A simple implementation of mean actor critic on Lunar Lander and Cart Pole. Mean actor critic reduces the variance of policy gradient by marginalizing over actions when computing the gradient. See the paper for more details: https://arxiv.org/abs/1709.00503