A Survey of Memory Models for Virtual Agents and Humans: From Psychological Foundations to Computational Architectures

Virtual Humans are an advanced class of Virtual Agents characterized by human-like embodiment and cognitive capabilities. Central to their adaptivity is the integration of computational cognitive architectures, in which memory models play a key role in learning, continuity, and contextual reasoning. This review bridges psychological theories of memory and their computational implementations by comparing symbolic and connectionist approaches and exploring new paradigms such as Memory-Augmented Neural Networks and Large Language Models. We propose a unified framework for analyzing the components of artificial memory - Working, Semantic, Episodic, Procedural, Spatial, and Autobiographical Memory - and review their applications in domains such as education, games, and social simulation. Finally, we discuss open challenges and future works.