A Path Planning Technique For Autonomous Mobile Robot Using Free-Configuration Eigenspaces

Shyba Zaheer & Tauseef Gulrez
International Journal of Robotics and Automation (IJRA), Volume (6) : Issue (1) : 2015 14
A Path Planning Technique For Autonomous Mobile Robot
Using Free-Configuration Eigenspaces
Shyba Zaheer s.shyba@gmail.com
Department of Electrical & Electronics Engineering
T.K.M. College of Engineering
Kerala, India
Tauseef Gulrez gtauseef@ieee.org
Virtual and Simulations of Reality (ViSOR) Lab,
Department of Computing, Macqaurie University
2109 NSW, Sydney, Australia.
Abstract
This paper presents the implementation of a novel technique for sensor based path planning of
autonomous mobile robots. The proposed method is based on finding free-configuration eigen
spaces (FCE) in the robot actuation area. Using the FCE technique to find optimal paths for
autonomous mobile robots, the underlying hypothesis is that in the low-dimensional manifolds of
laser scanning data, there lies an eigenvector which corresponds to the free-configuration space
of the higher order geometric representation of the environment. The vectorial combination of all
these eigenvectors at discrete time scan frames manifests a trajectory, whose sum can be
treated as a robot path or trajectory. The proposed algorithm was tested on two different test bed
data, real data obtained from Navlab SLAMMOT and data obtained from the real-time robotics
simulation program Player/Stage. Performance analysis of FCE technique was done with existing
four path planning algorithms under certain working parameters, namely computation time
needed to find a solution, the distance travelled and the amount of turning required by the
autonomous mobile robot. This study will enable readers to identify the suitability of path planning
algorithm under the working parameters, which needed to be optimized. All the techniques were
tested in the real-time robotic software Player/Stage. Further analysis was done using MATLAB
mathematical computation software.
Keywords: Free-configuration Space, Eigenvector, Motion Planning, Trajectory Planning.
.
1. INTRODUCTION
Motion planning is one of the most important tasks in intelligent control of an autonomous mobile
robot (AMR). It is often decomposed into path-planning and trajectory planning. Path planning is
referred as to generate a collision free path in an environment with obstacles. Whereas, trajectory
planning schedule the movement of a mobile robot along the planned path. Based on the
availability of information about environment, the path-planning algorithms are divided into two
categories, namely offline and online. Offline path planning of robots in environments uses
complete information about stationary obstacles and trajectory of moving obstacles, which are
known in advance. This method is also known as global path planning. When complete
information about environment is not available in advance, the mobile robot gets information
through sensors, as it moves through the environment. This is known as online or local path
planning. Essentially, online path planning begins its initial path offline but switches to online
mode when it discovers new changes in obstacle scenario. Classical approaches used in online
path planning are Potential Filed approach (PF), collision–cone approach, and vector field
histogram (VFH) method. Khatib [1] proposed the Artificial Potential Field (APF) approach which
is popular in mobile robotics. This approach is known for its mathematical elegance and simplicity

