A Fast Hybrid Reinforcement Learning Framework with Human Corrective Feedback