Turkish Journal of Electrical Engineering and Computer Sciences

Volume 25 Number 2 Article 2

1-1-2017

Obstacle avoidance for mobile robot based on improved dynamic

window approach
XIUYUN LI

FEI LIU

JUAN LIU

SHAN LIANG

Follow this and additional works at: https://journals.tubitak.gov.tr/elektrik

Part of the Computer Engineering Commons, Computer Sciences Commons, and the Electrical and
Computer Engineering Commons

Recommended Citation
LI, XIUYUN; LIU, FEI; LIU, JUAN; and LIANG, SHAN (2017) "Obstacle avoidance for mobile robot based on
improved dynamic window approach," Turkish Journal of Electrical Engineering and Computer Sciences:
Vol. 25: No. 2, Article 2. https://doi.org/10.3906/elk-1504-194
Available at: https://journals.tubitak.gov.tr/elektrik/vol25/iss2/2

This Article is brought to you for free and open access by TÜBİTAK Academic Journals. It has been accepted for
inclusion in Turkish Journal of Electrical Engineering and Computer Sciences by an authorized editor of TÜBİTAK
Academic Journals. For more information, please contact [email protected].
 Turkish Journal of Electrical Engineering & Computer Sciences Turk J Elec Eng & Comp Sci
(2017) 25: 666 – 676
http://journals.tubitak.gov.tr/elektrik/
⃝c TÜBİTAK
Research Article doi:10.3906/elk-1504-194

Obstacle avoidance for mobile robot based on improved dynamic window

approach

Xiuyun LI1,2 , Fei LIU3 , Juan LIU4 , Shan LIANG2,∗

1
Chongqing Vocational Institute of Engineering, Chongqing, P.R. China
2
College of Automation, Chongqing University, Chongqing, P.R. China
3
National Center for Robot Testing and Evaluation (Chongqing), Chongqing, P.R. China
4
China Coal Technology Engineering Group Chongqing Research Institute, Chongqing, P.R. China

Received: 22.04.2015 • Accepted/Published Online: 06.12.2015 • Final Version: 10.04.2017

Abstract: The dynamic window approach has the drawback that it may result in local minima and nonoptimal motion
decision for obstacle avoidance because of not considering the size constraint of a mobile robot. Thus, an improved
dynamic window approach is proposed, which takes into account the relation between the size of the mobile robot and
the free space between obstacles. A laser range finder is employed to improve the ability of sensing and prediction of the
environment, in order to avoid being trapped in a U-shaped obstacle, such as a box canyon. By applying the proposed
method, the local minima problem can be solved and the optimal path can be obtained. The effectiveness and superiority
is proven by theoretic analysis and simulations.

Key words: Improved dynamic window approach, obstacle avoidance, mobile robot, laser range finder

1. Introduction
For autonomous mobile robots, obstacle avoidance is the basic requirement for moving safely and executing
tasks [1]. The two most used methods for avoiding obstacles are the artificial potential field (APF) and grid-
based vector field histogram (VFH) [2–5]. The APF will control the motion by use of the sum of attractive and
repulsive forces from goal and obstacles, respectively. However, it compresses all the environment information
into a virtual force, which causes the loss of valuable information about obstacle distribution and results in
the local minima problem [6]. Furthermore, it may lead to an unstable state in narrow spaces [2]. The VFH
has better performance in narrow openings, but it does not consider the robot’s size and kinematics [4]. Some
other methods include the curvature-velocity method and basic dynamic window method [7-10], which may also
result in local minima problems due to the lack of information about the connectivity of free space.
The dynamic window method is a widely used method for obstacle avoidance, which has good performance
when the mobile robot moves at high velocity [8–10]. However, the local minima problem exists and may
prevent the robot from real-time stable and smooth avoidance of the obstacle. Traditional strategies such as
wall-following and modified APF methods may not solve this problem fast [1,11,12]. In addition, the size of
the robot is not considered in these methods, which may lead to unfeasible paths and brings negative effects
for global optimal path planning. In [14], Seder and Petrovic proposed a solution that integrates a focused
D* search algorithm and dynamic window local obstacle avoidance algorithm with certain adaptations for the
∗ Correspondence: [email protected]

