Display Publications by [Year] [Type] [Topic]
Hundreds of thousands of hours of delay, costing millions of dollars annually, are reported by airports in the US alone. The task of managing delay may be seen as a multiagent congestion problem with tightly coupled agents who collectively impact the system. Reward shaping is effective at improving agent learning for soft constraint problems by reducing learning noise caused by agent interactions, so we extend those results to hard constraints that cannot be easily learned, and must be enforced algorithmically. We present an agent partitioning algorithm in conjunction with reward shaping to simplify the learning domain. Our results show that an autonomous partitioning of the agents using system features leads to up to 1000x speed over simple hard constraint enforcement, as well as up to a 30\% improvement in performance over a greedy scheduling solution corresponding to hundreds of hours of delay saved in a single day.
@inproceedings{tumer-curran_aamas13,
author = {W. Curran and K. Tumer},
title = {Addressing Hard Constraints in the Air Traffic Problem through Partitioning and Difference Rewards (extended abstract)},
booktitle = {Proceedings of the Twelveth International Joint Conference on Autonomous Agents and Multiagent Systems},
month = {May},
address = {Minneapolis, MN},
abstract={Hundreds of thousands of hours of delay, costing millions of dollars annually, are reported by airports in the US alone. The task of managing delay may be seen as a multiagent congestion problem with tightly coupled agents who collectively impact the system. Reward shaping is effective at improving agent learning for soft constraint problems by reducing learning noise caused by agent interactions, so we extend those results to hard constraints that cannot be easily learned, and must be enforced algorithmically. We present an agent partitioning algorithm in conjunction with reward shaping to simplify the learning domain. Our results show that an autonomous partitioning of the agents using system features leads to up to 1000x speed over simple hard constraint enforcement, as well as up to a 30\% improvement in performance over a greedy scheduling solution corresponding to hundreds of hours of delay saved in a single day. },
bib2html_pubtype = {Refereed Conference Papers},
bib2html_rescat = {Multiagent Systems, Reinforcement Learning, Air Traffic Control},
year = {2013}
}
Generated by bib2html.pl (written by Patrick Riley ) on Wed Apr 01, 2020 17:39:43