Humans perform abductive reasoning routinely. We hypothesize about what happened in the past to explain an observation made in the present. This is frequently needed to model the present, too. In this paper we describe an approach to equip robots with the capability to abduce hypotheses triggered by unexpected observations from sensor data. This is realized on the basis of KnowRob, which provides general knowledge about objects and actions. First we analyze the types of environment changes that a robot may encounter. Thereafter we define new reasoning methods allowing to abduce past events from observed changes. By projecting the effects of these hypothetical previous events, the robot gains knowledge about consequences likely to expect in its present. The applicability of our reasoning methods is demonstrated in a virtual setting as well as in a real-world scenario. In these, our robot was able to abduce highly probable information not directly accessible from its sensor data.