Using a Logarithmic Mapping to Enable Lower Discount Factors in Reinforcement Learning