Insect Inspired View Based Navigation
Exploiting Temporal Information
Efstathios Kagioulis
(B) , Andrew Philippides , Paul Graham ,
James C. Knight , and Thomas Nowotny
Centre for Computational Neuroscience and Robotics, University of Sussex,
Brighton BN1 9QJ, UK
e.kagioulis@sussex.ac.uk
Abstract. Visual navigation is a key capability for robots. There is a
family of insect-inspired algorithms that use panoramic images encountered during a training route to derive directional information from
regions around the training route and thus subsequently visually navigate. As these algorithms do not incorporate information about the
temporal order of training images, we describe one way this could be
done to highlight this information’s utility. We benchmark our algorithms
in a simulation of a real world environment and show that incorporating temporal information improves performance and reduces algorithmic
complexity.
Keywords: Visual navigation · Visual homing · Bio-inspired robotics
1 Introduction
Foraging ants have impressive visual navigation capabilities [26]. They can use
celestial visual information as a compass for Path Integration (PI) [19] and use
learnt visual memories to guide complex routes [15] between their nest and a
food source. The ability to travel long routes, guided by stored visual scenes,
prompts the question of how the visual memories are organised in the small
brains of insects [2].
We and others have developed a series of algorithms mimicking ant route
navigation in which ants use a set of views or snapshots experienced during a first
(PI-mediated) training route to later navigate by rotating and comparing the
rotated current view with the stored training route views [2,13,22]. Because the
route memories are stored when travelling facing forwards, adopting a heading
which is familiar to them implicitly means that the agent is likely near the
route and facing in a similar direction as when last near that location. In this
way, finding familiar views means recovering the correct direction to move in and
thus routes can be successfully navigated. While these algorithms vary in the way
This work was supported by EPSRC grants EP/P006094/1 and EP/S030964/1 and a
University of Sussex doctoral scholarship.
c
Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 204–216, 2020.
https://doi.org/10.1007/978-3-030-64313-3_20
Exploiting Temporal Information
Efstathios Kagioulis
(B) , Andrew Philippides , Paul Graham ,
James C. Knight , and Thomas Nowotny
Centre for Computational Neuroscience and Robotics, University of Sussex,
Brighton BN1 9QJ, UK
e.kagioulis@sussex.ac.uk
Abstract. Visual navigation is a key capability for robots. There is a
family of insect-inspired algorithms that use panoramic images encountered during a training route to derive directional information from
regions around the training route and thus subsequently visually navigate. As these algorithms do not incorporate information about the
temporal order of training images, we describe one way this could be
done to highlight this information’s utility. We benchmark our algorithms
in a simulation of a real world environment and show that incorporating temporal information improves performance and reduces algorithmic
complexity.
Keywords: Visual navigation · Visual homing · Bio-inspired robotics
1 Introduction
Foraging ants have impressive visual navigation capabilities [26]. They can use
celestial visual information as a compass for Path Integration (PI) [19] and use
learnt visual memories to guide complex routes [15] between their nest and a
food source. The ability to travel long routes, guided by stored visual scenes,
prompts the question of how the visual memories are organised in the small
brains of insects [2].
We and others have developed a series of algorithms mimicking ant route
navigation in which ants use a set of views or snapshots experienced during a first
(PI-mediated) training route to later navigate by rotating and comparing the
rotated current view with the stored training route views [2,13,22]. Because the
route memories are stored when travelling facing forwards, adopting a heading
which is familiar to them implicitly means that the agent is likely near the
route and facing in a similar direction as when last near that location. In this
way, finding familiar views means recovering the correct direction to move in and
thus routes can be successfully navigated. While these algorithms vary in the way
This work was supported by EPSRC grants EP/P006094/1 and EP/S030964/1 and a
University of Sussex doctoral scholarship.
c
Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 204–216, 2020.
https://doi.org/10.1007/978-3-030-64313-3_20