That's interesting, but
this paper
suggests to use derivative of jerk instead, do you think it's worth to try? Maybe even both can be used, I've seen that in a different paper too,
Not sure about usability of this one(values are ridiculous). If used alone it will just make two spikes in different directions and in my opinion this will cause more troubles than help. If used with jerk, it will increase starting amplitude, but decrease it's duration. Looks like it is not usable with trapezoidal motion profile.
Have some results on jerk feedforward in work, X axis is a bit too tight and FF3 value used for it was too big. I had to reduce it by half, because machine became too jerky on engraving work. But in overall it did a bit better with one problem on Z axis.
Also I did a check of snap(jounce) feedforward, values for it were 10^-9, but it didn't help in any way, increasing it only did worse.
