• Sorted by Date • Classified by Publication Type • Classified by Topic • Grouped by Student (current) • Grouped by Former Students •
Peleg Shefi. Bugs with Features: Resilient Vision-Based Collective Motion. Master's Thesis, Bar Ilan University,2025.
Collective motion is a widespread natural phenomenon in which individuals move in an ordered fashion without centralized control, and only highly localized interactions are possible. While such movement, in nature, is often considered robust to individual failures, most models of collective motion lack this resilience.Specifically, when vision-based perception is utilized (which is inherently imperfect), collective motion becomes brittle. This thesis addresses two symptoms of imperfect perception. First, we show that inability to estimate ego-motion leads to breakdown of order when agents can fail.Inspired by observations of locusts, we introduce intermittent pauses in individual motion, where the pause duration is tied to the perception of neighbors' movements. Then, we show that occluded perception and inaccurate distance estimation are just as disruptive. We show that combining information about the horizontal and vertical sizes of perceived neighbors can practically restore order.We demonstrate collective motion relying solely on monocular vision sensing, using an Avoid-Attract model that accounts for realistic constraints such as visual occlusions. Through physics-based simulations of swarm robots, we demonstrate that this approach significantly enhances the swarm's resilience to both individual failures and visual perception errors, without compromising its functionality or coherence.
@mastersthesis{peleg-msc,
author = {Peleg Shefi},
title = {Bugs with Features: Resilient Vision-Based Collective Motion},
school = {{B}ar {I}lan {U}niversity},
year = {2025},
wwwnote = {},
abstract = {
Collective motion is a widespread natural phenomenon in which individuals move in an ordered fashion without centralized control, and only highly localized interactions are possible.
While such movement, in nature, is often considered robust to individual failures, most models of collective motion lack this resilience.
Specifically, when vision-based perception is utilized (which is inherently imperfect), collective motion becomes brittle. This thesis addresses two symptoms of imperfect perception.
First, we show that inability to estimate ego-motion leads to breakdown of order when agents can fail.
Inspired by observations of locusts, we introduce intermittent pauses in individual motion, where the pause duration is tied to the perception of neighbors' movements.
Then, we show that occluded perception and inaccurate distance estimation are just as disruptive. We show that combining information about the horizontal and vertical sizes of perceived neighbors can practically restore order.
We demonstrate collective motion relying solely on monocular vision sensing, using an Avoid-Attract model that accounts for realistic constraints such as visual occlusions.
Through physics-based simulations of swarm robots, we demonstrate that this approach significantly enhances the swarm's resilience to both individual
failures and visual perception errors, without compromising its functionality or coherence.
},
}
Generated by bib2html.pl (written by Patrick Riley ) on Mon Feb 03, 2025 16:33:37