Deciding Whether to Plan to React a Dissertation Submitted to the Department of Computer Science and the Committee on Graduate Studies of Stanford University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

Intelligent agents that operate in real-world real-time environm have limited resources. An agent must take these limitations into ac when deciding which of two control modes planning versus reaction control its behavior in a given situation. The main goal of this thesis develop a framework that allows a resource-bounded agent to decid planning time which control mode to adopt for anticipated possible ru contingencies. Using our framework, the agent: (a) analyzes a com (conditional) plan for achieving a particular goal; (b) decides which o anticipated contingencies require and allow for preparation of re responses at planning time; and (c) enhances the plan with pre reactions for critical contingencies, while maintaining the size of the the planning and response times, and the use of all other critical resou the agent within task-specific limits. For a given contingency, the decisi plan or react is based on the characteristics of the contingency, the ass reactive response, and the situation itself. Contingencies that may occur same situation compete for reactive response preparation because o agent's limited resources. The thesis also proposes a knowledge represe formalism to facilitate the acquisition and maintenance of knowledge in in this decision process. We also show how the proposed framework ca adapted for the problem of deciding, for a given contingency, wheth prepare a special branch in the conditional plan under development leave the contingency for opportunistic treatment at execution time. We a theoretical analysis of the properties of our framework and demonstrate them experimentally. We also show experimentally that framework can simulate several different styles of human reactive beh described in the literature and, therefore, can be useful as a basi describing and contrasting such behaviors. Finally we demonstrate th framework can be applied in a challenging real domain. That is: (a knowledge and data needed for the decision making within our fram exist and can be acquired from experts, and (b) the behavior of an age uses our framework improves according to response time, reliability resource utilization criteria.