2017 25th Mediterranean Conference on Control and Automation (MED)
July 3-6, 2017. Valletta, Malta
Collaborative mapping and navigation for a mobile robot swarm
Ioannis Arvanitakis and Anthony Tzes
Abstract— The collaborative navigation of a team of ground
mobile robots in an unknown planar environment towards fixed
stationary targets is the subject of this paper. Each mobile robot
is equipped with a limited field of view limited-range finder and
a magnetometer to infer its orientation. Each target location is
known, and a switching objective function initially guarantees
the individual robots exploration towards its target area and
afterwards safely guides each robot towards its target. During
the collective exploration, the robots exchange their maps in
a collaborative manner. If a robot detects the target that is
assigned to another robot, it communicates this information to
the corresponding robot and continues to explore its target.
The other robot rather than performing exploration towards
its already discovered target, it attempts to join its map with
that of the reporting robot. Simulation results that prove the
efficiency of the proposed scheme are presented.
I. I NTRODUCTION
Autonomous navigation of mobile robots gathers signifi-
cant attention from researchers over the years due to ongoing
technological advances [1]. Area coverage [2–4] and explo-
ration, surveillance, search and rescue missions, are some of
the tasks-at-hand, where the robots should move efficiently
in the environment, avoiding obstacles during motion and
keeping under consideration physical constraints.
Motion planning for known environments has been exten-
sively researched over the past few decades [5], producing
solutions such as the Artificial Potential Fields [6], the
vector field histogram [7], probabilistic roadmaps [8] and
Rapidly-exploring Random Trees (RRT) [9]. In uncertain
or unknown environments [10, 11], where information about
the environment is taken through on-board sensors, methods
such as the Dynamic Window Approach [12] have been
utilised. These methods, that rely upon local real-time ob-
stacle avoidance, while effective, prove to be inefficient [13]
in the aforementioned environments.
In most cases, sensory information about the environment
is utilized for an on-line map building process; the problem
of navigating towards the unknown target area contains the
sub-problem of exploration. One of the first methods is the
frontier based exploration, where a frontier is defined as
the boundary between explored and unexplored space. The
robot then attempts to move towards the closest frontier to
its position [14]. The frontier concept has been widely used
to research new exploration methods [15, 16].
Ioannis Arvanitakis is with the Department of Electrical & Computer
Engineering, University of Patras, Rio 26500, Greece.
Anthony Tzes is with the New York University Abu Dhabi, Electrical and
Computer Engineering, P.O.Box 129188, Abu Dhabi, United Arab Emirates.
This work has received partial funding from the European Union Horizon
2020 Research and Innovation Programme under the Grant Agreement No.
644128, AEROWORKS.
Corresponding author’s e-mail: anthony.tzes@nyu.edu
978-1-5090-4533-4/17/$31.00 ©2017 IEEE 696
Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on June 12,2020 at 05:41:54 UTC from IEEE Xplore. Restrictions apply.
Utilizing a team of mobile robots instead of a single
robot, can greatly reduce the completion time of the task-at-
hand [17]. In many cases in multi robot exploration schemes,
each member of the swarm is assigned with a specific
area of responsibility via Voronoi - most commonly used
- tessellation of the environment [18, 19]. In other cases a
centralized approach is used for the selection and assignment
of frontiers to each robot, with the help of an appropriate cost
function [20, 21].
The subject of this article is the navigation of a mobile
robot swarm in an unknown environment to known target
locations that need to be discovered. This problem mimics
the foraging problem [22], as an extension of the previous
work by the authors in [23]. A team of mobile robots is
employed, with each robot equipped with a range sensor
of limited field of view and range. The robots exchange
information of the explored space creating a collective map.
The primary contribution lays within the use of a switching
objective function, where its optimization aims initially the
exploration of the target positions from an equal number of
robots from the swarm. Each member is initially assigned
with a specific target. The selection of the frontier for
exploration for each robot is done independently through the
minimization of a cost function. If a robot has discovered
its corresponding target position then the control law for the
specific robot switches to a distance based objective function
to reach it.
The paper is structured as follows: In Section II mathemat-
ical preliminaries and problem statement is given, followed
by Section III where the control law is calculated, the per-
formance and weighting functions are found and the frontier
selection cost function is given. In Section IV simulation
studies that prove the efficiency of the proposed law are
presented, while in Section refs5 conclusions are drawn.
II. P ROBLEM F ORMULATION
A. Mathematical Preliminaries
Consider a path-connected topological space A ⊂ Rn . The
boundary of A is denoted as ∂A, while {Bi }, i = 1, ..., N
denotes a collection of subsets. Spaces A, B are considered
disjoint if A ∩ B = ∅.
The Minkowski sum of two spaces A, B can be defined
as the space given by A ⊕ B = {a + b|a ∈ A, b ∈ B}.
Given the collection of all paths {γk } that connect two
arbitrary points p1 , p2 ∈ A, the length of the shortest path
defines the geodesic metric dg (p1 , p2 ) and the resulting path
is called the geodesic path.
Definition 1: Consider two subspaces B, C ⊆ A. The
geodesic set distance is defined as the minimum geodesic
distance of all points q1 ∈ B from all points q2 ∈ C , i.e.  
   
