Efficient Reinforcement Learning via Self-supervised learning and Model-based methods