Estimation And Control Of Visitation Distributions For Reinforcement Learning