Autonomous Exploration and Mapping using a Mobile Robot

Running ROS

Rachael N. Darmanin and Marvin Bugeja

Department of Systems and Control Engineering, University of Malta, Msida, Malta

Keywords: Wheeled Mobile Robots, ROS, Autonomous Exploration, SLAM.

Abstract: This paper proposes a practical solution to the autonomous exploration and mapping problem using a single
mobile robot. Moreover, the authors implement the proposed scheme within the Robot Operating System
(ROS), and validate it experimentally using PowerBot, a real wheeled mobile robot equipped with a 2D laser
scanner. In essence, the proposed scheme integrates an efficient particle-filter-based SLAM algorithm, two
different exploration strategies, and a path-planning and navigation module. The modular nature of the pro-
posed scheme is intentional and advantageous. It allows the authors to compare the two exploration strategies
under investigation objectively and with ease. Finally, hypotheses testing is also used to strengthen the results
of the comparative analysis.

1 INTRODUCTION vironment as possible, and in the shortest time pos-

sible. Other researchers argue that rather than just
Exploration and mapping by an autonomous agent exploratory actions, accurate mapping requires ex-
have been of particular interest to the robotics com- ploitation actions, such as place revisiting actions
munity for decades. Research shows that the au- (Makarenko et al., 2002). This gave rise to Active
tonomous mapping of an unknown environment by a SLAM strategies, that seek to improve the localiza-
mobile robot is a non-trivial task (Thrun et al., 2005). tion estimate of the robot rather than explore as much
Global Positioning Systems (GPS) and the use of bea- of the environment as possible in the shortest time.
cons for mapping an unknown environment may facil- However, both exploration and Active SLAM strate-
itate the task of mapping (Dissanayake et al., 2011). gies ultimately provide the robot with a sequence of
However, such systems suffer from a number of in- locations that it needs to visit in order to meet the
herent limitations, namely, the inaccessibility of GPS specified exploration criteria.
indoors and the laborious setting up of strategically For a robot to autonomously explore and map an
placed beacons. Hence, the concept of Simultaneous environment, it must follow a set of navigation in-
Localization and Mapping (SLAM), was developed in structions that allow it to proceed from one location
order to amalgamate sensory data into a world model to the next safely, even in the presence of obstacles.
(or a map). Through probabilistic techniques, SLAM During the years, several path planning and motion
algorithms are capable of processing sensory data to control algorithms have been proposed which enable
estimate the location of a robotic system and a map of the robot to move to the chosen location. The integra-
its environment at the same time. tion of a SLAM algorithm, an exploration or Active
In robotics, maps are mainly acquired to enable SLAM strategy and a path planning/navigation algo-
autonomous navigation. However, mapping using rithm allows the robot to autonomously explore and
robots can also be useful when it comes to explore map an unknown environment using just sensory data.
and map dangerous environments, or places that are The purpose of this work was to develop a sin-
inaccessible to human beings. Hence, autonomous gle robot system, running Robot Operating System
exploration strategies were developed, which enable (ROS), that can autonomously explore and map un-
the robot to determine the next location to explore known environments efficiently. The system needed
in order to improve its map and localization esti- to be modular, in that the SLAM, path planning and
mates. Exploration strategies place an emphasis on motion control modules, can be used with differ-
autonomously mapping as much of an unknown en- ent exploration strategies, without any modifications.

208
Darmanin, R. and Bugeja, M.
Autonomous Exploration and Mapping using a Mobile Robot Running ROS.
DOI: 10.5220/0005962102080215
In Proceedings of the 13th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2016) - Volume 2, pages 208-215
ISBN: 978-989-758-198-4
Copyright c 2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
 Autonomous Exploration and Mapping using a Mobile Robot Running ROS

