Semi-reactive planner based on a possible models action formalization

Abstract

We describe the basics of a semi-reactive planner that is based on a possible models action formalization. By semi-reactive planner, we mean a planning system that is able to incrementally generate plans in a classical way, thereby extending, when necessary, the compiled knowledge available for reactive execution. The action formalization, which is a correction of the possible worlds approach by Ginsberg and Smith, allows one to cope with lacking information, ambiguity and context dependency in the planning domain by considering different courses of action. This enables the planner to handle certain forms of uncertainty at planning time. Plans developed under this model are complied into finite state automata for execution, as a basis for reacting to less expected effects at execution time. Moreover, we explain how replanning fits into the framework, thereby opening it towards handling unexpected effects which occur during plan execution, and for which no reaction plan is available.