AI Applied to Real Time Control: A Case Study

Abstract

The application of AI technology to a particular problem in real-time control of complex machinery is described. The problem tackled, that of dynamic positioning of a highly manoeuvrable ship, is seen as being representative of a range of important machine control problems and has been studied with a view to inferring the practical relevance of AI to such problems.

The particular class of applications where this work applies is those where current multi-variable control theory is applicable, but where the system operates in a number of states the selection of the correct control strategy for a given state (and the selection of parameters to fine tune a given strategy) is non-trivial.

An important result of this work comes out of our experience in representing the knowledge by which the control strategy selection is made. An initial approach using a straightforward rule based formalism proved unworkable. The alternative employed is based upon the use of plan scripts to represent the major system states, with explicit representation of plan goals and assumptions.