The second talk in this symposium was by Martin Egelhaaf on how the blowfly brain processes self-generated visual input. His first main point was that optic flow must first be dissected into translational and rotational movement components. Flies don't fly smooth turns but rather rapid movements, so-called body-saccades. Flight body-saccades serve to separate translation and rotation components.
In order to get to the neural substrate of vision in blowflies, the Egelhaaf lab fixes the flies and records electrophysiologically from motion-sensitive neurons in the optic lobes. To generate realistic motion stimuli, they have constructed a virtual reality arena of 7000 LEDs all around the fixed fly.
Interestingly, they do not study self-motion as the title and his introduction may suggest, they only play realistic scenes for the fly and then record the responses in the neurons. They do not have any evidence that the neurons respond the same way when they perceive actual self-motion stimuli.
The main message of the talk was that the flies reduce behavioral complexity (i.e. saccades and straight flight instead of smooth turns) in order to reduce the complexity of the visual input.

I asked him after the talk how he can be sure that the neurons he recorded from actually respond in the same way to truly elf-generated stimuli as he had shown for exafferent stimuli. He conceded that nobody can be sure. He went on to make a fairly good case that any central commands act downstream of the lobula plate where he recorded. I guess that's fair enough for now.