| Computing architectures based on software agents are an important trend in computing technology today. Within such architectures, there are many applications in which planning procedures can play a significant operational role. Planning procedures are characterized by steps of operators that transform an initial state into a desired goal state.;In this dissertation, a graph planning method called "I&G Graphplanner" is presented. I&G stands for Initial and Goal. The I&G Graphplanner is the reasoning engine of a distributed software agent architecture, MAGSIC. The multi-agent architecture that uses situation calculus (MAGSIC) is proposed in Chapter 3. The I&G Graphplanner is based on an existing planning method, Graphplan [BLU & FUR 97], with the following improvement: it solves the problem of explosive operator activations that is associated with forward planners. As a direct result, the planning speed is increased. In this dissertation, the algorithm of the I&G Graphplanner, including its implementation in Appendix A, and the machine run results of the planning algorithm are described.;As the environment to host the I&G Graphplanner, MAGSIC is a proposed agent architecture. It uses Task Programming Language (TPL) and employs the Knowledge Query and Manipulation Language (KQML) as an inter-agent communication language. In dynamic and incompletely known worlds, TPL automates agent-level tasks by finding a sequence of actions that constitutes a legal execution of some high-level programs for most well-defined tasks and reverts to detailed planning when no such task is found. The architecture is partially implemented. |
