Efficient Exploration in Resource-Restricted Reinforcement Learning