Hierarchical Plan Formation and Failure Recovery in the Flight Domain

Abstract

Expert planners have always been confronted with the domain complexity problem. This problem surfaced with the early works in automatic problem solving and has been a formidable obstacle ever since. The flight domain places additional demands on the planner due to the domain's complex and dynamic nature. This research presents a planning architecture that can perform both planning and recovery planning in the flight domain.

The key to a robust planner that can function in a complex and dynamic domain is the vertical decomposition of the domain knowledge. The decomposition is semantic in nature where the flight domain is decomposed into four minidomains. Each minidomain models the flight domain along some degree of global viewpoint, hence, the vertical hierarchy of semantic levels. The decomposition of the flight domain into four nearly independent minidomains reduces the domain complexity confronted by the planner. The explicit modeling of the different degrees of global viewpoint further reduces the domain complexity geometrically. The vertical decomposition of the domain knowledge results in shallow planning and shallow recovery planning.