Strong Cycling Planning aims at generating iterative plans that implement trial-and-error strategies, where loops are allowed only so far as there is a chance to reach the goal. In this paper, we tackle the problem of Strong Cyclic Planning under Partial Observability, making three main contributions. First, we provide a formal definition of the problem. We point out that several degrees of solution are possible and equally interesting, depending on the admissible delay between achieving the goal and detecting that it has been achieved. Second, we present a family of planning algorithms that tackle the different versions of the problem. Third, we implement the algorithms using efficient symbolic representation techniques, and experimentally compare their performances.