2018
We investigate the viability of collecting demonstrations, for learning robot control, in Virtual Reality (VR). If acceptable demonstrations can be collected offline using VR, problems with affecting the production line can be avoided and large amounts of realistic data from the actual production environment can be used. We show that the relatively low detail data that can be captured from the production line using a RGB-D camera provides sufficient information for an expert to perform acceptable demonstrations in Virtual-Reality. However, performance is not yet on par with what can be achieved through real world observations. We propose several future improvements that will help close the gap.