Perhaps it only works for certain classes of solutions? Like
the horseshoe solutions or the orbits around Lagrange points?
Those are cases that a human physicist also could predct with
relatively little effort..
--
Jos
[Moderator's note: see also
https://arXiv.org/abs/1909.05272 -P.H.]