Representing and Learning Quality-Improving Search Control Knowledge

Generating good, production-quality plans is an essential element in transforming planners from research tools into real-world applications , but one that has been frequently overlooked in research on machine learning for planning. This paper describes quality , an architecture that automatically acquires operational quality-improving control knowledge given a domain theory, a domain-speciic metric of plan quality, and problems which provide planning experience. The framework includes two distinct domain-independent learning mechanisms which differ in the language used to represent the learned knowledge, namely control rules and control knowledge trees, and in the kinds of quality metrics for which they are best suited. quality is fully implemented on top of the prodigy4.0 nonlinear planner and its empirical evaluation has shown that the learned knowledge is able to substantially improve plan quality. Although the learning mechanisms have been developed for prodigy4.0, the framework is general and addresses a problem confronted by any planner that treats planning as a constructive decision-making process.