Comparisons in Self-Supervised Environmental Modelling

Terry Stewart

Abstract:

The field of ALife contains many experiments involving agents making decisions based on experientially learned predicted outcomes of actions in various situations. In this work, we investigate a specific cognitive architecture comprised of a developing self-supervised model of the environment and an action selection system that uses the model. Other researchers' works are shown to be instantiations of this cognitive architecture. Given this framework, a detailed comparison is carried out within many different environments and situations, while varying the particular parameters within the different approaches. The goal is to develop a better understanding of the advantages and disadvantages of different implementations of the modules within this particular cognitive architecture.