Proceedings of the 2020 International Conference on Advanced Mechatronic Systems, Hanoi, Vietnam, December 10 - 13, 2020

Comparison of ROS Local Planners with

Differential Drive Heavy Robotic System
Isira Naotunna Theeraphong Wongratanaphisan
dept. of Mechanical Engineering dept. of Mechanical Engineering
Chiang Mai University (Center of Mechatronics System and Innovation)
Chiang Mai, Thailand Chiang Mai University
[email protected] Chiang Mai, Thailand
[email protected]

Abstract—This work aims to analyze the performance of the differential drive heavy robotic system. In this study, the
ROS local planners with a differential drive heavy robotic navigation stack is developed based on the data provided by
system. Intel Realsense D435i depth camera and T265 tracking Intel Realsense D435i depth camera and T265 tracking
camera are used as the primary sensor source for navigation camera setup.
with real-time appearance-based mapping and localization
technique. This work has studied the performance of DWA, The outline of this paper is divided into five main sections.
TEB, and EBand local planners under two experiments. Sections II consist of the studies that have been previously
Obtained results are used to analyze the local planners based on conducted to analyze the performance of the ROS navigation
how well it follows the global planners, their obstacle avoidance stack and a brief explanation about the local planner
capability, time consumption, goal-reaching accuracy, and the algorithms. Then, the experimental setup is explained in
quality of motion generation. Obtained results indicate that section III before discussing the experiment and obtained
DWA and TEB local planners are more suitable for the large results in section IV. The conclusion and future works in
differential drive robot setup. section V.
Keywords—Dynamic-Window, Elastic Band, Timed Elastic II. BACKGROUND STUDY
Band, Robot Operating System
A. Related Work
I. INTRODUCTION Even though ROS is a popular platform to develop
Within the past few decades, the interest to develop robotic autonomous navigation tasks in mobile robotics, only a
systems that are dedicated to autonomous navigation has been handful number of studies have been done to investigate the
rapidly growing. The fast-moving advancements in path planning techniques of ROS. These studies address the
communication, sensors, and computing technologies have performances of various areas of path planning techniques
provided significant progress in the field of autonomous such as local planners, global planners, goal-reaching
navigation in robotics. The concept of autonomous navigation accuracy, control velocity generation, and repeatability.
covers many areas, such as map development, localization,
A systematic analysis of global and local planners for
obstacle avoidance, and path planning. The navigation
different scenarios in simulation and real-world applications
algorithms use sensory information to determine a suitable
is presented in [1]. The study is done using a Turtlebot
trajectory and move accurately within its environment without
platform, and the results show that the TEB local planner had
collision. However, the development of a custom navigation
the best performance. A similar ROS local planner study is
algorithm for a robot from scratch is quite a challenge.
presented in [2]. This experiment is done in a simulated
Therefore, the Robot Operating System (ROS) has become a
environment using a small scale differential drive robotic
popular open-source software platform among researchers
system. The impact of the local planner on the accuracy of
around the world for robot development. Navigation of mobile
reaching the goal is presented in [3]. In this work, TEB, DWA,
robots is well documented in ROS with the developed
and EBand local planners are analyzed using a TurtleBot
software libraries to address various tasks in robot navigation.
platform in both simulated and real-world environments, and
ROS official packages are adequate in everyday robotic tasks,
the test results indicate that the DWA performed the best
and ROS provides an Application Programming Interface
repeatability, EBand had the highest tolerance, and TEB local
(API) to build custom packages or to communicate with
planner had the best speed of action. As per the above
external systems and equipment.
literature, a majority of the studies on path planners are done
ROS navigation stack is a popular ROS library which is using small scale robots or simulated robots. A slightly
used to perform navigation tasks in mobile robot different study when compared to those above, can be found
development. In the ROS navigation stack, the path planning in [4]. This study analyzes the performance of the TEB local
strategy is divided into two sections; global planner and local planner using an actual Ackremen type vehicle.
planner. Global planning provides the optimal path based on
According to the existing studies on the path planning
the provided map, and local planner recalculates the path to
strategies of mobile robotic systems, it is not easy to find a
avoid dynamic obstacles. Generally, the ROS navigation stack
performance analysis of a ROS-based path planning
consists of inbuilt local and global planner libraries which can
techniques with a 3D vision system. Also, most of the existing
be customized according to the robot configuration. Dynamic-
studies are done using the Turtlebot research platform or in a
Window Approach (DWA) local planner, Timed Elastic Band
simulated environment. Therefore, this study will be useful to
(TEB) local planner, and classical Elastic Band (EBand) local
identify the behaviour of ROS-based local planners with a 3D
planner are the most commonly used local planner libraries
vision-based localization system and a heavy robotic system.
with ROS navigation stack. The purpose of this study is to
analyze the performance of these three local planners using

