Content uploaded by Theodore Manikas
Author content
All content in this area was uploaded by Theodore Manikas on Jul 30, 2014
Content may be subject to copyright.
Content uploaded by Theodore Manikas
Author content
All content in this area was uploaded by Theodore Manikas on Jul 16, 2014
Content may be subject to copyright.
26 IEEE Instrumentation & Measurement Magazine December 2007
E
ngineers and scientists use instrumentation and mea-
surement equipment to obtain information for spe-
cific environments, such as temperature and pressure.
This task can be performed manually using portable gauges.
However, there are many instances in which this approach
may be impractical; when gathering data from remote sites
or from potentially hostile environments. In these applica-
tions, autonomous navigation methods allow a mobile robot
to explore an environment independent of human presence
or intervention. The mobile robot contains the measurement
Theodore W. Manikas, Kaveh Ashenayi, and Roger L. Wainwright
Genetic Algorithms
for Autonomous
Robot Navigation
1094-6969/07/$25.00©2007IEEE
device and records the data then either transmits it or brings it
back to the operator.
Sensors are required for the robot to detect obstacles in the
navigation environment, and machine intelligence is required
for the robot to plan a path around these obstacles [1]. The use
of genetic algorithms is an example of machine intelligence
applications to modern robot navigation. Genetic algorithms
are heuristic optimization methods, which have mechanisms
analogous to biological evolution. This article provides initial
insight of autonomous navigation for mobile robots, a descrip-
tion of the sensors used to detect obstacles and a description of
the genetic algorithms used for path planning.
Autonomous Robot Navigation
Autonomous navigation implies that a robot must decide how
to travel through a given environment [2]. While basic infor-
mation may be available to the robot about the navigation area
boundaries, unknown obstacles may exist within the naviga-
tion area. This is called an uncertain environment: the robot
must be able to detect and maneuver around these obstacles to
reach its target point.
The navigation environment in which the robot and ob-
stacles both exist is called the “world space.” A path between
the starting and target points that avoids collisions with ob-
stacles is said to be “feasible.” Robot navigation methods need
to solve the path-planning problem, which is to generate a
“feasible” path and optimize this path with respect to specifi c
criteria.
Many path-planning methods use a grid-based model to
represent the world space (Figure 1). In this model, the world
space is partitioned into grids, where the size of each grid
depends on the particular specifi cations of the autonomous
December 2007 IEEE Instrumentation & Measurement Magazine 27
Fig. 1. World space example with grid system representation.
Fig. 2. Example of a robot sensing obstacles in the navigation environment.
Note that the switching points
are part of the evolutionary
process and vary from
chromosome to chromosome.
28 IEEE Instrumentation & Measurement Magazine December 2007
vehicle and the navigation area. An obstacle may occupy one
or more grids, depending on the size of the obstacle relative
to the size of the vehicle. Squares, fully or partly blocked by
obstacles, are identified as occupied cells when detected in
the search space. The robot can move on all free cells, where
the center of the robot moves along an imaginary line from the
center of one cell to the center of another cell, either in a vertical
or horizontal direction.
Sensing the Navigation Environment
Autonomous navigation requires the robot to interact with its
environment and to adapt to changing conditions. This means
that sensors must be mounted on the robot to detect and locate
obstacles in the navigation space (Figure 2). Types of sensors
that are commonly used in robot navigation include contact,
orientation, proximity (capacitive and magnetic), imaging,
ultrasonic and infrared sensors, as well as laser range finders
and cameras.
Since each sensor type has its strengths and weaknesses
(suitable for a different application), an autonomous robot sys-
tem will typically combine various sensor types for effective
obstacle mapping. For example, proximity detectors are used
to identify items in the proximity (short range) of the robot.
Orientation sensors such as gyroscopes and global positioning
systems (GPS) are used to provide the robot with data on its
orientation and direction.
Imaging sensors such as IR cameras and Omni cameras
are used to provide visual data in visible as well as the IR
frequency ranges (to detect temperature differentials). This
information, in combination with other sensors, can be
used to help the robot to avoid obstacles such as potholes
and bodies of water in an unknown terrain. The imaging
sensors can be used for medium- and long-range detec-
tion.
