Curiosity-Driven Reinforcement Learning with Homeostatic Regulation