as the path is found with very little computation. However, the drawback of this algorithm is that
the robot may become stagnant or trapped when there is a cancellation of equal magnitudes of
attractive and repulsive forces. Till date many variants of the potential field approach like escape-
force algorithm [2], trap recovery model, adaptive virtual target algorithm etc. have been
proposed. Path planning problems can also be solved by VFH approach [3]. At every instant, a
polar histogram is generated to represent the polar density of obstacles around a robot. The
robot’s steering direction is chosen based on the least polar density and closeness to the goal. In
a given environment, the polar histogram must be regularly regenerated for every instant and
hence this method is suited for environments with sparse moving obstacles. Another commonly
used online approach is based on the collision cone concept [4]. The Collision of a robot can be
averted if the relative velocity of it with respect to a particular obstacle falls exterior to the collision
cone. Another online approach for obstacle avoidance is dynamic windows approach [5]. The
dynamic window contains the feasible linear and angular velocities taking into consideration the
acceleration capabilities of a robot. Then the velocity at the next instant is optimized for obstacle
avoidance subject to vehicle dynamics. With classical techniques, the optimum result requires
more computational time due to incomplete information of the environment. This classical
approach can be combined with heuristic approaches like genetic algorithm (GA) and particle
swarm optimization (PSO) [6]. Another class of online path planning algorithms are sampling
based path planning algorithms like rapidly evolving random trees (RRT) and probabilistic
roadmap methods PRM [7]. The idea of connecting points sampled randomly from the state
space is essential in both RRT and PRM approaches.
A limitation of the classical planning algorithm is that a complete model of the environment is
needed before the planner can proceed. A solution to this problem is a sensor based path
planner. Sensor based planner allows robots to work autonomously in unknown environments.
Specifically the robot is able to move to a given configuration without prior knowledge of the
environment. The motion of the robot is generated step by step while more and more knowledge
about the environment is accumulated incrementally. A great variety of sensors that can be used
for robots includes tactile sensor, vision sensor, sonar sensor etc. A well-known sensor based
technique is “Bug” family. These path planners incorporates sensing and path planning as an
integrated solution. Bug algorithms assume only local knowledge of the environment and a global
goal. The most commonly used sensor based robot path planners are Bug1, Bug2, TangentBug,
DistBug, and VisBug [8] .Bug 1 and Bug 2 uses tactile sensors while tangent bug, and Distbug
uses range sensors. These techniques require its own position by using odometry, goal position
and range sensor. In this paper, a novel sensor based path planning technique is proposed,
namely, free-configuration eigenspace (FCE). This approach tends to find principal components
(Eigenvectors) spanning the low dimensional space (Eigenspace) of high order scanning data.
Integrating the highest eigenvector in time steps can produce a collision free trajectory.
In this paper, we have compared our proposed FCE path planning approach with other well-
known path planning techniques. Also, we have analyzed the path produced by any path planner
based on the following parameters. Path Length: distance of the path from start to finish.
Computation time: algorithm’s total execution time excluding time spent during driving (i.e. from
start to goal). Turning: the amount of turning which is performed along the path from start to
finish. Memory requirements: the amount of global memory reserved by the algorithm. For good
path planner it is assumed that all these parameter should be as small as possible.
This paper is organized as follows, in section 2, the Problem formulation and proposed technique
is presented. In Section 3 the materials and methods used are explained. Section 3.3 describes
the detection of eigensapce and trajectory generation algorithms. Section 4 explains about the
experimental set up and implementation of the FCE algorithm. In section 5 a detailed
performance analysis done with existing four planning algorithms namely, APF, A*, RRT and
PRM and section 6 with conclusion.

International Journal of Robotics and
2. PROBLEM FORMULATION
2.1 Free-Configuration Space
Finding a collision free path problem can be formulated as a search in a confi
Let A be a robot, moving in a Euclidean space
be fixed rigid bodies distributed in
configuration of A is a specification of the position of every point in
WF is a Cartesian coordinate system. The configuration space of
with all possible configurations of
by A(q). A path from an initial configuration
map with [ ] C→1,0:τ )1(τ
of obstacles denoted by BB ,, 21
{)( =i qBC
Which is called as obstacleC . The union of all
Ufree CC −=
A collision free path between two configurations is any continuous path
FIGURE 1: Configuration Space 2D Model
arrows inside the scan area are the eigenvectors and the red arrows outside the laser scan region show the
2.2 Proposed FCE Technique
To obtain a collision free path, we have utilized
Eigenspaces” which explains the underlying
represented by an eigenvector which corresponds to th
vectorial combination of all these eigen
trajectory. To realize the above
critical hypotheses:
Robotics and Automation (IJRA), Volume (6) : Issue (1) : 2015
PROBLEM FORMULATION
pace Concept
Finding a collision free path problem can be formulated as a search in a configuration space [9
Euclidean space.
N
RW = , where N = 2 or 3. Let
be fixed rigid bodies distributed in W are called as obstacles and are the closed subsets of
is a specification of the position of every point in A with respect to
is a Cartesian coordinate system. The configuration space of A is the space denoted by C,
with all possible configurations of A. The subset of W occupied by A at configuration
A path from an initial configuration initq to a goal configuration goalq is a continuous
goalq= and initq=)0(τ . The workspace contains a finite number
nB,,K . Each obstacle maps in C to a region:
}0)(| ≠∩∈ iBqAC (1)
. The union of all obstacleC is the region U
n
i iBC1
)(=
and the set
U
n
i iBC1
)(= (2)
A collision free path between two configurations is any continuous path [ ]→1,0:τ
Configuration Space 2D Model - The blue line represents the laser scan area, whereas the red
obstacle region.
FCE Technique
ollision free path, we have utilized our paradigm [13] of “Free-C
” which explains the underlying low-dimensional manifolds of laser scanning data,
an eigenvector which corresponds to the free-configuration space and
combination of all these eigenvectors at discrete time scan frames manifests a
To realize the above methodology of path planning, we have set forth the basis of two
16
guration space [9]:
nBBB ,,, 21 K
subsets of W. A
with respect to WF , where
is the space denoted by C,
at configuration q is denoted
is a continuous
The workspace contains a finite number
and the set
(2)
freeC
The blue line represents the laser scan area, whereas the red
Configuration
dimensional manifolds of laser scanning data,
configuration space and the
frames manifests a
we have set forth the basis of two