Ultrasonic sensors are widely used for obstacle detection as
a result of their simplicity and relatively low cost [3]. An ultra-
sonic sensor system emits a sonar signal and receives its echo
from an object. The object distance is determined by measuring
the time difference between the signal emission and the echo
reception.
While ultrasonic sensors are useful for measuring the
distance of an obstacle from the robot, they are less suc-
cessful at detecting object profiles (such as edges) because
of sonar wave reflections. To overcome this limitation, IR
sensors may be used in addition to ultrasonic sensors [4].
An infrared sensor system emits a pulse of infrared light
that is reflected back from an object. The angle of this reflec-
tion that is received at the sensor yields information about
the object distance and profile. One disadvantage of IR sen-
Fig. 3. Flow chart of a basic genetic algorithm.
Fig. 4. Burchardt’s path chromosome structure [8].
December 2007 IEEE Instrumentation & Measurement Magazine 29
sors is that they can become
severely corrupted by other
light sources. Therefore, a
robot navigation system
should typically contain
several types of sensors, as
each sensor has its advan-
tages and disadvantages.
Planning Navigation Paths Using
Sensor Information
The path-planning problem is often divided into global and
local planning approaches. Global planning optimizes the
overall traveled distance, while local planning determines
how to navigate around obstacles. The global approach con-
structs a world model based on sensory information and uses
this model to plan a global path. Since the construction and
maintenance of a global map may become computationally
complex for a robot, the local approach is often used to plan
paths around specific obstacles. Local path planning uses the
obstacle location and profile information obtained from the
sensors to determine a feasible navigation path for the robot.
The path-planning problem has been shown to be NP (Non-
deterministic Polynomial-time)-hard [5], which means that
many of the traditional methods can become computation-
ally intractable. Genetic algorithms have been shown to be
effective in solving NP-hard problems; thus, they have often
been used for local path planning in contemporary robot
navigation algorithms [7].
Genetic algorithms are heuristic optimization methods,
the mechanisms of which are analogous to biological evolu-
tion [6]. Figure 3 shows a flow chart that describes the basic
operations of a genetic algorithm. The algorithm starts with
a population of individuals (chromosomes). Each individual
represents a possible solution for the given problem. For robot
navigation, an individual may represent a path between the
starting and target points. Each individual is assigned a fitness
value, based on how well the individual meets the problem ob-
jectives. Using these fitness values, individuals are selected to
be parents. These parents form new individuals, or offspring,
via crossover and mutation. Parent selection, crossover, and
mutation operations continue for numerous iterations (gen-
erations) until the algorithm
converges to an optimal or
near-optimal solution.
The dominant task in
developing a genetic algo-
rithm to solve a particular
problem is the develop-
ment of the chromosome
structure and the operators
that process this structure, such as the fitness function and
crossover and mutation operators. For robot navigation, the
goal is to determine a feasible path as quickly as possible.
Therefore, the chromosome must be able to represent a valid
path in the navigation space. In addition, the structures of
the chromosome and fitness function must allow for efficient
computation. These factors govern the development of ge-
netic algorithm path planners.
Path Planning Using Genetic
Algorithms
Various genetic algorithm methods have been developed
to solve the path-planning problem for autonomous robot
navigation. Using the grid-based model, the world space
can be viewed as a set of (x,y) coordinate points, or as a set
Fig. 5. (a) Sample path; (b) fixed-length chromosome representation.
Fig. 6. (a) Chromosome with switching points; (b) resultant path.
The navigation environment
in which the robot and
obstacles both exist is
called the “world space.”
30 IEEE Instrumentation & Measurement Magazine December 2007
of rows and columns. Often
the grid is viewed as a square,
in which the number of rows
equals the number of columns
[7]. The chromosome struc-
ture used by many genetic
algorithm path planners is a
value-encoded scheme: x and
y coordinate information for
the path points is contained
in the chromosome structure.
