Certain insect behaviours appear to function as reflexes when tested with restricted
sensory stimuli. When more complex sensory situations are considered it becomes ap¬
parent that these behaviours are not independent; they must interact within the nervous
system of the animal or through feedback. This dissertation describes experiments and
modelling undertaken to address the question of how optomotor following interacts
with phonotaxis in the cricket.
Paths of crickets walking in response to calling song and optical flow stimuli were
recorded on an open-loop trackball and in an arena. The results were used to guide the
development of a new model of co-ordination between the auditory and visual systems,
implemented on a miniature robot.
Five initial hypotheses were investigated, based on a previous robotic modelling
study: inhibition of the optomotor response, modulation of the optomotor gain, chaining of two behavioural subsystems, summation at the motor output, and efference copy.
Experiments with the trackball allowed the first three possibilities to be rejected, but
summation and efference copy were more difficult to distinguish between. The first
evidence of a possible efference copy-type mechanism comes from modified trackball
experiments where a closed feedback loop was established for the visual system. Modulation of open-loop behaviour was observed that depended on the sign of previously
experienced visual feedback, suggesting adaptation of an internal signal. However the
same effect could be explained by sustained activity in the visual system. The second
piece of evidence supporting the efference copy theory came from the robot model;
when the summation mechanism was implemented the optomotor response caused the
robot to over-compensate after turning to sound. This behaviour was not observed
from crickets in the arena.
Implementation of an efference copy-based mechanism on the robot is accomplished using a recurrent network of spiking neurons (liquid state machine). It is
proposed that the circuitry of the mushroom bodies might permit such computation
in insects. It is shown that the neural network can learn to predict and cancel out self-generated optical flow signals in the robot during phonotaxis, whilst reproducing the
behaviour seen on the trackball under equivalent conditions.