Agent coordination with regret clearing

Sequential single-item auctions can be used for the distributed allocation of tasks to cooperating agents. We study how to improve the team performance of sequential single-item auctions while still controlling the agents in real time. Our idea is to assign that task to agents during the current round whose regret is large, where the regret of a task is defined as the difference of the second-smallest and smallest team costs resulting from assigning the task to the second-best and best agent, respectively. Our experimental results show that sequential single-item auctions with regret clearing indeed result in smaller team costs than standard sequential single-item auctions for three out of four combinations of two different team objectives and two different capacity constraints (including no capacity constraints).