Using a Parameterized Memory Model to Modulate NPC AI


Li, W., Balint, T. and Allbeck, J.M. Using a Parameterized Memory Model to Modulate NPC AI. In Proceedings of Intelligent Virtual Agents, 2013, pages 1-14.


While there continues to be exciting developments in research related to virtual characters, improvements are still needed to create plausibly human-like behaviors. In this paper, we present a synthetic parameterized memory model which includes sensory, working, and long-term memories and mechanisms for acquiring and retrieving memories. With the aid of this model, autonomous virtual humans are able to perform more reasonable interactions with objects and agents in their environment. The memory model also facilitates emergent behaviors, enhances behavioral animation, and assists in creating heterogeneous populations. To demonstrate the e ectiveness of the memory model, we also provide an example in a 3D game environment and have conducted a user study in which we found general guidance in determining parameter values for the memory model, resulting in NPCs with more human-like game playing performances.

Paper (pdf)