Control Knowledge to Improve Plan Quality

<p> </p><p>Generating production-quality plans is an essential element in transforming planners from research tools into real-wor ld applications. However most of the work to date on learning planning control knowledge has been aimed at improving the <em>efficiency of planning</em>; this work has been termed “speed-up learning”. This paper focuses on learning control knowledge to guide a planner towards better solutions, i.e. to improve the <em>quality of the plans</em> produced by the planner, as its problem solving experience increases. We motivate the use of quality-enhancing search control knowledge and its automated acquisition from problem solving experience . We introduce an implemented mechanism for learning such control knowledge and some of our preliminary results in a process planning domain.</p> <p></p>