ROS facilitates this modular implementation since it (DWA) (Fox et al., 1997), among others, have been
allows a set of independent processes to communicate proposed to allow a robot to navigate autonomously
through a message passing structure. in the environment. During the years, there have
The main contributions of this work are: 1) the de- been many advancements in the field of path planning
velopment and experimental evaluation of a new ROS algorithms that can be used seamlessly within the
package consisting of two autonomous exploration integrated problem of autonomous exploration and
strategies for mobile robots. To the best of the au- mapping. Moreover, the Dijkstra, A* and D* algo-
thors’ knowledge, no other ROS package offering the rithms are all global path planners, in that they plan a
same functionality is available to date; 2) a compara- path from the current robot location to the destination
tive study of the two mentioned exploration strategies, based on the current global map (the whole map ac-
based on physical experiments in a real-life environ- quired so far). The DWA was designed to make use of
ment. the velocity control space to search for control com-
The rest of this paper is structured as fol- mands that must be sent to the robot in order to reach
lows. Section 2 provides a review of the exist- a goal destination. Hence, the DWA does not make
ing autonomous exploration and mapping systems. use of a global map, but rather it uses the local map
Section 3 describes the main hardware used in this obtained through sensory data.
implementation and the underlying software frame- Perhaps the most complex problem that must be
work that makes up the whole robotic system. solved in the task illustrated in Figure 1, is SLAM. In
Section 4, presents a set of experimental results while autonomous exploration, the map estimate produced
Section 5 contains a comparative analysis, backed up by SLAM is used by the exploration algorithms in
by statistical hypothesis testing, of the two explo- order to identify which is the best location that the
ration strategies under investigation. Finally, conclu- robot must visit next. Overall, SLAM algorithms
sions from this work are drawn in Section 6. may be organized into three main categories as fol-
lows: Extended Kalman Filter (EKF) SLAM, Particle
Filter SLAM, and Graph Based SLAM. EKF SLAM
was for many years the preferred choice since it re-
2 RELATED WORK quires a simple and straight-forward implementation
and performs well in small areas having a limited
Being aware of one’s environment means that: one number of landmarks. However, since EKF SLAM
is able to perceive the environment, localize himself bases its estimate on just one hypothesis and assumes
in it, and be able to navigate effectively in this envi- that the posterior is normally distributed, it may be-
ronment. In the field of mobile robotics, this involves come overconfident in its estimate, leading to less ac-
learning maps of an unknown environment, which is curate mapping (Dissanayake et al., 2011). Hence,
the integration of three non-trivial tasks that must be Particle Filtering became the preferred probabilis-
solved concurrently (Stachniss et al., 2005). Figure 1 tic technique for many researchers. Particularly, in
illustrates this concept as the integration of mapping, (Thrun et al., 2004), the authors integrate the common
localization and path planning and motion control. Sequential Importance Resampling, (SIR) filter with
EKF to produce FastSLAM, which results in a faster
algorithm, when compared to the common SIR filter
on its own. Furthermore, Rao-Blackwellized Particle
Filters are more commonly used to solve the SLAM
problem since they attempt to reduce the degeneracy
and particle depletion problems presented by less ad-
vanced particle filters such as the Sequential Impor-
tance Sampling (SIS) and SIR filters (Grisetti et al.,
2007). In (Grisetti et al., 2007) the authors propose
two improved techniques for Rao-Blackwellized Par-
ticle Filter SLAM, that make the algorithm more effi-
Figure 1: Illustration of the integrated approaches, adapted cient and accurate in estimating the most likely pose
from (Makarenko et al., 2002). as well as having a reduction in the particle depletion
problem. The implementation of this improved Rao-
Given a known map, several path planning algo- Blackwellized Particle Filter is often referred to as
rithms such as Dijkstra’s algorithm (Dijkstra, 1959), GMapping, which is directly accessible online from
A* algorithm (Dechter and Pearl, 1985), D* algo- openslam.org. Moreover, the use of Graph-Based
rithm (Stentz, 1994) and Dynamic Window Approach

