A Visually-Based Evolvable Control Architecture for Agents in Interactive Entertainment Applications

摘要

A visually-based evolvable control architecture for agents in interactive entertainment applications is presented. Agents process images of their local surroundings according to evolved image processing operators. A behaviour-based framework of action rules encapsulates image processing parameters and action parameters into discrete behavioural modules. These modules are interconnected and retain internal state through the dynamics of these internal connections. This novel control architecture has a wide behavioural range and is specified in an evolvable framework which allows agents for entertainment applications to be evolved as opposed to explicitly designed. The results of several demonstrations and experiments are presented to showcase the possibilities of this control architecture.