This paper describes a system for learning domain-dependent knowledge in the form of goal-indexed Hierarchical Task Networks (HTNs). DLIGHT is a goal-directed incremental learning algorithm which observes solution traces and generates rules for solving problems. One of the main challenges in learning this kind of knowledge is determining a good level of generality. Analytical methods, such as explanation-based macro-operator learning, construct very specific structures that guarantee a successful execution when applicable but generalize poorly to new problems. Previous goal-directed learning approaches produce hierarchical rules with more relaxed preconditions, but the learned knowledge suffers from over-generality. Our approach builds on one such approach but it strikes a better balance between generality and specificity. This is done by carrying out a goal-dependency analysis to determine the structure of the hierarchy and precondition of each rule to follow the successful solutions more closely while maintaining generality. We hypothesize that this algorithm produces HTNs that generalize well and can solve problems efficiently. We evaluate the system's behavior experimentally in several planning scenarios and conclude with related work and future research paths.

Knowledge-based planning methods offer benefits over classical techniques, but they are time consuming and costly to construct. There has been research on learning plan knowledge from search, but this can take substantial computer time and may even fail to find solutions on complex tasks. Here we describe another approach that observes sequences of operators taken from expert solutions to problems and learns hierarchical task networks from them. The method has similarities to previous algorithms for explanation-based learning, but differs in its ability to acquire hierarchical structures and in the generality of learned conditions. These increase the method's capability to transfer learned knowledge to other problems and supports the acquisition of recursive procedures. After presenting the learning algorithm, we report experiments that compare its abilities to other techniques on two planning domains. In closing, we review related work and directions for future research.