Multi-Agent Collaboration in Adversarial Environments∗

This paper presents a strategy for developing dis-tributed algorithms for a class of problems by representing the problems as constrained opti-mizations and then employing a generic tech-nique for deriving distributed optimization algo-rithms. In the distributed constrained optimiza-tion algorithm, multiple agents collaborate to op-timize global, system-wide, objective functions while satisfying global constraints though each agent has only limited local information. The strategy can be employed in dynamic systems in which the numbers and locations of agents vary arbitrarily as computations proceed. This paper deals with problems in which agents operate in adversarial or uncertain environments and where the goal is to maximize the strength of the weak-est link in a system.