EANT+KALMAN: An Efficient Reinforcement Learning Method for Continuous State Partially Observable Domains