A multi-agent architecture for collaborative serious game applied to crisis management training: Improving adaptability of non played characters

Abstract

Serious Games (SG) are more and more used for training in various domains, and notably in the crisis management domain. Crisis management can gather several hundred stakeholders, which can present various difficulties when organizing field exercises. Serious Game constitutes a more practical alternative with specific benefits concerning detailed players' actions tracking during a simulated exercise. Moreover, Non Played Characters (NPC) can be used to adapt the crisis management exercise perimeter to the available stakeholders as well as specific training objectives. In this paper we present a general Multi-Agent System (MAS) architecture providing support to the behavioral simulation as well as the monitoring and assessment of human players. To each NPC is associated a so called Game Agent designed to reproduce the behavior of the actor simulated. The Game Agents are based on a deliberative model (Belief Desire Intention) with added editing features to facilitate the scenario design phase. Thus an Agent editor allows a designer to configure agents' behaviors illustrated in this paper with the case of crisis management scenario. The behavior simulation was implemented within the preexisting SIMFOR project, a serious game for training in crisis management.