very cool. reminded me of the difference of exploring bifurcation points of non-linear functions instead of using pure randomness according to this paper by tom mudd Nonlinear Dynamics In Musical Interactions - Open Research Online. maybe with the MLPRegressor. Probably also cool to use this to drive the x and y axis of the MLP to control a bunch of Synth paramters, or both.
BTW, Fb1_ODE from miSCellaneous_lib is an interface for numerical resolving of ODE systems (with a reliable symplectic procedure as default but it allows to choose other procedures as well). Pendulum examples are included in the help file as well, though not the double pendulum.
I had tried to scaled down the time step, but now I realized I was only scaling down the acceleration (as in the double pendulum example above), thinking it would propagate to the velocity and position, but alas. Again some FM for a demo: