Robustness for MultiAgent Systems in Hostile Environments

Abstract: A MultiAgent System (MAS) is composed of multiple interacting agents, where the system purpose is met through achievement of individual agent goals. In a hostile environment, agents are subject to debilitation; resultant inability to achieve goals may have wide ranging consequences across the system. Approaches such as plan repair or conditional planning are invoked when a goal has been selected and failure has occurred; addressing the consequences of debilitation rather than the root causes of why actions fail to achieve the desired result. This work proposes a model relating capabilities to potential goals, including those from agent relationships. It is intended to improve robustness by allowing specification of pre-emptive, maintenance behaviour to prevent current debilitations from being al-lowed to cause future problems in MAS operations