07-04-2012, 05:05 PM
Computing a Network of ASRs using a Mobile Robot Equipped with Sonar Sensors
project 01.pdf (Size: 310.04 KB / Downloads: 27)
INTRODUCTION
In recent years the idea of computing a network of local
(metrical) spaces as a representation of the environment that an
autonomous agent has experienced is gaining momentum. This
idea, commonly known as computing a hybrid map, is favored
by both the pragmatists (i.e. those researchers interested in
building mobile robots) and the theorists (i.e. those researchers
interested in developing computational theories of cognitive
maps).
THE EXPERIMENT
The robot we use is a Pioneer 2 robot from ActivMedia and
it came with a ring of 8 sonar sensors. The robot is positioned
somewhere in the corridor in an office environment and is
allowed to explore the environment until it is told to stop. No
modification of the environment is done. That is, things that
already existed in the environment (such as rubbish bins,
flower pots, cabinets, etc.) remain there and doors leading into
offices are close or open depending on the time of the
experiment.
ALGORITHMS FOR COMPUTING ASRS
the initial surfaces computed from the sonar
readings obtained by the robot traversing the environment
along the path as shown in Fig. 1. As can be seen, there are too
many spurious surfaces. We pre-process the input by removing
outliers and insignificant surfaces. Fig. 3 shows the actual
surfaces used as input for computing ASRs.
DISCUSSION
The results obtained show a network of ASRs is computed.
However, it is not often the case that the split is made at a point
that humans think is reasonable. However, it is interesting to
observe that the same number of ASRs is produced for both the
forward and return journey. This is not often the case. We have
computed ASRs in the forward journey at different times and
some of them showed very different ASRs are computed. This
is within the expectation of the theory [1].