1. H1: In the exploration process t
paths which lie in that space can form an exploratory
2. H2: The free-configuration space path shows a distinct pattern in the sensor data (surrounded
by obstacles) and can be learnt to pr
According to the hypothesis:
• The single vector formulation of the trajectory point can be described as
free
C I
♦ Consequently th
3. MATERIALS AND METHODS
3.1 Laser Sensor Model
Range sensing is a crucial element of any obstacle avoidance system.
obstacle detection are 2-D LADAR (i.e., a laser that scans in one plane). 2
widely used sensor for obstacle detection
measurement of time-of-flight (TOF)
by its range and bearing, with the resolution of
range of 5 meters. Where the center position
terms of the world coordinates frame. The incoming data are numbered as
180,,1,0 K=j are the distance value from the central position of the laser range finder to the
object. The object position (Obj
position of the laser scanner to the object. The object position (
following where dr is the distance from the centre of the SICK
cθ is the angle of the mobile robot measured counter clockwise from the positive x
+= xx AMRObj
=y AMRyObj
: In the exploration process there lies a free-configuration space, such that the discrete
paths which lie in that space can form an exploratory-trajectory
configuration space path shows a distinct pattern in the sensor data (surrounded
by obstacles) and can be learnt to produce a better quality exploratory map.
The single vector formulation of the trajectory point can be described as
)
max
(
λ
E
free
C
ML
M =I
Consequently the vectorial sum of all the free-configuration eigen
AND METHODS
Range sensing is a crucial element of any obstacle avoidance system. Sensors suitable for
D LADAR (i.e., a laser that scans in one plane). 2-D laser scanners are
d sensor for obstacle detection. The SICK LMS is a laser scanner based on the
flight (TOF) as shown in Fig.2 (a). The Laser Cartesian space is defined
by its range and bearing, with the resolution of
0
5.0 , having
0
180 of frontal AMR scan with the
rs. Where the center position of the AMR is represented as (AMR
world coordinates frame. The incoming data are numbered as
are the distance value from the central position of the laser range finder to the
yx ObjObj , ) is determined by measuring distance from the central
position of the laser scanner to the object. The object position ( yx ObjObj , ) is determined by the
is the distance from the centre of the SICK LMS to that of the mobile robot and
s the angle of the mobile robot measured counter clockwise from the positive x-






−
Π
++
Π
+ cdcj r
j
r θθ cos
2180
cos
(5)






−
Π
+−
Π
− cdcj r
j
r θθ sin
2180
sin
(6)
17
pace, such that the discrete
configuration space path shows a distinct pattern in the sensor data (surrounded
The single vector formulation of the trajectory point can be described as;
(3)
eigenspaces;
(4)
Sensors suitable for
D laser scanners are
The SICK LMS is a laser scanner based on the
. The Laser Cartesian space is defined
of frontal AMR scan with the
), Yx AMRAMR in
world coordinates frame. The incoming data are numbered as jr , where
are the distance value from the central position of the laser range finder to the
from the central
) is determined by the
LMS to that of the mobile robot and
-axis.
(5)
(6)

