Curriculum-based Sample Efficient Reinforcement Learning for Robust Stabilization of a Quadrotor