Managing Deliberation and Reasoning In Real-Time Ai Systems

Abstract

This paper describes some recent research on architectures for situated (embedded) systems that need to deliberate and reason in real time. One of the most difficult problems in the design of such architectures is how to manage the reasoning performed by such a system while still meeting the real-time constraints of the problem domain. We present an architecture, based on the Procedural Reasoning System (PRS), that provides mechanisms for the management and control of deliberation and reasoning in real-time domains. In particular, we show how deliberation and reasoning strategies can be represented in the form of metalevel plans, and describe an interpreter that selects and executes these in a way that retains bounded reaction time. In addition, this approach allows us to represent different types of situated systems by varying the metalevel deliberation strategies. Finally, we provide some statistical measures of performance for one such type of situated system applied to a complex real-time application.