Hg (C, B) = min min dg (q2 , q1 ) . A= Si (t)
q2 ∈C q1 ∈B i∈IN t

Definition 2: Consider a point r ∈ A. The visibility Each robot is governed by the following kinodynamics
subspace of A from r is defined as a subset Av (r; R),
containing all points q, so that the geodesic path connecting    
ṙi ui
r and q is a straight line and has length less or equal to = (2)
R > 0, i.e. θ̇i ωi

Av (r; R) = {q ∈ A; dg (r, q) = r − q ≤ R, ∧ where ui ∈ R2 , ωi ∈ R.

r + λ(q − r) ∈ A ∀ λ ∈ [0, 1]} . (1) The objective is for each robot to:
1) maximize the following function

H(xi ; pit ) = f (p)φ(p)dp , if pit ∈
/ Ar i (3)
Si

during exploration phase, and

2) maximize the function,
1
H(xi ; pit ) = , if pit ∈ Ari , (4)
dg (pit , ri )
during navigation to the goal position.
Functions f (p) : Si → R+ and φ(p) : Si → R+ are the
performance and weighting functions respectively [23].
Fig. 1: Visibility subspace from an arbitrary position
III. PATH P LANNING
Consider a non path connected topological space W that A. Control Law Derivation
consists of a collection of disjoint subspaces, and a point
r ∈ W. Then let W r ⊂ W be the disjoint subspace such as Theorem 1: Consider a robot with a sensing pattern of a
r ∈ Wr. circular sector with field of view angle ψi and range Ri ,
In×m and On×m denote the n×m identity and zero matrix governed by kinodynamics (2). If pit ∈
/ Ari , the control law
respectively. Let IN denote the set of all natural numbers up that maximizes the objective function (3) is given by
to N , i.e. IN = {1, 2, . . . , N } while L[∂Ak ] denotes the   k


length of the boundary segment ∂Ak . u ∂p T
= f (p)φ(p) |p∈∂Sj,i
c ndp +
ω ∂xi
j=1 ∂S c
B. Problem Statement j,i

2
Let a path connected space, Ω ⊂ R be the unknown area m 1

∂p T
of interest. Let a swarm consisting of N mobile robots, where fij1 (υ)φ1ij (υ) |  υndυ +
∂ x̃ p∈∂Sj,i
xi = [ri , θi ]T , i ∈ IN be i-th robot’s state vector, where ri ∈ j=1 0
Ω and θi ∈ R be the position and orientation respectively. 2 1

∂p T
For each robot, a goal position pit ∈ Ω is assigned. The fij2 (υ)φ2ij (υ) |p∈∂Sj,i
v υndυ . (5)
robots are equipped with a range sensor of circular sector j=1 0
∂ x̃
pattern Cs (ri ; θi , ψi ), with a sensing limit Ri and a field of
view angle ψi , centered around its current heading, defined where fij1 = f (aij +υ(bij −aij )), φ1ij = φ(aij +υ(bij −aij )),
as j ∈ Im , i ∈ IN , fij2 = f (ri +υ(cij −ri )), φ2ij = φ(ri +υ(cij −
 
