The female cricket recognizes the calling song and can identify the position
of the male. The female needs to recognize the sound of the male of the
same species and distinguish it from the sounds of males of other species
and from miscellaneous noise from other places. Furthermore, when a large
number of conspecific male crickets can be heard, the females need to decide
which male to choose. Many ethological, anatomical, and neurophysiological
investigations have been conducted on this attraction to a specific sound
(phonotaxis). Important clues for the female to recognize the calling song
are the carrier frequency and the temporal pattern. The male makes
sounds by crossing his 2 wings and rubs them together like a file rubbing
a comb. The sound is further magnified by resonance through the wings.
The female chooses repeated burst signals at a specific frequency. Some
researchers have built a cricket robot to understand the choice mechanism.
Here, we introduce a part of it to show how to understand the nervous system
using a robot.
The cricket robot is based on a scaled-down version of the Khepera robot．This
robot has a sound-processing circuit designed to mimic the unique sound
sensory system of the cricket. Crickets have a tympanic organ on each of
their forelimbs, which is connected to the other via a trachea and also
connected to the spiracles on the body surface. As a result, the vibration
of the eardrum reflects the pressure difference between the sound on the
outer and inner surfaces. Sounds coming from different directions cause
a pressure differential between the 2 sides of the eardrum that depends
on the distances from which the sounds are propagated. This causes signal-phase
cancelation; thus, the amplitude of the eardrum vibrations reflects the
direction of the sound. Hence, despite the relatively short distance between
the left and right auditory receptors in crickets, they show strong directional
sensitivity. The robot uses the same mechanism by using the difference
between signals from 2 microphones slightly apart from each other and programmable
electrical delay. To match this robot with the carrier frequency of the
cricket attraction song, the ear interval and internal delay were adjusted.
This measure was found to reduce the capacity to identify a direction of
a sound source with a different carrier frequency. This secondary effect
also helped to enhance the carrier frequency selectivity. These results
demonstrated that the robot was capable of moving toward the sound source
with the correct frequency without using a frequency filter.
In this robot, sound was processed by a neural network. Some versions of
this network were implemented, making allowances for the more detailed
biological process. The details of the neural circuit in the cricket are
not fully understood; however, the general principles of the method are
explained below by explaining 1 of the simpler networks. The pair of ascending
neurons (ANs) identified in the anterior thoracic ganglia of the cricket
is thought to be critical in controlling phonotaxis. ANs directly receive
inputs from the auditory receptors. The AN1 (one type of ANs) firing response
is the same as the calling song; the firing rate and response time are
consistent with the amplitude. These characteristics were reproduced in
the robot using the most basic “integrate and fire” neuron model. In this
simple model, AN was connected to 2 motor neurons (MNs) as output neurons.
Each AN has an excitatory connection with MN ipsilaterally and an inhibitory
connection with the other MN. The synaptic connection is expressed by a
certain weight for a single connection (synaptic weight), and the weight
indicates the strength of the inter-synapse transmission of each firing.
The weight changes dynamically. When fast spike sequences occur, the synapses
are suppressed (reduction in synaptic weight) and the restoration requires
the spike sequence in the presynaptic neuron to be intermitted. Inhibition
is not a negative input but an open/close switching control to suppress
the effect of synaptic excitation. These characteristics are closer to
the actions of a real synapse than to those of an artificial neural network.
In this circuit, the MN spike was used to control the direction of the
robot. A sound signal received by the left ear induces the excitation of
AN on the same side. This leads to the excitation of MN on the same side
and simultaneously inhibits the excitation of AN on the opposite side.
Thus, whichever AN fires first controls the response. In addition, because
the synapse is suppressed, only the initial portion of the AN activity
will make a major contribution to the excitation of MN. Based on this principle,
a wide range of behavior observed in the cricket can be explained by testing
the robot with regard to stimulus and behavior. For example, by testing
the robot with pseudo-cricket songs with different syllable repetition
intervals (SRIs), SRI has been shown to be a crucial clue for recognizing
the calling song in crickets.
Investigation of the nervous system
mechanism using such robots is ongoing. This case study introduces a robot as a
means of exploring hypotheses on insect behavior. The results open new lines of
inquiry, and these will require biological experiments leading to a better
robot model. By testing the robot in the same framework as the animal studied,
it is possible to know whether our understanding of the system is sufficient to
reproduce animal behavior.
昆虫ミメティックス Insect Mimetics(2008)，針山孝彦，下澤楯夫，pp. 791-797