0
Automatic Parallel Parking
Y.
K.
Lo,
A.
B.
Rad,
C.
W.
Wongand
M. L.
Ho
Depment
of
Elech-ical Engineering
The Hong Kong Polytechnic University
Hung Hom, Kowloon,
Hotig Kong
Abstract
An automated parallel parking strategy for a vehicle-like
robot
is
presented. This study addresses general
casts
of
parallel
parking within a rcctangular spacc. The procedure consists
of
three phases. In scanning phase. Infrared sensors
in
the robot
arc usedtoscanningthe parking environment inorder tosearch
a suitable parking position. Then maneuvcring path
is
generated for different parking space
in
the
next phase, starting
phase. The robot moves backward
to
the edge ofthe parking
space and starts
its
parking strategy.
In
maneuver tracking
phase. the robot follows the inaneuvering path
to
the parking
position. It depends on the width of the rectangular space
which has been scanned in the previous phase. This strategy
has been implemented in a vehicle-like robot and
is
developed
for an assistant to help human drivers in the future.
1.
Introduction
An automated parallel parking strategy for
a
vehicle-like robot
is
presented. This study
is
part
ofthc
research project
in
t,hc
development of the autonomous vehicle control. I’arallel
parking
is
difficult for human driver especially Ibr beginners.
Therefore.
this
type
or
problems altracts a great deal
of
attention from the research community. This research
is
derived
from the
study ofmotion-planning
ofrobots. There exists many
algorithms in robot motion planning and they were difficult to
apply those algorithms into fourwheels vehicle parking cases.
The research
on
this
topic
can
be
clissifird
into
two
groups:
I-stabilization of the vehicle to a point by means of feedback
state;
2-
planning a feasiblc path to reach a point and following
the path. In the lormer category, Yasunobu and Murai
[I]
have
proposcd
a
controllcr based on human experience to develop a
hierarchical fuzzy control and predictive fuvy control for
vehicle parking. The vehicle
is
controlled moving point
by
point. That algorithm generates the mancuvering path point by
point from the human knowledge base
in
the predictive fuzzy
controller. Researchcn
1101
only concerning the classical
four
wheels vehicle. trailer vehicle are also considered. Jenkin and
Yuhas
121
have reported
a
simplified neural network controller
by
decomposition. The neural controller
is
decomposed
by
subtasks. The neural controller
is
trained based on the
kinematics data, however the decomposed neural network
require
less
training time thus it has simplified the training
process. Kin,jo. Wang and Yamamoto
[3]
have used Genetic
Algorithm (GA)
to
oplimk
neural
Contmller
for controlling
trailer truck. Initially theneural controller
is
produced randomly
and GA changes
the
weighting
afthe
controller
to
a
suitable
due. The algorithms discussed above are based on the
kinematics data to formulate intelligent controllers.
Figure
I
Maneuvering
of
Parallel Parking
For the path planning category. Paromtchik and Laugier
141
presented an approach
to
parallel parking for a nonholonomic
vehicle. In thatapproach. a parking space
is
scanned before the
vehicle moves backward into
its
parking bay. The vehicle
follows
a
sinusoidal
path
in backward motion. while the fbrward
motion
is
along
a
straight line withoot sideways displacement.
In this approach.
the
possible collision during reverse betwecn
thc vehicle and thc longitudinal bnundar) ofthe parking space
is
not discussed. Murray and Sastry
[5]
worked on steering
a
nonholonomic system between arbitrary points by means
of
sinusoids.
Automatic parallel parking involves inany problems. such as
recognition
of
driving circumstances. maneuvering path
planning. communication and vehicle control. This paper
focuses
on
the maneuvering path planning. ‘The system works
in
three phases. In scanning phases: the parking circumstance,
is scanned by infrared
srnsors
after the parking command
is
activated. It then
gucs
to
ncxt
phasr. starling phase. after
a
suitable parking space has been detected. The maneuvering
path
is
also
produced according to the scanned information.
The robots
movrs
backward
to
the edge ofthr parking position
and begins its parking strategy. In order to avoid potential
collision.
the
robot starts
at
thc
suitable position which
depends on different dimension
of
parking space. In the final
phase. inaneuvertrackiiig phase. the robot followsthc path
10
a
desired parking posi!ion. The parameters ofthe maneuvering
path have beenproduced off-line.
A
database has
been
built for
different ~ir~~mstan~e
such
as
the
longitudinal and
latrral
dimension ofparking space, thedimension ofvehicle-like robot.
its
specification ofthe stcrring
angle
(maximum turning angle)
and lateral displacement from the aside car.
1190