142
C. Goldsmith et al.
of insects [8]. Drosophibot serves as a testbed for replicating neuromechanical
behaviors on a mechatronic platform, which we believe will eventually lead to
robust, adaptive locomotion.
One feature of the nervous system that makes animals so adaptable is their
ability to process sensory information in a task- or context-dependent way. One
thoroughly studied example of this processing in the control of posture and
locomotion is the apparent “reflex reversal” of insect joint control reflexes [5].
Briefly summarizing this behavior, stretching of the femoral chordotonal organ
(fCO) (signifying joint flexion) in a standing insect’s femur-tibia (FTi) joint will
result in a resistance reflex (RR) that attempts to halt the joint motion. However,
in an actively walking insect the same stimulus will cause an active reaction (AR)
where the muscles allow the flexion to proceed [3,9]. The exact mechanisms in the
nervous system that cause this reflex reversal are currently unknown; however,
several studies have hypothesized the importance of the groups of non spiking
interneurons (NSI) between the sensory neurons receiving stimuli from the fCO
and the slow extensor tibiae motorneuron (SETi) [12,13]. In particular, Sauer
et al. in ref. [11] found that blocking Chlorine ions (i.e. inhibition) in the fCO
sensory cells leads to changes in the NSI activities and eventually the MNs in
response to fCO stretching. They suggest that the nervous system may switch
between the RR in the resting state and the AR in the active state by selectively
inhibiting or disinhibiting fCO afferent cells. Driesang et al. in ref. [7] similarly
found that the NSI properties appear unchanged between these two contexts but
NSI activity changes, implying that their inputs are changing.
Traditional electrophysiological experimentation methods lack the precision
to definitively pinpoint the role of these sensory afferents in reflex reversal [11].
However, by using a neuromechanical model we can selectively inhibit or disinhibit position- or velocity-sensitive afferents with arbitrary precision, theoretically changing NSI activity as observed in the animal and transitioning our
joint controller function from the RR to the AR. B¨ assler et. al. have previously
attempted to functionally model reflex reversal in ref. [1] and used the model for
simulated joint control. However, as the morphology of the NSI sub-networks had
not been characterized at the time of the model’s development, the system lacks
morphological accuracy. Sauer et. al. later recorded the connectivity of the NSI
in ref. [13] and developed a model of the sub-system based on their data, but did
not use the model to control a limb or modify it to observe the response. To our
knowledge, no dynamic neuromechanical models of this particular sub-network
currently exist from which to observe the closed loop joint behavior.
To better understand the reflex reversal mechanism and how it could be
applied to robotics, we constructed a neuromechanical model of this reflex loop
and explored how its parameter values give rise to these distinct behaviors. In
this manuscript, we present the development of our simulation based on previous mappings of the connections to the SETi in stick insects. We validate the
model’s ability to replicate known biological behavior, then use the model to
selectively inhibit groups of position and velocity sensory neurons and observe
changes to the joint’s reflex in open and closed loop cases. Interestingly, when in
C. Goldsmith et al.
of insects [8]. Drosophibot serves as a testbed for replicating neuromechanical
behaviors on a mechatronic platform, which we believe will eventually lead to
robust, adaptive locomotion.
One feature of the nervous system that makes animals so adaptable is their
ability to process sensory information in a task- or context-dependent way. One
thoroughly studied example of this processing in the control of posture and
locomotion is the apparent “reflex reversal” of insect joint control reflexes [5].
Briefly summarizing this behavior, stretching of the femoral chordotonal organ
(fCO) (signifying joint flexion) in a standing insect’s femur-tibia (FTi) joint will
result in a resistance reflex (RR) that attempts to halt the joint motion. However,
in an actively walking insect the same stimulus will cause an active reaction (AR)
where the muscles allow the flexion to proceed [3,9]. The exact mechanisms in the
nervous system that cause this reflex reversal are currently unknown; however,
several studies have hypothesized the importance of the groups of non spiking
interneurons (NSI) between the sensory neurons receiving stimuli from the fCO
and the slow extensor tibiae motorneuron (SETi) [12,13]. In particular, Sauer
et al. in ref. [11] found that blocking Chlorine ions (i.e. inhibition) in the fCO
sensory cells leads to changes in the NSI activities and eventually the MNs in
response to fCO stretching. They suggest that the nervous system may switch
between the RR in the resting state and the AR in the active state by selectively
inhibiting or disinhibiting fCO afferent cells. Driesang et al. in ref. [7] similarly
found that the NSI properties appear unchanged between these two contexts but
NSI activity changes, implying that their inputs are changing.
Traditional electrophysiological experimentation methods lack the precision
to definitively pinpoint the role of these sensory afferents in reflex reversal [11].
However, by using a neuromechanical model we can selectively inhibit or disinhibit position- or velocity-sensitive afferents with arbitrary precision, theoretically changing NSI activity as observed in the animal and transitioning our
joint controller function from the RR to the AR. B¨ assler et. al. have previously
attempted to functionally model reflex reversal in ref. [1] and used the model for
simulated joint control. However, as the morphology of the NSI sub-networks had
not been characterized at the time of the model’s development, the system lacks
morphological accuracy. Sauer et. al. later recorded the connectivity of the NSI
in ref. [13] and developed a model of the sub-system based on their data, but did
not use the model to control a limb or modify it to observe the response. To our
knowledge, no dynamic neuromechanical models of this particular sub-network
currently exist from which to observe the closed loop joint behavior.
To better understand the reflex reversal mechanism and how it could be
applied to robotics, we constructed a neuromechanical model of this reflex loop
and explored how its parameter values give rise to these distinct behaviors. In
this manuscript, we present the development of our simulation based on previous mappings of the connections to the SETi in stick insects. We validate the
model’s ability to replicate known biological behavior, then use the model to
selectively inhibit groups of position and velocity sensory neurons and observe
changes to the joint’s reflex in open and closed loop cases. Interestingly, when in