Design of memory architecture for autonomous virtual characters using visual attention and quad-graph

Most autonomous virtual characters use programmed information and always repeatedly perform actions with predefined patterns. When gamers interactively play with such characters, they lose their interest and feel that the characters are unfriendly and unrealistic. In this paper, we present a cognitive memory model for autonomous virtual characters, which manages recognized information by visual attention. The cognitive memory of autonomous virtual characters is comprised of visual memory. Visual memory stores the recognized information in a physical structure embodied by a quad-graph structure. Virtual characters assume adaptive actions for the situation by utilizing the stored information in the memory system. To effectively utilize memory capacity, we devise an algorithm of rapid visual attention, which is well suited to game environment. We have implemented our memory system experimentally in order to confirm its feasibility and availability. From the experiments, we confirm that autonomous virtual characters perceive the static and dynamic objects of a 3D virtual world by a visual attention mechanism with approximately 1.6 times improvement in time complexity; further, the virtual characters utilize the information of memory, searches the target object, and establishes a path plan.