(a)
FIGURE 2: (a)The Pioneer 2AT autonomous mobile robot
3.2 Dimensionality Reduction
Using the SICK-laser a large number of point feature
correspond to the same environmenta
to extract higher-order features
data patterns. Principal Components Analysis (PCA) is a powerful
bases functions that can be linearly combined to represent
been used extensively to cluster sets of point features in a map
represent each cluster of point features, generat
representation using just the four parameters of the endpoints of 2
the highest eigenvector is to be found by applying PCA to the sensor data which can be
integrated in discrete time scan to get a trajectory.
3.3 Eigenspace Detection Method
It is possible to produce a trajectory as vectorial combination of highest eigenvectors in discrete
time scan frame of sensor scanning data.
this direction shown by the eigen
vehicle to pass through. Hence it can be ascertaine
by the dimensionality reduction technique, with reference to the vehicle pose at that point along
with the sensor data. The pose of the mobile robot is denoted by the position and orientation (x, y,
θ) in the map. In the beginning we compute the
which is commonly used for constructing range data maps in robotics.
map, sensor’s position and robot’s orientation, while
accurately. The algorithm for eigen
evident that the highest eigenvector
the vehicle can easily manure without hitting the obstacle and hence this space
as free-configuration eigenspace
(b)
autonomous mobile robot (AMR) carrying an onboard SICK laser
(b) A single laser scan plot output.
Reduction Technique
laser a large number of point features can be obtained and
the same environmental structure. The dimensionality reduction technique enables
order features in lower dimensional manifold representations which capture the
data patterns. Principal Components Analysis (PCA) is a powerful tool that com
s functions that can be linearly combined to represent a collection of data [1
been used extensively to cluster sets of point features in a map. We can extract line segments to
represent each cluster of point features, generating very efficient representations, e.g., a
using just the four parameters of the endpoints of 2-d line segment
vector is to be found by applying PCA to the sensor data which can be
scan to get a trajectory.
ethod
frame of sensor scanning data. But to ensure the easy manoeuvring of the robot along
is direction shown by the eigenvector, one must make sure that there is enough space
vehicle to pass through. Hence it can be ascertained by plotting the all the eigenvector
) in the map. In the beginning we compute the visibility model, which is the data2D laser
commonly used for constructing range data maps in robotics. In order to construct such
nsor’s position and robot’s orientation, while obtaining laser scans should be estimated
The algorithm for eigenspace detection is described below. From the Fig.
eigenvector shows the direction of maximum free space through which
the vehicle can easily manure without hitting the obstacle and hence this space w
space.
18
SICK laser scanner
and all of them
reduction technique enables
which capture the
tool that computes a set of
a collection of data [16]. PCA has
e can extract line segments to
esentations, e.g., a
d line segment. In this case,
vector is to be found by applying PCA to the sensor data which can be
the robot along
vector, one must make sure that there is enough space for the
vectors obtained
visibility model, which is the data2D laser sensor,
In order to construct such
should be estimated
space detection is described below. From the Fig. 3 it is
shows the direction of maximum free space through which
e have named

Algorithm 1: Finding Free-Configuration Space
Input: The Pose data (x, y, theta),
Output: Eigenspace (Eigenvectors)
Begin
Initialize Pose (x, y, theta), n, p
M = [ ]
for Time = 1 to n
for angle =-90 : 0.05: 90
M(time, angle) = d(Time, angle )
end
end
Calculate the Covariance of M
Calculate the Eigenvalues (V) and Vectors
for i = 0,1,…,m
for angle =-90 : 0.5:90
sin(
cos(
max
max
EE
EE
y
x
×=
×=
λλ
λλ
end
end
Plot the eigenvectors from
End
(a)
FIGURE 3: (a) Principle components extracted from the laser data, the green vectors are the smaller
eigenvectors, whereas the red vector is the highest principal component,
3.4 Autonomous Robot Trajectory
According to the hypotheses in low
eigenvector which will represent
highest eigenvector in discrete time can be integrated to get an exploratory trajectory .The
Configuration Space
The Pose data (x, y, theta), sensor scanning data (matrix)
vectors)
Initialize Pose (x, y, theta), n, p
90
Calculate the Eigenvalues (V) and Vectors (E) of the Covariance of M
90 : 0.5:90
)sin(
)cos(
angle
angle
Plot the eigenvectors from the corresponding vehicle pose,
(a) (b)
components extracted from the laser data, the green vectors are the smaller
eigenvectors, whereas the red vector is the highest principal component, (b) the RED Vector
towards maximum the free area.
rajectory Generation Method
hypotheses in low-dimensional manifolds of laser scanning data, there lies an
represent free area or obstacle area. Assuming this to be free area, the
vector in discrete time can be integrated to get an exploratory trajectory .The
19
components extracted from the laser data, the green vectors are the smaller
(b) the RED Vector indicating
dimensional manifolds of laser scanning data, there lies an
be free area, the
vector in discrete time can be integrated to get an exploratory trajectory .The

