An anytime algorithm for dynamic multi-agent task allocation problems

This paper explores the dynamic multi-agent task allocation problem in an open and dynamic environment. Some existing works on task allocation have not been able to cater for the extra needs of scalability and robustness in large scale complex systems. Furthermore, some algorithms highly rely on the agents' organisation structure. We proposes a novel algorithm based on the notions of potential contributions, values and utilities of the individual agents and the system. A game structure is constructed to model the task allocation problem, based on which a novel algorithm POT is then proposed to search for a Nash equilibrium solution that maximises the system's potential utility. We carry out comprehensive empirical studies to demonstrate that POT achieves a higher system utility compared with the state-of-the-art GreedyNE algorithm.