COBAI : A Generic Agent-based Model of Human Behaviors Centered on Contexts and Interactions

This paper presents a generic agent-based model for human behaviors in simulations. COBAI (Context-Based Agent Interactions) is based on a previous model featuring contexts as a source of behaviors. We kept the base principles of this model. Agents cannot act alone: contexts give them behaviors to adopt. Agents can be influenced by several contexts and choose behaviors to adopt. Agents possess character attributes that adjust this choice. This mechanism results in the ability to control realism both at the level of individual agents and groups of agents. The previous model presented some limitations. For instance, a behavior could only result from a single context, limiting the variety and realism of behaviors. This paper presents a new model with features addressing this issue. We present a new behavior selection architecture, allowing the execution of several simultaneous behaviors and behaviors resulting from a combination of contexts. We introduce the new concepts of resources, skills, tools, modalities, and incomplete behaviors. Using this new architecture, we define groups of agents with task distribution. We apply the model to a case study of an emergency crisis developed in Unity.