Efficient Allocation of Agent Groups for Complex Tasks in Real Cost Environments

In this paper we analyze and propose solutions for complex task allocation problems that have predetermined and known overall payments for any given task. Here a particular task is considered to be complete if all its relevant subtasks are assigned to agents with the necessary capabilities, and the total costs of all the agents falls within a preset limit. In previous work we prove that the problem is NP-hard and that for the general case of the non-cooperative setting, no protocol achieving the efficient solution can exist that is individually rational and budget balanced. Moreover, we show that although efficient protocols may exist in some settings, these will inevitably besetting-specific. Therefore, in this paper we analyze more specific, but nevertheless important, settings for which we develop protocols with the following main advantages. First, we prove these protocols to be individually rational, budget balanced and in equilibrium. Second, the performance of the protocols are evaluated via extensive experiments that show that they outperform previous solutions in this area in terms of efficiency and stability. Third, these protocols are proved to be polynomial in the number of subtasks and agents. Finally, as all proposed protocols are strategy proof or Bayesian Nash incentive compatible, the equilibrium agents' strategies are simply to declare their real costs and capabilities.