis, W.; Junior, O. Sensors applied to automated guided vehicle position control. Int. J. Adv. Manuf. Technol. 2021, 113, 21–34.
[CrossRef]
3. Ramalepa, L.P.; Jamisola, R.S. A Review on Cooperative Robotic Arms with Mobile or Drones Bases. Int. J. Autom. Comput. 2021,
18, 536–555. [CrossRef]
4. Suparjon, S. Evaluation of Layout Design, Operation and Maintenance of Multi Automated Systems Guided Vehicles (AGV):
A Review. Int. J. Mech. Eng. Technol. Appl. 2022, 3, 1–7. [CrossRef]
5. Thanh, V.N.; Vinh, D.P.; Nghi, N.T.; Nam, L.H.; Toan, D.L.H. Restaurant serving robot with double line sensors following
approach. In Proceedings of the 2019 IEEE International Conference on Mechatronics and Automation (ICMA), Tianjin, China,
4–7 August 2019.
6. Moshayedi, A.J.; Roy, A.S.; Sambo, S.K.; Zhong, Y.; Liao, L. Review On: The Service Robot Mathematical Model. EAI Endorsed
Trans. AI Robot. 2022, 1, 1–19. [CrossRef]
7. Liu, S.; Tian, G.; Zhang, Y.; Duan, P. Scene recognition mechanism for service robot adapting various families: A cnn-based
approach using multi-type cameras. IEEE Trans. Multimed. 2021, 24, 2392–2406. [CrossRef]
8. Wu, Q.; Liu, Y.; Wu, C. An overview of current situations of robot industry development. In Proceedings of the 4th Annual
International Conference on Wireless Communication and Sensor Network, Wuhan, China, 15–17 December 2017.
9. Tzafestas, S.G. Mobile robot control and navigation: A global overview. J. Intell. Robot. Syst. 2018, 91, 35–58.
10. Atiyah, A.N.; Adzhar, N.; Jaini, N.I. An overview: On path planning optimization criteria and mobile robot navigation. J. Phys.
Conf. Ser. 2021, 1988, 1–10. [CrossRef]
11. Nessrine, K.; Nahla, K.; Safya, B. Reinforcement Learning for Mobile Robot Navigation: An overview. In Proceedings of the 2022
IEEE Information Technologies & Smart Industrial Systems (ITSIS), Paris, France, 15–17 July 2022.
12. Hart, P.E.; Nilsson, N.J.; Raphael, B. A formal basis for the heuristic determination of minimum cost paths. IEEE Trans. Syst. Sci.
Cybern. 1968, 4, 100–107. [CrossRef]
Mathematics 2023, 11, 4424 20 of 20

13. Stentz, A. Optimal and efficient path planning for partially-known environments. In Proceedings of the 1994 IEEE International
Conference on Robotics and Automation, San Diego, CA, USA, 8–13 May 1994.
14. Karaman, S.; Frazzoli, E. Incremental sampling-based algorithms for optimal motion planning. Robot. Sci. Syst. VI 2010, 104, 267–274.
15. Fox, D.; Burgard, W.; Thrun, S. The dynamic window approach to collision avoidance. IEEE Robot. Autom. Mag. 1997, 4, 23–33.
[CrossRef]
16. Borenstein, J.; Koren, Y. Real-time obstacle avoidance for fast mobile robots. IEEE Trans. Syst. Man Cybern. 1989, 19, 1179–1187.
[CrossRef]
17. Borenstein, J.; Koren, Y. The vector field histogram-fast obstacle avoidance for mobile robots. IEEE Trans. Robot. Autom. 1991, 7,
278–288. [CrossRef]
18. de Lima, D.A.; Pereira, G.A.S. Navigation of an autonomous car using vector fields and the dynamic window approach. J. Control
Autom. Electr. Syst. 2013, 24, 106–116. [CrossRef]
19. Ballesteros, J.; Urdiales, C.; Velasco, A.B.M.; Ramos-Jimenez, G. A biomimetical dynamic window approach to navigation for
collaborative control. IEEE Trans. Hum. Mach. Syst. 2017, 47, 1123–1133. [CrossRef]
20. Missura, M.; Bennewitz, M. Predictive collision avoidance for the dynamic window approach. In Proceedings of the 2019
International Conference on Robotics and Automation (ICRA), Montreal, Canada, 20–24 May 2019.
21. Lin, Z.; Taguchi, R. Improved dynamic window approach using the jerk model. In Proceedings of the 22nd International
Conference on Control, Automation and Systems, Busan, Republic of Korea, 27–30 November 2022.
