Distributed Algorithm with Resilience for Multi-Agent Task Allocation

This paper develops an Iterated Auction-Consensus Algorithm (IACA) for resilient task allocation on a distributed multi-agent platform. The IACA enables each agent in the network to bid for tasks of interest at the auction stage and then utilizes the idea of consensus to resolve conflicts among agents' bids. Such a combination of auction and consensus enables each agent to repeatedly add tasks one by one to its task bundle until reaching its maximum bundle length. The major advantages of the IACA are twofold. One is the simplicity of implementation without requiring complicate decision logic. The other one is resilience, when there are malicious agents trying to block the normal agents' bids by giving infinitely large bids to all tasks, IACA is able to recover immediately after the removal of malicious agents. We investigate and compare costs for communications and computations by providing quantitative measures for distributed task allocation algorithms. Simulations are given for studying the tradeoff between communication and computational costs and the resilience of IACA in the presence of malicious agents.