2016
Regular Strategies and Strategy Improvement: Efficient Tools for Solving Large Patrolling Problems
KUČERA, Antonín a Tomáš LAMSERZákladní údaje
Originální název
Regular Strategies and Strategy Improvement: Efficient Tools for Solving Large Patrolling Problems
Autoři
KUČERA, Antonín (203 Česká republika, garant, domácí) a Tomáš LAMSER (203 Česká republika, domácí)
Vydání
New York, Proceedings of the 2016 International Conference on Autonomous Agents & Multiagent Systems, od s. 1171-1179, 9 s. 2016
Nakladatel
ACM
Další údaje
Jazyk
angličtina
Typ výsledku
Stať ve sborníku
Obor
10201 Computer sciences, information science, bioinformatics
Stát vydavatele
Spojené státy
Utajení
není předmětem státního či obchodního tajemství
Forma vydání
tištěná verze "print"
Odkazy
Kód RIV
RIV/00216224:14330/16:00088484
Organizační jednotka
Fakulta informatiky
ISBN
978-1-4503-4239-1
ISSN
UT WoS
000465199800134
Klíčová slova anglicky
patrolling games; strategy synthesis
Příznaky
Mezinárodní význam, Recenzováno
Změněno: 13. 5. 2020 19:32, RNDr. Pavel Šmerk, Ph.D.
Anotace
V originále
In patrolling problems, the task is to compute an optimal strategy for a patroller who moves among vulnerable targets and aims at detecting possible intrusions. Previous approaches to this problem utilize non-linear programming to synthesize (sub)optimal patroller's strategies, which has a negative impact on their scalability. Further, the solution space is usually restricted to positional strategies or to strategies dependent on a bounded history of patroller's moves. In this paper we introduce regular strategies that utilize deterministic finite-state automata to collect some information about the whole history of patroller's moves, and show that regular strategies are strictly more powerful than strategies dependent on a bounded history. Further, we design a strategy improvement technique for regular strategies which completely avoids solving large non-linear programs. Intuitively, we start with some regular strategy, and then improve this strategy by performing a finite number of rounds, where each round produces another regular strategy obtained by combining the ``old'' one with a solution of a certain linear program. Our experiments demonstrate that, compared to the existing methods, our approach is applicable to patrolling problems of considerably larger size, and can quickly produces strategies of very good quality.
Návaznosti
GA15-17564S, projekt VaV |
|