978-1-7281-6530-1/20/$31.00/ ©2020 IEEE 

Authorized licensed use limited to: DRDO-R&DE. Downloaded on December 26,2022 at 06:47:08 UTC from IEEE Xplore. Restrictions apply.
B. Local Planner Algorithms
The algorithms of the local planner techniques which are
discussed in this study are briefly explained below.
a) DWA Local Planner: A dynamic window approach
of 1997 is presented by [5]. According to their work, DWA
performs a sample-based optimization to produce a pair of
linear and angular velocity values which represents a robot's
optimal circular trajectory. The trajectories are simulated
according to the specified horizon lengths based on the
motion model of the robot. Because of the constant control
action along the defined horizon, the DWA planner is unable
to produce reverse motion. The principle of operation of
DWA local planner consists of five main steps. The first step
is the velocity sampling in the velocity space. Secondly, for a
short time, a forward simulation is performed to every
velocity pair to identify the possible robot's motion. As a third Fig. 1. Robot setup
step, the trajectory scores are calculated from each forward
simulation using the cost function. Finally, the best trajectory The Intel Realsense D435i depth camera and the T265
is selected, and the associated velocity value is sent to the tracking camera are the primary sensor sources for
base controller. This algorithm is implemented in the autonomous navigation. Both cameras are connected to the
dwa_local_planner package in ROS [6]. Nvidia TX1 via a USB hub, and the communication is
b) EBand Local Planner: EBand local planner is established via Realsense 2.3.5 SDK. Since all tasks related
to autonomous navigation is done in the ROS environment,
developed based on the elastic band theory [7]. This
the Realsense ROS wrapper library is used to collaborate
algorithm imitates the elastic behaviour of a rubber band, and
Realsense cameras with ROS.
the algorithm generates a collision-free path based on two
essential components; contraction force and repulsion force. B. Experiment Description
The contraction force is responsible for generating the The experiment is performed in a selected area on the first
shortest path between the start point and goal point while the floor of the main engineering building at Chiang Mai
repulsion force repels the path from the obstacles. This University. A 2D map of the selected area is shown in Fig. 2.
algorithm is developed in ROS as the eband_local_planner The selected floor area is approximately 145 m2, and it consists
package [8]. of a room, a doorway, a corridor, and a vacant space. The
c) TEB Local Planner: TEB local planner is performance of the local planners is investigated by
implemented using the Timed Elastic Band Algorithm, which navigating the robot between four defined waypoints which
is a modified algorithm of the elastic band approach [9]. This are namely O1, O2, O3, and O4. The 2D coordinates (in
method also has similar characteristics but, instead of using meters) of each point with respect to the map frame are (0.0,
0.0), (5.0, 2.0), (13.2, 2.5), and (22.5, 7.0) respectively. The
contraction and repulsion forces, this algorithm optimizes
robot starts to move from point O1, and then it reaches points
every moment of trajectory deformation and minimizes the
O2, O3, and O4, respectively. Finally, the robot returns to its
target cost function. This algorithm requires information initial position O1 from the point O4. During the testing of
about kinematics, dynamics, geometric shape, acceleration, each local planner, the robot performs 20 trials on this route.
and velocity limits. TEB local planner is available in ROS as These 20 trials are also divided into two sections. The first ten
the Teb_local_planner package [10]. trials run without any additional obstacles, and the next ten
trials run with obstacles. During each trial, the data is recorded
III. METHOD
into a rosbag file.
A. Robot Description
The experiment is performed using a differential drive
mobile robotic system which is developed to carry a Baxter
manipulator, as shown in Fig. 1. Since this experiment focuses
only on the local planner study, the manipulator system is not
used. The approximate weight of the entire system is about
100 kg, and it is 0.75 m long, 0.7 m wide and 1.5 m tall. Nvidia
TX1 development board is used as the main computer in the
system, and an Arduino MEGA 2560 which communicates
with NVidia TX1 via ROS Serial package is used to control
the mobile robot that is driven by two brushed DC motors.
AMS's AS5147P magnetic encoders are used to get the
velocity feedback from the motors. Two serially connected
12V lead-acid batteries power the system. Also, the setup
consists of a MeanWell TS100 inverter to generate the 230V Fig. 2. 2D map of the experimental area
AC voltage required to operate the Baxter manipulator
system.


