A virtual training system includes a 3D scanner in recording space and a camera. A scan of the recording space are received and combined to make a sequence of models of the recording space while a tasks is being performed. The model states are presented in sequence to a trainee in a playback space and the actions of the trainee are detected. Deviation of the trainee or objects in the playback space from the trainer and objects in the recording space may be detected and alerts generated. Scanning of the recording space may be accelerated by making a full scan followed by partial scans. The partial scans may be performed in response to detected movement.