This thesis proposes a new design approach for building software agents that reflect believable behaviors in simulated virtual environments. Observing a specific agent's actions usually leads to defining its overall behavior schema, which is done according to the actions it performs while attempting to reach a goal and, the choices this agent decides to make because of specific personality traits. The contribution of this thesis is directly related to the design phase wherein an agent's behavior is to be defined. Defining an agent's behavior consists of reasoning and individuality. This thesis focuses on the introduction of a new action selection mechanism to an agent behavior's individuality. The main concepts behind the design of our action selection mechanism are motivations and alternatives. We look at motivations as the independent expressions of a specific agent's traits that influence its upcoming action selection. We look at alternatives as possible resolutions computed by the reasoning part of an agent's behavior. In our approach, the motivations influencing agents' behaviors are domain-free. Therefore, we were able to utilize our work in contributing to the CoCoA project by providing a concrete motivation-based action selection mechanism. Further to the interaction-oriented approach promoted by the CoCoA project, the action selection mechanism we propose makes it possible to provide the means to have reliable behavioral engine that is the same for all agents.
Thesis of the team SMAC defended on 04/11/2010