666
 LI et al./Turk J Elec Eng & Comp Sci

avoidance of moving obstacles. Brock and Khatib proposed a global DWA that combines motion planning
and real-time obstacle avoidance to achieve robust high-velocity, goal-directed, reactive motion for a robot in
unknown and dynamic environments [15]. In [16], Arras et al. introduced an approach to obstacle avoidance
and local path planning for polygonal robots, where a reduced dynamic window dealing with robot shape and
dynamics is utilized.
Moreover, convergence to the goal was theoretically analyzed in [17–20]. Ögren and Leonard integrated
the convergent Koditschek scheme with the fast reactive dynamic window approach (DWA) to make the robot
attain the goal configuration [17,18]. Vista et al. proposed an objective function for the DWA that holds
convergence to the goal [19]. Berti et al. proposed improvements over the dynamic window approach (I-DWA),
utilized for computing in real-time autonomous robot navigation [20]. In [20], an objective function involving
Lyapunov stability criteria was presented to guarantee the global and asymptotic convergence to the goal while
avoiding collisions. In [21], a modified DWA was proposed for obstacle avoidance where stereovision is used. By
defining a strict search space, the error of the stereoprocessing is controlled and steering control is smoothed,
thus producing the optimal velocities for the robot. However, the local minima problem may still occur and the
stereo increases the time and space complexity.
In this paper, an improved DWA is proposed to solve the obstacle avoidance problem. We first detect and
analyze the connectivity of free space in front of the robot by a laser ranger finder (LRF), which can prevent
the situation of being trapped in a U-shape obstacle. The basic DWA is then modified by considering the size
of the robot in order to improve the safety and save time.
The rest of this paper is organized as follows: Section 2 describes the models of the environment and
the robot. Section 3 describes the proposed improved dynamic window approach. In Section 4, simulations are
conducted and the result is presented and discussed. Finally, the conclusion and future work are addressed in
Section 5.

2. Problem statement
One of the classic obstacle avoidance methods is the dynamic window approach, which was first proposed by
Fox et al. in 1997 [8]. The method is derived directly from the motion dynamics of the robot, in which the
limitations of velocity and acceleration are considered. First, a velocity vector (ν , w) is defined where ν is the
heading velocity and w is the rotational velocity. Then, to avoid colliding with obstacles, an admissible speed
set is defined as
√ √
Va = {(v, w)|v ≤ 2dist(v, w)v̇b ∧ w ≤ 2dist(v, w)ẇb }, (1)

where dist(ν , w) is the closest distance from the robot to the obstacle, and v̇b and ẇb are corresponding
accelerations for breakage. Eq. (1) can guarantee that the robot stops without hitting an obstacle.
When considering the constrained accelerations of the motors, the whole search can be reduced to the
dynamic window that involves only the velocities that can be reached within the next time interval. The
dynamic window Vd is defined as

Vd = {(v, w)|v ∈ [va − v̇t, va + v̇t] ∧ w ∈ [wa − ẇt, wa + ẇt]}, (2)

where t is the time interval during which v̇ and ẇ are applied and (va , wa )is the actual velocity. Eq. (2)
shows that within the next time interval all the area the robot can reach is inside the dynamic window and
thus obstacle avoidance outside the window can be ignored. As shown in Figure 1, the restricted search space

667
 LI et al./Turk J Elec Eng & Comp Sci

is Vr = Vs ∩ Va ∩ Vd , where Vs contains all the available velocities. Finally, we choose a suitable value from Vr
as the velocity in next time interval. To optimize the movement, we build an object function

90 cm/s

-90 deg/s 90 deg/s

Figure 1. The dynamic window.

G(v, w) = α · heading(v, w) + β · dist(v, w) + γ · vel(v, w), (3)

which will be maximized. In Eq. (3), heading(ν , w) is used to measure the progress towards the goal, which
is maximal when the robot moves directly to the goal. dist(ν , w) is the closest distance to the obstacle and
vel ( ν , w) is the forward velocity. α , β , and γ are the weights.
This dynamic window approach has been proven to be well suited for robots operating at high speed and
thus is widely applied. However, the local minima problem will occur since this method only takes into account

Figure 2. Application of the dynamic window approach.

668
 LI et al./Turk J Elec Eng & Comp Sci

the admissible velocities within the dynamic window, while the connectivity of the free space is not considered.
Moreover, due to not considering the size of the robot, the path planned may not be optimal. In the case shown
in Figure 2, when the DWA is utilized, the robot will abandon the preplanned path (the dashed arrow) because
of the casual occurrence of obstacle A. Then the robot moves to the goal by following the newly planned path
(the red line), which is much longer than the preplanned path. It is feasible for the robot to pass the opening
between obstacles A and B . To solve this problem, we will take into account the size of the robot and the
real-time sensing information.

