A Synthetic Nervous System Model of the Insect
Optomotor Response
Anna Sedlackova, Nicholas S. Szczecinski (B) , and Roger D. Quinn
Case Western Reserve University, Cleveland, OH 44106, USA
nss36@case.edu
Abstract. We seek to increase the sophistication of our insect-like hexapod robot
MantisBot’s visual system. We assembled and tested a benchtop robotic testbed
with which to test our dynamical neural model of the insect visual system. Here
we specifically model wide-field vision and the optomotor response. The system is
composed of a Raspberry Pi with a camera outfitted with a 360° lens. The camera
sits on a motorized turntable, which represents the “robot”. Above the turntable
sits another motorized system that rotates a drum with printed patterns around the
camera, which represents the visual “background”. The camera downsamples the
visual scene and sends it to a synthetic nervous system (SNS) model of the insect
optic lobe. The optic lobe is columnar. Each column detects changes in receptor
intensity (retina), inhibits adjacent columns to increase dynamic range (lamina),
compares time-delayed activities of adjacent columns to detect motion (medulla),
then pools the motion of each column in a directionally-specific connectivity to
compute the direction and speed of the wide-field scene (lobula plate). Our robotic
model successfully encodes lateral wide-field visual speed into the activity of a pair
of opposing Lobula Plate Tangential Cells (LPTCs). Furthermore, the optomotor
response can be recreated by using the LPTCs to stimulate the neck motor neurons
(MNs), producing a real-time, closed-loop dynamical model of the optomotor
response.
Keywords: Synthetic nervous system · Optomotor response · Insect · Vision
1 Introduction
Despite their miniature brains of less than a million neurons, insects are able to solve
complex vision tasks- locating prey, avoiding obstacles, tracking prey or mates - all with
the use of environmental cues [1]. Insects can still outperform man-made robots and
systems in visual tasks despite their limited metabolic and computational power [2]. The
basis of this performance is the optic lobe of the insect brain, which uses a parallelized
system of functionally-distinct layers to process visual input. Insects’ ability to compute
optic flow, also called “wide-field vision,” has been thoroughly studied [3], but gaps in
the knowledge remain. To test how well the current understanding of these networks
can explain their function [4], we built an anatomically-constrained model of the optic
lobe using our “synthetic nervous system” (SNS) approach, and use it as the basis of a
robotic model of the insect optomotor response.
© Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 312–324, 2020.
https://doi.org/10.1007/978-3-030-64313-3_30
Optomotor Response
Anna Sedlackova, Nicholas S. Szczecinski (B) , and Roger D. Quinn
Case Western Reserve University, Cleveland, OH 44106, USA
nss36@case.edu
Abstract. We seek to increase the sophistication of our insect-like hexapod robot
MantisBot’s visual system. We assembled and tested a benchtop robotic testbed
with which to test our dynamical neural model of the insect visual system. Here
we specifically model wide-field vision and the optomotor response. The system is
composed of a Raspberry Pi with a camera outfitted with a 360° lens. The camera
sits on a motorized turntable, which represents the “robot”. Above the turntable
sits another motorized system that rotates a drum with printed patterns around the
camera, which represents the visual “background”. The camera downsamples the
visual scene and sends it to a synthetic nervous system (SNS) model of the insect
optic lobe. The optic lobe is columnar. Each column detects changes in receptor
intensity (retina), inhibits adjacent columns to increase dynamic range (lamina),
compares time-delayed activities of adjacent columns to detect motion (medulla),
then pools the motion of each column in a directionally-specific connectivity to
compute the direction and speed of the wide-field scene (lobula plate). Our robotic
model successfully encodes lateral wide-field visual speed into the activity of a pair
of opposing Lobula Plate Tangential Cells (LPTCs). Furthermore, the optomotor
response can be recreated by using the LPTCs to stimulate the neck motor neurons
(MNs), producing a real-time, closed-loop dynamical model of the optomotor
response.
Keywords: Synthetic nervous system · Optomotor response · Insect · Vision
1 Introduction
Despite their miniature brains of less than a million neurons, insects are able to solve
complex vision tasks- locating prey, avoiding obstacles, tracking prey or mates - all with
the use of environmental cues [1]. Insects can still outperform man-made robots and
systems in visual tasks despite their limited metabolic and computational power [2]. The
basis of this performance is the optic lobe of the insect brain, which uses a parallelized
system of functionally-distinct layers to process visual input. Insects’ ability to compute
optic flow, also called “wide-field vision,” has been thoroughly studied [3], but gaps in
the knowledge remain. To test how well the current understanding of these networks
can explain their function [4], we built an anatomically-constrained model of the optic
lobe using our “synthetic nervous system” (SNS) approach, and use it as the basis of a
robotic model of the insect optomotor response.
© Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 312–324, 2020.
https://doi.org/10.1007/978-3-030-64313-3_30