Donald Bren School of Information and Computer Sciences

A framework is proposed for the investigation of planning systems that must deal with bounded uncertainty. A definition of this new class of contingency planners is given. A general, complete contingency planning algorithm is described. The algorithm is suitable to many incomplete information games as well as planning situations where the initial state is only partially known. A rich domain is identified for the application and evaluation of contingency planners. Preliminary results from applying our complete contingency planner to a portion of this domain are encouraging and match expert level performance.