3. Improved dynamic window approach

3.1. LRF-based traverse feasibility analysis
The LRF is a time of flight-based sensor that provides precise distance of obstacles within a range. Compared
with image-based sensors, the measurement is not influenced by light, and it has the advantage of high accuracy
[13]. A typical LRF, SICK LMS200, is used to check the environment information. With each turn the LMS200
can get 181 values from the distance to obstacles with an angular range from 0 ◦ to 180 ◦ and the resolution is
1 ◦ . The maximum measurement distance is 8 m and the distance resolution is 15 mm. The data obtained can
be expressed using a polar coordinate method, e.g., the i th distance value si is

si = (ρi , θi )T , i = 1, ..., N, (4)

where ρi is the distance between the sensor and the obstacle, and θi is the relative angle. The model of the
LRF is shown in Figure 3.

Figure 3. The model of a laser range finder.

The environment is described in a 2-dimension Cartesian coordinate system. Therefore, we use Eqs. (5)
and (6) to map si into the robot’s coordinate system and further convert it into the global coordinate system
(GCS):
{
x′ = d + ρ sin θ
, (5)
y ′ = −ρ cos θ

{ √
x = xR + √ x′2 + y ′2 cos(α + α′ )
, (6)
y = yR + x′2 + y ′2 sin(α + α′ )

where (x′ , y ′ ) is the location of the obstacle in the robot’s coordinate system (RCS), and d is a constant that
is the difference between the original points of the RCS and the sensor’s coordinate system (SCS), which is
obtained by the robot’s physical structure. (xR , yR ) and (x, y) represent the robot’s coordinates in RCS and
′
GCS. α is the angle of the robot’s heading direction, and α′ = arctan( xy′ ). The coordinates GCS ( X−O−Y ),

669
 LI et al./Turk J Elec Eng & Comp Sci

RCS(XR −OR −YR ) , and SCS(XS −OS −YS ) are shown in Figure 4. Finally, the SCS is converted into a grid-
based map by

Figure 4. Convention of three coordinates.

{
xg = int( wx
) · w + int( w2 )
, (7)
yg = int( wy ) · w + int( w2 )

where w is the width of the grid and (xg , yg ) is the grid’s coordinate in the GCS.

3.2. The improved dynamic window approach

The key to the improved dynamic window approach is the sensing of obstacles. We use a LRF, SICK LMS200,
to measure the distance to obstacles. By using the scanned data, the analysis of traverse feasibility is conducted
as shown in Figures 5 and 6. In Figure 5, the LMS200 is mounted in the front of the robot. The scanning area
is divided into 6 regions, which are A, A′ , B , B ′ , C , and C ′ respectively. For simplicity, we use AA′ , BB ′ ,
and CC ′ to describe the regions A and A′ , B and B ′ , and C and C ′ , respectively. Thus, the angular ranges
of AA′ , BB ′ , and CC ′ are obtained as

C C'
B B'

A A'

Figure 5. Partition of scanning regions. Figure 6. The robot and the laser beams.

670
 LI et al./Turk J Elec Eng & Comp Sci


 Ra(AA′ ) = [0, 30◦ ) ∪ (150◦ , 180◦ ]
Ra(BB ′ ) = [30◦ , 70◦ ) ∪ (110◦ , 150◦ ] , (8)

Ra(CC ′ ) = [70◦ , 90◦ ) ∪ (90◦ , 110◦ ] = [70◦ , 110◦ ]

where Ra(·) represents the angular range of the corresponding region.

Figure 6 shows the model of the robot and laser’s beam. The robot is modeled as rectangular, which
is a typical model for robots. It is approximated to be a cylinder of size D × L and a safety clearance radius
δ for movement, where δ depends on the velocity vector (v, w) . In addition, three beams, li , lj , and lk , are
shown in Figure 6. We define d as the distance between the two points produced by beams li and lj shooting
on obstacles, and then we obtain
√
d= ρ2i + ρ2j − 2 cos(θj − θi )ρi ρj . (9)

However, if one beam (e.g., lk ) cannot shoot anything, it means that the obstacle is out of the laser’s range,
and then we denote ρk = ∞ .
As shown in Figure 7, with the partition of scanning regions AA′ , BB ′ , and CC ′ , the process of traverse
feasibility analysis is executed as follows:

dCC '
C C'
dBB '
B B'
A A'

Figure 7. Traverse feasibility analyses.

