The current data recorder is very basic. It simply dumps the raw data to disk.
It will make sense to store received observations in more structured and compressed formats, like video.
This is a prerequisite to creating datasets for imitation learning.