The Resilience Game: A New Formalization of Resilience for Groups of Goal-Oriented Autonomous Agents

Groups of autonomous robots should be resilient. They should have the ability to cope with unknown events, long-lasting alterations to the environment, degradation of capacities, robot losses, and changes to communication networks. This paper presents a multiagent resilience formulation for goal-based agents. The formulation applies to mixed motive groups where agent goals have commonalities but are not perfectly aligned. Resilient groups must not only be resilient to chance exogenous perturbations but also intentional endogenous perturbations among the agents. Defining resilience using expected utilities leads to a new way of looking at multiagent resilience, namely the resilience game. The resilience game makes it possible to use the notion of equilibrium from game theory to evaluate how the intentional stances of agents determine when multiagent algorithms are resilient. A guided diffusion of innovations problem is used to demonstrate how the resilience game provides insight into the effectiveness of various joint algorithms.