# Physics that learns from a recording

> A ball's physics step (gravity, drag, bounce) is unrolled into one Handwired neural net and fitted to a recorded path by gradient descent. The constants converge, a new throw follows the real ball, and the floor stays exact. A generic net fitted to the same path can't handle a new throw.

Source: https://handwired.wiki/pages/physics.html

<p class="lede">Keep your simulation. Let a recording set its constants.</p>

*(Interactive demo: open https://handwired.wiki/pages/physics.html)*