Automatically Acquiring Planning Templates from Example Plans

General-purpose planning can solve problems in a variety of domains but can be quite inefficient. Domain-specific planners are more efficient but are difficult to create. In this paper, we introduce template-based planning, a novel paradigm for automatically generating domain-specific programs, or templates. We present the DISTILL algorithm for learning templates automatically from example plans and explain how templates are used to solve planning problems. D ISTILL converts a plan into a template and then merges it with previously learned templates. Our results show that the templates automatically learned by D ISTILL compactly represent its domain-specific planning experience. Furthermore, the templates situationally generalize the given example plans, thus allowing them to efficiently solve problems that have not previously been encountered.