Authorized licensed use limited to: DRDO-R&DE. Downloaded on December 26,2022 at 06:47:08 UTC from IEEE Xplore. Restrictions apply.
C. ROS Navigation Stack Setup of the Robot c) Navigation Stack Configuration: The navigation
In order to perform the navigation tasks, the ROS stack has to be appropriately configured to perform
navigation stack needs to be appropriately set up according to autonomous navigation. Localization and path planning are
the robot's requirement. In the developed system, the primary the two main sections of autonomous navigation. According
sensor sources of navigation tasks are the Realsense D435i to the navigation stack setup shown in Fig. 5, the robot
depth camera and the T265 tracking camera. This section localization is also performed using the RTAB-Map. RTAB-
briefly explains the robot transformation (TF) tree setup, Map provides robot localization based on the database
simultaneous mapping and localization (SLAM) method, and generated during the mapping process, and the T265 tracking
the move_base node setup. camera provides the odometry data. The D435i camera is the
a) Robot TF Tree: Generally, a REP 105 coordinate primary observation source of the ROS navigation stack. The
frame setup is used when setting up the robot's TF tree [11]. point-cloud data, and the fake laser scan data that is generated
In the traditional REP 105 setup, the odometry frame is from the depth images are used to update local and global
always directly linked to the base frame of the robot. In that costmaps. Based on the costmap data, local and global paths
scenario, the odometry data is provided from the wheel are calculated for collision-free motion. In path planning, the
encoder or an Internal Measurement Unit (IMU). However, Dijkestra algorithm is used as a global planner that is
in this system, the robot's base frame is connected to the available in the navfn ROS package [13].
odometry frame of the T265 camera, as shown in Fig. 3. It d) Navigation Stack Parameters: All three local
helps to provide accurate odometry readings from T265 planners are tested with the same global and local parameters.
tracking camera rather than getting the odometry readings Since the robot setup is a heavy system, the acceleration and
from the wheel encoders. velocity limits are set to low values to ensure better safety.
b) Map Generation: Map generation is done using the The essential local planner parameters used in this
Realtime Appearance Based Mapping (RTAB-Map) package experiment are shown in Table I. With this robot setup,
[12]. RTAB-Map is an RGB-D, Stereo, and Lidar Graph- slightly higher goal tolerances have to be set to avoid
Based SLAM approach based on an incremental appearance- unnecessary oscillations from the robot. In DWA local
based loop closure detector. RGB-D data from the D345i planner, the sim time parameter provides the time interval
camera is used to construct the map using RTAB-Map. The for trajectory optimization. The required velocity samples for
constructed 2D occupancy grid map of the experimental area trajectory generation can be set using vx_samples and v_theta
is shown in Fig. 4. parameters. The path_distance_bias and goal_distance_bias
parameters define how much local planner stay close to the
global path and the local goal.
The proportional and damping gain of the Eband local
planner parameters need to be tuned to obtain better control of
the robot. Otherwise, the robot oscillates along the path
instead of reaching the goal location. When setting the TEB
local planner parameters, the most important thing is to select
the proper footprint model. The footprint model of the robot
is defined as a circular footprint, even though the actual
footprint of the robot is rectangular. Since the wheels of the
robot are located along the centre axis of the robot's base
frame, selecting a circular footprint model helps to maintain
better control of the robot.

Fig. 3. Robot TF Tree

Fig. 5. General setup of the robot’s navigation stack

Fig. 4. Occupancy grid map generated by RTAB-Map


Authorized licensed use limited to: DRDO-R&DE. Downloaded on December 26,2022 at 06:47:08 UTC from IEEE Xplore. Restrictions apply.
 TABLE I. LOCAL PLANNER TUNING PARAMETERS

Parameter Category Parameters

Maximum linear velocity:0.4ms-1
Velocity and acceleration Maximum angular velocity:0.4 rads-1
parameters Linear acceleration limit:1.5 ms-2
Angular Velocity limit:1.5 rads-2
xy goal tolerance: 0.22 m
Goal tolerances
yaw goal tolerance: 0.1 rad
Sim time: 4
vx_samples:20
DWA planner parameters v_theta samples:40
path distance bias: 32
goal distance bias: 24
Proportional gain: 4.0
EBand planner parameters Damping gain: 3.5
Control rate: 10 Hz
Footprint radius: 0.35m
TEB planner parameters Minimum obstacle distance: 0.6m
Maximum backward velocity:0.1ms-1