following algorithm describes trajectory formation using eigen
trajectory and the actual trajectory with laser scans
Algorithm 2: Trajectory connectivity in Free
Input: Robot Pose data, sensor
Output: Highest Eigenvector Sensor data in discrete time scan
Begin
Initialise Pose (x, y, theta), n, p
M = [ ]
for Time = 1 to n
for angle =-90 : .05:90
end
end
Sort the diagonal of E in descending order
Taking the Maximum of E and its corresponding V
Plot E and V from the initial Pose(x,y,
sin(
cos(
max
max
angleEE
EE
y
x
×=
×=
λλ
λλ
Plot the eigenvectors from the corresponding vehicle pose ,Pose(x,y,theta)
End
FIGURE 4: Green & red lines showing the trajectory formed
4. EXPERIMENTAL SETUP and
4.1 Player/Stage Data Simulation
The proposed technique was applied to the robotic
system (RTOS) Ubuntu 8.04 Hardy
RAM. The environment was created in
robotic software, Player/Stage as shown in Fig
that contain positions of environment floor plan, a pioneer 2DX autonomous mobile robot, laser
sensor, odometry sensor. The mobile robot is defined as a non
can move around in the environment and can sense the obstacle by measuring the laser distance
rajectory formation using eigenvectors. The obtained eigen
jectory and the actual trajectory with laser scans for Navlab test bed data are shown in Fig.4.
Trajectory connectivity in Free-Configuration Space
, sensor measurement data M
vector Sensor data in discrete time scan
Initialise Pose (x, y, theta), n, p
Sort the diagonal of E in descending order
Taking the Maximum of E and its corresponding V
V from the initial Pose(x,y,theta) as follows
)
)
angle
angle
Plot the eigenvectors from the corresponding vehicle pose ,Pose(x,y,theta)
red lines showing the trajectory formed by the Eigenvectors in discrete
laser data.
EXPERIMENTAL SETUP and SYSTEM DESCRIPTION
Data Simulation
applied to the robotic simulator running under real-
ystem (RTOS) Ubuntu 8.04 Hardy–heron on 2.0 GHz Intel dual core processor having 3 GB of
RAM. The environment was created in real time (distributed under GNU) 2D autonomous m
Stage as shown in Fig. 5. A configuration definition XML file was created
The mobile robot is defined as a non-holonomous (pioneer2DX) , which
environment and can sense the obstacle by measuring the laser distance
20
vectors. The obtained eigenvector
are shown in Fig.4.
e time scan of
-time operating
heron on 2.0 GHz Intel dual core processor having 3 GB of
time (distributed under GNU) 2D autonomous mobile
nfiguration definition XML file was created
holonomous (pioneer2DX) , which
environment and can sense the obstacle by measuring the laser distance

from its Centre point of gravity. TCP/IP sockets are used to for the communication between the
mobile robotic agent and the robotic software server. For experimental purposes, the mobil
robot’s Laser data and the odometer readings was recorded using player/stage data commands,
processed/ reduced online as shown in the Fig.
was implemented in MATLAB (product of Mathworks Inc.)
segments and each segment’s consecutive laser sensor readings were taken. The latent values
were found for these chunks of laser data using PCA analysis. T
eigenvector) were integrated in time and the
robot and environment is shown in the Fig
eigenvector trajectory in blue
(a)
(c)
FIGURE 5: (a) 2Dview of the environment with the laser mounted autonomous mobile robot. (b) Simulated
laser plot. (c) Graphical User Interface client program “playerv” that visualizes laser data from a player
server, (c) The Actual tr
4.2 Real Time Data Simulation
The scenario of Navlab SLAMMOT Datasets, Carnegie Mellon University USA as s
in yellow Loop-2. The laser data for the entire loop was divided in to four segments and within
each segment consecutive laser sensor readings was taken. The latent value
PCA analysis and the latent values (highest eigen
trajectory was obtained .This new trajectory and the actual trajectory was compared and an error
graph was plotted as shown in Fig. 6 (c , d, e, g) . It was
point of gravity. TCP/IP sockets are used to for the communication between the
mobile robotic agent and the robotic software server. For experimental purposes, the mobil
processed/ reduced online as shown in the Fig. 5.The proposed trajectory detection methodology
(product of Mathworks Inc.). The actual trajectory was divided in
and each segment’s consecutive laser sensor readings were taken. The latent values
were found for these chunks of laser data using PCA analysis. The latent values (highest
vector) were integrated in time and the new trajectory was obtained. The scenario of the
robot and environment is shown in the Figure.5(c) and the actual trajectory (in red) and the
(b)
(d)
(a) 2Dview of the environment with the laser mounted autonomous mobile robot. (b) Simulated
server, (c) The Actual trajectory & learned trajectory (red).
imulation
SLAMMOT Datasets, Carnegie Mellon University USA as s
The laser data for the entire loop was divided in to four segments and within
onsecutive laser sensor readings was taken. The latent values were found using
e latent values (highest eigenvector)were integrated in time and the new
as shown in Fig. 6 (c , d, e, g) . It was concluded from the Fig. 6(b) that at
21
point of gravity. TCP/IP sockets are used to for the communication between the
mobile robotic agent and the robotic software server. For experimental purposes, the mobile
.The proposed trajectory detection methodology
l trajectory was divided into
and each segment’s consecutive laser sensor readings were taken. The latent values
he latent values (highest
new trajectory was obtained. The scenario of the
(c) and the actual trajectory (in red) and the
(a) 2Dview of the environment with the laser mounted autonomous mobile robot. (b) Simulated
SLAMMOT Datasets, Carnegie Mellon University USA as shown in Fig. 6
The laser data for the entire loop was divided in to four segments and within
s were found using
vector)were integrated in time and the new
6(b) that at the