Figure 4 shows the chromo-
some structure used by Burchardt and Salomon [8]. The first
gene of the chromosome contains the value of the chromo-
some length, which indicates the number of path points. Each
subsequent pair of genes contains the (x,y) coordinates for
each path point. The path fitness is based on both path length
and feasibility, with a significant penalty for paths that collide
with obstacles. This fitness function helps the genetic algo-
rithm identify a collision-free path through the world space.
Alvarez, Caiti, and Onken [7] use a similar representation,
with (x,y,z) coordinates, since their genetic algorithm was de-
veloped for an autonomous undersea vehicle. Path fitness is
determined by feasibility and the estimated energy required
by the vehicle to travel this path.
Encoding path points into a chromosome is an intuitive
and straightforward approach. However, differences in path
lengths may result in differences in the number of chromo-
some genes. For a genetic algorithm to function efficiently, it
is desirable for all chromosomes to have the same number of
genes, especially for crossover and mutation operations. Fig-
ure 5 shows an example of a fixed-length chromosome, based
on the model of Geisler and Manikas [9]. Given a navigation
environment that is modeled by N rows, a path in that environ-
ment is represented by a chromosome with N genes. Each gene
position (locus) corresponds to an x-coordinate, while each
gene value (allele) corresponds to a y-coordinate within that
column. Figure 5(a) shows a path in a 6 × 6 navigation space,
modeled by the chromosome shown in Figure 5(b). This chro-
mosome represents a path that starts at point (1,4) and travels
along intermediate points (2,4), (3,2), (4,6), and (5,5) to reach its
target at point (6,1).
While this chromosome structure is easier to handle for a
genetic algorithm, a main limitation of this structure is that it
requires all paths to be x-monotone: x
i+1
> x
i
. Depending on the
obstacle configuration in the world space, this restriction may
not allow the genetic algorithm to find a feasible path. Sedighi
et al. [10] address this limitation by allowing paths to switch
between x-monotone and y-monotone (y
i+1
> y
i
) orientations.
This is accomplished by adding “switching points” to the
chromosome structure to identify where the path switches
orientation.
Figure 6 shows an example for this model. The chromo-
some structure (Figure 6[a]) contains two parts: path-loca-
tion and path-switch. The path-location part works similar
to the way it does in the Geisler and Manikas model, except
that it represents both x- and y-
monotone subpaths within the
main path. For x-monotone ori-
entation, each locus represents
a column index, while each al-
lele represents a row within that
column. For y-monotone ori-
entation, each locus represents
a row index, while each allele
represents a column within that
row. There are two switching
points that identify the loci on
the path-location part where the path orientation switches.
Note that the switching points are part of the evolutionary
process and vary from chromosome to chromosome. Up to
two switching points are allowed. However, a given chromo-
some may evolve only one switch point, or perhaps none.
Figure 6(b) shows the navigation path that corresponds
to the chromosome shown in Figure 6(a). Note that the world
space origin is in the upper left corner, as per the specifica-
tion of Sedighi et al. [10]. The path always starts at (1,1) with
y-monotone orientation. The direction bit for locus 1 is 1, so
the navigation direction to the next point is horizontal, then
vertical. The next point is identified by locus 2 in the path-lo-
cation chromosome, with an allele of 9. Since the orientation
is y-monotone, this corresponds to row 2, column 9, or point
(2,9) in the world space. The path continues through points
(3,8) to (4,2).
The first switching point is at locus 4 of the path-location
chromosome. Thus, at point (4,2) the orientation changes
from y-monotone to x-monotone. The next point is identified
by locus 5 in the path-location chromosome, with an allele of
10. Since the orientation is now x-monotone, this corresponds
to column 5, row 10, or point (10,5) in the world space. The
path continues to point (6,6), where the next switch point
occurs.
The second switching point is a locus 6 of the path-location
chromosome, so the path now changes back to a y-monotone
orientation. This means that locus 7, allele 8 corresponds to
row 7, column 8, or point (7,8) in the world space. The path
continues through points (8,10) and (9,9) to the target point
(10,10). As with the other genetic algorithm path planners,
path feasibility is important when evaluating an individual.
