Knowledge-based entity prediction (KEP) is a novel task that aims to improve machine perception in autonomous systems. KEP leverages relational knowledge from heterogeneous sources in predicting potentially unrecognized entities. In this paper, we provide a formal definition of KEP as a knowledge completion task. Three potential solutions are then introduced, which employ several machine learning and data mining techniques. Finally, the applicability of KEP is demonstrated on two autonomous systems from different domains; namely, autonomous driving and smart manufacturing. We argue that in complex real-world systems, the use of KEP would significantly improve machine perception while pushing the current technology one step closer to achieving the full autonomy.