corners, the direction of eigenvector obtained deviated from the actual trajectory where as in a
straight line segment, the eigenvector
(a)
(c)
(e ) (f)
FIGURE 6: (a)The Aerial Photo is downloaded from and the copyrighted property of GlobeXplorer, LLC.
(“GlobeXplorer”). The trajectory (Loop
obtained after applying PCA to the laser data
segments vs laser-data chunks and their error bounds for PCA with real
4.3 Experimental Result Analysis
We have tested the developed
result plots for the real time data
constructed eigenvector trajectory (in red) is perfectly matching
actual trajectory. But, failed to produce goo
vector obtained deviated from the actual trajectory where as in a
eigenvector is very much close to the actual trajectory.
(b)
(d)
(e ) (f)
(a)The Aerial Photo is downloaded from and the copyrighted property of GlobeXplorer, LLC.
(“GlobeXplorer”). The trajectory (Loop-2) in yellow, was used for laser data-set collection. (b)The trajectory
obtained after applying PCA to the laser data-set. (c).(d),(e) (f)error-plots show the respective trajectory
data chunks and their error bounds for PCA with real-trajectory
esult Analysis
developed algorithm on real time and simulated laser data set.
result plots for the real time data, as shown in Fig. 6(b) & Fig. 6(d), it is evident tha
vector trajectory (in red) is perfectly matching on the longer segments of
failed to produce good results on the corners. In order to get
22
vector obtained deviated from the actual trajectory where as in a
(a)The Aerial Photo is downloaded from and the copyrighted property of GlobeXplorer, LLC.
set collection. (b)The trajectory
plots show the respective trajectory
trajectory.
and simulated laser data set. From the
it is evident that the
on the longer segments of the
a clear picture

about the corner misalignment of
2D map of the laser data. It is also evident from Fig
deviating away from the actual trajectory of the robot.
FIGURE 7: The 2D map obtained from plotting the outliers of laser data and the
5. PERFOMANCE ANALYSIS
The path produced by any path planner can be
Length: distance of the path from start to finish.
time excluding time spent driving.
path from start to finish. Memory requirements
algorithm. The following section
planners APF, RRT, PRM and
scenarios Room1 and Room2 and found the estimated values for all the
consideration. The analysis was done using
below tables summaries the results obtained
(a)
about the corner misalignment of the trajectory, we plotted the Navlab data trajectories with
It is also evident from Fig. 7, at the corners the eigenvector
deviating away from the actual trajectory of the robot.
The 2D map obtained from plotting the outliers of laser data and the actual trajectory
the eigenvector trajectory(blue).
PERFOMANCE ANALYSIS
The path produced by any path planner can be analysed on the following parameters.
: distance of the path from start to finish. Computation time: algorithm’s total execution
g time spent driving. Turning: the amount of turning which is performed along the
Memory requirements: the amount of global memory reserved by the
The following section illustrate the result performance analysis done
and A* with the proposed FCE method. We have considered
oom2 and found the estimated values for all the parameters under
. The analysis was done using MATLAB mathematical computation software
esults obtained:
(b)
23
trajectories with in the
eigenvector trajectory is
actual trajectory (black) and
on the following parameters. Path
algorithm’s total execution
is performed along the
: the amount of global memory reserved by the
done existing path
considered two
parameters under
MATLAB mathematical computation software. The

`
(c) (d)
(e) (f)
(g) (h)