209
ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

techniques in SLAM is fast gaining popularity. In among others, seek to compare such strategies. In
Graph-Based SLAM, each node in the graph repre- (Amigoni, 2008), the author shows that the simpler
sents an estimated robot pose or the estimated posi- algorithms, such as the NBV approach, yield the best
tion of a landmark, while the edges connecting these results in the shortest time possible, and the robot
nodes encode in them sensor observations that are travelling the shortest distance. However, in (Carlone
constrained by the estimated robot pose. The perfor- et al., 2014), the authors show that when the focus is
mance of Graph-Based SLAM is said to be compara- placed on the quality of the map, then those strate-
ble to that of Particle Filter-Based SLAM algorithms gies that involve Active SLAM produce the best re-
such as GMapping (Burgard et al., 2009). sults, even if not in the shortest time possible. Such
Although the task of choosing the next loca- works indicate that the choice of exploration or Ac-
tion that the robot should visit seems to be intu- tive SLAM algorithm is highly dependent on the ap-
itive, several researchers have developed several dif- plication and on the criteria that shall be used in the
ferent exploration and Active SLAM algorithms for evaluation of the results.
this purpose. One very popular and simple explo-
ration strategy is the Nearest Frontier approach (Ya-
mauchi, 1997). This algorithm simply analyses an 3 SYSTEM DESIGN AND
Occupancy Grid Map and detects all potential borders
(called candidate frontiers) between the cells marked
IMPLEMENTATION
as OPEN (free from obstacles) and those marked as
UNKNOWN (not yet explored). Such a map is seg- The concept of autonomous exploration and map-
mented into cells, where each cell is assigned with a ping represented in Figure 1 was implemented in
probability of occupancy, with zero probability mean- ROS on PowerBotT M , a differentially-driven wheeled
ing that it is free from obstacles, and with a probabil- mobile robot designed and manufactured by Adept
ity of one, meaning that it is OCCUPIED. Normally, Mobilerobots for research and rapid prototyping.
a probability of 0.5 means that there is no informa- PowerBot is equipped with a SICK LMS200 laser
tion about the occupancy of the cell, and hence it is scanner, an Advanced Robotics Control and Opera-
still UNKNOWN (Yamauchi, 1997). The distance be- tions (ARCOS) robot controller for low-level motion
tween the current robot pose and each frontier, is anal- control, as well as an on-board computerfor algorithm
ysed, and the frontier closest to the robot is chosen as development. The on-board computer is interfaced
the next location to explore. with the robot controller via Advanced Robotics In-
terface for Applications (ARIA), which is a library of
Another algorithm that is designed to explore as functions that are capable of handling the lowest level
much of the environment as possible is the Next Best details of the client-server interactions, such as con-
View (NBV) approach proposed in (Gonzalez-Baños trolling the speed of the robot.
and Latombe, 2002). In this case, the evaluation cri- Robot Operating System (ROS) is a software
teria of the candidate destinations, attempt to strike a framework that is ideal for the development of robotic
balance between the amount of unknown area that the applications. It consists of a number of software pack-
robot can explore from that location, and the distance ages that are used to perform particular tasks, for ex-
that the robot must travel to arrive there. ample, SLAM. ROS facilitates modular implementa-
Unlike the Nearest Frontier and the NBV strate- tion of different functions, in that processes that are
gies just described, in (Makarenko et al., 2002) and currently running on the system (called nodes), in-
(Stachniss et al., 2005), among others, the authors teract between themselves through a message passing
propose techniques that do not only focus on explor- system, as shown in Figure 2. Each packet of data is
ing as much of the environment as possible, but in sent between nodes according to the topic that it has
addition aim to improve the localization estimate of been assigned. The topic defines the type of data mes-
the robot (Active SLAM). Particularly, in (Stachniss sage that is sent. Furthermore, a node may either be
et al., 2005), the authors propose a technique that a publishing node (broadcasting data) or a subscriber
makes use of the entropy of a Rao-Blackwellized Par- node (receiving data). Hence any node subscribing
ticle Filter together with the distance from the current to the same topic shall receive the same data being
robot pose in order to evaluate candidate destinations. broadcast by the publisher. For instance, in Figure 2
To date, the robotics community is still in de- the publishing node “talker” is communicating with
bate on which of these exploration strategies is best the subscriber node “listener” by sending data mes-
to solve the problem of autonomous exploration and sages over the topic “chatter”. These interactions are
mapping. To this end, several works such as those controlled by the master node called “roscore”.
presented in (Amigoni, 2008; Carlone et al., 2014), The modularity of ROS leads to a rather simple

210
 Autonomous Exploration and Mapping using a Mobile Robot Running ROS

was set to be updated every three seconds.

