Font Size: a A A

A Counterplanning Approach Based On Goal Driven Theory

Posted on:2008-10-08Degree:MasterType:Thesis
Country:ChinaCandidate:Y Q DongFull Text:PDF
GTID:2178360215979146Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
Classical planning problems make assumptions about the world as follows: the goals of the planner are a logical description of world states; the actions taken by the planner are the only sources of change in the world; each action can be described by the conditions under which it can be applied and its effects on the world. However, the adversarial planning is one of the most complicated fileds. Generally,the characteristic of the environment inwhich an agent lies during the adversarial process, is inaccessible, nondeterministic, dynamic, nonepisodic, and multi-agent. Therefore, the conventional methods in the classical plannning is not appropriate very well for some adversarial problems.Previous adversarial planning approaches solve problems generally based on the hierarchically decomposition of the HTN Planning (Hierarchical Task Network Planning). Conventional adversarial planning approaches mentioned above still have restrictions.Some new definitions based on goal driven are introduced in this paper, such as basic defense tree, goal metric and etc. The goals in the basic defense trees of both adversaries are taken as the basic reference of predicting and recognizing the plans of the opponent agent and making our counter plans during the adversarial process. This paper proposes an approach that in every current state, an agent compares the key values of the inhomogeneous goals that appear in all the possible plans of the opponent agent. Through this approach, an agent can discover the most important opportunity and problem in time; predict and recognize the plans of the opponent agent effectively. Also,the approach we present here can deal with significant contingency events intently by applying the key value, defense value that are both in goal metric, and a new algorithm to them. Then an agent needs not to deal with all the contingency events blindly as conventional approaches did.The improved goal driven method herein can be applied widely in the adversarial planning domains such as intrusion detection, battle, adversarial game and etc.
Keywords/Search Tags:Intelligent planning, plan recognition, adversarial planning, basic defense tree, goal metric
PDF Full Text Request
Related items