SOME EFFICIENCY AND CONSISTENCY ISSUES IN AUTOMATIC PLANNING SYSTEMS
Abstract
A wide variety of problems are of the type where some goals are achieved by a sequence of actions. Such problems are referred to as planning problems in artificial intelligence. The process of planning the sequence of actions that achieve the goal is basically a search process which needs to be guided in an intelligent manner in order to avoid computational excesses. In this thesis, an efficient network representation of the planning environment is first developed. A flow propagation technique is used on this network to perform deduction. This deductive capability is integrated with a myopic planning algorithm to provide the latter with more foresight. A multi-robot planning system is also proposed as a means to achieving greater efficiency in planning processes. Without coordination, the local actions of the robots could be globally inconsistent. A control strategy that enables the robots to cooperate in a consistent manner is then presented. Finally, the state space approach to generating plans is used to show how multi-period optimization problems could be solved efficiently using AI techniques. The key to efficient solution lies in the ability of the planning system to formally model problem specific characteristics which are exploited in the solution process. Traditional solution methods for the multi-period optimization problems have tried to use problem specific characteristics but only in an ad hoc way.
Degree
Ph.D.
Subject Area
Management
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.