SMS dit: Deep variational reinforcement learning for POMDPs