... and the dictionary being just a set of self-references would be completely useless.
You've just noticed that knowledge (not only regarding the term) is based on a set of tautologies. Kant already suggested that. In such case we can take it to the extreme: you are asking a useless question, since any answer would be based on facts you already know.
But the essential fact behind is that knowledge is not just a set of entities (Kant's "concepts"), but also a set of logical relationships (a concept in a dictionary is a relationship with other concepts: a dictionary is just a set of references). Your question, therefore, deals not about finding new concepts, but moreover new relationships between concepts you already have. Your dictionary lacks of some relationships among already existing terms.
A secondary fact is related to the most criticized issue regarding Kant: language can approximately reflect knowledge, but it usually with a quite poor level of precision. When you walk the sand and feel something, you might try to express it, but it is actually impossible to use language for that. Art is better for such purpose (you can use music, sculpture, hand shoe making), but it is far from being precise. And Kant dismissed the imprecisions of language. Reason cannot factually be described by language with precision, nor the facts raising from it. For example, 'yellow' could or not be a concept; raises moreover a feeling, than cannot be expressed with language. In such case, reason is not only like a dictionary, a set of references to other concepts, but also a set of references with feelings, desires, faculties, etc.
Consider a third fact, in addition: even scientific knowledge is not such thing if it is written in a book which no one reads. So, even scientific knowledge has a load of subjectivity. You would ask "wasn't it objective???", yes, since objectivity is shared subjectivity (any scientific fact depends on a lot of subjective elements and beliefs, and it is objective when such subjective elements are shared between two individuals). Absolute objectivity is impossible, it would imply that a martian would agree with all science. But perhaps martians have other subjective ideas, perceptions, capabilities. Absolute objectivity implies complete denial of subjectivity, and that's impossible (see Berkeley, Hume).
Having such elements in mind, perhaps you get an answer. Knowledge comes to be a subjective (fact 3) model of reality, consisting in relationships (fact 1) between concepts and physiological faculties (fact 2). Epistemology implies the attempt to model such subjectivity, by means of the physiological capabilities we have. So, 'epistemology is the study of knowledge' is correct, whilst considering the three aforementioned facts. If such facts are ignored ---concepts would just be isolated objects, concepts would be absolutely precise, concepts would be absolutely objective---, yes, in such case, robotic-epistemology is impossible.
Epistemology is the mirror reason uses to model itself. Your argument would be equivalent to state that one cannot look oneself using a mirror because mirror contents are not reality. That's arbitrary. While a mirror shows not reality, it usually provides enough information for the goals it serves to.