Software agents using simulation for decision-making

Abstract

Though autonomy is one of the major concepts of agent technology, the question how to implement this autonomy is mainly up to the agent developer. To show autonomous behaviour the agent has to make decisions regarding which actions to take next in order to meet its design objectives. This article suggests the use of simulation by an agent in its decision-making process. While simulation-based planning has already been proposed, we take it one step further and also use it for the control of the system the agent is reasoning about and for enabling the agent to take measures proactively. To explore this idea's potential we have implemented an agent test-bed to experimentally compare simulation-based decision-making with a rule-based implementation in a manufacturing control scenario. Both the implementation and the simulation study are presented in this article. Based on how the corresponding agents performed in reducing the production cycle time and in reasoning about the probability of meeting the delivery due date, we conclude that the use of simulation for such planning and control purposes is a promising, intuitive and competitive approach that is well suited to supplement other approaches that are already in use.