High-Level Planning and Control with Incomplete Information Using POMDPs

Contents

We develop an approach to planning with incomplete information that is based on three elements: 1. a hlgh-level language for describing the effects of actions on both the world and the agent’s beliefs 2. a semantics that translates such descriptions into Partially Observable Markov Decision Processes or POMDPs, and 3. a real time dynamic programming algorithm that produces controllers for such POMDPs. We show that the resulting approach is not only clean and general but that may be practical as well. We have implemented a shell that accepts high-level descriptions of POMDPs and produces suitable controllers, and have tested it over a number of problems. In this paper we present the main elements of the approach and report empirical results for a challenging problem of planning with incomplete information.