This paper describes an integrated solution to the problem of describing and interpreting goals for robots in open uncertain domains. Given a formal specification of a desired situation, in which objects are described only by their properties, general-purpose planning and reasoning tools are used to derive appropriate actions for a robot. These goals are carried out through an online combination of hierarchical planning, state-estimation, and execution that operates robustly in real robot domains with substantial occlusion and sensing error.