Memory as an Active Component of a Behavioral Animation System

Abstract

Our research is interested in behavioral animation among virtual reality applications. A major concern in this field is animating background actors and modeling their interactions. The aim is to provide virtual agents behaviors enabling them to evolve an autonomous and coordinated way in dynamic environments.

The behavior is modeled through the standard perception/decision/action loop where the characteristics of the decision module determine the agent abilities. The artificial intelligence “cognitive” agents have reasoning capabilities upon symbolic representations of the objects surrounding them, the way humans do. The artificial life agents possess the reactive and adaptive features from life imitation techniques.

The canvas of behavioral animation combine both approaches in order to obtain autonomous, coherent, reactive and adaptive agents. The so-called “hybrid” agents are for the most cognitive agents including reactive features. Two properties follow: they handle symbolic information and they store it in a memory regarded as a passive module.

We propose a different approach, focused on memory. We consider memory as an active component of cognition and reasoning or intelligence as the emergent expression of its operating. While keeping an “artificial life” view, we propose an original hybrid architecture which avoids the traditional reactiveness/cognition dichotomy and relies on distributed implicit mental representations.

Our model is a neural networks based architecture where two dimensions are considered: Whereas a vertical dimension models the procedural perception/action associations which form the reactiveness of a behavior, a horizontal dimension introduces the semantic concept association.