- POMDPs offer one of the richer models for sequential decision making. But this expressiveness comes with a price, both theoretical and practical. In this paper we seek to address the practical difficulty of solving POMDPs by focusing on symbolic, goal-oriented POMDP models, exploiting these features within a simple, generic, and novel online replanning architecture. This architecture leverages recent improvements in contingent planning, addressing the problem of selecting the next action at each decision epoch by solving a contingent planning problem derived from the original POMDP and the current belief state. While this is work in progress, we provide some initial empirical results.