Knowledge-Enabled Generation of Semantically Annotated Image Sequences of Manipulation Activities from VR Demonstrations.