An Abstraction­Based Approach to Interleaving Planning and Execution in Partially­Observable Domains

James Helwig and Peter Haddawy Decision Systems and Artificial Intelligence Laboratory Department of Electrical Engineering and Computer Science University of Wisconsin­Milwaukee

Introduction

An intelligent reactive planning agent for partially ob­ servable stochastic domains requires a number of di­ verse capabilities. First, the agent must be able to intelligently allocate its resources. This means that it must be able to decide how much time to allocate to deliberation in a way that is responsive to the environ­ ment in which the agent finds itself. It must also be able to decide when to sense and how much time and effort to spend sensing.

The agent must have flexible planning abilities. The planning algorithm must be anytime but should not be myopic. The planner must be able to reason under uncertainty due to incomplete information and due to the stochastic nature of available actions. It should be able to incorporate infomation obtained from sensors. It should be able to update the space of possible plans once an action is chosen for execution.

Finally, the agent must be capable of coordinate planning and acting. This means that it must be able to recognize when an action is completed and should be able to deliberate while acting. It also may need to translate between the high­level action representa­ tion used by the planner and low­level commands to its effectors.

Since the question of meta­level control to determine optimal deliberation times has been well studied, in this paper we focus on the required planning capabil­ ities, as well as on the coordination of planning with execution.