Can Small Scale Search Behaviours
Enhance Large-Scale Navigation?
Fabian Steinbeck
(B) , Paul Graham , Thomas Nowotny ,
and Andrew Philippides
University of Sussex, Brighton, UK
f.steinbeck@sussex.ac.uk
Abstract. We develop a spiking neural network model of an insectinspired CPG which is used to underpin steering behaviour for a
Braitenberg-like vehicle. We show that small scale search behaviour, produced by the CPG, improves navigation by recovering useful sensory
signals.
Keywords: CPG · Braitenberg · Navigation
1 Introduction
Motor control is fundamental to the adaptive behaviour of natural and artificial
systems, with one role being the production of active movement strategies to
acquire and use key sensory information. A key neural component of motor
systems are central pattern generators (CPGs) [1], for instance, those involved
in the control of swimming movements of lamprey [2] or movement control in
invertebrates [3]. CPGs have also been shown to be useful components for motor
control in bio-inspired robots [4]. In insect brains, the Lateral Accessory Lobe is a
key pre-motor area that has been shown to generate CPG like outputs in certain
conditions. It is a conserved brain structure and is fundamental to a range of
sensori-motor behaviours such as pheromone search in moths [5] and phonotaxis
in crickets [6]. Computational models of the LAL have been developed to obtain
a better understanding of how the LAL network can generate “flipflop” activity
(Adden et al. (in prep.)) or how the LAL may contribute to pheromone plume
tracking [7]. Here we develop a minimal spiking neural network model of the
LAL, based on our previously developed general steering framework [8]. Our
aim is to demonstrate that the model produces outputs that drive two distinct
behavioural modes. Firstly, in the presence of sensory information it should
output a steering signal that is proportional to that sensory information, so the
location of a stimulus relative to the agent drives the steering. Secondly, in the
absence of reliable sensory information, the network should produce a rhythmic
output that can drive search patterns. We explore the adaptive properties of this
network, by situating it in a simple Braitenberg-style animat (Fig. 1).
This work was funded by the EPSRC (grant EP/P006094/1).
c
Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 338–341, 2020.
https://doi.org/10.1007/978-3-030-64313-3_32
Enhance Large-Scale Navigation?
Fabian Steinbeck
(B) , Paul Graham , Thomas Nowotny ,
and Andrew Philippides
University of Sussex, Brighton, UK
f.steinbeck@sussex.ac.uk
Abstract. We develop a spiking neural network model of an insectinspired CPG which is used to underpin steering behaviour for a
Braitenberg-like vehicle. We show that small scale search behaviour, produced by the CPG, improves navigation by recovering useful sensory
signals.
Keywords: CPG · Braitenberg · Navigation
1 Introduction
Motor control is fundamental to the adaptive behaviour of natural and artificial
systems, with one role being the production of active movement strategies to
acquire and use key sensory information. A key neural component of motor
systems are central pattern generators (CPGs) [1], for instance, those involved
in the control of swimming movements of lamprey [2] or movement control in
invertebrates [3]. CPGs have also been shown to be useful components for motor
control in bio-inspired robots [4]. In insect brains, the Lateral Accessory Lobe is a
key pre-motor area that has been shown to generate CPG like outputs in certain
conditions. It is a conserved brain structure and is fundamental to a range of
sensori-motor behaviours such as pheromone search in moths [5] and phonotaxis
in crickets [6]. Computational models of the LAL have been developed to obtain
a better understanding of how the LAL network can generate “flipflop” activity
(Adden et al. (in prep.)) or how the LAL may contribute to pheromone plume
tracking [7]. Here we develop a minimal spiking neural network model of the
LAL, based on our previously developed general steering framework [8]. Our
aim is to demonstrate that the model produces outputs that drive two distinct
behavioural modes. Firstly, in the presence of sensory information it should
output a steering signal that is proportional to that sensory information, so the
location of a stimulus relative to the agent drives the steering. Secondly, in the
absence of reliable sensory information, the network should produce a rhythmic
output that can drive search patterns. We explore the adaptive properties of this
network, by situating it in a simple Braitenberg-style animat (Fig. 1).
This work was funded by the EPSRC (grant EP/P006094/1).
c
Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 338–341, 2020.
https://doi.org/10.1007/978-3-030-64313-3_32