Leveraging historical experience to evaluate and adapt courses of action

Abstract

In many planning domains there may be multiple potential solutions to a given problem. Each solution may require different resources, involve more or less risk, and result in desirable or undesirable effects. Reuse of historical plans is a strategy that can be employed to solve planning problems. While the retrieval of similar historical plans can be facilitated with sophisticated annotation and search engines, evaluating the usefulness of historical plans tends to be subjective, is context sensitive, and difficult when no single historical plan can be used to develop a new plan. Course of action (COA) evaluation is a method that can be used to compare a set of alternative solutions. An agent-based tool called MICCA (Mixed-Initiative Course of Action Critic Advisors) can aid human operators or software agents in evaluating and adapting historical plans for use in achieving one or more objectives in some current or future hypothetical world state. In this paper we introduce MICCA and describe how case base reasoning (CBR) and generative planning techniques are utilized to support COA evaluation and adaptation. © 2013 Springer-Verlag.