ri )), j ∈ I2 , i ∈ IN and
 
 − tan−1 (θi + ψ2i ) 1 0  
Cs (ri ; θi , ψi ) = ri + p≤ . ∂p 1 0 −Ri sin(ϕij + θi )
tan−1 (θi − ψ2i ) −1 0 |p∈∂Sj,i
c = , (6)
∂xi 0 1 Ri cos(ϕij + θi )
At any time instance, a sector visibility subspace ∂p  
b −a 
|  = − riji −aijij υI2×2 O2×1 , (7)
∂xi p∈∂Sj,i
Si (t) = Ω v (ri ; Ri ) ∩ Cs (ri ; θi , ψi )  
∂p 0 0 −υri − cj,i  sin(ϕij + θi )
|p∈∂Sj,i
v = .(8)
created by the range sensor of each robot is defined which ∂xi 0 0 υri − cj,i  cos(ϕij + θi )
describes the instantaneous sensed area of the robot at time
instant t. The robots are able to share their instantaneous
sensed area Si (t), creating the cumulative sensed area of the The full proof of the theorem is ommited and instead a
swarm A, which is defined as brief outline of the proof is given in the following section.

697

Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on June 12,2020 at 05:41:54 UTC from IEEE Xplore. Restrictions apply.
 By differentiating (3) with respect to xi = [ri , θi ]T and the navigation function is defined over a subspace of Ari ,
using the Leibniz integral rule [24] we obtain ArRi defined as

∂H ∂p T ArRi = Ari  R (12)
= fφ ndp , (9)
∂xi ∂xi R = {q ∈ R2 : q ≤ ρ} ,
∂Si

where n is the outward unit normal vector to ∂Si . where ρ > 0 is a safety distance.
Boundary ∂Si can be decomposed into four individual B. Frontier Cost, Performance and Weighting Functions
disjoint collections of segments: a) a collection of l-segments Selection
o
that belong to visible area boundaries {∂Sl,i } ⊆ Ω , b) a
c With the calculation of the control law, functions f (p)
collection of k-circular arcs {∂Sk,i } created by the limited
 and φ(p) can be defined. The overall scheme is based on a
visibility range, c) a collection of m-line segments {∂Sm,i }
frontier based exploration process and for this reason initially
created by visibility constraints denoted as {aim , bim }, and
v the frontier cost function must be introduced.
d) two line segments {∂S2,i } created by the limited field of
Boundary ∂A is decomposed into the: a)part of the sensed
view of the sensor, denoted as {rI , ci2 }.
area boundaries, {∂Aol } ⊆ ∂Ω , and b) free boundaries
From the above, ∂Si can be written as
{∂Afk } that contain segments created by visibility constraints
l
 k
 m
 2

o c  v
or field of view limits.
∂Si = ∂Sj,i + ∂Sj,i + ∂Sj,i + ∂Sj,i . (10) In the case of a single robot, the frontier selection scheme
j=1 j=1 j=1 j=1
would involve: a) the proximity of the frontier to the target, b)
Equation (9) is thus transformed to the proximity of the robot to the frontier, and c) the accessi-
  bility to new unexplored areas. In the collaborative mapping
∂H 
l
∂p T  k
∂p T
= fφ ndp + fφ ndp + case however, it should be noted that during exploration,
∂xi j=1
∂xi j=1
∂xi the i-th robot’s target may be discovered at some point by
o
∂Sj,i c
∂Sj,i

m   another robot, and thus for the target will that pit ∈ A \ Ari .
 ∂p T 2
∂p T
fφ ndp + fφ ndp . (11) For this reason, two distinct cases are considered, and for
j=1
∂xi j=1
∂xi each case an appropriate frontier selection cost function is
 v
∂Sj,i
∂Sj,i
constructed for frontier exploration selection.
Finally, calculating for each term of (11) the Jacobian The complimentary unexplored space W is defined as
matrix ∂p/∂xi = [∂p/∂ri ∂p/∂θi ], by considering that ⎛ ⎞
we are dealing with a static environment and calculating the  2  
W = R \A ⎝ ∂Afj ⎠ ,
relative movement of an arbitrary point in the remaining free
j
boundaries with respect to robot position and orientation,
results in that consists of a collection of simply connected disjoint
 
k subspaces.
u ∂p T Initially, the case where pit ∈ W is considered. In this case
= f (p)φ(p) |p∈∂Sj,i
c ndp +
ω ∂xi the frontier search is limited to the frontiers that are bound-
j=1 c
∂Sj,i i
aries of the disjoint subspace W pt . After the selection of
m

