Avatars Animation using Reinforcement Learning in 3D Distributed Dynamic Virtual Environments

The animation problem of avatars or virtual creatures using learning, involves research areas such as Robotics, Artificial Intelligence, Computer Sciences, Virtual Reality, among others. This work presents a Machine Learning approach using Reinforcement Learning and a Knowledge Base for animating avatars. This Knowledge Base (ontology) provides the avatar with semantic definition and necessary awareness of its internal structure (skeleton), its behavior (personality, emotions and moods), its learned skills, and also of the rules that govern its environment. In order to animate and control the behavior of these virtual creatures in 3D Distributed Dynamic Virtual Environments, we use Knowledge-Based Conscious and Affective Personified Emotional Agents as a type of logical agents, within the GeDA-3D Agent Architecture. We focus on the definition of minimum conscience of the avatars. The conscience and cognitive processes of the avatars allow them to solve the animation and behavior problems in a more natural way. An avatar needs to have minimum conscience for computing the autonomous animation. In our approach, the avatar uses the Knowledge Base first as a part of its conscience, and second to implement a set of algorithms that constitute its cognitive knowledge.