17-03-2014, 10:36 AM
Abstract
This paper presents a proposal for a visual homing
algorithm inspired by the behaviours of social insects. The
homing method presented is based on an affine motion
model which parameters are estimated by a best matching
criteria. In the matching phase no attempts are made to
recognise objects or to extract 30 models of the scene.
Hypotheses and perspectives about the use of single
landmarks by bees are introduced. Tests and results are
presented.
1.- Introduction
A navigation task is normally composed of two phases:
a coarse approach to the goal and a precise positioning to it
(homing).
Entomological studies about social insects (bees, ants,
etc.) have discovered some mechanisms of visual
navigation and landmarks use that can be useful in robotics
[Santos-Victor et al. 93, 941. In fact, several interesting
considerations could be made regarding the ability of many
insects to return to precise locations for foraging or for
finding home [Wehner 921.
According to experiments, the strategy used by bees in
order to be able to reach a known point can be summarised
in the following two points:
bees store images (snapshots) and remember the
apparent dimension and the position of landmarks
[Cartwright, Collett 831 surrounding a place;
bees remember the shape, the pattern and the colour
of a landmark [Could 861.
Experiments on ants and bees suggested in fact that an
insect fixes the location of landmarks surrounding a place
by storing a sort of snapshot of the landmarks taken from
that place; the snapshot taken from the environment is
considered as a constellation of objects and it does not
encode explicitly the distance between landmarks or
between landmarks and target but, instead, the position of
each landmark is labelled by its compass bearing
[Cartwright, Collet 871.
This use of landmarks is quite different from the
classical one, where a landmark has fixed and known
position and it is usually labelled with its distance from an
important place.
In order to return to specific places, landmarks are used
by social insects in one (of two different methods: deadreckoning
and near-by landmark [Snyder 971.
The former is used, for example, by bees when they
search a zone for feeding. In this case, information about
distance and direction are provided by the dance of the
nestmates [von Frisch 711.
The near-by landmark method is used when the deadreckoning
does not have the required degree of precision
and in general bees use it for homing or for approaching
precisely the feeding area.
An interesting model introduced to explain how bees
exploit the near-by landmark phase navigation was
proposed by Cartwright and Collett in 1983, subsequently
refined in 1987. The results of this model, simulated on a
computer, strongly resemble the actual behaviour of bees
[Wittmann 951.
In the model, bees seem to learn some information
concerning the landmarks surrounding a place through a
two dimensional picture (snapshot) memorised from that
place together with their orientation.
The steps involved for homing are then:
matching phase: the bee compares the snapshot
stored of the place surrounding the goal with the
actual snapshot;
near-by landmark n,avigation phase: the differences
in position and in dimensions between the landmarks
of the two images drives the bee for positioning.
The matching and the navigation phases have been
implemented and tested iInd they are presented in the
following paragraphs.
Interesting aspects concerning the use of distinct
landmarks instead of the: whole image (snapshot) are
considered in paragraph 4.
This paper presents a proposal for a visual homing
algorithm inspired by the behaviours of social insects. The
homing method presented is based on an affine motion
model which parameters are estimated by a best matching
criteria. In the matching phase no attempts are made to
recognise objects or to extract 30 models of the scene.
Hypotheses and perspectives about the use of single
landmarks by bees are introduced. Tests and results are
presented.
1.- Introduction
A navigation task is normally composed of two phases:
a coarse approach to the goal and a precise positioning to it
(homing).
Entomological studies about social insects (bees, ants,
etc.) have discovered some mechanisms of visual
navigation and landmarks use that can be useful in robotics
[Santos-Victor et al. 93, 941. In fact, several interesting
considerations could be made regarding the ability of many
insects to return to precise locations for foraging or for
finding home [Wehner 921.
According to experiments, the strategy used by bees in
order to be able to reach a known point can be summarised
in the following two points:
bees store images (snapshots) and remember the
apparent dimension and the position of landmarks
[Cartwright, Collett 831 surrounding a place;
bees remember the shape, the pattern and the colour
of a landmark [Could 861.
Experiments on ants and bees suggested in fact that an
insect fixes the location of landmarks surrounding a place
by storing a sort of snapshot of the landmarks taken from
that place; the snapshot taken from the environment is
considered as a constellation of objects and it does not
encode explicitly the distance between landmarks or
between landmarks and target but, instead, the position of
each landmark is labelled by its compass bearing
[Cartwright, Collet 871.
This use of landmarks is quite different from the
classical one, where a landmark has fixed and known
position and it is usually labelled with its distance from an
important place.
In order to return to specific places, landmarks are used
by social insects in one (of two different methods: deadreckoning
and near-by landmark [Snyder 971.
The former is used, for example, by bees when they
search a zone for feeding. In this case, information about
distance and direction are provided by the dance of the
nestmates [von Frisch 711.
The near-by landmark method is used when the deadreckoning
does not have the required degree of precision
and in general bees use it for homing or for approaching
precisely the feeding area.
An interesting model introduced to explain how bees
exploit the near-by landmark phase navigation was
proposed by Cartwright and Collett in 1983, subsequently
refined in 1987. The results of this model, simulated on a
computer, strongly resemble the actual behaviour of bees
[Wittmann 951.
In the model, bees seem to learn some information
concerning the landmarks surrounding a place through a
two dimensional picture (snapshot) memorised from that
place together with their orientation.
The steps involved for homing are then:
matching phase: the bee compares the snapshot
stored of the place surrounding the goal with the
actual snapshot;
near-by landmark n,avigation phase: the differences
in position and in dimensions between the landmarks
of the two images drives the bee for positioning.
The matching and the navigation phases have been
implemented and tested iInd they are presented in the
following paragraphs.
Interesting aspects concerning the use of distinct
landmarks instead of the: whole image (snapshot) are
considered in paragraph 4.