Automatically Defined Swarms for Task Allocation

摘要

Insect inspired task allocation schemes have received significant attention as a way to control agents in dynamic or uncertain domains. This is largely because such mechanisms rely on only simple definitions of agent behavior, a small amount of communication and a high-degree of fault tolerance. However, it is often difficult to conceptualize the appropriate learning and decision rules for these agents since in the case of swarm-intelligence approaches, the focus is not on an individual agent's ability to optimize its behavior, but on the resulting performance of the entire complex system. Although there have been successes in a variety of domains in the past, many of these approaches have required considerable effort by the researcher to tailor the canonical definition to the specific problem at hand. This paper presents a generalized framework for solving multi-agent task allocation problems using the insect-inspired model. I then show that because of the inherent simplicity of the agent's design, we can automatically define these learning and decision rules. A multi-robot task allocation experiment has been defined and performed. The results show how these automatically-defined behaviors outperform existing manually defined behaviors. What follows is a reusable and automatic approach to developing customized insect inspired agent behaviors for use with any dynamic task allocation problem.