(i) (
(k)
FIGURE 8: Trajectories obtained from different Path
of Room-2. c) Trajectories obtained from FCEalgorithm from scenario of Room
points trajectory, where as green is the FCE algorithm t
from scenario of Room-2. The red is normal way points trajectory, where as green is the FCE algorithm
trajecorty. e) Trajectory obtainedfrom APF method for scenario of Room
method forscenario of Room-2. g) Trajectory obtained from RRT method fromscenario of Room
Trajectory obtained from RRT method from scenario of Room
method from scenarioof Room-1. j) Smoothed trajecto
2. k) Trajectory obtained from A* method from scenario of Room
Source = [70 70]; in Y, X format
Goal = [400 400] in Y, X format
Technique Scenario
RRT Room1
PRM Room1
APF Room1
A* Room1
FCE Room1
) (j)
) (l)
Trajectories obtained from different Path-planning alogrithms. a) Scenario of Room
2. c) Trajectories obtained from FCEalgorithm from scenario of Room-1. The red is normal way
points trajectory, where as green is the FCE algorithm trajecorty. d) Trajectories obtained from FCE algorithm
2. The red is normal way points trajectory, where as green is the FCE algorithm
trajecorty. e) Trajectory obtainedfrom APF method for scenario of Room-1. f) Trajectory obtained fr
2. g) Trajectory obtained from RRT method fromscenario of Room
Trajectory obtained from RRT method from scenario of Room-2. i) Smoothed trajectory obtained from PRM
1. j) Smoothed trajectory obtained from PRM method from scenario of Room
method from scenario of Room-1.l) Trajectory obtained from A
scenario of Room-2.
[70 70]; in Y, X format.
[400 400] in Y, X format.
Scenario Path length(cm) Processing time Max-turn(cm
Room1 5.646315e+002 1.513718e+002 0.1412
Room1 4.491235e+002 1.848314e+001 -0.0019
Room1 5.889301e+002 1.846913e+000 3.00
Room1 4.740559e+002 4.430567e+001 0.1474
Room1 10.91235e+002 1.98314e+001 1.7471
TABLE 1: Result for Room1.
25
)
planning alogrithms. a) Scenario of Room-1. b) Scenario
1. The red is normal way
rajecorty. d) Trajectories obtained from FCE algorithm
2. The red is normal way points trajectory, where as green is the FCE algorithm
1. f) Trajectory obtained from APF
2. g) Trajectory obtained from RRT method fromscenario of Room-1. h)
2. i) Smoothed trajectory obtained from PRM
ry obtained from PRM method from scenario of Room-
1.l) Trajectory obtained from A* method from
turn(cm
-1)

Source = [140 260]; in Y, X format.
Goal = [260 140] in Y, X format.
Technique Scenario Path
length(cm)
Processing
time
Max-turn(cm
-1)
RRT Room2 3.347247e+000 3.012290e+000 0.0004
PRM Room2 2.806335e+002 3.273510e+000 0.0854
APF Room2 3.234583e+002 1.037137e+000 -0.0204
A* Room2 2.482843e+002 5.420330e+001 0.1574
FCE Room2 6.234583e+002 2.012290e+000 2.900
TABLE 2: Result for Room2.
5.1 Result Analysis
As shown in Table-1 & 2, performance analysis on different path planning algorithms shows that
the PRM technique performs better in terms of turning and path length. But it is probabilistic
complete. But RRT is faster as compared to PRM and produce fine path with minimum turning.
Even though A*shows an optimal path, the computational cost is high and the clearance space
from the obstacle is low. The APF algorithm requires less computational time but the path length
depends on the set value of the potential forces and it suffers from local-minima problem. In case
of FCE, the path length and turning value are comparatively larger than all other methods.A good
path is relatively short, keeps some clearance distance from the obstacles, and is smooth. Result
analysis shows APF and proposed FCE technique is better on this attributes.
6. CONCLUSIONS
This paper proposes a novel path planning problem of an autonomous mobile robot navigating in
a structured unknown environments, using free-configuration eigenspaces (FCE). The results are
very much consistent with the hypothesis we laid for our research in the beginning, i.e. trajectory
formation in correspondence to the highest eigenvector of the free-configuration laser data
always results into a better exploration of the unknown area. Consequently by adapting to the
similar trajectory formations, the autonomous mobile robot has better tendency towards the map-
building process. The trajectories maximizing the map building process could be formed by the
intramural property of the laser sensor dually responsible for the map building process, the vector
sum of all these highest vectors obtained from the laser data result into the trajectory which
maximizes the information of the map. These individual eigenvectors were machine learned and
the predicted new trajectory vectors facilitated the autonomous robot to maneuver at considerably
higher speed. A limitation of our proposed approach is the assumption of simple obstacle
geometry. This can be tackled by situation aware sensing descriptors, which may result into
dimensionality scaling and require algorithms that scale well.
7. REFERENCES
[1] O. Khatib, “Real-time obstacle avoidance for manipulators and mobile robots,” The
international journal of robotics research, vol. 5, no. 1,pp. 90–98, 1986.
[2] P. Vadakkepat, T. H. Lee, and L. Xin, “Application of evolutionary artificial potential field in
robot soccer system,” in IFSA World Congress and 20th NAFIPS International Conference,
2001. Joint 9th, pp. 2781–2785, IEEE, 2001.
[3] 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, 1991.
[4] A. Chakravarthy and D. Ghose, “Obstacle avoidance in a dynamic environment:A collision
cone approach,” IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems
and Humans, vol. 28, no. 5, pp. 562–574, 1998.

