In the discussion of the dynamical model (sect. 5.2), the authors mention that it would be possible to choose sensible defaults for A x and B, but more satisfactory to estimate them from input data. I would like to know how much more satisfactory, and whether from an implementation standpoint, choosing 'sensible' defaults could mean avoiding the need for input data of 'typical motions', which presumably requires operator knowledge - because it seems that many applications would benefit from as little bootstrapping as possible.