22. Stefek, A.; Van Pham, T.; Krivanek, V.; Pham, K.L. Energy comparison of controllers used for a differential drive wheeled mobile
robot. IEEE Access 2020, 8, 170915–170927. [CrossRef]
23. Meng, Z.; Wang, C.; Han, Z.; Ma, Z. Research on SLAM navigation of wheeled mobile robot based on ROS. In Proceedings of the
5th International Conference on Automation, Control and Robotics Engineering (CACRE), Dalian, China, 18–20 September 2020.
24. Hassan, N.; Saleem, A. Analysis of Trajectory Tracking Control Algorithms for Wheeled Mobile Robots. In Proceedings of the
2021 IEEE Industrial Electronics and Applications Conference (IEACon), Georgetown, Malaysia, 22–23 November 2021.
25. Wen-lan, W.; Bai, X. P and Feedforward Control for Mobile Robot. In Proceedings of the 2nd International Conference on Electrical
Engineering and Computer Technology (ICEECT 2022), Suzhou, China, 23–25 September 2022.
26. Liu, Y.; Bai, K.; Wang, H.; Fan, Q. Autonomous Planning and Robust Control for Wheeled Mobile Robot with Slippage
Disturbances Based on Differential Flat. IET Control. Theory Appl. 2023. [CrossRef]
27. Yang, D.; Bi, S.; Wang, W.; Yuan, C.; Qi, X.; Cai, Y. DRE-SLAM: Dynamic RGB-D encoder SLAM for a differential-drive robot.
Remote Sens. 2019, 11, 380. [CrossRef]
28. Jiang, H.; Sun, Y. Research on global path planning of electric disinfection vehicle based on improved A* algorithm. Energy Rep.
2021, 7, 1270–1279. [CrossRef]
29. Khan, M.A.; Baig, D.-E.; Ali, H.; Ashraf, B.; Khan, S.; Wadood, A.; Kamal, T. Efficient System Identification of a Two-Wheeled
Robot (TWR) Using Feed-Forward Neural Networks. Electronics 2021, 11, 3584. [CrossRef]
30. Mushtaq, Z.; Qureshi, M.; Zohaib, A.; Akmal, M. Estimation of Real-Time Wheeled Mobile Robot (Differential Drive) Motion
& Pose with Obstacle Avoidance. In Proceedings of the 2022 19th International Bhurban Conference on Applied Sciences and
Technology (IBCAST), Islamabad, Pakistan, 16–20 August 2022.
31. Zhao, Y.; Zhu, Y.; Zhang, P.; Gao, Q.; Han, X. A Hybrid A* Path Planning Algorithm Based on Multi-objective Constraints. In
Proceedings of the 2022 Asia Conference on Advanced Robotics, Automation, and Control Engineering (ARACE), Qingdao,
China, 26–28 August 2022.
32. Gong, K.; Xu, Z.; Zhang, X. Bounded-DWA: An Efficient Local Planner for Ackermann-driven Vehicles on Sandy Terrain. In
Proceedings of the 2023 IEEE International Conference on Real-time Computing and Robotics (RCAR), Datong, China, 17–20
July 2023.
33. Quigley, M.; Conley, K.; Gerkey, B.; Faust, J.; Foote, T.; Leibs, J.; Wheeler, R.; Ng, A.Y. Ros: An open-source robot operating system.
In Proceedings of the ICRA Workshop on Open Source Software, Kobe, Japan, 12–17 May 2009.
34. Vaughan, R. Massively multi-robot simulation in stage. Swarm Intell. 2008, 2, 189–208. [CrossRef]
35. An Index of ROS Robots. Available online: https://robots.ros.org/ (accessed on 18 October 2023).
36. Mobile Industrial Robots. Automate Your Internal Transportation. Available online: https://www.mobile-industrial-robots.com/
(accessed on 19 October 2023).
37. Moving Robot. Available online: https://www.hansrobot.net/product-center/yidongjiqiren/ (accessed on 19 October 2023).
38. Automated & Autonomous Mobile Robots|Robotnik®. Available online: https://robotnik.eu/products/mobile-robots/
(accessed on 19 October 2023).
39. AITEN AGV (China) Official Site|—Satisfying Every Real Demand in Factory. Available online: https://www.szaiten.com/en/
ProductIndex/ (accessed on 19 October 2023).
40. AMR Autonomous Mobile Robots|AMS, Inc. Available online: https://www.ams-fa.com/autonomous-mobile-robots/ (accessed
on 19 October 2023).

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.