The design and implementation of a case-based planning framework within a partial-order planner
Laurie H. Ihrig · 1996
The aim of case-based planning (CBP) is to improve the efficiency of plan generation by taking advantage of previous problem-solving experience. Information is stored in a case library about each planning episode as problems are solved. When a new problem is encountered, a judgment is made about its similarity to these previous experiences. Similar episodes retrieved from the library are used as guidance in solving the new problem. This dissertation describes the design and implementation of a framework for the case-based planner, DerSNLP (Derivational SNLP). It argues that as a partial-order planner which searches in the space of plans, DerSNLP has a greater capability for exploiting its previous experience than its state-space counterparts. It analyses this advantage and provides empirical evidence for the superiority of plan-space planning in case adaptation. One of the most difficult tasks in CBP is the retrieval of a case which is similar enough to be applicable to a new problem. Although it is demonstrated that DerSNLP is more likely to be able to adapt a particular case to a new problem context, there remain instances where a case judged to be applicable turns out to provide the wrong guidance. DerSNLP has therefore been extended to incorporate explanation-based learning (EBL) techniques that allow it to explain and learn from the case retrieval failures it encounters. These techniques are used to refine judgments about case similarity in response to feedback when a wrong decision has been made. An empirical evaluation of this learning component demonstrates the advantage of learning from case failure.