Although there are some packages that implement
the Nearest Frontier exploration approach in ROS,
these presented certain limitations, such as not being
Figure 2: ROS architecture of nodes and topics. fully autonomous, and thus could not be used for the
intended purpose. Hence, the Nearest Frontier ap-
design that emerges from the integration of the fun- proach proposed in (Yamauchi, 1997) and the Next
damental concepts and the system block diagram de-
Best View approach proposed in (Gonzalez-Baños and
picted in Figure 3. Going from the block diagram in Latombe, 2002) were both implemented from scratch
Figure 3 to a system designed in ROS requires the
as ROS nodes in a new package called “exploration”.
identification of the specific package to be used in To the best of the authors knowledge, this is the first
place of each component in the block diagram. The
time that these two algorithms are being made avail-
parameters of each package must then be tuned for able in ROS.
the specific robot platform and application.
For both exploration strategies, the potential can-
didates for exploration emerge through the use of
frontier detection which was also proposed in (Ya-
mauchi, 1997). Frontier detection is performed in two
steps:
1. The search for frontier cells, and
2. The grouping of said frontier cells into frontiers
Figure 3: ROS architecture of nodes and topics. comprising adjacent frontier cells only.
In the search of frontier cells, the algorithm traverses
To perform SLAM, the gmapping package, which
the map data and creates an vector of frontier cells.
implements the Rao-Blackwellized particle-filter-
These frontier cells are then grouped into frontiers ac-
based approach developed in (Grisetti et al., 2007),
cording to their adjacency. Each frontier is then stored
was chosen due to its ability to map an environment
in a two-dimensional vector which is used by the ex-
in an accurate and efficient manner. In contrast to
ploration algorithm for further analysis. Eventually,
EKF-based SLAM, this technique uses multiple hy-
the exploration process terminates when the frontier
potheses represented by the particle set in the parti-
detection algorithm can no longer detect frontiers that
cle filter. The node, called “slam gmapping”, sub-
are at least ten cells long. This applies to both explo-
scribes to the raw odometry data and laser scan data
ration approaches that were examined in this work.
over the “/RosAria/pose” and “/scan” topics, respec-
The difference among the two approaches lies
tively. The package “ROSARIA” acts as a wrapper
in how they evaluate the candidate destinations and
for the ARIA libraries in ROS. It allows the transmis-
choose the location that the robot must explore next.
sion of high level commands to the robot controller
In the Nearest Frontier approach, a global path is
from ROS. In fact, the odometric pose estimate of the
planned from the current robot pose to each candidate.
robot is published by the “RosAria” node. In turn, the
The candidate that is closest to the robot (i.e. the one
“slam gmapping” node publishes the map data, which
having the shortest path) is then chosen. On the other
is then used by the exploration algorithm to compute
hand, the Next Best View approach implements the
the next location that the robot needs to explore.
following equation to rank the potential candidates,
The published map is an Occupancy Grid Map
vector. Moreover, some of the “slam gmapping” pa-
rameters had to be tuned in order to extract the best g(q) = A(q)exp(−λL(q, qk )) (1)
performance in the given environment with the given where g(q) represents the ranking function, A(q) is
robot. For example: two particular parameters were the total unknown area that is expected to be covered
used to enable the SLAM package to execute a scan by the laser scanner from a candidate location, and
every time the robot translates by 0.2 metres or rotates L(q, qk ) is the length of the path planned from the cur-
by 0.1 radians. The settings were chosen in an attempt rent robot pose to each candidate location, using the
to obtain the best map with the available computa- global occupancy grid map published by the SLAM
tional power, as more frequent scans naturally require algorithm.
more computational power. Furthermore, the laser The path length, L(q, qk ) is computed in the same
range was also tuned in accordance with the speci- manner as it is computed in the Nearest Frontier ap-
fications of the laser scanner, namely the maximum proach. This means that the same global path planner
range setting was set to five metres. Finally, the map used to plan accessible paths for navigation is used to

