With the advent of artificial agents in everyday life, it is important that these agents are guided by social norms and moral guidelines. Notions of obligation, permission, and the like have traditionally been studied in the field of Deontic Logic, where deontic assertions generally refer to what an agent should or should not do; that is they refer to actions. In Artificial Intelligence, the Situation Calculus is (arguably) the best known and most studied formalism for reasoning about action and change. In this paper, we integrate these two areas by incorporating deontic notions into Situation Calculus theories. We do this by considering deontic assertions as constraints, expressed as a set of conditionals, which apply to complex actions expressed as GOLOG programs. These constraints induce a ranking of "ideality" over possible future situations. This ranking in turn is used to guide an agent in its planning deliberation, towards a course of action that adheres best to the deontic constraints. We present a formalization that includes a wide class of (dyadic) deontic assertions, lets us distinguish prima facie from all-things-considered obligations, and particularly addresses contrary-to-duty scenarios. We furthermore present results on compiling the deontic constraints directly into the Situation Calculus action theory, so as to obtain an agent that respects the given norms, but works solely based on the standard reasoning and planning techniques.
