https://flecart.github.io/notes/proximal-policy-optimization/