Abstract
When agents like mobile robots discover that the world is not as expected after carrying out a sequence of actions, they are interested in what action failures or unnoticed actions could have actually occurred, which would help them rectify the situation. For this purpose, we investigate a kind of history‐based diagnosis which is appropriate for explaining what went wrong in dynamic domains. It turns out that there are often many diagnoses which are quite similar and differ only in the objects they refer to. In this paper we show how these instances can be compactly represented by introducing so‐called diagnosis templates. We formalize this approach for an action theory based on the situation calculus and discuss a prototypical implementation of a diagnostic system which generates diagnosis templates according to certain preference criteria.