211
ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

plan a path for the robot in an incomplete occupancy grid map that is being built by SLAM. Moreover, hav-
grid map. The length of this path is then computed. ing just a global path planned does not enable the
The unknown area that the robot can potentially “see” robot to avoid dynamic obstacles such as people walk-
from a candidate location, A(q), is calculated by a ray ing around it. To mitigate this problem, the package
casting technique, which is also used in (Gonzalez- makes use of a local path planner, which plans inter-
Baños and Latombe, 2002). Ray casting attempts to mediate paths using a local map. This local map is
simulate the data obtained through a single scan of another costmap which is constructed using the laser
the laser scanner. Given a scan resolution, the max- scan data directly. Therefore, if a dynamic obstacle
imum laser range and the scanning angle of the real appears in the local path planned of the robot, this
laser scanner, a number of rays are projected from a would be detected by the laser scanner and the local
candidate location. If the ray encounters a known cell path planner changes course accordingly. However,
(OPEN or OCCUPIED), then the length of the ray the local path planner always seeks to guide the robot
from that point to the maximum ray length is cropped as close to the global path as possible. For this effect,
out. Consequently, the endpoint of the ray is the point the global and local costmap parameters were tuned to
at which the ray hits a known cell, and therefore, the the dimensions of the robot. Furthermore, the global
length of that ray is the Euclidean distance between path planner adopted in this case is the Dijkstra’s Al-
the candidate location coordinates and the coordinates gorithm (Dijkstra, 1959), while the local path plan-
of the ray endpoint. Thus, the unknown area that ner adopted is the Dynamic Window Approach (Fox
may be visible from a candidate point is the summa- et al., 1997). The implementation of these path plan-
tion of the lengths of the rays projected outwards over ners were both available in ROS.
the unknown cells only, as seen in (2), where (xc , yc )
are the map coordinates of the candidate location and
(xend , yend ) are the coordinates of the ray endpoint. 4 RESULTS
q One of the aims of this study was to compare the
A(q) = ∑ (xend − xc )2 + (yend − yc )2 (2) mapping accuracy of two exploration strategies which
No. o f rays
were implemented as part of a complete autonomous
Furthermore, the parameter λ plays an important role exploration and mapping system in ROS. The first
in the selection of the next best view since it allows set of results presented in Section 4.1 shows how the
the robot to either prefer shorter routes over a larger different exploration strategies choose different ex-
area of visibility, or vice-versa. If the value of λ ploratory locations from a set of candidates. Further-
is small, the algorithm prefers the candidates from more, Section 4.2 presents a set of results obtained
which a larger unknown area may be explored and through a number of real life autonomous exploration
hence gives them a higher rank. If λ is large, the algo- and mapping experiments. The map obtained in each
rithm gives a higher rank to those candidates that are experiment was compared to a ground truth map, il-
closer to the robot. Eventually the candidate that has lustrated in Figure 4, which was obtained by manually
the largest ranking value, g(q), is chosen as the next steering the robot in the environment. Hence, the Ac-
location that the robot must navigate to and explore. ceptance Index metric could be calculated. This com-
parative metric was introduced in (Carpin, 2008). The
As soon as the next destination is chosen, the robot
acceptance index, ω, is calculated as the ratio of the
must navigate to that location. To do so, path plan-
agreement between the known cells of a ground truth
ning and motion control algorithms are required to
map, M1, and a resulting map, M2, to the total sum
allow the robot to navigate effectively in an environ-
of agreed and disagreed cells, as shown in (3). When
ment cluttered with both static and dynamic obstacles.
ω = 0 it means that the two maps are completely dis-
For this purpose, the package “move base” was used.
tinct and when ω = 1 it means that the two maps are
This package requires the pose of the robot within the
completely identical.
map to be able to plan a global path from the current
robot pose to the goal destination. In order to take the
agr(M1, M2)
dimensions of the robot into consideration, the path ω(M1, M2) = (3)
is planned in a global costmap that inflates the cost agr(M1, M2) + dis(M1, M2)
around each OCCUPIED cell. If the robot traverses
the inflated circle around such a cell, it may be either 4.1 Case Study
close to colliding with an obstacle or in fact, it may be
in collision, depending on the cost of each cell. This In this case study, an incomplete map of the Univer-
global costmap is constructed on the static occupancy sity of Malta Faculty of Engineering ground floor was

