A Bayesian Approach to Robust Inverse Reinforcement Learning