COMPOSER is one of a growing number of techniques for learning to plan. Like other approaches, it embodies a number of simplifications to overcome the complexities of learning. These simplifications introduce tradeoffs between learning efficiency and effectiveness. In this paper we relate COMPOSER to our general framework of simplifications for learning to plan [Gratch92a]. This discussion illustrates how such a framework may be used to analyze a particular approach, highlighting the learning system’s strengths and weaknesses.
Download Full PDF Version (Non-Commercial Use)