212
 Autonomous Exploration and Mapping using a Mobile Robot Running ROS

location since it offered the smallest path length. The

algorithm of interest is NBV. When the value of λ was
small (i.e. {0.02, 0.15, 0.5}), Candidate 3 was always
preferred since it offered the largest area of visibil-
ity even though it was not the closest location to the
robot. When λ was equal to 1.0, the algorithm pre-
ferred Candidate 1 since it is the location closest to the
robot. Hence, one can conclude that as the value of λ
Figure 4: Ground truth map obtained by manually steering increases, the NBV approach starts preferring candi-
the robot in the laboratory environment. dates that are closer to the current robot pose and thus,
this approach becomes similar to the Nearest Frontier
used, where three candidate locations were detected approach.
by the frontier detection technique as shown in Figure
5. Candidate 3 shows a frontier that is made up of a 4.2 Completely Autonomous
collection of connected smaller frontiers. This is due Exploration and Mapping Results
to the diffused laser rays in that area which resulted in
several gaps of unknown cells between the free cells. In order to validate and compare the algorithms,
For this scenario, the ranking values of each approach the whole system was used in a set of four ex-
are tabulated in Table 1. In order to show how the periments where for each experiment, seven trials
NBV approach can be used to rank candidates differ- were recorded. In each experiment, the robot au-
ently, the parameter λ was set to four different values tonomously explored, navigated and mapped a con-
from the set {0.02, 0.15, 0.5, 1.0}. trolled environment (a laboratory) consisting of sev-
eral static obstacles such as desks and chairs. In
the first experiment the Nearest Frontier approach
was used while in the other three experiments, the
NBV approach was used with different values of
λ = {0.15, 0.5, 1.0}. These values of λ were cho-
sen heuristically depending on the visual inspection
of the robot behaviour. Figure 6 illustrates a typical
map resulting from one trial for each of the adopted
approaches while Table 2 presents the average (across
the seven trials in each experiment) acceptance index,
distance travelled and total exploration time for each
approach.
Figure 5: Incomplete map of the University of Malta Fac-
ulty of Engineering ground floor showing three exploratory
candidates, together with the respective path length (in me-
tres) from the current robot pose and the expected unknown
area (in metres squared) that is visible from each candidate.

Table 1: Case Study results showing the value of the rank-

ing function for each candidate, {1, 2, 3}, where the ranking
function of the Nearest Frontier is the path length, which
(a) Nearest Frontier (b) NBV λ = 0.15
must be minimized, and that of the NBV is g(q) as shown in
(1), which must be maximized. Note that the ranking value
of the chosen candidate for each approach is italicized.
Approach 1 2 3
Nearest Frontier 4.84 35.65 7.04
NBV λ = 0.02 23.70 40.55 144.1
NBV λ = 0.15 12.64 0.39 57.74
NBV λ = 0.5 2.33 1.49 ×10−6 4.92 (c) NBV λ = 0.5 (d) NBV λ = 1.0
NBV λ = 1.0 0.21 2.67 ×10−14 0.15
Figure 6: Sample map results for each approach showing
the robot trajectory. ‘A’ and ‘B’ indicate the initial and final
As can be seen in Table 1, the Nearest Frontier robot poses respectively.
approach chose Candidate 1 as the next exploratory

