Design choices for agent-based control of AGVs in the dough making process

In this paper we consider a multi-agent system (MAS) for the logistics control of Automatic Guided Vehicles (AGVs) that are used in the dough making process at an industrial bakery. Here, logistics control refers to constructing robust schedules for all transportation jobs. The paper discusses how alternative MAS designs can be developed and compared using cost, frequency of messages between agents, and computation time for evaluating control rules as performance indicators. Qualitative design guidelines turn out to be insufficient to select the best agent architecture. Therefore, we also use simulation to support decision making, where we use real-life data from the bakery to evaluate several alternative designs. We find that architectures in which line agents initiate allocation of transportation jobs, and AGV agents schedule multiple jobs in advance, perform best. We conclude by discussing the benefits of our MAS systems design approach for real-life applications.