1
W , and limiting the search to that of {∂Arki ,f } ⊆ {∂Afk }
pit
∂p T
fij1 (υ)φ1ij (υ) |  υndυ +
i
that belong both in Ari and W pt , the geodesic Hausdorff
∂xi p∈∂Sj,i r ,f
j=1 0 distance Hg (pit , ∂Aki ) of a frontier from the target can be

2 1 used, which relates with the distance a robot will need to
∂p T
fij2 (υ)φ2ij (υ) |p∈∂Sj,i
v υndυ . traverse in the unknown area to reach the corresponding
j=1 0
∂xi target. Furthermore in space Ari the geodesic Hausdorff
distance of the robot from a frontier Hg (ri , ∂Akri ,f ) is
where fij1 = f (aij +υ(bij −aij )), φ1ij = φ(aij +υ(bij −aij )), calculated, which estimates the cost of moving towards a
j ∈ Im , i ∈ IN , fij2 = f (ri +υ(cij −ri )), φ2ij = φ(ri +υ(cij − frontier. Lastly, frontier length is taken into account in the
ri )), j ∈ I2 , i ∈ IN and the respective Jacobians are given cost function, which takes in this case the following form
from (6), (7) and (8).
As mentioned in subsection II-B, this control input is   −1
applied to the i-th robot until the target area is within ∂Acri ,f = arg min wi1 L ∂Ajri ,f +
Ari , at which point the navigation function switches to j

that of equation (4) and a gradient descent method can wi2 Hg (pt , ∂Ajri ,f ) + wi3 Hg (ri , ∂Akri ,f ) (13)
,
construct the final path. It should be noted that while a
geodesic distance based navigation function such as that of where wij ∈ [0, 1], j = 1, 2, 3 are weights assigned to each
equation (4) effectively guides a robot in non-convex, path part of the cost function.
connected spaces, it could potentially lead the robot into In the case where pit ∈ A \ Ari , the robot will instead
i
collision with the area boundaries ∂Ω [13]. To avoid this, try and expand Ari in order to connect with Apt . For this

698

Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on June 12,2020 at 05:41:54 UTC from IEEE Xplore. Restrictions apply.
 pi ,f i
reason, initially the frontier ∂Ac t of Apt that is closest to Two robots were utilised, with the initial positions and
the free frontiers of Ari is found by utilising the following targets depicted in Fig. 2 with green dot and cross for the
equation first robot, and maroon dot and cross for the second robot
respectively. The first robot has a range sensor of R = 1.8m
pi ,f
∂Afp = arg min Hg (∂Aj t , ∂Akri ,f ) , and ψ = 0.87rad, while the second has a range sensor of
j R = 1.6m and ψ = 1.4rad. At each time instant, each
robot moves according to control law (5) with a maximum
and term Hg (pit , ∂Akri ,f ) in equation (13) is substituted by
pi ,f
translational velocity of ν = 0.2m/sec and angular velocity
term Hg (∂Ac t , ∂Akri ,f ), and the respective cost function of ω = 0.2rad/sec. The weights of equations (13)-(14) are
changes to selected as w11 = 1, w12 = 0.8 and w13 = 0.4 for the first
  −1 and w21 = 0.8, w22 = 0.9 and w33 = 0.1 for the second
∂Arci ,f = arg min wi1 L ∂Ajri ,f +
j robot respectively. Boundaries of Ai at each step are archived
using an OctoMap [25] method with a grid resolution of
pi ,f
wi2 Hg (∂Ac t , ∂Arj i ,f ) + wi3 Hg (ri , ∂Arj i ,f ) (14)
,
0.05m.
Equations (13)-(14) comprise the switching frontier selection In Fig. 3, the evolution of the navigation towards the target
cost function. area is seen, where for illustration purposes the boundaries
Performance function f (p) implicates the exploration pro- of the area Ω, ∂Ω are visualised with dashed black lines.
cess into the objective given by (3) and weighting function The collective sensed area is depicted with ’light blue’,
φ(p) implicates the navigation towards the desired position boundaries of the area ∂Ao are depicted with black, while
as mentioned in [23]. The performance function remains with red the frontiers are depicted, and blue depicts the
unchanged from the definition given as selected frontier for each robot given from equations (13)-
(14). As may be seen in Figs. 3(a) - (b) despite the limited
1 field of view of the sensor and the orientation of the robots,
f (p) = . (15)
Hg (p, ∂Acri ,f ) + 1 they are able to move efficiently in exploring their selected
frontiers. As may be seen in Figs. 3(c),Robot 2 discovers
The weighting function φ(p) is modified to accomodate the
p1t and Robot 1 is adjusted to this event by switching to
respective switching strategy in the frontier selection as
equation (14). In Fig. 3(d) - (e) the switching to the shortest
1 path towards target takes effect as the two areas become
φ(p) = , (16)
dg (y,
i ) + 1 connected the target of Robot 1 is within its subspace,
y = arg min Hg (
i , ∂Acri ,f ) . (17) p1t ∈ Ar1 . As seen in Fig. 3(f) the resulting paths for each
y∈Aci
r ,f robot are sufficiently far from the boundaries of the area ∂Ω
 to account for safe and fast navigation, without danger of
pit , if pit ∈ W

