Insect-Inspired Navigation Algorithm for an Aerial Agent Using Satellite Imagery

Abstract

Humans have long marveled at the ability of animals to navigate swiftly, accurately, and across long distances. Many mechanisms have been proposed for how animals acquire, store, and retrace learned routes, yet many of these hypotheses appear incongruent with behavioral observations and the animals’ neural constraints. The “Navigation by Scene Familiarity Hypothesis” proposed originally for insect navigation offers an elegantly simple solution for retracing previously experienced routes without the need for complex neural architectures and memory retrieval mechanisms. This hypothesis proposes that an animal can return to a target location by simply moving toward the most familiar scene at any given point. Proof of concept simulations have used computer-generated ant’s-eye views of the world, but here we test the ability of scene familiarity algorithms to navigate training routes across satellite images extracted from Google Maps. We find that Google satellite images are so rich in visual information that familiarity algorithms can be used to retrace even tortuous routes with low-resolution sensors. We discuss the implications of these findings not only for animal navigation but also for the potential development of visual augmentation systems and robot guidance algorithms.

High-throughput simulations indicate feasibility of navigation by familiarity with a local sensor such as scorpion pectines

“…The hypothesis appears congruent with 61 numerous behavioral observations [29][30][31][32][33][34][35][36]. Furthermore, there are now robots and 62 simulated agents that navigate autonomously via familiarity [1,18,[37][38][39], and some 63 studies [40, 41] have suggested how neural tissue, such as the central complex and 64 mushroom bodies [42-51], might be organized to accommodate familiarity-based 65 navigation. 66 June 10, 2020 5/34Navigation by familiarity with a local sensor 67In this paper, we consider the hypothesis that the dense fields of peg sensilla on pectines 68 are analogous to the tightly packed ommatidia in compound eyes, detecting matrices of 69 chemical and textural information that are used for navigation by familiarity.…”

“…When they have different 172 chemical identities, the summand is the sum of their concentrations. This yields the 173 intuitive result that there is a bigger difference between a high concentration of sensors in [1,37,52]. In an effort to increase realism, some researchers [1,28,40] The code for our model and visualizations is open-source and publicly available [56]; 183 details on its optimized implementation can be found in section 4 of S1 Appendix.…”

“…where it is necessary to distinguish between many more glimpses, having more chemical 454 information could become unambiguously critical for navigational success. 455 We propose, however, that these results can be explained through a fundamental 456 interplay in familiarity navigation between "complexity" -the need to distinguish 457 between glimpses in different places and orientations -and "continuity" -the need 458 for small offsets to minimally affect familiarity [37,52] small compared to the size of the sensor, continuity will be low but the landscape will continuity. In the case of β = 1.0, texture is completely ignored and the imbalance is 479 even more pronounced, causing navigation to become essentially impossible (see 480 Figure 5).…”

Autonomous Visual Navigation of an Indoor Environment Using a Parsimonious, Insect Inspired Familiarity Algorithm

“…The environmental information accesses the appropriate stored neural engram and links the appropriate scene sequence. We recently used a scene familiarity based approach to develop a simple algorithm for recapitulating complex routes in a visual information landscape obtained from 2D images from a downward facing field of view (satellite) set to approximately 250 meters above the Earth [ 18 ]. This approach successfully recapitulated routes using visual information not only from complex landscapes such as cities, but also from areas with much less visual information such as lakes and deserts.…”

“…In previous work, successful recapitulation of learned routes in simulated scenes was related to several factors, including sensor resolution, distance of objects/information from sensor, degree of navigational rotation, and the amount of visual information within the landscape [ 15 , 16 , 18 – 20 ]. However, there is a dearth of information detailing the interplay between these factors and navigation success in natural environments.…”