Specifying the behavior of an intelligent autonomous robot or agent is a non-trivial task. The question is: how can the knowledge of the domain expert be encoded in the agent program? Qualitative representations in general facilitate to express the knowledge of a domain expert. In this paper, we propose a semantics for qualitative fluents in the situation calculus. Our semantics is based on fuzzy sets. Membership functions define to which degree a qualitative fluent belongs to a particular category. Especially intriguing about a fuzzy set semantics for qualitative fluents is that the qualitative ranges may overlap, and a value can, at the same time, fall into several categories.