[5] D. Fox, W. Burgard, and S. Thrun, “The dynamic window approach to collision avoidance,”
IEEE Robotics & Automation Magazine, vol. 4, no. 1, pp. 23–33, 1997.
[6] P. Vadakkepat, K. C. Tan, and W. Ming-Liang, “Evolutionary artificial potential fields and
their application in real time robot path planning,”in Evolutionary Computation, 2000.
Proceedings of the 2000 Congress on, vol. 1, pp. 256–263, IEEE, 2000.
[7] D. Gallardo, O. Colomina, F. Fl´orez, and R. Rizo, “A genetic algorithm for robust motion
planning,” in Tasks and Methods in Applied Artificial Intelligence, pp. 115–121, Springer,
1998.
[8] S. Tang, W. Khaksar, N. Ismail, and M. Ariffin, “A review on robot motion planning
approaches,” Pertanika Journal of Science and Technology, vol. 20, no. 1, pp. 15–29,
2012.
[9] H. M. Choset, Principles of robot motion: theory, algorithms, and implementation. MIT
press, 2005.
[10] E. Plaku, K. E. Bekris, B. Y. Chen, A. M. Ladd, and E. Kavraki, “Sampling-based roadmap
of trees for parallel motion planning,” IEEE Transactions on Robotics, vol. 21, no. 4, pp.
597–608, 2005.
[11] R. Al-Hmouz, T. Gulrez, and A. Al-Jumaily, “Probabilistic road maps with obstacle
avoidance in cluttered dynamic environment,” in Intelligent Sensors, Sensor Networks and
Information Processing Conference, pp. 241–245, IEEE, 2004.
[12] L. E. Kavraki, P. Svestka, J.-C. Latombe, and M. H. Overmars, “Probabilistic roadmaps for
path planning in high-dimensional configuration spaces,” IEEE Transactions on Robotics
and Automation, vol. 12, no. 4, pp. 566–580, 1996.
[13] T. Gulrez, S. Zaheer, and Y. Abdallah, “Autonomous trajectory learning using free
configuration-eigenspaces,” in IEEE International Symposium on Signal Processing and
Information Technology (ISSPIT), pp. 424– 429, IEEE, 2009.
[14] T. Gulrez and A. Tognetti, “A sensorized garment controlled virtual robotic wheelchair,”
Journal of Intelligent & Robotic Systems, vol. 74, no. 3-4, pp. 847–868, 2014.
[15] T. Gulrez, A. Tognetti, and D. De Rossi, “Sensorized garment augmented 3d pervasive
virtual reality system,” in Pervasive Computing, pp. 97– 115, Springer, 2010.
[16] I. Jolliffe, Principal component analysis. Wiley Online Library, 2005.
[17] A. Al-Odienat and T. Gulrez, “Inverse covariance principal component analysis for power
system stability studies,” Turkish Journal of Electrical Engineering & Computer Sciences,
vol. 22, no. 1, pp. 57–65, 2014.
[18] M. Tipping and C. Bishop, “Probabilistic principal component analysis, “Journal of the Royal
Statistical Society”, Series B, vol. 61, pp. 611–622, 1999.
[19] “Player stage.” http://playerstage.sourceforge.net/.
[20] T. Chaudhry, T. Gulrez, A. Zia, and S. Zaheer, “Bezier curve based dynamic obstacle
avoidance and trajectory learning for autonomous mobile robots,” in Proceedings of the
10th International Conference on Intelligent Systems Design and Applications, ISDA’10,
2010, pp. 1059–1065.

[21] S. Zaheer, T. Gulrez, “Beta-eigenspaces for autonomous mobile robotic trajectory outlier
detection,” in 2011 IEEE Conference on Technologies for Practical Robot Applications,
TePRA, pp. 31–34, 2011.
[22] P. Raja and S. Pugazhenthi, Review Optimal path planning of mobile robots: International
Journal of Physical Sciences Vol. 7(9), 23, pp. 1314 – 1320, February, 2012
[23] R. Kala, K. Warwick Multi-Vehicle Planning using RRT-Connect. Paladyn Journal of
Behavioural Robotics, 2(3): 134-144, 2011

A Path Planning Technique For Autonomous Mobile Robot Using Free-Configuration Eigenspaces

More Related Content

What's hot (19)

Viewers also liked (7)

Similar to A Path Planning Technique For Autonomous Mobile Robot Using Free-Configuration Eigenspaces (20)

Recently uploaded (20)

A Path Planning Technique For Autonomous Mobile Robot Using Free-Configuration Eigenspaces