In this study we develop and compare the performance of different agent control architectures based on learning through episodic memory for the design of Non-Player Characters (NPCs) in computer games. We focus on the Categorised Long-term Autobiographic Memory (CLTM) architecture, utilising abstracted notions of human autobiographic memory and narrative structure humans apply to their life stories. We also investigate the influence of remembering negative experience on agents' adaptivity. A large and dynamic virtual environment is created to examine different agent control architectures in an Artificial Life and bottom-up fashion. Agents' lifespan is measured in the experiments. Results show that CLTM architecture including remembering negative events can significantly improve the performance of a single autonomous agent surviving in the dynamic environment.
展开▼
