Neural Mechanisms of Leg Motor Control in Crayfish - Advancing Artificial Intelligence Through Biological Process Applications - page 26

Sensory Coding in Autonomous
Embodied Systems

in vitro , similarities of activation patterns among
the population suggest that movement monitor-
ing is based on a collegial sensory encoding of
the various biomechanical parameters. Direct
electrical synaptic connections between sensory
afferents coding for the same direction (El Manira
et al. , 1993) could play a substantial role in this
collegiality. As recently proposed (Rowland et
al. , 2007), it is likely that such multi-sensory
information enhances network computation in
such a way that the response latency is substan-
tially shortened.

One of the major challenges in studying the func-
tion of a sensory circuit is to understand what
features of natural stimuli are encoded in the
spiking of its neurons (Barlow, 1972). Movements
comprise both dynamic (displacements) and static
(successive positions) parameters that are encoded
into phasic and tonic discharge, the latter playing
also a role as an internal reference. Combining
the two components allows a rapid (phasic) and
referenced (tonic) spatial representation of joint
position and movement. In the crayfish, phasic
coding is a property of all CBCO sensory neurons,
and their firing frequency during actual movement
is always higher than during stationary plateaus,
indicating that the major information conveyed
by sensory afferents is dynamic rather than static.
Thus, motor control depends largely on dynamic
parameters, constituting a more economic system
from a computational point of view. Indeed, when
a perturbation occurs, a system based only on po-
sition encoding would require an incompressible
integration time to measure positional changes
for comparison with a precise pre-determined
scheme, before counteracting the perturbation.
In contrast, a motor system in which detection
is based on joint movement will be able to ad-
just faster the ongoing movement or posture to
compensate perturbations. However, since purely
phasic coding only gives information about joint
movement without accounting for joint position,
a corrective system based exclusively on move-
ment detection would result in position shifting.
Nature solved this problem by adding two types
of afferents that combine movement and position
monitoring, thereby providing the CNS with both
the dynamic and static parameters of joint move-
ment relative to the whole body.
In vivo , CBCO coding appears much more
complicated and less precise than in vitro stud-
ies have suggested. Although the population of
sensory afferents obviously remains as diverse as

SENSORy-MOTOR INTEGRATION IN
PARALLEL PATHWAyS

To simplify, the basic output of a neural network
that generates a behavior results from the use of
sensory inputs through a combination of circuit
characteristics and intrinsic neuronal proper-
ties. For the control of the CB joint in crayfish,
sensory-motor integration is mainly achieved
through parallel pathways that convey CBCO
sensory information to the motor centers. The
organization of this sensory-motor network has
been studied extensively (Le Bon-Jego & Cattaert,
2002; Le Bon-Jego et al. , 2004; and see Cattaert
& Le Ray, 2001) and involves an extremely well
organized and specific wiring.
CBCO release-sensitive afferents (sensing
upward leg movements) excite monosynapti-
cally the depressor motoneurons (Dep MNs), and
stretch-sensitive afferents (sensing downward
leg movements) directly stimulate the levator
motoneurons (Lev MNs). These monosynaptic
connections are therefore responsible for nega-
tive feedback reflexes. Two parallel polysynaptic
pathways, comprising a dozen interneurons, also
participate in integrating the sensory inflow to
the motor command in either a resistance or an
assistance reflex mode. Added to the forty CBCO
afferents, this sensory-motor network thus com-
prises a small number of neurons, which are all
accessible to intracellular recordings in vitro .