[lpribis]

I'm pretty sure he has used gradient descent in previous videos to optimize systems, maybe this time it was just easier to hand tune rather than set up an optimization feedback harness around MuJoCo.