TY - GEN
T1 - Automated verification of social law robustness for reactive agents
AU - Tuisov, Alexander
AU - Karpas, Erez
N1 - Publisher Copyright: © 2020 The authors and IOS Press.
PY - 2020/8/24
Y1 - 2020/8/24
N2 - Coordinating agents in a multi-agent system is an interesting and important challenge. One of the most effective methods of coordinating multiple agents is using a 'social law', which restricts some possible behaviors in order to ensure every agent can achieve its goal. Recent work has connected social laws with automated planning, and shown how to verify if a given social law is robust, that is, ensures each agent can achieve its goal regardless of the plans chosen by the other agents. This prior work assumed the agents choose a plan offline, and never modify it in response to the other agents' actions. In this paper, we address reactive agents, that is, agents that can reconsider their course of action during execution. This setting presents a new challenge, as agents now have the possibility of entering an infinite loop (a livelock) in which each agent replans in the same way in response to the other agents. We show how to verify if a given social law is robust in such a setting, and our main contribution is a compilation which eliminates the need to keep track of each agent's current plan and constitutes a backbone of the verification algorithm.
AB - Coordinating agents in a multi-agent system is an interesting and important challenge. One of the most effective methods of coordinating multiple agents is using a 'social law', which restricts some possible behaviors in order to ensure every agent can achieve its goal. Recent work has connected social laws with automated planning, and shown how to verify if a given social law is robust, that is, ensures each agent can achieve its goal regardless of the plans chosen by the other agents. This prior work assumed the agents choose a plan offline, and never modify it in response to the other agents' actions. In this paper, we address reactive agents, that is, agents that can reconsider their course of action during execution. This setting presents a new challenge, as agents now have the possibility of entering an infinite loop (a livelock) in which each agent replans in the same way in response to the other agents. We show how to verify if a given social law is robust in such a setting, and our main contribution is a compilation which eliminates the need to keep track of each agent's current plan and constitutes a backbone of the verification algorithm.
UR - http://www.scopus.com/inward/record.url?scp=85091752523&partnerID=8YFLogxK
U2 - https://doi.org/10.3233/FAIA200369
DO - https://doi.org/10.3233/FAIA200369
M3 - منشور من مؤتمر
T3 - Frontiers in Artificial Intelligence and Applications
SP - 2386
EP - 2393
BT - ECAI 2020 - 24th European Conference on Artificial Intelligence, including 10th Conference on Prestigious Applications of Artificial Intelligence, PAIS 2020 - Proceedings
A2 - De Giacomo, Giuseppe
A2 - Catala, Alejandro
A2 - Dilkina, Bistra
A2 - Milano, Michela
A2 - Barro, Senen
A2 - Bugarin, Alberto
A2 - Lang, Jerome
T2 - 24th European Conference on Artificial Intelligence, ECAI 2020, including 10th Conference on Prestigious Applications of Artificial Intelligence, PAIS 2020
Y2 - 29 August 2020 through 8 September 2020
ER -