213
ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

Table 2: Autonomous exploration and mapping results

showing the average acceptance index, average distance
travelled (metres) and the average exploration time for each
approach (seconds).
Acceptance Distance Exploration
Index Travelled Time
Nearest
0.82 25.41 132.48
Frontier
NBV
0.86 30.95 156.28
λ = 0.15
NBV
0.84 40.44 140.90
λ = 0.5
NBV Figure 7: Boxplot of the acceptance index distributions.
0.83 35.15 160.56
λ = 1.0
Table 3: ANOVA statistical test: The Hypotheses.
Visual inspection of the resulting maps, compared H0 The NF approach and the NBV approach
to the ground truth map in Figure 4, and trajectories, with λ = {0.15, 0.5, 1.0} all perform equally
reveals that the Nearest Frontier approach keeps the well.
robot from visiting certain locations that are quite far H1 Some of the approaches have better perfor-
from its current pose. This then leads to unclear fea- mance than the others in terms of map accu-
tures and regions in the map, as shown in Figure 6a. racy.
On the other hand, in general, with the NBV approach
the robot seems to cover a larger portion of the envi-
dex performance of the four experiments under test.
ronment, hence leading to more accurate maps. It is
Hence, this means that H0 was rejected. However, this
important to note that the two approaches are subject
test does not reveal where the significant difference
to the same termination condition, specified in Sec-
occurs, and hence the Tukey HSD post-hoc test was
tion 3, to end exploration. Furthermore, in this study,
performed. The outcome of this test shows that while
although the focus is placed on the map accuracy, the
there is a significant difference between the NF algo-
distance travelled by the robot and the total time of ex-
rithm and the NBV algorithm with λ = 0.15, there
ploration were also recorded to analyse the efficiency
is no significant difference among the other groups.
of the algorithms.
Since the mean of NBV with λ = 0.15 is larger than
that of NF, then the NBV approach with λ = 0.15
is significantly better than NF. Furthermore, the lack
5 DISCUSSION of significant difference between NF and NBV with
λ = 0.5, and NF and NBV with λ = 1.0 indicates that
A question arises, of whether it is better to use a sim- for higher value of λ the NBV algorithm converges
ple and fast exploratory algorithm like the Nearest to the NF algorithm. One may argue that this is in
Frontier (NF) approach, or the more complex and less line with the theoretical expectations since as λ in-
time-efficient Next Best View (NBV) approach. To creases, the NBV approach prefers the closer candi-
evaluate the map accuracy objectively, the differences dates (i.e. shorter path lengths) over candidates that
between the acceptance indices of the seven trials for offer a higher area of visibility. Furthermore, one
each approach were statistically tested. The boxplot must note that the parameter λ is application-specific.
in Figure 7 illustrates the distribution of these indices Thus, the threshold value at which the NBV algorithm
for each of the four adopted approaches. Intuitively, is not statistically different from the NF approach may
Figure 7 indicates that the best accuracy was obtained be different for different environments.
by the NBV approach since this exhibited the high- From the means tabulated in Table 2, one can im-
est mean and the smallest variance. However, as the mediately identify a difference between the distance
value of λ is increased, the NBV approach starts con- travelled by the NF approach and the NBV approach.
verging towards the NF approach. This intuition was This difference is also apparent in the time taken to
verified statistically by using the One-Way ANOVA finish the exploration and mapping process. Although
hypothesis test, where the null (H0 ) and alternative such metrics have been used to compare different ex-
(H1 ) hypotheses were formed as shown in Table 3. ploratory algorithms, it is important to note that in this
The main result of ANOVA reported that there case, the distance travelled and the time taken do not
exists a significant difference in the acceptance in- depend on just the adopted strategy. The actual dis-

