EVOLUTION
OF SENSORIMOTOR NEURAL NETWORKS TO PERFORM ACTIVE MOTION CAMOUFLAGE
Project funded by EPSRC
PhD student employed on this grant. Andrew Anderson
The objectives of this project may be
summarised as follows;
·
To develop the first
explicit neural model for active motion camouflage.
·
Determination of the
optimal, biologically plausible, sensorimotor neural architectures and learning
rules for a system to achieve active motion camouflage.
·
Visualisation of the
behaviour of the best performing motion camouflage sensorimotor systems.
As the prey, denoted by the dark circles in
the figure on the left, passed by a genuine fixed point in the environment, P,
the lines of sight connecting the prey’s eyes to this point determine the set
of constraint lines which define the optic flow characteristic of a stationary
object. Camouflage may be accomplished by ensuring that the predator's
trajectory, shown by the textured circles, is such that it always moves so that
it lies along a set of constraint lines consistent with some chosen fixed
position in the environment. This selected position can be anywhere along the
initial camouflage constraint line, that is the line joining the initial
positions of predator and prey. The camouflage trajectory may be achieved by
requiring that at each stage of the approach the predator moves to ensure that
it consistently views the prey frontally and moves radially away from its
initial position, so moving forwards along the camouflage constraint lines.
This strategy ensures that as the approach is made the predator’s image,
projected onto the prey’s retina, is the same as would arise from the predator
remaining stationary at its initial position. The first requirement, of viewing
the prey frontally, requires the predator to be able to co-ordinate sensory
visual input with motor output to correct orientation. To achieve the second
requirement, that of a strictly radial movement towards the prey, the predator
must learn to co-ordinate the angular movements required by the orienting step
with corrective lateral motions such that the ratio of the two is a constant,
and equal to the distance from the initial position.