Download PDFOpen PDF in browser

Insect-Inspired Visual Navigation On-Board Autonomous Robots

EasyChair Preprint 2323

2 pagesDate: January 6, 2020

Abstract

Small-brained insects are expert at many tasks that are currently difficult for robots especially in dealing with real-world, dynamic environments. In particular, the speed and robustness of insect learning is in stark contrast to many AI methods which take long times to train and require large amounts of labelled data. For example, the desert ant, Melophoros bagoti, is a champion navigator, able to visually navigate routes of up to 100 m through complex habitats [1] with a performance far above current map-based methods such as SLAM (simultaneous localization and mapping). Remarkably, these ants are able to learn the information needed to visually navigate 10’s of metres after only a single exposure to the route information [2]. They achieve this feat despite brains of only ~500,000 neurons and a kilo-pixel visual system (see Fig 1A for examples of low-resolution images), and this makes them ideal inspiration for engineers seeking to design resource-light control algorithms for small robots whose size makes power efficiency paramount [1]. In this spirit, we have developed a visual route navigation algorithm which replicates many aspects of ant navigation with biologically plausible computation and memory requirements (Fig 1A and [3-4]). These resource constraints mean first, that our algorithms can be used on-board fully autonomous small robots, and second, that they are amenable to swarm robotic approaches.

Keyphrases: Autonomous Robotics, Bio-inspired robotics, ants, visual navigation

BibTeX entry
BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:
@booklet{EasyChair:2323,
  author    = {Andrew Philippides and James Knight and Alex Dewar and Norbert Domcsek and Efstathios Kagioulis and Stefan Meyer and Thomas Nowotny and Paul Graham},
  title     = {Insect-Inspired Visual Navigation On-Board Autonomous Robots},
  howpublished = {EasyChair Preprint 2323},
  year      = {EasyChair, 2020}}
Download PDFOpen PDF in browser