• Sorted by Date • Classified by Publication Type • Classified by Topic • Grouped by Student (current) • Grouped by Former Students •
David L. Krongauz. Vision-Based Collective Motion: A Locust Inspired Reductionist Approach. Master's Thesis, Bar
Ilan University,2023.
Naturally occurring collective motion is an omnipresent, fascinating phenomenon that has been long studied within different scientific fields. Swarming individuals are believed to aggregate, coordinate and move utilizing only local social cues projected by conspecifics in their vicinity. Major theoretical studies of this phenomenon assume perfect information availability, where agents rely on complete and exact knowledge of inter-agent distances and velocities. However, the sensory modalities that are responsible for the acquisition of the environmental information in nature are often ignored. Vision plays a central role in animal perception, and in many cases of collective motion it is assumed to be the sole source of social information. We investigate a vision-based collective motion model, i.e., a model relying on visually available parameters only, inspired by the case study of locust marching bands. We address two major challenges: the estimation of distance and velocity, and visual occlusions. We consider and compare three strategies an agent can use to interpret partially occluded visual information. In silico experiments were conducted in two different frameworks: the first simulating physical entities in a square arena, and the second simulating simplified two-dimensional agents moving under various geometrical conditions. The results show the feasibility of our model and its three occlusion handling approaches for overcoming occlusion. While all the models display convergence to an ordered state, they differ in the respective computational requirements they demand from an agent: the least computationally demanding approach, in which no object detection is taking place, shows slower convergence to order. This is mostly apparent in geometrically constrained environments, which may hint as to the requirements from biological swarming species in natural settings.
@mastersthesis{krongauz-msc,
author = {David L. Krongauz},
title = {Vision-Based Collective Motion: A Locust Inspired Reductionist Approach},
school = {{B}ar {I}lan {U}niversity},
year = {2023},
OPTkey = {},
OPTtype = {},
OPTaddress = {},
OPTmonth = {},
OPTnote = {Available at \url{http://www.cs.biu.ac.il/~galk/Publications/b2hd-krongauz-msc.html}},
OPTannote = {},
wwwnote = {},
abstract = { Naturally occurring collective motion is an omnipresent, fascinating phenomenon that has been long studied within different scientific fields. Swarming individuals are believed to aggregate, coordinate and move utilizing only local social cues projected by conspecifics in their vicinity. Major theoretical studies of this phenomenon assume perfect information availability, where agents rely on complete and exact knowledge of inter-agent distances and velocities. However, the sensory modalities that are responsible for the acquisition of the environmental information in nature are often ignored. Vision plays a central role in animal perception, and in many cases of collective motion it is assumed to be the sole source of social information.
We investigate a vision-based collective motion model, i.e., a model relying on visually available parameters only, inspired by the case study of locust marching bands. We address two major challenges: the estimation of distance and velocity, and visual occlusions. We consider and compare three strategies an agent can use to interpret partially occluded visual information. \textit{In silico} experiments
were conducted in two different frameworks: the first simulating physical entities in a square arena, and the second simulating simplified two-dimensional agents moving under various geometrical conditions. The results show the feasibility of our model and its three occlusion handling approaches for overcoming occlusion. While all the models display convergence to an ordered state, they differ in the respective computational requirements they demand from an agent: the least computationally demanding approach, in which no object detection is taking place, shows slower convergence to order. This is mostly apparent in geometrically constrained environments, which may hint as to the requirements from biological swarming species in natural settings.
},
}
Generated by bib2html.pl (written by Patrick Riley ) on Fri Apr 19, 2024 19:01:33