In the future mobile robots could be used in our offices as service robots. helping us with tasks such as picking up and delivering mail, returning borrowed books, or bringing coffee. To manage these service tasks in an unpredictable environment, the robot must have the ability to move robustly without bumping into people or furniture and to recognize its current location. The robot must also have high-level control that can reason about its given goals and find new ways to achieve its goals if something goes wrong. That is, it must "deliberate." Moreover, since the real world is unpredictable and the robot's sensors give noisy and sometimes incomplete information. the robot must be able to do so under uncertainty.

The initial motivation of the research presented here is to study the problems involved in deliberation under uncertainty. Although the term "deliberation" is often applied to different components of robot architectures, we have found that this concept is not well defined in the literature. Therefore this work presents our analysis of deliberation itself. as a necessary prerequisite to a deeper study of specific techniques that are needed in deliberation. This analysis leads us to propose a working definition of deliberation. During the analysis we make a few preliminary experiments on a simulated mobile robot. These experiments give us a deeper understanding of the problems involved in deliberation and help us forming the working definition.

This document is the midway milestone in an ongoing study of deliberation under uncertainty, intended to result in a Ph.D. thesis.