The Dreaming Variational Autoencoder for Reinforcement Learning Environments