Event
MSSE Thesis Defense: Akshay Bapat, "Decentralized Task Algorithms for Multi-Agent Systems"
Wednesday, April 15, 2020
12:00 p.m.
Zoom online meeting
Akshay Bapat
abapat@umd.edu
https://umd.zoom.us/j/5485097669
M.S. Thesis Defense: Akshay Bapat
Development of Decentralized Task Allocation Algorithms for Multi-Agent Systems with Very Low Communication
Abstract
Existing decentralized task allocation algorithms perform poorly when the communication availability is very low. Although previous work has considered task allocation algorithms in the presence of imperfect communication, the case of very low communication has not yet been addressed. In this thesis, we study the cases when the instantaneous probability p of a successful message between agents satisfies p << 0.01. We simulate the communication based on three models: Bernoulli model, Gilbert-Elliot model and Rayleigh Fading model.
We present two new algorithms: the Spatial Division Playbook Algorithm and the Travelling Salesman Playbook Algorithm, which work by assuming that communications may not happen, but then derive advantages whenever communications are successful. We evaluate these algorithms on a target visit scenario, comparing performance with three state-of-the-art algorithms - ACBBA, DHBA and PIA - across five levels of very low communication availability and six levels of the number of targets. Our results show that the algorithms perform better than the other algorithms and reduce the time required to ensure all targets are visited.
Existing decentralized task allocation algorithms perform poorly when the communication availability is very low. Although previous work has considered task allocation algorithms in the presence of imperfect communication, the case of very low communication has not yet been addressed. In this thesis, we study the cases when the instantaneous probability p of a successful message between agents satisfies p << 0.01. We simulate the communication based on three models: Bernoulli model, Gilbert-Elliot model and Rayleigh Fading model.
We present two new algorithms: the Spatial Division Playbook Algorithm and the Travelling Salesman Playbook Algorithm, which work by assuming that communications may not happen, but then derive advantages whenever communications are successful. We evaluate these algorithms on a target visit scenario, comparing performance with three state-of-the-art algorithms - ACBBA, DHBA and PIA - across five levels of very low communication availability and six levels of the number of targets. Our results show that the algorithms perform better than the other algorithms and reduce the time required to ensure all targets are visited.
Committee Members
Dr. Jeffrey W. Herrmann (Chair) - Department of Mechanical Engineering
Dr. Michael W. Otte - Department of Aerospace Engineering
Dr. Nikhil Chopra - Department of Mechanical Engineering
Zoom instructions
Join Zoom Meeting
https://umd.zoom.us/j/5485097669
Meeting ID: 548 509 7669
One tap mobile
+19294362866,,5485097669# US (New York)
+13126266799,,5485097669# US (Chicago)
Dial by your location
+1 929 436 2866 US (New York)
+1 312 626 6799 US (Chicago)
+1 669 900 6833 US (San Jose)
+1 253 215 8782 US
+1 301 715 8592 US
+1 346 248 7799 US (Houston)
Meeting ID: 548 509 7669
Find your local number: https://umd.zoom.us/u/aeeXAgajfo