IV. EXPERIMENT AND RESULTS

As explained earlier, this experiment is categorized into
two sections. The first experiment is done without any random
obstacles in the experimental area, and the second experiment
Fig. 6. Generated local planners without obstacles
is done by adding obstacles at different locations on the map.
All the local planners are tested with ten trials in each
experiment, and the results are analyzed offline using the the walls. Therefore, a larger inflation radius is more suitable
recorded data. for EBand approach to increase the safety of the navigation.
Table II shows the overall performance of each local planner
With this data, the local planners are analyzed based on the during obstacle-less navigation.
following methods. First, the performance of the local
planners is analyzed based on the Root Mean Square Based on the average RMSD values obtained from the
Deviation (RMSD) between global and local planners. It can experimental data, the global planner following accuracy of
be used to identify how accurately the local planner follows DWA and TEB local planners are high compare to the EBand
the global planner. If the coordinates of the global planner and local planner. By considering the average time to complete the
the local planner are given by (Xg, Yg) and (Xl, Yl) respectively, path, it can be observed that DWA local planner takes less
the RMSD value can be calculated for n number of trials as time to complete the compared to the TEB and EBand local
shown in (1). In this experiment, the target reaching accuracy planners. Even though EBand local planner generates the
is measured using only the data taken at the point O1. Once shortest path, the generated maximum velocity from the
the robot completes the task by travelling on the path Eband local planner is 0.25 ms-1 which is less than the
O1→O2→O3→O4→O1, the robot's final position and the maximum set velocity of 0.3 ms-1. The velocity can be
orientation is measured based on the odometry data provided increased by increasing the proportional and derivative gains
by the Intel Realsense T265 camera. Additionally, the in the EBand local planner parameters. However, for higher
generated velocity commands and the time taken to complete gains, the robot oscillates along the robot trajectory.
the task are also studied to presents a better comparison. Therefore, the existing Eband local planner parameters are
described in Table I are used to generating the optimum
motion under the defined speed limits.
ଵ ଵ
  ൌ ට σ୬ଵሺܺ௚ െ ܺ௟ ሻଶ ൅ σ୬ଵሺܻ௚ െ ܻ௟ ሻଶ  
୬ ୬ TABLE II. OVERALL PERFORMANCE OF LOCAL PLANNERS WITHOUT
OBSTACLES
A. Performance of the Local Planners without Obstacles Local planner
Fig. 6 shows a sample set of generated local planners Performance Measurement
DWA EBAND TEB
during the experiment. The Figure includes the local planner
Average RMSD value 0.47 0.74 0.41
generated during the path O1→O2→O3→O4 and path
O4→O1 separately. According to Fig. 6, it can be seen that all The average time is taken to
252 s 285 s 257 s
complete the task
the local planners offset to a certain degree from the exact goal Minimum distance from final
points. One of the main reasons for that is the slightly higher 0.19m 0.12m 0.25m
goal point O1
goal tolerances which are used with this system to avoid Maximum distance from final
0.41m 0.35m 0.48m
unnecessary oscillations. Other than that, the localization goal point O1
errors also affect the accuracy of reaching the goal. However, Minimum angular error from
errors that occur when reaching the goal are not more than 0.4 desired orientation at goal point 0.01 rad 0.01rad 0.02rad
O1
m in any of the trials performed during the experiment. It can Maximum angular error from
be considered as an acceptable range for this robot setup. desired orientation at goal point 1.3rad 1.1rad 1.5rad
O1
EBand local planner generates the shortest path compared Number of successfully
to the other two. Even though the path is short, sometimes the 10 10 10
completed trials out of 10 trials
robot has to travel through the paths which are very close to