i = (18) collision, both during the exploration phase and during the
arg min pi
t ,f
dg (q, pit ) . geodesic based navigation function phase.
q∈∂Ac

It must be noted that dg (y,

i ) refers to space W. This
selection gives greater importance in neighbourhoods of
∂Acri ,f that are closer to
i than neighbourhoods further In this paper a novel method of collaborative mapping
away from it.
IV. S IMULATION STUDIES
The efficiency of the proposed scheme is verified through
a simulation scenario. An area for navigation was created that
is depicted in Fig. 2, in which the rectangle encapsulating
the convex hull of Ω is of 14 m × 12 m.
Fig. 2: Ω-sample area for navigation
V. C ONCLUSIONS
and navigation in unknown environments by a swarm of
mobile robots is presented. A swarm of mobile robots is
considered, with each robot equipped with a ranged sensor of
limited field of view and range. A specific target is assigned
for each robot that needs to be discovered in the unknown
area, while the sensorial information about the environment,
collected from each robot is exchanged to create a collective
map of the explored space. By utilization of a switching
objective function that initially aims to discover each robot’s
target area, a frontier exploration method occurs. The frontier
is selected by utilising a switching cost function that takes
into consideration the possible discovery of the target area
of a robot by another member of the swarm. As soon as
the target area of a robot is inside its’ disjoint explored
subspace, the navigation function switches to a geodesic
based function that ensures navigation to the explored area.
Simulation results prove that the proposed control scheme
navigates efficiently the swarm towards the respective target
positions.