(1) Region AA′ is a protection area, which is used to prevent the robot from bumping into the obstacles on
its two sides. Denote ρ1 and ρ181 as the two measured distances of two beams of 0 ◦ and 180 ◦ . If ρ1 = ∞
and ρ181 = ∞ , the robot can move safely. If ρ1 = ∞ and ρ181 ̸= ∞, the robot should rotate left, and if
ρ1 ̸= ∞ and ρ181 = ∞ , the robot should rotate right. When ρ1 ̸= ∞ and ρ181 ̸= ∞ , the robot needs to
stop or move backwards if
(ρ1 + ρ181 ) ≤ c(D + δ), (10)

where c is a parameter that is determined by the robot’s physical structure. When (ρ1 + ρ181 ) > c(D + δ),
if ρ1 ≤ ρ181 , the robot will rotate left; otherwise, it will rotate right.

(2) Region BB ′ is a proximal prediction area. First we find out the minimal values ρmin
B and ρmin
B ′ in divisions
B and B ′ , respectively. Then with Eq. (9) we can obtain dBB ′ . Then the robot can move forward if
√
dBB ′ > c2 (D + δ)2 . (11)

Otherwise, the robot cannot pass this narrow area and path replanning is needed.

671
 LI et al./Turk J Elec Eng & Comp Sci

(3) Region CC ′ is a distal prediction area, which implies the connectivity and describes the width dCC ′ that
the robot can pass through the narrow area between obstacles of the distal region in the robot’s forward
motion. In order to obtain dCC ′ , it is necessary to search the first noninfinity values ρ1C and ρ1C ′ in
region C and C ′ , respectively. If ρ1C ≤ ρ1C ′ , we find out the minimal value ρmin ′
C ′ in region C and thus
we obtain dCC ′ with ρ1C and ρmin 1 min 1 1
C ′ and corresponding angles θC and θC ′ . However, if ρC > ρC ′ , dCC ′
is computed with ρmin
C and ρ1C ′ and the corresponding angles θC1 min
′ and θC . Similarly, if
√
dCC ′ > c2 (D + δ)2 , (12)

the robot can move forward; otherwise, it needs to replan the path. If a noninfinity value is not found in
region CC ′ , the free space in this region is wide enough to be passed through. However, if this happens
in regions BB ′ and CC ′ , there may exist a local minimum area, and thus the robot should replan a local
path to avoid the local minima.

When a dynamic obstacle is found that does not exist in the environment map, the obstacle avoidance
mechanism works, which can be described as follows:
Step 1: The robot avoids collision with unforeseen obstacles by analyzing region AA′ .
Step 2: Predict the feasibility of passing through the planned path by analyzing region BB ′ . If neither
ρmin
B nor ρmin min
B ′ exists, go to step 3. If both ρB and ρmin
B ′ exist and Eq. (11) becomes true, go to step 3;
otherwise, go to step 4.
Step 3: If neither ρmin
C nor ρmin min
C ′ exists, go to step 4. If both ρC and ρmin
C ′ exist and Eq. (12) is true,
then go to step 5; otherwise, go to step 4.
Step 4: Replan the local path to avoid collision with obstacles since the opening gap between the obstacles
is too narrow to pass through.
Step 5: Return to step 2.
In order to improve the ability to pass the openings between obstacles by traverse feasibility analysis
above, a new feasibility evaluation function, f (v, w, s) , is defined as

f (v, w, s) = α · heading(v, w) + β · dist(v, w) + γ · vel(v, w) + εwidth(v, w, s), (13)

where heading(v, w) is the heading direction, dist(v, w) is the closest distance to obstacles, and vel(v, w) is
the velocity vector. s is the sensing data, ε is the weight, and width(v, w, s) is
{
l, if dAA′ > c(D + δ)or(ρ = ∞) ∈ AA′
width(v, w, s) = , (14)
0, if dAA′ ≤ c(D + δ)

where l is

 c2 (D+δ)2

 , if dBB ′ > c(D + δ)and dCC ′ > c(D + δ);
 dBB ′ dCC ′

l= c(D+δ)
dBB ′ > c(D + δ)and (ρ = ∞) ∈ CC ′ ; . (15)

 dBB ′ , if


0, else.

The parameter width(v, w, s)determines the possibility to pass through the opening area between obstacles.
The situation width(v, w, s) = 0 implies that the robot cannot pass through the opening area between obstacles,
and thus it needs to replan a new path. When width(v, w, s) ̸= 0 , the robot is able to pass through and the
motion is controlled by parameters α , β , γ , and ε.

