The reference picture of passing over a bump includes the way the

>bump approaches and the way it feels while going over it, the way
>the bicycle sounds, the way the handlebars feel, the way the feet
>on the pedals feel. The reference picture comes to contain
>standing up as one element, which one learns by not standing up
>There's absolutely no reason to treat feedforward as a separate

I wouldn't be confident of this point before having a model that
actually does all this stuff, which I haven't noticed that anybody does.
Meanwhile, this brand of `feedforward' (which I would prefer to call
`anticipation', seems like a useful descriptive category. *I* don't
yet understand the detailed workings of any mechanisms which perform
(apparent) anticipation, though it's obvious that living systems do it.