Partial agreement task assignment algorithm for secure plan consensus in multi-agent system

Abstract

This paper presents a secure market based task assignment algorithm for decentralized multi-agent system. In a decentralized algorithm, individual agents calculate their solution with imperfect information of environments. To achieve agreement in the system, the plan consensus solves local task assignment problems and reach a conflict-free solution through the plan exchange. The Consensus Based Bundle Algorithm(CBBA), which is well-known consensus based plan auction method, guarantees feasible and conflict-free solution. However, in the plan consensus protocol, each agent determines tasks only through bidding for each task like maximum bid consensus. It is vulnerable to malicious data to attack a consensus process or solution result. Therefore, providing security to the decentralized agent system is a major issue. The main contribution of this paper is an algorithm, termed partial-agreement consensus-based bundle algorithm, that proposed for two types of attacks, exaggerated score attack and changing score attack. Partial agreement of the agent with some of the mission plans resists attacks that are intended to degrade the overall plan. The security solution is a more robust task allocation algorithm with a partial agreement on the plans and task bidding data.