A Cognitive Model for Learning Autonomous Nonplayer Characters

A cognitive model is proposed to construct autonomous nonplayer characters with learning (LANPCs) by defining three functions, respectively sensing, reasoning and learning and effecting. The cognitive model is the product of the three functions. LANPCs can acquire knowledges from game world by the function sensing, where a discretization method based on linguistic variables can be used to solve the possible uncountable worlds and keep inequality reasoning. LANPCs can also change the game world by the function effecting which can map discrete attributes into continuous procedures. The function reasoning and learning based on Q-learning guides LANPCs how to obtain knowledge. The superior goal-oriented action (GOA) technique in game AI can be well described by the model. The model is built on the AI Interface Standards Committee´s (AIISC) world interfacing to separate AI logic from game engine. An experiment implements the model