Of course, the main objective of the ARL architecture is to get an interactive system. Therefore, consider the case where user B has stepped out of the scene and his vision system is disabled. The system's role is now to synthesize B's presence in a continuous way instead of merely predicting a small step into the future or filtering user B's signal. Of course, without loss of generality, we are assuming that user B is to be impersonated. However, by symmetry, all the models apply equally well if A leaves the room and B is interacting alone with the system.