Authorized licensed use limited to: DRDO-R&DE. Downloaded on December 26,2022 at 06:47:08 UTC from IEEE Xplore. Restrictions apply.
 According to the position and orientation errors measured
at the final goal location O1, it can be seen that Eband has a
comparatively higher accuracy of goal-reaching, and TEB
local planner has the maximum error deviation. The position
and orientation errors can also be improved by reducing the
goal tolerances and reducing the footprint size. However, as
explained before, reducing the goal tolerance to a lower value
creates unwanted oscillations when reaching the goal points.
Also, to increase safety, the footprint size of the navigation
stack is set to 0.8 m X 0.8 m which is higher than the actual
robot's footprint size, which is 0.75 m X 0.7 m. That is also
has a slight effect on the accuracy of reaching the goal.
However, when compared with the size of the robot, the
position errors and orientation errors of the existing setup is
acceptable.
B. Performance of the Local Planners with Obstacles
Fig. 7 shows the linear velocity commands generated from
each local planner during the experiment. The DWA and TEB
local planners have better control of the robot compared to the
EBand local planner. That is clearly understood by observing
the velocity plot shown in Fig. 7. Since the proportional and
the derivative gains of the EBand controller has to be
manually tuned, parameter tuning of EBand local planner is
slightly tricky when compared to that of the TEB and DWA
local planners. Fig. 8. Generated local planners with obstacles

The second experiment is performed by placing two

generate backward motions. Generally, the DWA local
obstacles in random locations on the map. An example set of
planner avoids the obstacles by making a quick turn when it
local paths generated with obstacles is shown in Fig. 8. During
detects the obstacles. However, especially in narrow areas,
the trials, the same measurements are taken for each local
this quick turn would not be enough to create the path to avoid
planner. Out of the ten trials performed by each local planner,
the obstacle. EBand local planner follows a concept similar to
the TEB local planner was able to complete all the trials
TEB local planner in obstacle avoiding. However, the reactive
successfully while DWA local planner completed eight trials,
time of EBand local planner is comparatively low. Thus, it is
and EBand local planner completed seven trials. From this
understood that the EBand local planner also has limitations
experiment, it is observed that at some stages, DWA and
in avoiding obstacles. When observing the paths O4→O1, all
EBand local planners got stuck in recovery behaviours while
the local planners have avoided the obstacles before the robot
avoiding the obstacles.
reaches the obstacle. That is because the global costmap gets
When looking at the generated path, it can be observed that updated after detecting each obstacle when the robot travels
the TEB local planner has travelled a longer distance when from O1 to O4. Therefore, the global plan from O4 to O1 does
compared to the EBand and the DWA local planners. not go through the previously detected obstacle locations, and
Generally, the TEB local planner always maintains a this helps the robot to avoid the obstacles before it reaches it.
minimum distance from the obstacles, and the path is planned
Table III shows the overall performance of the local
accordingly. A minimum distance can be set during parameter
planners with the obstacles. The RMSD values are
tuning, and for this robot setup, it is 0.6 m. Also, TEB local
comparatively higher than that of the previous experiment due
planner can generate backward motions when it cannot
to the continuous change of path during obstacle avoidance.
generate the forward paths. Therefore, the TEB local planner
When considering the overall time taken to complete the task,
has the best capability to avoid obstacles without getting stuck
the DWA local planner is the quickest. However, DWA local
in the recovery state. DWA local planner also performed well
planner can only complete 7 trials out of 10. Therefore, TEB
except on a few occasions where the robot got stuck in
local planner can be considered as the best, even if it takes
recovery behaviour. The main problem of using the DWA
more time to complete the task.
local planner with differential drive robots is its inability to