672
 LI et al./Turk J Elec Eng & Comp Sci

4. Simulation and result

Simulations are conducted in MATLAB and parameters are set as α = 0.5, β = 0.2, γ = 0.1, ε = 0.2, ω = 10
cm, D = 6 cm, and L = 10 cm. The robot moves at the speed of 15 cm/s when the free space in front of the
robot is wide enough to pass. When closing to any obstacle, the robot moves at the speed of 10 cm/s, and when
the robot needs to rotate, the forward velocity and rotation velocity are set as 5 cm/s and π /4 rad s −1 . In
simulations, the DWA and the proposed improved dynamic window approach (IDWA) are applied respectively
for each case. Simulation results are shown in Figures 8–10.

Figure 8. Result of simulation I.

Figure 9. Result of simulation II.

Figures 8a and 8b show the result that applies DWA and IDWA, respectively. The major difference
occurs when facing obstacles C and D. By using DWA, the robot moves around obstacle C according to the
optimal decision. However, using IDWA, Figure 8b shows that the robot can pass the narrow passage between
obstacles C and D after taking into account the size of robot. Thus, the path becomes much longer according
to DWA.
The result of simulation II is shown in Figures 9a and 9b, which use DWA and IDWA, respectively.
Compared with simulation I, the space between C and D becomes larger and there is a U-shaped trap between

673
 LI et al./Turk J Elec Eng & Comp Sci

obstacles E and F. We can see from Figure 9 that the robot can pass through the gap between C and D in both
cases. However, when facing the U-shaped local minima, DWA cannot make the robot sense this trap before
entering it, which is shown in Figure 9a. Thus, the robot must turn around and replan the path to the goal,
which costs more time and makes the robot traverse longer. In Figure 9b, by employing IDWA, the robot will
find the U-shaped trap before moving into that area and replan the path to the goal in a timely manner, which
saves a lot of time.

Start
C E
point A
B

D F Goal
(a) DWA
Start
C E
point A
B

D F Goal

(b) IDWA
Figure 10. Result of simulation III.

Figure 10 shows the result of simulation III, in which the start point, the goal, and the distribution of
obstacles are different from simulations I and II. The result indicates that when facing obstacles A and B, by
using IDWA, the robot can avoid moving into the local minima and thus uses less time and path length than
when using DWA. Furthermore, Figure 10a shows that the robot finds it impossible to pass through the gap
between obstacle F and the border. However, Figure 10b shows that by considering the size and the detection
of the gap between obstacle F and the border, the robot can avoid wasting time to get to the border and find
the feasible path between E and F.
The three simulations above show that IDWA can solve the local minima problem while avoiding obstacles
and thus produces an optimal path to the goal. Moreover, we list the time cost and the total path length in
each case in the Table, in which comparison implies that by applying IDWA, the robot spends less time, and
the path length is much shorter. In all, IDWA proves to be better than DWA.

Table. Comparison of simulation results.

DWA IDWA
Simulations
Path length (cm) Time (s) Path length (cm) Time (s)
I 512 54 430 42
II 486 46 430 37
III 544 68 506 52

5. Conclusion and future work

This paper proposes an improved DWA to solve the problem of obstacle avoidance. The major modification
lies in the consideration of the size of robot and the development of a new feasibility evaluation function of the
width of the opening space between obstacles. The simulations show that it can solve the local minima problem

674
 LI et al./Turk J Elec Eng & Comp Sci

and improve the ability to deal with narrow passages, and thus proves to be more capable of global optimal
path planning while avoiding obstacles. In future work, a more complicated environment will be employed to
verify and improve the proposed strategy. Finally, we will apply it to a real robot.

Acknowledgment
This work was financially supported by the Natural Science Foundation of the Chongqing Municipal Education
Commission, China (Grant No. KJ1403209).