214
 Autonomous Exploration and Mapping using a Mobile Robot Running ROS

tance travelled by the robot does not only depend on cle filters using Kullback-Leibler Divergence. Journal
the destination, but it also depends on the local tra- of Intelligent and Robotic Systems, 75(2):291–311.
jectories planned by the local path planner, which ac- Carpin, S. (2008). Fast and accurate map merging for multi-
counts for the robot kinematic constraints. Hence, the robot systems. Autonomous Robots, 25(3):305–316.
length of the travelled path is typically longer than Dechter, R. and Pearl, J. (1985). Generalized best-first
that which is considered as the global path, L(q, qk ) in search strategies and the optimality of A*. Journal
of the ACM (JACM), 32(3):505–536.
both algorithms. Furthermore, the time taken to finish
the whole process does not only depend on the algo- Dijkstra, E. W. (1959). A note on two problems in connex-
ion with graphs. Numerische mathematik, 1(1):269–
rithm or the distance travelled. It also depends on the 271.
speed of the robot while it is navigating. This speed Dissanayake, G., Huang, S., Wang, Z., and Ranasinghe, R.
may be reduced significantly if the robot passes close (2011). A review of recent developments in Simul-
to an obstacle, which is a precaution taken by the lo- taneous Localization and Mapping. In 6th Interna-
cal path planner in order to avoid collision with obsta- tional Conference on Industrial and Information Sys-
cles. For these reasons, this study mainly focused on tems, pages 477–482.
the accuracy of the map obtained by the two different Fox, D., Burgard, W., Thrun, S., et al. (1997). The
algorithms which were implemented from scratch as Dynamic Window Approach to collision avoidance.
ROS packages for the first time. IEEE Robotics & Automation Magazine, 4(1):23–33.
Gonzalez-Baños, H. H. and Latombe, J.-C. (2002). Naviga-
tion strategies for exploring indoor environments. The
International Journal of Robotics Research, 21(10-
6 CONCLUSIONS 11):829–848.
Grisetti, G., Stachniss, C., and W.Burgard (2007). Improved
In this paper, a modular design concept was used techniques for grid mapping with Rao-Blackwellized
particle filters. IEEE Transactions on Robotics,
in order to implement a robotic system that can au-
23(1):34–46.
tonomously explore, navigate and map an unknown
Makarenko, A. A., Williams, S. B., Bourgault, F., and
environment in ROS. Furthermore, this work con- Durrant-Whyte, H. F. (2002). An experiment in inte-
tributes a new package to the ROS community. This grated exploration. In Proc. of the IEEE/RSJ Int. Conf.
package consists of the implementation of two explo- on Intelligent Robots and Systems, (IROS), pages 534–
ration algorithms which can be used independently 539.
of the navigation and the SLAM components. More- Stachniss, C., Grisetti, G., and Burgard, W. (2005).
over, the experimental evaluation of the Nearest Fron- Information gain-based exploration using rao-
tier (NF) and the Next Best View (NBV) approach re- blackwellized particle filters. In Robotics: Science
vealed that in general, the NBV approach produces and Systems, volume 2, pages 65–72.
more accurate maps than the NF approach. Further- Stentz, A. (1994). Optimal and efficient path planning for
partially-known environments. In Proc. of IEEE In-
more, from this study one can also conclude that as ternational Conference on Robotics and Automation,
the parameter λ in the NBV approach is increased, the 1994., pages 3310–3317.
NBV algorithm converges to the NF approach. More- Thrun, S., Burgard, W., and Fox, D. (2005). Probabilistic
over, the results clearly confirm that the best choice robotics, chapter 10, pages 309 – 330. MIT press.
of an exploration strategy, is highly dependent on the Thrun, S., Montemerlo, M., Koller, D., Wegbreit, B., Ni-
problem at hand and the environment in question. eto, J., and Nebot, E. (2004). FastSLAM: An efficient
solution to the simultaneous localization and mapping
problem with unknown data association. Journal of
Machine Learning Research, 4(3):380–407.
REFERENCES Yamauchi, B. (1997). A frontier-based approach for au-
tonomous exploration. In IEEE International Sympo-
Amigoni, F. (2008). Experimental evaluation of some ex- sium on Computational Intelligence in Robotics and
ploration strategies for mobile robots. In IEEE In- Automation, 1997., pages 146–151.
ternational Conference on Robotics and Automation
(ICRA), 2008, pages 2818–2823.
Burgard, W., Stachniss, C., Grisetti, G., Steder, B., Kum-
merle, R., Dornhege, C., Ruhnke, M., Kleiner, A., and
Tardós, J. D. (2009). A comparison of SLAM algo-
rithms based on a graph of relations. In IEEE/RSJ In-
ternational Conference on Intelligent Robots and Sys-
tems, (IROS)., pages 2089–2095.
Carlone, L., Du, Jingjing, K., Ng, M., Bona, B., and Indri,
M. (2014). Active SLAM and exploration with parti-

215