Pro-active Advice to Improve the Efficiency of Self-Organizing Emergent Systems

Abstract

An advisor autonomously improves the efficiency of self-organizing emergent multi-agent systems at runtime. It identifies tasks that the system has to perform recurringly. If the system performs these tasks inefficiently, the advisor will create so-called exception rules for the agents that enable them to perform these tasks more efficiently in the future. In contrast to the previously presented concept of ignore rules, a type of exception rule which only keeps the agents from doing certain tasks, in this paper we present the concept of pro-active rules. This type of exception rule allows the advised agents to already prepare for tasks before they are even announced to the system. Our experimental evaluation shows that a combination of these two rule types for the domain of dynamic pickup and delivery problems utilizes the advantages of both, improves previously badly handled problem instances, and additionally offers slight improvements for randomly created instances.