Abstract
Episodic memory is an important component of “minds” of many long-living virtual agents, for instance non-player characters from role-playing games. So far, research on episodic memory modeling in the context of these agents has focused mostly on producing the memory content online, that is, when the agent is being simulated.
In this paper, we address a complementary issue: automatic generation of the memory content off-line. The intended use is generating memories that anticipate the start of the simulation, that is, generating history of an agent.
We propose here a complex design method enabling a designer to specify high-level requirements on the agent’s history and use planning to generate this history according to these requirements. We also detail that part of this method that concerns itself with the planning, and describe a prototype implementation and the first results.