References

[1] Liu F, Liang S, Xian X, Bi H. Oscillation elimination for mobile robot based on behavior-memorizing. ICCI Express
Letters 2011; 5: 3109-3115.
[2] Koren Y, Borenstein J. Potential field methods and their inherent limitations for mobile robot navigation. In:
Proceedings of the IEEE Conference on Robotics and Automation; 7-12 April 1991; Sacramento, CA, USA. New
York, NY, USA: IEEE. pp. 1398-1404.
[3] Borenstein J, Koren Y. The vector field histogram-fast obstacle avoidance for mobile robots. IEEE J Robot Autom
1991; 7: 278-288.
[4] Morales Y, Carballo A, Takeuchi E, Aburadani A, Tsubouchi T. Autonomous robot navigation in outdoor cluttered
pedestrian walkways. J Field Robots 2009; 26: 609-635.
[5] Ulrich I, Borenstein J. VFH: Local obstacle avoidance with look ahead verification. In: IEEE International Con-
ference on Robotics and Automation; 24–28 April 2000; San Francisco, CA, USA. New York, NY, USA: IEEE. pp.
2505-2511.
[6] Cai Z, Zheng M, Zou X. Real-time obstacle avoidance for mobile robots strategy based on laser radar. J Cent South
Univ 2006; 37: 324-329.
[7] Simmons R. The curvature-velocity method for local obstacle avoidance. In: Proceedings of the 1996 IEEE Inter-
national Conference on Robotics and Automation; 1996; Piscataway, NJ, USA. New York, NY, USA: IEEE. pp.
3375-3382.
[8] Fox D, Burgard W, Thrun S. The dynamic window approach to collision avoidance. IEEE Robot Autom Mag 1997;
4: 23-33.
[9] Kiss D, Tevesz G. Advanced dynamic window based navigation approach using model predictive control. In: 2012
International Conference on Methods and Models in Automation and Robotics; 27–30 August 2012; Miȩdzyzdroje,
Poland. New York, NY, USA: IEEE. pp. 148-453.
[10] Saranrittichai P, Niparnan N, Sudsang A. Robust local obstacle avoidance for mobile robot based on dynamic window
approach. In: International Conference on Electrical Engineering and Electronics, Computer, Telecommunications
and Information Technology; 15–17 May 2013; Krabi, Thailand. New York, NY, USA: IEEE. pp. 1-4.
[11] Yan Y, Zhang Y. Collision avoidance planning in multi-robot based on improved artificial potential field and rules.
In: Proceedings of the 2008 IEEE International Conference on Robotics and Biomimetics; 21–26 February 2009;
Bangkok Thailand. New York, NY, USA: IEEE. pp. 1026-1031.
[12] Nooraliei A, Iraji R. Robot path planning using wavefront approach with wall-following. In: International Conference
on Computer Science and Information Technology; 8–11 August 2009; Beijing, China. New York, NY, USA: IEEE.
pp. 417-420.
[13] Ren L. Obstacle perception and obstacle-avoiding strategy research of mobile robot based on laser range finder.
Heilongjiang, China: Harbin Institute of Technology, 2007.
[14] Seder M, Petrovic I. Dynamic window based approach to mobile robot motion control in the presence of moving
obstacles. In: Proceedings of IEEE International Conference on Robotics and Automation; 10–14 April 2007; Rome,
Italy. New York, NY, USA: IEEE. pp. 1986-1991.

675
 LI et al./Turk J Elec Eng & Comp Sci

[15] Brock O, Khatib O. High-speed navigation using the global dynamic window approach . In: IEEE International
Conference on Robotics and Automation; 10–15 May 1999; Detroit, MI, USA. New York, NY, USA: IEEE. pp.
341-346.
[16] Arras K O, Persson J, Tomatis N, Siegwart R. Real-time obstacle avoidance for polygonal robots with a reduced
dynamic window. In: Proceedings of the 2002 IEEE International Conference on Robotics & Automation; 11–15
May 2002; Washington, DC, USA. New York, NY, USA: IEEE. pp. 3050-3055.
[17] Ögren P, Leonard NE. A convergent dynamic window approach to obstacle avoidance. IEEE T Robotic Autom
2005; 21: 188-195.
[18] Ögren P, Leonard NE. A tractable convergent dynamic window approach to obstacle avoidance. In: IEEE/RS
International Conference on Intelligent Robots and Systems; 29 October–3 November 2002; Lausanne, Switzerland.
New York, NY, USA: IEEE. pp. 595-600.
[19] Vista FP 4th, Singh AM, Lee DJ, Chong KT. Design convergent dynamic window approach for quadrotor navigation.
Int J Precision Eng 2014; 15: 2177-2184.
[20] Berti H, Sappa AD, Amennoni OE. Improved dynamic window approach by using Lyapunov stability criteria. Lat
Am Appl Res 2008; 38: 289-298.
[21] Zhang HQ, Dou LH, Fang H, Chen J. Autonomous indoor exploration of mobile robots based on door-guidance and
improved dynamic window approach. In: Proceedings of the 2009 IEEE International Conference on Robotics and
Biomimetics; 19–23 December, 2009; Guilin, China. New York, NY, USA: IEEE. pp. 408-413.

676