An Approximate Bayesian Reinforcement Learning Approach Using Robust Control Policy and Tree Search