Pushing the Envelope: Programming Reasoning Agents

Abstract

Formal theories of actions have evolved in recent years into high-level programming languages for robots. While these languages allow to implement complex strategies in a declarative, concise and modular fashion, they are often doubted to be sufficiently efficient for practical purposes. In this paper we push the envelope of reasoning agents, thereby making a case for state-based solutions to the frame problem. We analyze the computational behavior of the logic-based agent programming language FLUX and show that it scales up well to problems which require reasoning about the performance of several thousand actions.