Additional goals are to minimize the total path length and the
number of turns required by the robot.
Conclusion
This paper has provided an overview of applications of
genetic algorithms to autonomous robot navigation. Many
of these methods have been tested using simulated environ-
ments. The genetic algorithm optimization method has been
previously shown to be effective in solving NP-hard prob-
lems such as path planning. However, these algorithms may
take some time to converge to an optimal solution, which will
affect the speed of robot navigation. After the genetic algo-
rithm path planners have been verified using simulations,
Orientation sensors such
as gyroscopes and global
positioning systems
(GPS) are used to provide
the robot with data on its
orientation and direction.
December 2007 IEEE Instrumentation & Measurement Magazine 31
these algorithms will need to be tested on actual robots and
modified as necessary to ensure acceptable real-time naviga-
tion performance.
References
[1] K. Warwick and S.J. Nasuto, “Historical and current machine
intelligence,” IEEE Instrum. Meas. Mag., vol. 9, (no. 6), pp. 20–26,
Dec 2006.
[2] J.-C. Latombe, Robot Motion Planning, Boston, MA: Kluwer
Academic Publishers, 1991.
[3] S. Choi, T. Jin, and J. Lee, “Obstacle avoidance algorithm for visual
navigation using ultrasonic sensors and a CCD camera,” Artif.
Life Robotics, vol. 7, (no. 3), pp. 132–135, Sep 2003.
[4] M. Mazo, “An integral system for assisted mobility [automated
wheelchair],” IEEE Robotics Automation Mag., vol. 8, (no. 1), pp.
46–56, Mar 2001.
[5] Y.K. Hwang and N. Ahuja, “Gross motion planning—A survey,”
ACM Comput. Surv., vol. 24, (no. 3), pp. 219–291, Sep 1992.
[6] M. Mitchell, An Introduction to Genetic Algorithms, Cambridge, MA,
MIT Press, 1996.
[7] A. Alvarez, A. Caiti, and R. Onken, “Evolutionary path planning
for autonomous underwater vehicles in a variable ocean,” IEEE J.
Oceanic Eng., vol. 29, (no. 2), pp. 418–429, Apr 2004.
[8] H. Burchardt and R. Salomon, “Implementation of path planning
using genetic algorithms on mobile robots,” in Proc. 2006 IEEE
Congress Evolutionary Computation, Piscataway, NJ, IEEE, 2006, pp.
1831–1836.
[9] T. Geisler and T.W. Manikas, “Autonomous robot navigation
system using a novel value encoded genetic algorithm,” in 45th
IEEE Int. Midwest Symp. Circuits and Systems, Piscataway, NJ, IEEE,
2002, pp. 45–48.
[10]K.H. Sedighi, K. Ashenayi, T.W. Manikas, R.L. Wainwright,
and H.-M. Tai, “Autonomous local path planning for a mobile
robot using a genetic algorithm,” in Proc. 2004 IEEE Congress
Evolutionary Computation, Piscataway, NJ, IEEE, 2004, pp.
1338–1345.
Theodore W. Manikas (theo-
dore-manikas@utulsa.edu) has
been an assistant professor of
electrical engineering at The
University of Tulsa since 2000.
Prior to that, he was a systems
analyst at the NSABP Biostatis-
tical Center, University of Pitts-
burgh. His research interests
include genetic algorithms,
robotics, and VLSI design. He
is a registered Professional Engineer in Oklahoma.
Kaveh Ashenayi (kash@utulsa.
edu) has been with The Univer-
sity of Tulsa since 1986. He is
currently a professor with the
Electrical Engineering Depart-
ment. His research interests in-
clude intelligent systems, robot-
ics, control, and power systems.
He is a registered Professional
Engineer in Oklahoma.
Roger L. Wainwright (rogerw@
utulsa.edu) has been with The
University of Tulsa since 1974.
He is currently a professor of
computer science and the chair
of the Mathematical and Com-
puter Sciences Department. His
research interests include analy-
sis of algorithms, combinato-
rial optimization, genetic algo-
rithms, and robotic navigation.
