Information Technology Reference
In-Depth Information
Improving Area Center Robot Navigation Using
a Novel Range Scan Segmentation Method
José Manuel Cuadra Troncoso 1 , José Ramón Álvarez-Sánchez 1 ,
Félix de la Paz López 1 , and Antonio Fernández-Caballero 2
1 Dpto. de Inteligencia Artificial - UNED - Madrid, Spain
{jmcuadra,jras,delapaz}@dia.uned.es
2 Departamento de Sistemas Informáticos-UCLM-Albacete
caballer@dsi.uclm.es
Abstract. When using raw 2D range measures to delimit the border
for the free area sensed by a robot, the noise makes the sensor to yield a
cloud of points, which is an imprecise border. This vagueness pose some
problems for robot navigation using area center methods, due to free
area split points locations. The basic method, when locating split points,
does not take into account environmental features, only the raw cloud
of points. In order to determine accurately such environmental features
we use a novel range scan segmentation method. This method has the
interesting characteristic of being adaptive to environment noise, in the
sense that we do not need to fix noise standard deviation, even different
areas of the same scan can have different deviations, e. g. a wall besides
a hedge. Procedure execution time is in the order of milliseconds for
modern processors. Information about interesting navigational features
is used to improve area center navigation by means of determining safer
split points and developing the idea of dynamic split point. A dynamic
split point change its position to a new feature if this new feature is
considered more dangerous than the one marked by the split point.
1
Introduction
In range sensor measurements noise sources may be very disparate, like: intrinsic
sensor error, sensor calibration errors and the environment itself. The noise por-
tion due to the environment cannot be neglected, but it can be the main noise
source. For example, laser measures over a garden hedge carry much more error,
in standard deviation terms, than measures over a plain wall. Although devices
technical specifications usually carry information about intrinsic device noise
and calibration procedures, we have no prior information about environmental
noise, so we need to make assumptions about its behavior. If we want to develop
a procedure being able to handle worlds with surfaces of different characteristics
at a time we have to reduce or to relax our assumptions about noise. As more
general our noise model as more flexible and robust navigation we get.
Filtering noise we can get an ideal representation of the world as a set of line
segments and curved segments, this procedure is known as scan segmentation or
 
Search WWH ::




Custom Search