This paper presents a method for detecting invariant task structures that are commonly shared by the human operator and the autonomous robot in a VR-based (i.e., virtual reality based) mobile robot teleoperation system. We call such a property of having common structures as task-morphology, and we discuss how this property functions well to establish the naturalistic task sharing between the two who have different perception-acting capabilities. For this purpose, we analyze their continuous task performance and segment them into some qualitatively different phases using Adaptive Resonance Theory (ART).