Using behaviour patterns to generate scripts for computer role-playing games

Abstract

Character behaviours in computer role-playing games have a significant impact on game-play, but are often difficult for story authors to implement and modify. Many computer games use custom scripts to control the behaviours of non-player characters (NPCs). Therefore, a story author must write fragments of computer code for the hundreds or thousands of NPCs in the game world. The challenge is to create non-repetitive (more entertaining) behaviours for the NPCs without investing substantial programming effort to write custom non-trivial scripts for each NPC. Consequently, current computer games mostly rely on simplistic non-interactive behaviours for NPCs. This research describes the design and implementation of a novel behaviour model for interacting NPCs, based on generative design patterns, that requires no manual script writing. In this model, NPCs assume different roles during the story and select behaviours based on static probabilities or dynamic motivations. We also devised a reinforcement learning algorithm, ALeRT, based on Sarsa(λ) and we extended our behaviour model to support behaviour selection based on learning. In our model, an NPC can exhibit proactive, reactive, or latent behaviours that may be independent or collaborative. This behaviour architecture supports behaviours that can be interrupted and resumed based on priorities. The implementation of this model produces scripting code for BioWare Corp.'s Never-winter Nights computer role-playing game.