699

Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on June 12,2020 at 05:41:54 UTC from IEEE Xplore. Restrictions apply.
 (a)
(d)
Fig. 3: Evolution of the robot navigation towards the target location with respect to the actual area
R EFERENCES
[1] E. Garcia, M. Jimenez, P. De Santos, and M. Armada, “The evolution
of robotics research,” Robotics Automation Magazine, IEEE, vol. 14,
no. 1, pp. 90–103, March 2007.
[2] Y. Stergiopoulos, M. Thanou, and A. Tzes, “Distributed collaborative
coverage-control schemes for non-convex domains,” IEEE Transac-
tions on Automatic Control, vol. 60, no. 9, pp. 2422–2427, 2015.
[3] Y. Kantaros, M. Thanou, and A. Tzes, “Distributed coverage control
for concave areas by a heterogeneous robot-swarm with visibility
sensing constraints,” Automatica, vol. 53, pp. 195–207, 2015.
[4] Y. Stergiopoulos and A. Tzes, “Spatially distributed area coverage op-
timisation in mobile robotic networks with arbitrary convex anisotropic
patterns,” Automatica, vol. 49, no. 1, pp. 232–237, 2013.
[5] S. M. LaValle, Planning Algorithms. Cambridge, U.K.: Cambridge
University Press, 2006.
[6] O. Khatib, “Real-time obstacle avoidance for manipulators and mobile
robots,” in IEEE International Conference on Robotics and Automa-
tion, vol. 2, Mar 1985, pp. 500 – 505.
[7] J. Borenstein and Y. Koren, “The vector field histogram-fast obstacle
avoidance for mobile robots,” IEEE Transactions on Robotics and
Automation, vol. 7, no. 3, pp. 278–288, June 1991.
[8] L. Kavraki, P. Svestka, J.-C. Latombe, and M. Overmars, “Probabilistic
roadmaps for path planning in high-dimensional configuration spaces,”
IEEE Transactions on Robotics and Automation, vol. 12, no. 4, pp.
566–580, Aug 1996.
[9] J. J. Kuffner and S. LaValle, “Rrt-connect: An efficient approach
to single-query path planning,” in IEEE International Conference on
Robotics and Automation, vol. 2, 2000, pp. 995–1001.
[10] N. Du Toit and J. Burdick, “Robot motion planning in dynamic,
uncertain environments,” IEEE Transactions on Robotics, vol. 28,
no. 1, pp. 101–115, Feb 2012.
[11] B. Tovar, R. Murrieta-Cid, and S. LaValle, “Distance-optimal navi-
gation in an unknown environment without sensing distances,” IEEE
Transactions on Robotics, vol. 23, no. 3, pp. 506–518, June 2007.
[12] O. Brock and O. Khatib, “High-speed navigation using the global
dynamic window approach,” in IEEE International Conference on
Robotics and Automation, vol. 1, 1999, pp. 341–346 vol.1.
[13] A. Valero-Gomez, J. V. Gomez, S. Garrido, and L. Moreno, “The path
to efficiency: Fast marching method for safer, more efficient mobile
robot trajectories,” IEEE Robotics Automation Magazine, vol. 20,
no. 4, pp. 111–120, Dec 2013.
700
Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on June 12,2020 at 05:41:54 UTC from IEEE Xplore. Restrictions apply.
(b) (c)
(e) (f)
[14] B. Yamauchi, “A frontier-based approach for autonomous exploration,”
in IEEE International Symposium on Computational Intelligence in
Robotics and Automation, Jul 1997, pp. 146–151.
[15] A. Haumann, K. Listmann, and V. Willert, “Discoverage: A new
paradigm for multi-robot exploration,” in IEEE International Confer-
ence on Robotics and Automation (ICRA), May 2010, pp. 929–934.
[16] D. Haumann, A. Breitenmoser, V. Willert, K. Listmann, and R. Sieg-
wart, “Discoverage for non-convex environments with arbitrary obsta-
cles,” in IEEE International Conference on Robotics and Automation
(ICRA), May 2011, pp. 4486–4491.
[17] E. Şahin, “Swarm robotics: From sources of inspiration to domains of
application,” in International workshop on swarm robotics. Springer,
2004, pp. 10–20.
[18] S. Bhattacharya, R. Ghrist, and V. Kumar, “Multi-robot coverage and
exploration on riemannian manifolds with boundaries,” The Interna-
tional Journal of Robotics Research, vol. 33, no. 1, pp. 113–137, 2014.
[19] D. Haumann, V. Willert, and K. Listmann, “Discoverage: From
coverage to distributed multi-robot exploration*,” IFAC Proceedings
Volumes, vol. 46, no. 27, pp. 328 – 335, 2013.
[20] W. Burgard, M. Moors, C. Stachniss, and F. E. Schneider, “Co-
ordinated multi-robot exploration,” IEEE Transactions on Robotics,
vol. 21, no. 3, pp. 376–386, June 2005.
[21] K. M. Wurm, C. Stachniss, and W. Burgard, “Coordinated multi-robot
exploration using a segmentation of the environment,” in IEEE/RSJ
International Conference on Intelligent Robots and Systems. IEEE,
2008, pp. 1160–1165.
[22] L. Bayindir and E. Şahin, “A review of studies in swarm robotics,”
Turkish Journal of Electrical Engineering & Computer Sciences,
vol. 15, no. 2, pp. 115–147, 2007.
[23] I. Arvanitakis, K. Giannousakis, and A. Tzes, “Mobile robot navigation
in unknown environment based on exploration principles,” in 2016
IEEE Conference on Control Applications (CCA), Sept 2016, pp. 493–
498.
[24] H. Flanders, “Differentiation under the integral sign,” American Math-
ematical Monthly, vol. 80, no. 6, pp. 615–627, 1973.
[25] A. Hornung, K. M. Wurm, M. Bennewitz, C. Stachniss, and W. Bur-
gard, “Octomap: an efficient probabilistic 3d mapping framework
based on octrees,” Autonomous Robots, vol. 34, no. 3, pp. 189–206,
2013.