Agent-based simulation to analyze business office activities using reinforcement learning

This paper attempts to clarify team behavior in cooperative organizations by agent-based simulations. We focus attention on both the roles of manages and the initiatives of staffs, and then model them using agent-based model concepts. This enables us to investigate phenomena in organizations at micro-level and macro-level. Besides, we formulate the task processing of each member in real organizations as learning for maze problem. The advantages of applying maze problem for our simulation model are as follows: It is possible to describe agents who acquire skills by reinforcement learning and to represent environmental uncertainty by changing block placements dynamically. Several computational experiments show what the whole organization behaves from microscopic points of view. At the same time, the authors confirm that the ability to adapt environments under uncertainty is different from the characters of organization.