Influence of inter-agent variation on system redundancy in multiagent systems

The goal of our research is to understand and model the effects of inter-agent variation in adaptively building and maintaining a back up pool of agents in multiagent systems (MAS). Redundancy is important in maintaining system stability during pertubations. We consider a decentralized MAS to which tasks are assigned. Specifically, we consider tasks in which agents gain experience by performing the tasks, and where past experience improves future performance. We refer to redundancy as the presence of a pool of extra agents with work experience. The motivation for our research is drawn from how we, as humans, build a back up team of workers with experience. Consider the following example of a football team. The gameday roster consists of 45 players of whom eleven play at any given time. Let it be the case that the team has eleven best players. The team achieves best performance each time it plays with these eleven. If, unfortunately, one or more members of this best team are unable to play, then the team no longer has the best performance. However, loss of a team’s performance is seldom seen due the presence of the back up players with experience for the eleven best players. Thus, the presence of back up players with experience contributes to maintaining team performance. The key to our investigation is balancing performance versus redundancy, that is, balancing the optimization of team performance versus the creation of a back up pool. When the best players always play, the team performance is at its best. The team, however, is unstable without the back up pool during perturbations. When the players in the back up pool play, the back up players gain experience. The team, however, is not capable of best performance. We investigate a method that uses inter-agent variation to balance the team performance and system redundancy. We assume a response threshold-based MAS [1]. Each agent has a threshold at which it responds to a task’s stimulus. Agents with