Fig. 7. Linear velocity generation of local planners


Authorized licensed use limited to: DRDO-R&DE. Downloaded on December 26,2022 at 06:47:08 UTC from IEEE Xplore. Restrictions apply.
 TABLE III. OVERALL PERFORMANCE OF LOCAL PLANNERS WITH global planning technologies can also be studied to make a
OBSTACLES
better comparison.
Local planner
Performance Measurement
DWA EBAND TEB
ACKNOWLEDGEMENT
Average RMSD value 0.72 0.98 0.93 This research is supported by the Department of
Mechanical Engineering at Chiang Mai University in
The average time is taken to
complete the task
285 s 302 s 322 s Thailand.
Minimum distance from final goal
0.18m 0.15m 0.28m REFERENCES
point O1
Maximum distance from final goal [1] M. Pittner, M. Hiller, F. Particke, L. Patino-Studencki and J.
0.52m 0.38m 0.54m
point O1 Thielecke, "Systematic Analysis of Global and Local Planners for
Minimum angular error from 0.01 Optimal Trajectory Planning," ISR 2018; 50th International
0.01rad -0.01rad
desired orientation at goal point O1 rad Symposium on Robotics, Munich, Germany, 2018, pp. 1-4.
Maximum angular error from [2] M.S.Kim, R.Delgado, and B.W.Choi, "Comparative Study of ROS on
1.5rad 1.3rad 1.2rad
desired orientation at goal point O1 Embedded System for a Mobile Robot," Journal of Automation,
Number of successfully completed Mobile Robotics & Intelligent Systems, vol.12,2018.
8 7 10
trials out of 10 trials [3] B. Cybulski, A. Wegierska and G. Granosik, "Accuracy comparison of
navigation local planners on ROS-based mobile robot," 2019 12th
International Workshop on Robot Motion and Control (RoMoCo),
V. CONCLUSION AND FUTURE WORK Poznań, Poland, 2019, pp. 104-111, doi:
10.1109/RoMoCo.2019.8787346.
The main objective of this study is to analyze the
[4] P. M. Plaza, A.Hussein, D.Martin, and A. D. Escalera, "Global and
performance of the ROS local planners with a differential Local Path Planning Study in a ROS-Based Research Platform for
drive heavy robot. After conducting the tests and analyzing the Autonomous Vehicles," Journal of Advanced Transportation, 2018.
results, it can be concluded that both TEB and DWA local [5] D. Fox, W. Burgard and S. Thrun, ”The dynamic window approach to
planners are an effective path planning strategy for a collision avoidance,” in IEEE Robotics & Automation Magazine, vol.
differential drive heavy robotic system. However, when 4, no. 1, pp. 23-33, March 1997. doi: 10.1109/100.580977
comparing the TEB and DWA local planners, it is hard to [6] " Wiki.ros.org.(2020) .dwa_local_planner - ROS Wiki", Wiki.ros.org,
indicate the specific comparison winner. Each of these 2020. [Online]. Available: http://wiki.ros.org/dwa_local_planner.
[Accessed: 9- Nov- 2020].
planners should be considered under different criteria. When
considering the smoothness of the motion, repeatability, and [7] S. Quinlan and O. Khatib," Elastic bands: connecting path planning and
control," [1993] Proceedings IEEE International Conference on
time consumption, the DWA local planner is significant. The Robotics and Automation, Atlanta, GA, USA, 1993, pp. 802-807 vol.2.
goal-reaching accuracy of the EBand local planner is doi: 10.1109/ROBOT.1993.291936.
comparatively higher than the others. However, when it comes [8] “Wiki.ros.org. (2020). eband_local_planner - ROS Wiki”. [online]
to the obstacle avoidance capability and the higher reactivity, Available at: http://wiki.ros.org/eband local planner [Access: 9- Nov-
the TEB local planner can be distinguished as the best option. 2020].
Therefore, TEB local planner can be considered as a better [9] C. Roesmann, W. Feiten, T. Woesch, F. Hoffmann and T. Bertram,"
option for a large differential drive robot. Trajectory modification considering dynamic constraints of
autonomous robots," ROBOTIK 2012; 7th German Conference on
This experiment is conducted with the same navigation Robotics, Munich, Germany, 2012, pp. 1-6.
parameters for each local planner. It might affect some of the [10] “Wiki.ros.org. (2020). teb_local_planner” - ROS Wiki. [online]
results of local planners alongside with the localization errors. Available at: http://wiki.ros.org/teb local planner [Access: 9-Nov-
2020].
In this experiment, the navigation is completely done based on
[11] " Wiki.ros.org. (2020). REP 105 -- Coordinate Frames for Mobile
the data from the Realsense D435i camera and the T265 Platforms (ROS.org)", Ros.org, 2020. [Online]. Available:
tracking camera. The results would be more accurate if the https://www.ros.org/reps/rep-0105.html. [Accessed: 10- Nov- 2020].
localization technique is improved by adding a LIDAR sensor. [12] " Wiki.ros.org. (2020). rtabmap_ros - ROS Wiki", Wiki.ros.org, 2020.
[Online]. Available: http://wiki.ros.org/rtabmap_ros. [Accessed: 10-
For future work, local planners can be studied in a Nov- 2020].
dynamic environment with a large robot setup. Also, the [13] " Wiki.ros.org. (2020). navfn - ROS Wiki", Wiki.ros.org, 2020.
experiment can be improved by adding an external motion [Online]. Available: http://wiki.ros.org/navfn. [Accessed: 10- Nov-
capturing setup to measure the goal-reaching tolerances. 2020.
Furthermore, the performances of local planners with different


Authorized licensed use limited to: DRDO-R&DE. Downloaded on December 26,2022 at 06:47:08 UTC from IEEE Xplore. Restrictions apply.