A Coherence-Driven Action Selection in Dynamic Environments

Abstract

In this paper, we propose a coherence-driven approach to action selection in agents. The mechanism is inspired by the cognitive theory of coherence as proposed by Thagard. Based on a proposal to extend BDI agents with coherence, we interpret, how action se- lection can be viewed as a coherence-maximising problem. Contrasted against the classical BDI approach to action selection where actions are selected against a pre-determined set of beliefs and desires, this method provides us with a reasoning formalism that incorporates uncertainty and dynamism in the world model without loosing the type of formal qualities that make BDI-like architectures so attractive for testability and reliability reasons.