Layered and resource-adapting agents in the RoboCup simulation

Abstract

Layered agent architectures are particularly successful in implementing a broad spectrum of (sub-)cognitive abilities, such as reactive feedback, deliberative problem solving, and social coordination. They can be seen as special instances of boundedly rational systems, i.e., systems that trade off the quality of a decision versus the cost of invested resources. For sophisticated domains, such as the soccer simulation of RoboCup, we argue that a generalised framework that combines a layered design with explicit, resource-adapting mechanisms is reasonable. Based on the InteRRaP model, we describe a prototypical setting that is to guide and to evaluate the development of reasoning about abstract resources. These are representations of general interdependencies between computational processes. The realised soccer team, CosmOz Saarbrücken, participated successfully in the RoboCup-98 competition and confirmed that abstract resources are an appropriate modelling device in layered and resource-adapting agents.